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 |
|---|---|---|---|---|---|
miR-9 is an evolutionarily conserved miRNA that is abundantly expressed in Area X , a basal ganglia nucleus required for vocal learning in songbirds . Here , we report that overexpression of miR-9 in Area X of juvenile zebra finches impairs developmental vocal learning , resulting in a song with syllable omission , reduced similarity to the tutor song , and altered acoustic features . miR-9 overexpression in juveniles also leads to more variable song performance in adulthood , and abolishes social context-dependent modulation of song variability . We further show that these behavioral deficits are accompanied by downregulation of FoxP1 and FoxP2 , genes that are known to be associated with language impairments , as well as by disruption of dopamine signaling and widespread changes in the expression of genes that are important in circuit development and functions . These findings demonstrate a vital role for miR-9 in basal ganglia function and vocal communication , suggesting that dysregulation of miR-9 in humans may contribute to language impairments and related neurodevelopmental disorders .
Like humans developing speech and language , juvenile songbirds learn to sing by imitating the songs of an adult tutor early in life ( Doupe and Kuhl , 1999; Immelmann and Hinde , 1969; Marler and Tamura , 1964; Tchernichovski et al . , 2001 ) . Juvenile zebra finches ( pupils ) first hear and memorize a tutor’s song , and then , at about post-hatching day 30 ( 30 d ) , the pupils begin to vocalize . Initially , juvenile songs are highly variable . Through thousands of trial-and-error practice sessions guided by auditory feedback , the juvenile song gradually matures into a stereotyped adult song that resembles the tutor’s song ( Immelmann and Hinde , 1969; Konishi , 1965; Tchernichovski et al . , 2001 ) . The adult song , however , exhibits a residual level of variability . Depending on the social context , adult males sing a courtship song directed toward a female ( DS ) or an undirected song ( UDS ) when singing alone ( Sossinka and Bohner , 1980 ) . These two types of songs exhibit subtle differences in acoustic features , and the DS is more stereotyped than the UDS ( Hessler and Doupe , 1999; Jarvis et al . , 1998; Kao and Brainard , 2006; Murugan et al . , 2013 ) . The neural circuit that controls song behavior is organized into two pathways: a motor pathway and an anterior forebrain pathway ( AFP ) ( Figure 1A ) . The motor pathway controls song production , while the AFP , a cortical-basal ganglia circuit required for song learning and maintenance , guides motor output through a reinforcement mechanism ( Bottjer et al . , 1984; Brainard and Doupe , 2000; Doupe and Kuhl , 1999; Kao et al . , 2005; Nottebohm et al . , 1976 , 1982; Olveczky et al . , 2005; Simpson and Vicario , 1990 ) . Area X is a basal ganglia nucleus within the AFP . Lesions in Area X of juveniles prevent them from crystallizing their song ( Scharff and Nottebohm , 1991 ) . Area X receives dopaminergic innervation from the midbrain ventral tegmental area ( VTA ) , and it also modulates the activity of VTA dopaminergic neurons in response to auditory experience , implicating Area X in reinforcement learning ( Ding and Perkel , 2002; Gadagkar et al . , 2016; Gale and Perkel , 2006 , 2010; Lewis et al . , 1981 ) . MicroRNAs ( miRNAs ) are small , non-protein-coding RNA molecules that regulate gene expression post-transcriptionally . miR-9 is an evolutionarily conserved miRNA that is highly expressed in vertebrate brains ( Landgraf et al . , 2007; Luo et al . , 2012 ) . In the zebra finch , miR-9 is expressed in Area X . Its expression is regulated during developmental vocal learning and in adult males singing undirected songs ( Shi et al . , 2013 ) , suggesting that miR-9 plays an active role in these processes . miRNAs regulate gene expression by targeting the 3’ untranslated regions ( 3’UTRs ) of mRNAs , leading to mRNA degradation or suppression of protein synthesis . Many of the genes expressed in the nervous system have long 3’UTRs ( Mayr , 2016 ) , highlighting the importance of miRNAs in fine-tuning gene expression in nervous system development and function . FOXP1 and FOXP2 are a pair of paralogous transcription factors that have important roles in nervous system development . Deletions , mutations , and copy number variations of the FOXP1 gene have been implicated in a range of neural developmental disorders , including language delay , intellectual disability , and autism spectrum disorders ( ASDs ) ( Hamdan et al . , 2010; Horn et al . , 2010; O'Roak et al . , 2011 ) . Heterozygous mutations in the FOXP2 gene cause severe speech and language impairments ( Lai et al . , 2001; Vargha-Khadem et al . , 1995 ) , accompanied by structural and functional abnormalities in multiple brain regions . These regions include the basal ganglia ( Watkins et al . , 2002 ) , which is thought to be a component of the distributed neural circuitry that underlies speech and language ( Graham and Fisher , 2013 ) . FOXP1 and FOXP2 regulate the transcription of a large number of downstream genes , many of which are critically involved in neural differentiation , neurite outgrowth , synapse formation , and synaptic transmission ( Konopka et al . , 2009; Spiteri et al . , 2007; Tang et al . , 2012; Vernes et al . , 2007 , 2011 ) . Thus , the functional dosage of FOXP1 and FOXP2 needs to be tightly regulated . Recent in vitro studies have shown that miR-9 regulates the expression of FOXP2 by targeting specific sequences in its 3’UTR ( Fu et al . , 2014; Shi et al . , 2013 ) . miR-9 also regulates avian FoxP1 in embryonic chicken spinal cord ( Otaegi et al . , 2011 ) . These findings raise the possibility that miR-9 has a role in language development through regulating FOXP1 and/or FOXP2 . Taking advantage of the unique vocal behavior and the underlying neural circuitry of songbirds , we sought to assess the consequences of miR-9 overexpression in Area X of juvenile zebra finches on vocal learning and performance . For these studies , we overexpressed miR-9 using a lentiviral approach . We report here that overexpression of miR-9 in juvenile Area X profoundly impairs basal ganglia-dependent developmental vocal learning in juveniles and impairs song performance in adulthood . We further show that these behavioral deficits are accompanied by downregulation of FoxP1 and FoxP2 expression , disruption of dopamine signaling , and widespread changes in the expression of numerous genes that are important for neural circuit development and function .
The lentiviral vector that we used carried the mCherry fluorescent protein marker driven by the human ubiquitin promoter ( hUBC ) ( Edbauer et al . , 2010 ) . We made a miR-9-expressing virus ( lenti-miR-9 ) by inserting a miR-9 precursor sequence downstream of mCherry ( Figure 1B ) . The control virus ( lenti-control ) carried mCherry alone . When tested in vitro , these lentiviruses effectively infected 293T cells , and overexpression of miR-9 downregulated the FOXP1 and FOXP2 proteins ( Figure 1—figure supplement 1A and B ) . To test these viruses in vivo , we injected lenti-miR-9 or lenti-control virus into Area X of 25-day-old ( 25 d ) male juvenile zebra finches , and examined miR-9 expression levels four weeks later using quantitative real-time PCR ( qRT-PCR ) . In Area X injected with lenti-miR-9 , miR-9 expression was increased more than three-fold compared to expression in Area X injected with lenti-control; the expression of an unrelated miRNA , miR-124 , did not change ( Figure 1C and D , p < 0 . 0001 for miR-9 and p = 0 . 288 for miR-124 , n = 7 ) . These results indicate that our lentiviral approach allowed effective overexpression of miR-9 in Area X in an miRNA-specific manner . We examined whether and how overexpression of miR-9 in juvenile Area X impairs vocal learning . In these experiments , 23–25 d male juvenile zebra finches ( whose vocal learning was about to begin ) were injected bilaterally into Area X with either the control virus ( control pupils ) or the lenti-miR-9 virus ( miR-9 pupils ) . Each pupil was raised individually with an adult tutor from 30 d to 70 d . Pupils’ songs were recorded at 65 d , 80 d , and 100 d ( Figure 2A ) . A zebra finch song is made up of multiple renditions of a motif . A motif typically consists of 5–7 syllables rendered in a fixed sequence , with each syllable bearing distinct acoustic features ( Figure 2B ) . We first analyzed pupils’ songs recorded at 100 d ( when they became adults ) . On the global motif structure level , control pupils imitated their tutors’ song motif without syllable omission . By contrast , miR-9 pupils omitted some of their tutor's syllables and their song motifs were shorter than those of control pupils ( Figure 2B ) . We quantified this phenomenon by manually counting the number of omitted syllables and the number of syllables per motif . We found that 5 of 6 miR-9 pupils omitted tutor syllables ( Figure 2C , p = 0 . 0075; n = 6 ) , and that the average number of syllables per motif was reduced by 24% in miR-9 pupils compared to control pupils ( Figure 2D , p = 0 . 0325; n = 6 ) . We next examined how well the miR-9 pupils imitated the spectral structure and acoustic features of their tutors’ songs using the Song Analysis Program software ( SAP , [Tchernichovski et al . , 2000] ) . SAP computes a similarity score , which reflects how similar two song motifs are , thus indicating how well a pupil learns its tutor’s song . In quantifying motif similarity , we compared 20 pupil motif renditions to 10 tutor motif renditions , and averaged the 200 pairwise measurements for each animal . We found that miR-9 pupils exhibited a lower motif similarity score than control pupils; whereas the control pupils’ motif similarity score was comparable to that of untreated pupils , indicating that virus injection alone did not affect song learning ( Figure 3A , p = 0 . 002 , two-tailed Mann-Whitney U Test; n = 6 for control and miR-9 pupils; n = 4 for untreated control pupils ) . We next ranked the 200 pairwise measurements , and averaged the 10 highest values to obtain a maximum similarity score for each pupil . The maximum similarity score of miR-9 pupils was significantly lower than that of control pupils ( Figure 3—figure supplement 1 , p < 0 . 001; n = 6 ) , suggesting that even at their best performance , miR-9 pupils were not able to produce a good copy of their tutors’ song . To examine how well miR-9 pupils were able to imitate their tutors at the level of individual syllables , we quantified the syllable accuracy scores of control and miR-9 pupils . We found that miR-9 pupils imitated tutors’ syllables less accurately than did control pupils ( Figure 3B and C , p = 0 . 002 , two-tailed Mann-Whitney U Test; n = 6 ) . We also examined how individual acoustic features , including duration , mean frequency , goodness of pitch , frequency modulation , and Wiener entropy , differed between pupils and tutors . We found that the mean frequency and Wiener entropy differed between miR-9 pupils and tutors significantly more than between control pupils and tutors ( Figure 3D , p = 0 . 01 for mean frequency; p = 0 . 006 for Wiener entropy , two-tailed Mann-Whitney U Test; n = 6 ) . In addition , Wiener entropy differed significantly between miR-9 pupils and their tutors , but did not differ significantly between control pupils and their tutors ( p < 0 . 001 , t ( 24 ) = 8 . 245 for miR-9 pupils; p = 0 . 432 , t ( 41 ) = 0 . 794 for control pupils , paired t-test; control pupils: n = 42 syllable; miR-9 pupils: n = 25 syllables; n = 6 animals per group ) ( Figure 3E ) . To assess the effect of miR-9 overexpression in juvenile Area X on song performance in adulthood , we examined syllable sequence order in 100 d pupils’ songs . A careful examination of sonograms showed that the songs of miR-9 pupils often exhibited switching of syllable order , truncation of motifs , and/or syllable stuttering ( Figure 4A and B ) . We calculated syllable transition entropy to score these phenomena , where a higher transition entropy score reflects lower stereotypy of syllable sequences . We found that syllable transition entropy of miR-9 pupils was significantly higher than that of control pupils ( Figure 4C , p = 0 . 002 , two-tailed Mann-Whitney U Test; n = 6 ) . We also measured trial-by-trial performance variation in syllable acoustic features across multiple renditions of songs of adult ( 100 d ) miR-9 and control pupils . Among the acoustic features analyzed , duration , goodness of pitch , and Wiener entropy were significantly more variable in adult miR-9 pupils than in adult control pupils ( Figure 4D , p = 0 . 009 for duration; p = 0 . 015 for goodness of pitch; and p = 0 . 009 for Wiener entropy , two-tailed Mann-Whitney U test; n = 6 ) . These results indicate that overexpression of miR-9 in juvenile Area X leads to more variable song performance in adulthood . It is known that the expression of miR-9 in Area X is upregulated by singing an undirected song ( UDS ) but not by singing a female-directed song ( DS ) ; furthermore , the acoustic features of UDS are more variable than those of DS ( Jarvis et al . , 1998; Kao and Brainard , 2006; Leblois et al . , 2010; Murugan et al . , 2013; Shi et al . , 2013; Teramitsu and White , 2006 ) . These findings prompted us to examine the possibility that miR-9 plays a role in modulating song variability according to social context . We recorded both UDS and DS of the same adult pupils ( 100 d ) , and examined the trial-by-trial variability in the constant fundamental frequency ( cFF ) of the same set of syllables in the two song types using a previously established method ( Kao and Brainard , 2006; Leblois et al . , 2010; Murugan et al . , 2013 ) . In control birds , the variability in syllable cFF was greater in UDS than in DS . In miR-9 pupils , however , the syllable cFF remained variable in DS , abolishing the social context-dependent modulation of syllable variability ( Figure 5A and B , p = 0 . 006 for control pupils; p = 0 . 510 for miR-9 pupils , paired-t test; control pupils: n = 21 syllables , 6 animals; miR-9 pupils: n = 11 syllables , 6 animals ) . As juveniles are capable of producing an adult-like DS ( Kojima and Doupe , 2011 ) , we extended this analysis to 65 d and 85 d juveniles . Similar to the adults , the 65 d and 80 d control pupils produced a more stereotyped DS with reduced variability in cFF , whereas both 65 d and 80 d miR-9 pupils retained variability in DS ( Figure 5C , p < 0 . 05 for 65 d and p < 0 . 01 for 80 d and 100 d groups , respectively; two-tailed Mann-Whitney U test; at 65 d , control pupils: n = 8 syllables , 3 animals; miR-9 pupils: n = 6 syllables , 4 animals; at 80 d , control pupils: n = 11 syllables , 5 animals; miR-9 pupils: n = 9 syllables , 5 animals; at 100 d , control pupils: n = 14 syllables , 6 animals; miR-9 pupils: n = 10 syllables , 6 animals ) . Together , these results suggest that miR-9 plays a role in modulating social-context-dependent song variability . Vocal learning is a developmental process during which a highly variable juvenile song gradually matures into a stereotyped adult song that resembles the tutor’s song . To gain insight into the role that miR-9 may play in this process , we tracked the developmental trajectory of song learning by analyzing songs of pupils produced at 65 d and 80 d . At both 65 d and 80 d , miR-9 pupils imitated poorly , and their songs were less similar to the tutors’ song than those of control pupils ( Figure 6A , p = 0 . 0022 for 65 d , 80 d and 100 d songs , two-tailed Mann-Whitney U test; n = 6 per group ) . We wondered whether miR-9 overexpression caused a developmental delay , causing miR-9 pupils to require a longer time to learn their song . To assess this , we extended motif similarity analysis to songs of 150 d pupils . We found that at 150 d , the similarity score of miR-9 pupils was significantly lower than that of control pupils ( Figure 6A , p = 0 . 008 , two-tailed Mann-Whitney U Test; n = 6 ) . While the control pupils improved the similarity score of their song as they matured , miR-9 pupils did not improve their score from 65 d to 150 d ( Figure 6A , p = 0 for control pupils and p = 0 . 137 for miR-9 pupils , one-way ANOVA; n = 6 ) . We also measured syllable Wiener entropy changes during development . At 65 d , Wiener entropy of the control pupils and the miR-9 pupils was similar , but at both 80 d and 100 d , the Wiener entropy scores of miR-9 pupils were higher than those of the control pupils ( Figure 6B , p = 0 . 482 for 65 d , p = 0 . 028 for 80 d , and p = 0 . 020 for 100 d , Mann-Whitney U test ) . The Wiener entropy of control pupils’ syllables was reduced as they matured , whereas the Wiener entropy of miR-9 pupils’ syllables was not ( Figure 6B , p = 0 . 002 for control pupils and p = 0 . 709 for miR-9 pupils , one-way ANOVA; n = 6 per group ) . We also examined the trial-by-trial variability of syllable acoustic features including the duration , goodness of pitch , and Wiener entropy of songs produced during development . Syllable acoustic variability was reduced in songs of both control pupils and miR-9 pupils from 65 d to 100 d ( Figure 6C–E , p < 0 . 001 for duration , pitch , and Wiener entropy for both control and miR-9 pupils [except p < 0 . 01 for duration for miR-9 pupils]; one-way ANOVA; control pupils: n = 42 syllables , 6 animals; miR-9 pupils: n = 25 syllables , 6 animals ) . However , variability in each of these acoustic features was significantly higher in miR-9 pupils than in control pupils at all ages ( Figure 6C–E , p < 0 . 0001 for duration , pitch , and Wiener entropy for each age , two-tailed Mann-Whitney U test; control pupils: n = 42 syllables , 6 animals; miR-9 pupils: n = 25 syllables , 6 animals ) . These data indicate that miR-9 overexpression in juvenile Area X leads to higher variability in songs throughout maturation . This high level of trial-by-trial variation in acoustic features of miR-9 pupils may contribute to their inability to imitate the tutor’s song accurately . We wondered whether miR-9 overexpression affected the amount of song production , which subsequently contributed to impaired song learning . To assess this possibility , we examined the amount of singing by pupils at 65 d and 100 d . We found that at both 65 d and 100 d , miR-9 pupils sang slightly more syllables than control pupils , but the differences were not significant ( Figure 6—figure supplement 1 , p = 0 . 419 for 65 d; p = 0 . 109 for 100 d , two-tailed Mann-Whitney U Test; n = 6 ) . Thus , it is unlikely that the amount of singing contributed to the effect of miR-9 on song learning . To understand the molecular substrates underlying the impairments in vocal learning and performance described above , we examined changes in gene expression in Area X upon miR-9 overexpression . We first examined FoxP1 and FoxP2 mRNA and protein expression in Area X of juveniles four weeks after viral injection . We found that the expression levels of both FoxP1 and FoxP2 mRNAs were significantly downregulated in Area X injected with lenti-miR-9 compared to Area X injected with lenti-control ( Figure 7A , p = 0 . 006 for FoxP1 and p < 0 . 0001 for FoxP2 , n = 7 ) , whereas no change in expression of either FoxP1 or FoxP2 mRNA was found in tissue adjacent to Area X . We found that both FoxP1 and FoxP2 protein levels were also downregulated in Area X injected with the lenti-miR-9 virus compared to controls ( Figure 7B , p = 0 . 0007 for FoxP1 and p = 0 . 0037 for FoxP2 , n = 4 ) . The neurotransmitter dopamine plays an important role in modulating basal ganglia circuit plasticity and song stereotypy ( Ding and Perkel , 2002; Leblois et al . , 2010; Murugan et al . , 2013; Sasaki et al . , 2006 ) . Both the dopamine D1 ( D1R ) and D2 ( D2R ) receptors are expressed in Area X ( Kubikova et al . , 2010 ) , and D1R is regulated by FoxP2 ( Murugan et al . , 2013 ) . Therefore , we examined the expression levels of D1R and D2R in Area X in which miR-9 was overexpressed . We found that four weeks after viral injection , D1R was downregulated in Area X injected with the lenti-miR-9 virus compared to Area X injected with the lenti-control virus , whereas the expression of D2R was unchanged ( Figure 7C , p < 0 . 0001 for D1R and p = 0 . 3384 for D2R , unpaired t-test; n = 7 ) . DARPP-32 , a 32 kDa dopamine- and cAMP-regulated phosphoprotein , is a major signal transduction component acting downstream of dopamine receptors , and is highly expressed in striatal medium spiny neurons ( Greengard et al . , 1999; Murugan et al . , 2013 ) . We found that DARPP-32 protein level was significantly downregulated in Area X injected with the lenti-miR-9 virus compared to Area X injected with the lenti-control virus ( Figure 7D , p = 0 . 0022 , unpaired t-test; n = 4 ) . FOXP1 and FOXP2 regulate a large number of downstream transcriptional target genes that have important roles in neural circuit development and functions ( Konopka et al . , 2009; Spiteri et al . , 2007; Vernes et al . , 2011 ) . We curated zebra finch homologs of 37 FOXP1 and/or FOXP2 downstream genes based on published literature ( Bowers and Konopka , 2012; Konopka et al . , 2009; Spiteri et al . , 2007; Tang et al . , 2012 ) , and examined whether their expression levels changed upon miR-9 overexpression . We found widespread changes in gene expression: 26 of 37 tested genes changed their expression in juvenile Area X 4 weeks after miR-9 viral injection . Consistent with the fact that FOXP1 and FOXP2 are capable of bi-directional regulation of gene expression , either enhancing or repressing gene expression ( Li et al . , 2004 ) , we observed bi-directional changes in the expression of these downstream genes: 15 upregulated and 11 downregulated . The magnitudes of the changes , however , were moderate; a majority of these genes increased or decreased their expression by less than 50% . We grouped these 26 genes into functional modules: components of synaptic transmission ( ABAT , CADPS2 , GRIN2B , GRIK2 , GRM8 , KCNA4 ) , cell adhesion molecules important for dendrite growth and synapse formation ( CNTNAP2 , CTNNA3 , DISC1 , FRMPD4 , NRXN3 , SRPX2 , STX1A ) , transcriptional regulators ( ATRX , MEF2C , MTF1 , NeuroD6 , SOX5 ) , signaling molecules ( BDNF , IGF2 , NTRK3 , PTPRD , VLVDR ) , and peptidase and components of the ubiquitin protein degradation pathway ( MMP2 , UBE2L3 , UBE3A ) ( Figure 8A–E , p < 0 . 05 for all genes , t-test; n = 7 for all genes except for BDNF , CTNNA3 , NTRK3 , and STX1A , n = 4 ) . The symbols , names , functions , and possible associations with various neural developmental disorders of these genes are summarized in Supplementary file 1 . A fraction ( 11 of 37 ) of the FOXP1 and FOXP2 downstream genes that we tested did not , however , change in expression following miR-9 overexpression ( Figure 8—figure supplement 1 ) . Among the possible explanations are that the regulatory relationships between FOXP1 or FOXP2 and these genes may be species-specific , or that additional cellular regulatory mechanisms may have contributed to the gene expression levels that we observed .
We show here that overexpression of miR-9 in the basal ganglia nucleus Area X of juvenile zebra finches impaired developmental vocal learning and adult vocal performance . On the global motif structural level , the most pronounced impairment was that birds with miR-9 overexpression sang shorter song motifs , omitting some of the tutor syllables . This phenomenon in miR-9 pupils ( in which FoxP1 and FoxP2 were downregulated ) may mirror the limited vocabulary observed in human individuals carrying deletions in the FOXP1 gene , who typically exhibit a working vocabulary of fewer than 100 words at the age of seven ( Horn et al . , 2010 ) . Songs of birds with miR-9 overexpression in Area X exhibited higher trial-by-trial variability , which was reflected in a more variable sequence of syllable order , truncated motifs , and syllable stuttering . Syntax change has not been reported in prior studies in which FoxP2 was either knocked down or overexpressed in Area X ( Haesler et al . , 2007; Heston and White , 2015; Murugan et al . , 2013 ) . That miR-9 regulates the expression of both FoxP1 and FoxP2 and potentially of other genes in Area X may explain the robust deficits in vocal learning we observed here . Effects of miR-9 on song performance also occurred in a social context-dependent manner . Similar to birds with reduced FoxP2 expression in adult Area X ( Murugan et al . , 2013 ) , birds with miR-9 overexpression failed to modulate song variability when singing a directed song . Perhaps not coincidentally , miR-9 expression is higher in Area X when adult males naturally sing an undirected song ( Shi et al . , 2013 ) . It appears that , whether naturally or artificially induced , higher miR-9 levels in Area X render the circuit permissive to a more variable song or interfere with the production of a more stereotyped directed song . Vocal learning is a developmental process during which a less-structured and highly variable juvenile song gradually transitions to a stereotyped adult song that matches the tutor’s song ( Immelmann and Hinde , 1969; Tchernichovski et al . , 2001 ) . The effects of miR-9 overexpression on song imitation were apparent at 65 days after hatching , and these birds failed to improve their imitation thereafter ( Figure 6A ) . Acoustic features of syllables of miR-9 pupils also exhibited higher trial-by-trial variability throughout development ( Figure 6C–E ) . It is not clear whether or how higher song variability affects a pupil’s ability to match its song to a tutor’s song . In normal birds , the developmental vocal learning process is accompanied with a gradual increase in the miR-9 level in Area X , reaching its high point as juveniles become adults , and songs become stabilized; meanwhile , throughout the process , the FoxP2 level gradually decreases ( Haesler et al . , 2004; Shi et al . , 2013; Teramitsu et al . , 2004 ) . Our current data indicate that an artificially increased miR-9 level ( and reduced FoxP2 level ) does not result in premature stabilization of a juvenile’s song nor in an increase in similarity to the tutor’s song . Rather , miR-9 overexpression appears to lock Area X in a plasticity-dominant state , preventing further progression toward song stereotypy . Previous lesion studies have shown that lesions in Area X and lMAN , two nuclei in the anterior forebrain pathway , have different consequences on song development . While lesions in lMAN lead to a prematurely stabilized song , birds with Area X lesions never achieve song stability as control birds do ( Scharff and Nottebohm , 1991 ) . Although the experimental approaches used in these studies are unrelated ( molecular manipulation vs . electrolytic lesion ) , our observations are more aligned with the effects of lesions in Area X , suggesting that the effect of miR-9 overexpression ( gene manipulation ) might be restrained by circuit functions , in this case functions of Area X . According to the reinforcement learning theory , Area X and the AFP guide song motor learning through processes that evaluate the motor output patterns according to auditory feedback , and reinforce favorable motor actions . These processes involve dopamine signaling via D1R and D2R receptors expressed in Area X ( Kubikova et al . , 2010 ) and dopaminergic projections from the VTA to Area X ( Ding and Perkel , 2002; Doupe et al . , 2005; Doya and Tesauro , 1995; Gadagkar et al . , 2016; Lewis et al . , 1981 ) . Dopamine levels in Area X are higher when birds sing DS ( Sasaki et al . , 2006 ) . Blocking D1R pharmacologically or reducing D1R expression by knocking down FoxP2 in Area X results in a more variable DS ( Leblois et al . , 2010; Murugan et al . , 2013 ) . Our findings that miR-9 overexpression in Area X selectively downregulated D1R but not D2R provide further evidence of the importance of dopamine signaling in vocal learning and performance . Studies in mammals suggest that the D1R-expressing direct pathway and the D2R-expressing indirect pathway act in an opposing but coordinated manner in the basal ganglia to finely control the timing and synchronization of motor actions; imbalances between the two pathways may lead to movement and cognitive disorders ( Cazorla et al . , 2015; Gerfen and Surmeier , 2011 ) . The downregulation of D1R but not D2R in Area X , as a consequence of miR-9 overexpression , may have produced a functional imbalance between the two signaling pathways , which may have contributed to impaired song learning and performance . In normal birds , miR-9 expression in Area X is regulated during vocal development , and is upregulated by adult singing UDS ( Shi et al . , 2013 ) . Thus , the miR-9–FoxP2–dopamine signaling network provides a mechanism that dynamically adjusts the functional balance between the D1R and D2R signaling pathways , allowing birds to adapt to the changing maturation state of song in juveniles and to modulate song stereotypy according to the social context in which a song is produced . Many of the FOXP1 and FOXP2 downstream transcriptional target genes were identified by genome-wide chromatin immunoprecipitation assays , which depend on the binding of transcription factors FOXP1 or FOXP2 to promoter sequences ( Konopka et al . , 2009; Spiteri et al . , 2007; Vernes et al . , 2011 ) . The changes in expression of these downstream genes in Area X when FoxP1 and FoxP2 were downregulated by miR-9 provides further evidence for a regulatory relationship between these genes and FOXP1 or FOXP2 in the basal ganglia circuit critical for vocal communication . Among the genes we tested , many have direct roles in neural circuit development and functions , and their dysregulation has been implicated in various neural developmental disorders . For example , GRIK2 , GRIN2B , and GRM8 encode subunits of glutamate receptors . Their altered expression can affect synaptic transmission of Area X neurons . KCNA4 encodes a voltage-gated potassium channel ( Ovsepian et al . , 2016 ) , and mutations in KCNA4 have been identified in patients exhibiting linguistic disabilities , attention deficit hyperactivity disroder ( ADHD ) , and cognitive impairments ( Kaya et al . , 2016 ) . MMP2 , a member of the matrix metalloproteinase family , plays critical roles in synaptogenesis , dendrite remodeling , and neurogenesis ( Fujioka et al . , 2012 ) . In songbird brain , MMP2 has been implicated in angiogenesis and neurogenesis in HVC , another song nucleus essential for song learning and production ( Kim et al . , 2008 ) , suggesting that MMP2 may have a role in Area X circuit development as well . CNTNAP2 encodes a cell surface protein that belongs to the neurexin family , and is important for synaptic formation and clustering of K+ channels at synaptic terminals ( Poliak et al . , 1999 ) . FOXP2 is known to regulate CNTNAP2 expression by binding to a regulatory sequence in its first intron , and CNTNAP2 dysfunctions have been implicated in specific language impairments , intellectual disabilities , and ASDs ( Rodenas-Cuadrado et al . , 2014; Vernes et al . , 2008 ) . DISC1 ( Disrupted in schizophrenia 1 ) plays important roles in cell migration and dendrite development , and it has been linked to schizophrenia , bipolar disorder , depression , and ASDs ( Millar et al . , 2000; Thomson et al . , 2013 ) . The widespread changes in gene expression caused by miR-9 overexpression touch upon an array of cellular functions , including synaptic transmission , dendrite growth , synapse formation , gene regulation , neurotrophin ( BDNF ) signaling , and protein degradation . We only tested expression changes in a small fraction of FOXP1 and FOXP2 downstream genes; however , given the fact that two-thirds ( 26 of 37 ) of these genes changed expression , it is likely that many of the genes we did not test also changed their expression . Thus , a large number of these affected genes may have collectively contributed to the deficits in vocal behavior . We noted that the magnitudes of changes in gene expression were moderate . Intriguingly , a recent study of a large cohort of schizophrenia patients found moderate changes ( ranging from 10% to 40% ) in the expression of several hundred genes in the prefrontal cortex ( Fromer et al . , 2016 ) . Our observations support the emerging view that subtle but broad changes in gene expression might be a molecular signature underlying complex neural developmental and/or neural psychiatric disorders ( Fromer et al . , 2016; Purcell et al . , 2014 ) . Evidence implicating particular genes in language impairments and autism is often established through genome-wide analysis such as screening for mutations and/or copy number variations in human subjects ( O'Roak et al . , 2012; Sanders et al . , 2012; Yuen et al . , 2015 ) . Our findings showing that these genes are expressed in the basal ganglia , and that alterations in their expression are accompanied by impairments in vocal communication , provide additional evidence that these genes function in language processes and development . Among the genes we tested , CNTNAP2 and DISC1 have each been implicated in multiple disorders , including language impairments , ASDs , ADHD , intellectual disabilities , and schizophrenia ( Fromer et al . , 2016; Purcell et al . , 2014 ) , suggesting that these neural developmental disorders , cognitive impairments , and psychiatric disorders share common molecular substrates . It is possible that distinct but overlapping phenotypes are manifested depending on where and when these genes are expressed and how they are regulated , emphasizing the need to study these genes in the context of specific functional neural circuits . miR-9 is expressed in the embryonic human brain and has been shown to regulate human FOXP2 gene expression ( Fu et al . , 2014 ) . It is likely that miR-9 plays a role in human language development by fine-tuning FOXP1 and FOXP2 expression , thereby coordinating the expression of a large number of genes that are active in neural development and function . Both FOXP1 and FOXP2 mRNAs have long 3’UTRs containing numerous miRNA binding sites , suggesting that these genes can be regulated by many miRNAs in addition to miR-9 . Dysregulation of these miRNAs or of miRNA–FOXP1/FOXP2 interactions by genetic , environmental , or physiological factors , thus , may contribute to language impairments and related neurodevelopmental disorders .
Animal usage was approved by the Louisiana State University Health Sciences Center ( LSUHSC ) Institutional Animal Care and Use Committee . All experiments were conducted in male zebra finches ( Taeniopygia guttata ) . Animals were housed in a 7 a . m . – 7 p . m . light-dark cycle . Juveniles at specific ages were obtained from our breeding colony at LSU School of Medicine , with each bird given an ID at hatching . The lentiviral vector used ( a gift from Dr . M . Sheng ) carries an mCherry fluorescent marker driven by the human ubiquitin C promoter ( hUBC ) ( Edbauer et al . , 2010 ) . The zebra finch genome contains three genes encoding miR-9: miR-9–1 , miR-9–2 , and miR-9–3 ( Luo et al . , 2012 ) . We amplified a 300-nt genomic DNA fragment containing the miR-9–3 precursor sequence from the zebra finch genome and inserted it downstream of mCherry in the lentiviral vector to generate the miR-9-expressing virus . The lenti-control virus is an empty vector . For lentivirus packaging and production , viral vectors and packaging plasmids were transfected into 293LTV cells ( Cat . No . LTV-100 , Cell Biolabs ) using the calcium-phosphate method following the manufacturer’s instructions ( Clontech ) . Cell identity and the absence of mycoplasma contamination were confirmed by the vendor . Viral particles were harvested 48 and 72 hr after transfection . The crude supernatant was filtered through a 0 . 45 µm filter , spun at 2000 RPM , and collected and spun again by ultracentrifugation ( 25 , 000 rpm x 2 hr ) . The precipitated viral particles were resuspended in 50 µl PBS . Typically , we obtained virus suspensions with titers in the range of 1−5 × 109/ml . Juvenile birds were separated from their fathers at day 10 ( 10 d ) , and raised by their mother in a sound-proof chamber until 30 d . PCR was performed to determine the sex of juveniles ( see Supplementary Table 2 for primer sequences ) . In assigning animals for viral injection , each animal had an equal probability of being injected with the control or miR-9 virus . Viral injection was performed on males at about 25 d . Stereotaxic injection was performed using a stereotaxic device including a head holder ( Myneurolab ) and a hydraulic microinjector ( Narishige ) . The glass needles used for injection have an inner tip diameter of 25 µm . The stereotaxic coordinates for injection into juvenile Area X were: anterior/posterior 2 . 8 and 3 . 2 mm , dorsal ventral 4 . 2 and 4 . 4 mm , and medial/lateral 1 . 3 and 1 . 5 mm using the bregma point as a reference . Animals were anesthetized with ketamine and xylazine . Each Area X received a viral injection at six or eight sites , 120–150 nl viral suspension per site . To facilitate diffusion of viral particles , the injection needle was allowed to remain at the site for 3–5 min before removal . For behavioral experiments , virus was injected bilaterally . For gene expression experiments , the lenti-miR-9 virus and the lenti-control virus were injected into Area X of opposite hemispheres . To ensure that the impairments in vocal learning and performance that we observed were due to virally transduced miR-9 expression in Area X , and not due to Area X tissue damage caused by the injection process , we sacrificed the birds after the last song recording , and examined their Area X . All birds showed bilateral expression of virally transduced mCherry in Area X; the average area exhibiting strong mCherry signal accounted for 15–20% of total Area X volume . We also observed scattered cell bodies or dendrites showing an mCherry signal outside the core infected region but within Area X , suggesting that virally transduced gene expression spread beyond the core infected region . There was no difference in total Area X volume between the non-injected and injected groups , and there was no difference in total Area X volume between the lenti-control- and lenti-miR-9-injected groups ( Figure 1—figure supplement 2A ) . In a separate experiment , we also quantified the number of neurons in juvenile Area X one month after viral injection using immunostaining of the neuronal marker Hu . We found similar numbers of Hu+ neurons in Area X of non-injected animals and in Area X injected with the lenti-control virus or with the lenti-miR-9 virus ( Figure 1—figure supplement 2B ) . These results indicate that physical damage to Area X caused by viral injection , if any , was minimal; thus , the behavioral phenotypes that we observed are likely due to virally transduced miR-9 overexpression . By 30 days of age , the mother was removed and an adult male tutor was introduced to one miR-9-virus- or control-virus-injected juvenile ( pupil ) . Both the pupil and the tutor were kept together in a sound-proof recording chamber until after 70 d . Songs of pupils were recorded at 65 d , 80 d , 100 d , and 150 d using a microphone ( Technica AT803B ) , an eight-channel computer interface ( M-Audio 2626 ) , and SAP software version 1 . 02 . Undirected songs were recorded automatically over two days for each bird at each age . Directed songs were recorded ( from the same groups of birds ) manually in the morning . Males were induced to sing directed songs by presenting female birds in a nearby cage; the females were changed every 10 min . Songs were classified as directed songs when the male sang facing the female as observed by an experimenter . Zebra finches sing in bouts . A song bout typically contains multiple renditions of a motif , and a motif contains 5–8 distinct syllables that are rendered in a fixed sequence . A syllable is defined as a continuous segment of sound separated from another syllable by a silence gap , and each syllable can be quantitatively described by a set of distinct acoustic features using the software package Sound Analysis Pro ( SAP ) . To select songs for analysis ( for all song analyses described here unless otherwise stated ) , we manually sorted all song files recorded in one day from 8 a . m . to 12 p . m . and eliminated files representing cage noise . For each pupil , we typically selected 20 song files , approximately evenly spread across the entire set of song files ( e . g . , if there were 200 song files , the 1st , 11th , 21st , 31st , etc . were selected ) . To evaluate the accuracy of injections into Area X and to measure the size of the virally infected region within Area X , after the last song recording , birds were euthanized and brains were sliced into 80 µm sagittal sections . Bright light and fluorescent images were scanned for each section using an Olympus BX61VS microscope equipped with VS-ASW FL software and a 2X lens . The size of Area X and the virally infected region ( mCherry positive ) within Area X was measured for all Area-X-containing sections using Image J software . For Hu staining , lenti-control and lenti-miR-9 viruses were injected into Area X of opposite hemispheres at about 30 d , and animals were euthanized one month later . Animals were anesthetized with ketamine and xylazine , followed by perfusion with PBS and fixation with 4% paraformaldehyde in PBS . Fixed brains were sliced into 30 µm sagittal sections . For each hemisphere , 3–4 sections containing mCherry signal within Area X were stained with an antibody against Hu ( Cat#21271 , Life Technologies; 1:500 dilution ) , followed by a fluorescein-conjugated goat anti-mouse secondary antibody ( Cat#F2761 , Invitrogen; 1:500 dilution ) . Images were taken with a Zeiss Axioplan2 fluorescent microscope ( 40X lens ) . For each section , the number of Hu+ neurons ( green fluorescence ) in one or two microscope fields were counted using Image J . The experimenter was blind to treatment groups . All information related to statistical analysis is documented in the corresponding figure legends . Sample sizes were not statistically determined , but were similar to those generally employed in the field . In all figures ( unless otherwise stated ) , each circle represents data from one animal . For data presented in Figure 2 , Figure 3A , and Figure 5B , two investigators , one blind to the experimental groups , analyzed two different sets of song files . Two injected animals ( one control pupil and one miR-9 pupil ) were excluded from song behavior analysis because injection was outside of Area X . Data were assumed to have normal distributions , but this was not formally tested . Variance was assumed to be similar between groups , but this was not formally tested . | When a cell needs to make a protein , it makes a temporary copy of the corresponding gene so that the genetic code can be carried to its protein-making machinery . When the temporary copy of the code is no longer needed , the cell destroys it . This system is fine-tuned by other small stretches of genetic code called microRNAs , which speed up the destruction and so help to switch genes off faster . Two genes called FOXP1 and FOXP2 are known to have roles in speech and language development in humans . When these genes do not work properly , people have severe difficulties when speaking and understanding speech . But scientists know little about how the brain controls them . The brains of animals with backbones – like birds and mammals – make a microRNA called miR-9 . Scientists thought miR-9 may control how active the FOXP1 and FOXP2 genes are in the brain . Like humans , zebra finches communicate vocally . Young male birds learn to sing by imitating the song of an adult tutor , usually their father . The process is controlled by a brain region called “Area X” . Now , Shi et al . report on the role of miR-9 in vocal learning and singing in zebra finches . First , the gene for miR-9 was inserted into a virus-based genetic tool . Shi et al . then injected this virus into Area X of juvenile zebra finches , which delivered the gene to the brain cells and forced them to make excess miR-9 . A control group received empty virus with no miR-9 gene for comparison . The juvenile finches then grew up with an adult bird that taught them to sing . Shi et al . found that the birds that overproduced miR-9 did not learn as well as their normal counterparts . Their songs were shorter , they stuttered , and they missed out syllables , which meant that they simply sounded different to their tutors . These young birds also failed to change their tune in different situations , for example , when they met a female zebra finch . Examination of the birds’ brains four weeks after the viral injection showed that the bird versions of the FOXP1 and FOXP2 genes were less active . There were also changes in other genes involved in brain circuit development . Humans have a brain area like Area X , called the basal ganglia . The link between miR-9 and vocal learning provides a starting point to understand more about language in general . This could lead to improved understanding of conditions like stuttering , Tourette’s syndrome , dyslexia and autism spectrum disorders . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | miR-9 regulates basal ganglia-dependent developmental vocal learning and adult vocal performance in songbirds |
It is unclear whether the two hippocampal lobes convey similar or different activities and how they cooperate . Spatial discrimination of electric fields in anesthetized rats allowed us to compare the pathway-specific field potentials corresponding to the gamma-paced CA3 output ( CA1 Schaffer potentials ) and CA3 somatic inhibition within and between sides . Bilateral excitatory Schaffer gamma waves are generally larger and lead from the right hemisphere with only moderate covariation of amplitude , and drive CA1 pyramidal units more strongly than unilateral waves . CA3 waves lock to the ipsilateral Schaffer potentials , although bilateral coherence was weak . Notably , Schaffer activity may run laterally , as seen after the disruption of the connecting pathways . Thus , asymmetric operations promote the entrainment of CA3-autonomous gamma oscillators bilaterally , synchronizing lateralized gamma strings to converge optimally on CA1 targets . The findings support the view that interhippocampal connections integrate different aspects of information that flow through the left and right lobes .
Lateralization of certain neural functions in vertebrates is thought to bear evolutionary advantages ( Halpern et al . , 2005 ) . Earlier studies have mainly focused on finding anatomical correlates to behavioral asymmetries , but there has been little insight gained into the functional mechanisms underlying the differential routing and integration of activity in bilateral networks . For example , fMRI studies have shown bilateral or lateral activation of the same structures when subject performs different tasks ( Smith and Goodale , 2015; Lee et al . , 2016 ) . The bilateral function in the hippocampus is largely unknown and it is unclear to what extent the two lobes convey similar or complementary information , and whether they do indeed work together . However , the existence of important bilateral connections between the same and different hippocampal subregions does suggest some degree of integration and cooperation . In rodents , hippocampal lateralization is observed during certain memory tasks ( Klur et al . , 2009; Shipton et al . , 2014 ) , in the expression of synaptic plasticity ( Kohl et al . , 2011 ) , or following environmental enrichment ( Shinohara et al . , 2013 ) . In the human , such lateralization was reported during sequence disambiguation ( Kumaran and Maguire , 2006 ) and in cognitive navigation ( Iaria et al . , 2003; Iglói et al . , 2015 ) . Available data is however limited and it provides no mechanistic insights . Here we have used pathway-specific local field potentials ( LFPs ) ( Herreras et al . , 2015 ) in anesthetized rats to study the spontaneous transmission in the bilateral CA3→CA1 Schaffer segment during irregular non-theta EEG states , a pattern that in active animals is associated to sensory input during immobility and consummatory behaviors . The hippocampal CA3 region is an important hub for ascending and cortical pathways ( Vinogradova , 2001 ) , and its output is conveyed to numerous brain regions , directly or from the next station in the CA1 after bilateral integration ( Swanson et al . , 1981 ) . The left and right CA3 are connected reciprocally through the ventral hippocampal commissure ( VHC ) , and they also send excitatory inputs to the CA1 on both sides of the brain through an associational ( Schaffer ) -commissural system that converges on pyramidal cells ( PCs , Figure 1A ) . Hence , this system represents an ideal model to explore the flow of activity and its integration in bilateral networks . In addition , numerous electrophysiological studies show that the CA3 region behaves as a powerful gamma oscillator with inhibitory and excitatory local network components ( Traub et al . , 1996; Fisahn et al . , 1998; Bartos et al . , 2002; Csicsvari et al . , 2003; Hájos et al . , 2004; Mann et al . , 2005; Oren et al . , 2006; Hájos and Paulsen , 2009; Pietersen et al . , 2009; Gulyás et al . , 2010; Mann and Mody , 2010; Kohus et al . , 2016 ) . Whether and how the left and right CA3 gamma oscillators are coupled and how this influences the bilateral activity and performance of the hippocampus have not been explored . 10 . 7554/eLife . 16658 . 003Figure 1 . Experimental paradigm and clean out of the Schaffer and CA3som activities . ( A ) Functional characteristics of the bilateral CA3→CA1 segment: ( 1 ) an intrinsic gamma oscillator fueled by inhibition in each CA3 region produces gamma output from PCs; ( 2 ) The left and right CA3-PCs are interconnected through the ventral hippocampal commissure ( VHC , maroon arrows ) , enabling the coupling of CA3 gamma oscillators; ( 3 ) The excitatory outputs of CA3-PCs from both sides converge in each CA1 ( Schaffer and Commissural pathways ) . ( B ) Experimental setup . Two-shank linear arrays were located at homotopic sites of the dorsal left and right hippocampi . Recordings were acquired simultaneously and each group was analyzed separately by an Independent Component Analysis ( ICA ) . ( C ) ICA of a sample epoch across the CA1 and CA3/DG layers . In raw LFPs ( black traces ) , several bands of coherent voltage fluctuations are observed that indicate multiple activation in different synaptic territories ( three are outlined by filled ovals spanning the CA1 and the Dentate subfields , while small maroon ovals mark activity in the st . pyramidale of the CA3 ) . The ICA returns the spatially-coherent components and provides readout of the temporal dynamics free of a contribution by the others . A set of components or LFP-generators was obtained per shank , each with a characteristic spatial distribution or voltage weight ( Vwt ) that enabled matching between shanks . Details of the extraction are in Figure 1—figure supplement 1 . Colored traces from top to bottom: Schaffer ( cyan ) , CA3som ( maroon ) , lacunosum-moleculare ( green ) , and GCsom ( purple ) . The amplitudes are normalized ( 0 . 37:0 . 25:0 . 84:1 ) . In other figures voltage units are employed that were estimated for the sites with maximum power ( triangles in Vwt plots ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 00310 . 7554/eLife . 16658 . 004Figure 1—figure supplement 1 . Details of the extraction and ICA performance . To illustrate the performance of the ICA , we chose sample epochs in which two nearby sources produce rhythmic LFP waves of similar duration that makes the respective time courses hardly recognizable by the naked eye: Black , LFPs; color , ICA components . ( A ) Mixed contributions by the Schaffer ( cyan ) and lacunosum-moleculare ( l-m ) components ( green ) . A phase mismatch of LFP waves recorded in contiguous sites ( vertical lines ) indicates pathway co-activation ( recall that a single pathway produces proportional LFPs in all sites ) . The differences between time courses of LFP and ICA waves denote the net volume-conducted contribution that each pathway entered on the other’s site: this is removed by the ICA ( bicolor arrows ) . ( B ) Example of LFPs in the CA3/Dentate Hilus containing CA3som waves and DG-contributed potentials . Only the first CA3som wave , a , is visible as it does not coincide with any other , but the waves b and c overlap in space and time with Hilar waves . The respective spatial distributions are however different . Hilar waves are nearly identical in all recording sites and hence , deviations from the time course in some sites indicate the presence of additional sources . CA3som wave b occurs at a negative hilar deflection ( arrowheads: the difference of LFP and the dashed time courses mark the site-specific contribution of the CA3som wave ) , while wave c does so on a positive-going hilar wave , modifying the slope of LFPs at CA3som specific sites . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 004 We set out to compare the dynamic relationships of two network components that can be addressed on a cell-assembly basis using pathway-specific LFPs , the CA3 PC excitatory output reflected in CA1 Schaffer LFPs , and a somatic inhibitory input to the same population ( CA3som ) . We also studied the impact of excitatory bilateral convergence on the output of neurons in the next station , the CA1 , by comparing the waveform details of Schaffer gamma waves on both sides and cell firing of CA1 units . To extract the Schaffer and CA3som LFP activities from the field contributions added by other concurrent pathways , we applied a spatial discrimination method to multisite linear LFP recordings ( Figure 1B ) ( Herreras et al . , 2015 ) . This approach provides unprecedented temporal precision and specification of the origin such that each spatially-isolated activity reflects the envelope of postsynaptic currents elicited by an afferent population on another ( Makarova et al . , 2011; Fernandez-Ruiz et al . , 2012a; Martín-Vázquez et al . , 2013; Schomburg et al . , 2014; Herreras et al . , 2015 ) . Indeed , we previously reported that the Schaffer LFPs contain a regular succession of gamma waves ( gamma strings ) that reflect excitatory packages of extremely variable amplitude ( Fernandez-Ruiz et al . , 2012a , 2012b ) , these being elicited in CA1 PCs by the synchronous firing of CA3 PC assemblies at gamma intervals ( Fisahn et al . , 1998; Hájos and Paulsen , 2009; Takahashi et al . , 2010; Fernandez-Ruiz et al . , 2012a ) . The respective excitatory and inhibitory nature of Schaffer and CA3som gamma waves has also been established previously ( Hájos et al . , 2004; Benito et al . , 2014 ) . These studies indicate that Schaffer and CA3som gamma waves are generally larger and lead from the right lobe , while the CA3som component forms part of a local gamma oscillator that paces outgoing excitatory Schaffer packages . These form strings of gamma waves that may run unilaterally as seen upon VHC disconnection . In both sides , Schaffer waves may lead at different antero-posterior CA1 locations indistinctly , although they are all submitted to global bilateral asymmetric entrainment that optimizes bilateral convergence in the CA1 . This inference is supported by the preferred firing of CA1 pyramidal cells during bilateral waves in contrast to the preferred unilateral driving of putative interneurons .
The Schaffer activity contained strings of gamma waves , irregular fluctuations , and Sharp-Waves ( SPW ) , each reflecting different regimes of organized firing in CA3 PCs ( Benito et al . , 2014 ) . The occurrence and duration of gamma strings was unpredictable , ranging from a few waves to many seconds and they were no further characterized . To test the bilateral complementariness we first compared the time course of Schaffer activity as a whole ( containing interspersed gamma strings and irregular fluctuations ) in two pairs of left-right homotopic sites of the dorsal hippocampus ( Figure 2A and Figure 2—figure supplement 1 ) . Epochs with frequent SPWs were avoided or these were removed as they contribute disproportionately to the statistics . We found significant wide band coherence between pairs of sites that are 0 . 5 mm apart within antero-posterior ( a-p ) lamellar strips . This coherence is reduced and restricted to the gamma band in bilateral comparisons ( Figure 2—figure supplement 2 ) . The cross-correlation coefficient ( CC ) behaved similarly ( La-Lp , 0 . 84 ± 0 . 03; Ra-La , 0 . 57 ± 0 . 04; F ( 1 , 12 ) = 26 . 9; p<0 . 001; mean of epochs lasting 85–167 s , n = 7 ) . The mean τmax of CCs was similar for a-p sites ( 0 . 32 ± 0 . 2 ms ) but shifted -0 . 82 ± 0 . 3 ms ( F ( 1 , 12 ) = 8 . 5 , p=0 . 01 ) for La-Ra comparisons , with the right side leading . 10 . 7554/eLife . 16658 . 005Figure 2 . Functional asymmetry in the bilateral CA3-CA1 system . ( A ) Sample string of Schaffer-gamma obtained from four sites . Individual waves coincide regardless of their amplitude . Globally , Schaffer-gamma is larger on the right side . The scheme shows the location of recordings from a coronal view ( Figure 2—figure supplement 1 ) . ( B–E ) Representative experiment showing the features of individual waves compared pairwise within ( La , Lp ) and between hippocampi ( La , Ra ) ( n = 6623 pairs of waves in 167 s ) . The blue and red dots belong to the pairs when L or R waves were longer , respectively . The population statistics and additional examples are in Figure 2—figure supplements 2–3 . ( B ) Waves co-vary closely in the same side ( left ) and much less so between sides ( right ) : a , best fit tangent; r , correlation coefficient . The insets show superimposed averaged waves ( cal: 20 ms and 100 μV ) . ( C ) A string of Schaffer gamma shows unilateral waves in both sides ( triangles ) . In paired bilateral waves , either side may lead ( ovals ) . ( D ) Bilateral synchronization was measured from the start of the waves ( time lag ) . The positive and negative values indicate that L or R waves led , respectively . R waves preceded more often ( black bars ) , the bilateral lag being larger when R-waves were longer ( line subplot in blue ) . ( E ) The amplitude difference between paired waves in the right and left sides is plotted against their time lag . Larger waves on any side had a tendency to lead . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 00510 . 7554/eLife . 16658 . 006Figure 2—source data 1 . Spreadsheet containing measurements of the LFP generators and extracted waves for each experiment . Data are presented as the mean and s . e . m . for: cross- correlation index and τmax , amplitude and duration of extracted Schaffer and CA3som gamma waves , total number of paired waves ( ipsilateral and bilateral ) , percentage of unilateral waves , covariation of amplitude and duration of bilateral waves , lag between start time of bilateral waves or ipsilateral paired waves , lags between paired waves in subgroups of longer waves in L , R , anterior or posterior sites , cross-correlation between Schaffer and CA3som waves , and the covariation index of amplitude and duration . The data pertain to Figures 2 , Figure 2—figure supplement 2 , Figure 4 , and Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 00610 . 7554/eLife . 16658 . 007Figure 2—source data 2 . Schaffer LFP generators and extracted waves for the experiments used in Figures 2 and 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 00710 . 7554/eLife . 16658 . 008Figure 2—figure supplement 1 . Histological and electrophysiological localization of recording sites . ( A ) The location of recording sites was reconstructed from histological sections in six animals . The tracks left by the twin-shank linear probes were recognized by DiI marks in the tissue ( pictures of one illustrative experiment are shown below ) , and drawn on sagittal representations of the rat hippocampus taken from the atlas of Paxinos and Watson . The indicated lateral coordinate is only an approximation to actual site . The extent of recordings across the CA1 and CA3/Dentate subfields ( colored bars ) has been scaled and adjusted by the location of cell body layers in the CA1 and CA3 using the characteristic evoked potentials recorded simultaneously from all four shanks ( star: stimulus in the left CA3b soma layer ) . The st . pyramidal of the CA1 and CA3 are marked by filled and open triangles , respectively . Dots mark the stimulus artefact . ( B ) An epoch containing spontaneous LFPs and an evoked potential ( color code indicates the recording sites in A ) . The black traces belong to the CA3 soma layer where CA3som waves appear ( arrows ) . Cyan ovals mark the Schaffer potentials in the CA1 st . radiatum ( note that stimulation in the left CA3 also activates Schaffer fibers in the right-hand side through antidromic firing of the CA3 ) . Black ovals mark recurrent excitatory waves in the st . radiatum of the CA3 . Note also the correspondence of amplitude differences of equivalent waves originated in different subfields to recording sites: closer to the CA3 ( CA3som and st . radiatum CA3 waves ) or to the DG ( asterisks ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 00810 . 7554/eLife . 16658 . 009Figure 2—figure supplement 2 . Additional data and population statistics for the comparison of intra and interhippocampal CA3→CA1 Schaffer activity . ( A ) Intra ( upper row ) and interhippocampal ( lower row ) wide band spectral coherences between Schaffer-LFPs . Δt is the length of the epoch analyzed . SPWs were excluded manually to avoid adding an excessive weight of slow frequencies to the analysis . Blue areas correspond to statistically significant coherences ( gray areas mark the level of statistical significance; surrogate test , n=1000 ) . Interhippocampal Schaffer coherence was lower in all animals and centered in 40–45 Hz gamma frequency ( arrows ) . ( B , C ) Population statistics for the comparisons as in the representative experiment shown in Figure 1B and D of the main text . ( B ) Cross-correlation coefficient ( CC ) for the co-variance of amplitude between the paired waves in all seven experiments . The diagonal red line marks equal CC for intra than interhippocampal comparisons . All values lie below the diagonal , indicating a lower CC between L and R homotopic sites compared to the intrahippocampal sites . In a t-test of the data all values were significant ( p<0 . 001 ) except one ( blue square , p<0 . 05 ) . ( C ) Population statistics for the lag between the paired waves in the two hippocampal hemispheres ( La , Ra ) . Negative values indicate that waves in the R led to L . Black symbols stand for all pairs , while the blue and red symbols belong to the pair subclasses when R or L led , respectively . The mean confidence intervals at α = 0 . 01 are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 00910 . 7554/eLife . 16658 . 010Figure 2—figure supplement 3 . Additional sample traces of Schaffer activity simultaneously obtained from CA1 homotopic sites . R and L traces are depicted in black and cyan , respectively . The upper and lower fragments correspond to 1 s epochs ( taken 2 s apart ) of tight and loose bilateral co-variation , respectively . Note the coarse ( slow waves/groups of gamma waves ) and the fine ( individual gamma waves ) bilateral amplitude co-variation of the gamma oscillations in the upper traces , and the frequent mismatch of amplitude in paired ( bilateral ) waves in the lower traces , which however maintained tight L-R synchrony . Such epochs of tight and loose co-variation were intermingled and occurred unpredictably . Since individual Schaffer gamma waves reflect the size and firing synchronization of CA3 pyramidal cells forming a functional assembly , tight bilateral covariation indicates a sequence of CA3 assemblies that are parallel in both hemispheres of the hippocampus , while loose bilateral co-variation indicates lateralized strings of CA3 assemblies that notwithstanding , beat at a similar pace . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 010 Gamma strings of tight left-right amplitude co-modulation between paired waves ( Figure 2A and C ) alternate with others of loose covariation ( Figure 2—figure supplement 3 ) . Therefore , we quantified the features of individual waves obtained through a deconvolution approach ( see Figure 2—source data 1 and 2 , and Materials and methods ) and compared these between different sites . To avoid noisy LFP events , waves were only considered when they lasted >5 ms and reached >20 μV . The amplitude , duration and start time of individual Schaffer gamma waves reflect the size and synchronization of CA3 PC assemblies . Waves were designated as paired when they overlapped at two sites by at least 70% of their duration . Despite of a clear amplitude fluctuation at any site , paired waves were nearly identical along lamellar strips in one side ( Figure 2A and B; covariance was ρA = 0 . 78 ± 0 . 02 for the amplitude and ρD = 0 . 63 ± 0 . 02 for the duration in La:Lp comparison ) , while they showed less co-variation between homotopic La-Ra sites ( ρA = 0 . 5 ± 0 . 04 and ρD = 0 . 44 ± 0 . 03; see population statistics in Figure 2—figure supplement 2B , C ) . The matching waveforms and strong covariation of ipsilateral waves rule out the possibility that the modulation of amplitude over successive waves is due to a different site of origin of the waves and hence , the distance to the electrodes ( Benito et al . , 2014 ) . Rather , it is consistent with a coarse lamellar-like distribution of Schaffer fibers in the CA1 for all waves , regardless of their amplitude . The mean wave duration was the same on both sides ( R and L: 26 . 7 ± 0 . 7 ms; F ( 1 , 12 ) = 0 , p=0 . 96 ) . We found notable bilateral asymmetries . In particular , waves on the right side were larger in amplitude ( Ra: 299 ± 11 µV; La: 237 ± 15 µV; F ( 1 , 12 ) = 12 , p=0 . 005 , n = 7 animals: Figure 2B–D ) . Interestingly , bilateral waves were rarely initiated synchronously . Although either side may lead ( ovals in Figure 2C ) , it was more common that waves in the right side did so ( 55 . 6 ± 0 . 8% vs 44 . 4 ± 0 . 8% ) . On average , the R waves preceded with similar values between anterior or posterior sites ( Ra-La: 0 . 62 ± 0 . 14 ms; Rp-Lp: 0 . 67 ± 0 . 18 ms ) . We also found a notable proportion of unilateral waves ( Figure 2C ) , with a higher incidence on the right side ( R: 11 . 7 ± 2 . 4%; L: 5 . 5 ± 1% ) . These unilateral waves were smaller than bilateral ones in the right hemisphere ( 212 ± 20 µV; bilateral vs . unilateral: p=0 . 002 ) , but not in the left one ( 192 ± 20 µV; p=0 . 09 ) . To countercheck the overall primacy of Schaffer activity on one side we applied the Granger Causality ( GC ) test between the right and left Schaffer generators over 90 s epochs . Figure 3A shows R and L Schaffer activations in a representative experiment . The GC test confirmed that there are statistically significant and reciprocal relations between the R and L sides ( Figure 3B: p=10–6 and p=10–8 for L to R and R to L relations , respectively ) . Moreover , the functional relation peaked in the low-gamma frequency band ( 30–50 Hz ) and the right to left link was strongly dominant over time ( Figure 3C , D ) . This result was verified for all seven animals . 10 . 7554/eLife . 16658 . 011Figure 3 . Assessment of functional asymmetry with Granger causality and phase relations . ( A ) A short epoch of activations of the right and left Schaffer pathways . ( B ) F-statistics for Granger Causality ( GC ) test revealing significant reciprocal influence from R to L and from L to R sides . ( C ) Frequency dependence of GC . R to L relation exhibits a peak at gamma frequency . ( D ) Time-frequency display of the GC index . R to L relationship is stronger and more persistent . ( E ) Distribution of phases in the L side with onsets related to zero phases in the R side , i . e . , when field events begin . The mean phase lag of 0 . 22 rad ( corresponding to 0 . 95 ms time lag ) is highly significant . The population data is indicated in the text . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 011 Although the GC test confirmed a preferred right to left directionality , it provided no time lag between the Schaffer activities . We crosschecked the lag obtained from the start times of the paired gamma waves using an additional test to evaluate the phase shift between the Schaffer activities . Figure 3E shows the histogram of the phases in Schaffer activity in the left side with onsets related to zero phases in the right side ( corresponding to the beginning of LFP events in that side ) . The circular statistics confirmed the presence of a highly significant peak at Δϕ = 0 . 22 ± 0 . 04 radians . Thus , the right side indeed appears to lead the generation of activity in Schaffer pathways . By evaluating the mean frequency of the generators ( ~45 Hz ) we can estimate the corresponding time lag Δt = 0 . 95 ± 0 . 2 ms , in a good agreement with the value obtained by the analysis of paired Schaffer waves ( 1 . 12 ms for this experiment ) . Again this result was verified in all seven animals ( Δϕ = 0 . 12 ± 0 . 06 radians; Δt = 0 . 54 ± 0 . 26 ms ) , with the mean time lag being roughly equal to the τmax in the CCs . Since the duration of paired waves was not identical , we explored whether the leading site had any relation to the wave’s features . The more relevant results were obtained when paired left-right waves were sorted by the site at which they showed longer duration ( L or R , anterior or posterior ) ( Figure 2B , D , E ) . Notably , the right-left lag was accentuated when R-waves were longer ( a , p: 3 . 09 ± 0 . 06 and 3 . 03 ± 0 . 17 ms ) , whilst longer L-waves also led to righ-hand ones albeit by a smaller amount ( a , p: 1 . 97 ± 0 . 13 and 1 . 91 ± 0 . 23 ms , n = 7 , F ( 1 , 12 ) = 517 , p<0 . 001 ) ( Figure 2D , and Figure 2—figure supplement 2 ) . The data from different individuals indicated that these lags were robust for different antero-posterior locations of the shanks or when there was a slight a-p displacement between the right and left sides ( Figure 2—source data 1 and Figure 2—figure supplement 1 ) . We also found a tendency towards a lead by longer and larger waves irrespective of the leading side ( Figure 2E ) . These observations indicated an independent discharge by left and right CA3 assemblies that project to the anterior or posterior sites of CA1 indistinctly , while the overall lag towards the right-hand side supports a global right-to-left asymmetric entrainment . Such directionality is also supported by the fact that the mean lag from crossed Lp to Ra sites ( 1 . 13 ± 0 . 25 ms ) was roughly equal to the accumulated lag from right to left plus the lag used by a-p axonal conduction: Ra→La ( 0 . 62 ± 0 . 14 ms ) , La→Lp ( 0 . 52 ± 0 . 1 ms ) . In turn , the La to Rp lag became balanced ( –0 . 1 ± 0 . 24 ms ) as expected if the righ-to-left lag were absorbed by the opposite sign of the a-p lag: La→Ra ( -0 . 62 ± 0 . 14 ms ) , Ra→Rp ( 0 . 67 ± 0 . 33 ms ) . The preferred lead of longer/larger Schaffer gamma waves in either side might suggest that CA3 clusters on one side excite CA3 clusters on the other , which then set up delayed contralateral Schaffer waves in the CA1 . We explored this possibility by estimating the minimum time used by CA3 spikes on one side to get to CA3 PCs through the latency of the contralateral CA3 antidromic population spike elicited from the septal pole of the Left CA3 , which was 2 . 5 ± 0 . 1 ms ( n = 6 ) . We also estimated the shortest possible lag for the direct driving of CA3 to the contralateral CA1 through the latency of the evoked commissural fEPSP in CA1 , which amounted 5 . 8 ± 0 . 3 ms ( n = 7 , i . e . : three-fold the L-R lag for spontaneous paired waves ) . These results leave some room for a reduced fraction of the bilateral waves to be explained by the inter-hippocampal excitatory driving of unilaterally ignited CA3 clusters , although the presence of numerous unilateral waves points to additional mechanisms . Notably , side-independent instigation of Schaffer waves was also found in intra-hippocampal a-p comparisons . Thus , while on average Schaffer waves led in anterior sites ( 0 . 52 ± 0 . 06 and 0 . 67 ± 0 . 33 ms for the L and R sides , respectively; L-R: F ( 1 , 12 ) = 0 . 3 , p=0 . 6 ) in compliance with a global antero-posterior lamellar topology of Schaffer fibers , we found quite different lags depending on which a-p site had longer waves . For waves that were longer in anterior sites the a-p lag increased to 2 . 6 ± 0 . 3 and 2 . 2 ± 0 . 3 ms in the right and left sides , respectively ( R-L: F ( 1 , 12 ) = 0 . 65 , p=0 . 4; total vs . longer R: F ( 1 , 12 ) = 102 , p<0 . 001 ) . More notably , when posterior waves were longer they preceded anterior ones by 1 . 13 ± 0 . 5 and 1 . 8 ± 0 . 3 ms for the R and L sides , respectively ( R-L: F ( 1 , 12 ) = 1 . 24 , p=0 . 29; total vs . longer L: F ( 1 , 12 ) = 53 , p<0 . 001 ) . Thus the large CA3 PC assemblies initiate firing on any side and location , regardless of the overall bilateral coupling that maintains a global precedence on the right-hand side . To delve further into the mechanisms of bilateral entrainment of Schaffer gamma oscillations , we explored the ipsi- and bilateral expression of the gamma activity in the soma layer of the CA3 region itself , and its relationship to Schaffer activity ( outgoing CA3 quanta ) . This so-called the CA3som generator is known to be part of a gamma pacemaker in this region ( Traub et al . , 1996; Fisahn et al . , 1998; Bartos et al . , 2002; Csicsvari et al . , 2003; Hájos et al . , 2004; Mann et al . , 2005; Oren et al . , 2006; Hájos and Paulsen , 2009; Pietersen et al . , 2009; Gulyás et al . , 2010; Mann and Mody , 2010; Kohus et al . , 2016 ) . The CA3som gamma-paced wavelets were recorded over 1 to 3 contiguous electrodes in different animals and they appeared as positive non-overlapping events riding on a flat baseline ( Figure 1C and Figure 1—figure supplement 1B ) . Ipsilateral comparisons of CA3som LFPs between anterior and posterior sites in the two sides were only possible in two animals ( see Figure 4 for an illustrative experiment ) , although partial comparisons were obtained in another four . The data were pooled by side ( La-Lp and Ra-Rp: n = 7 ) or position ( La-Ra and Lp-Rp: n = 8 ) , and like the Sch LFPs , the CA3som LFPs displayed close-fitting within-side activities , whereas bilateral synchrony was far less marked and mismatches of individual waves were more frequent ( unilateral CA3som waves: 22 ± 1 . 2 and 26 ± 2 . 3% for the L and R sides , respectively; Figure 4A , La-Lp vs . Lp-Rp traces ) . Accordingly , the antero-posterior wide band coherence gave significant values for frequencies above 25 Hz in all cases , whilst none gave significant values for bilateral comparisons ( Figure 4A , spectral coherence plots ) . The CCs behaved similarly ( a-p: 0 . 66 ± 0 . 6; L-R: 0 . 27 ± 0 . 02; F ( 1 , 12 ) = 32 , p<0 . 001 , with a non-significant phase difference between a-p comparisons ( τmax: 0 . 1 ± 0 . 2 ms ) , but similar to Schaffer activity in R-L comparisons ( 0 . 42 ± 0 . 8 ms ) . The comparisons between extracted CA3som gamma waves yielded similar results as that for Schaffer activity ( Figure 4—figure supplement 1A ) . Thus , the amplitude covariance of bilateral waves was stronger in intra ( ρA = 0 . 62 ± 0 . 06 ) than interhippocampal waves ( ρA = 0 . 24 ± 0 . 04; F ( 1 , 14 ) = 42 , p<0 . 001 ) , although the amplitude did not differ in the left and right hemispheres ( La: 162 ± 26; Ra: 183 ± 28 μV; ( F ( 1 , 10 ) = 0 . 32 , p=0 . 58 ) . Also , on average R-waves led by 0 . 34 ± 0 . 1 ms ( n = 8 ) , and this lag increased to 2 . 72 ± 0 . 01 ms when the R-waves were longer ( total vs . longer R: F ( 1 , 14 ) = 297 , p<0 . 001 ) , and longer L-waves also preceded the right-hand ones by 2 . 28 ± 0 . 02 ms ( total vs . longer L: F ( 1 , 14 ) = 643 , p<0 . 001; ( Figure 2—source data 1 ) , matching the relationships to the Schaffer waves . 10 . 7554/eLife . 16658 . 012Figure 4 . CA3som gamma activity has weak bilateral coherence but is coupled to ipsilateral Schaffer . ( A ) Comparison of CA3som activities between pairs of sites . The histograms of spectral coherence only show significant values ( blue ) for ipsilateral comparisons . The sample traces show tight matching in superimposed activities at a-p sites ( upper traces ) and frequent mismatch in bilateral comparisons in the same epoch ( lower traces ) . Cyan and black traces correspond to the left and right sides . ( B ) Comparisons between Schaffer and CA3som activities ( blue and maroon traces , respectively ) . The spectral coherence showed significant values only at gamma frequency at all four sites . Sample traces show strong wave-to-wave coupling despite the poor amplitude covariation . The CC was strong and showed a marked left-shift that mostly originates from the different waveform of individual waves . All data were taken from the same animal ( see population statistics in the text , and additional analyses in Figure 4—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 01210 . 7554/eLife . 16658 . 013Figure 4—figure supplement 1 . Additional features of the CA3som activity and waves . ( A ) Statistics of extracted bilateral CA3som waves from a representative experiment . Waves co-vary closely in the same side ( left panel ) and much less so between sides ( middle panel ) : a , best fit tangent; r , correlation coefficient . The panel on the right shows the quantification of time lags between bilateral paired waves . Same color coding as in Figure 2 . See population statistics in the text . ( B ) Correlation between Schaffer and CA3som activities at the same site shows distinct dynamics over different time scales . The upper pairs of superimposed plots represents the time envelope of Schaffer ( cyan ) and CA3som activities ( maroon ) calculated with a different sliding windows ( 1 or 0 . 1 s ) . A notable reduction in the CC for the smaller time scale denotes mismatch of the convolved waves , as noted in the successive enlargements shown below . The lower sample traces show a more regular succession of the CA3som ( maroon ) activity: although smaller , the CA3som waves continue to appear when Schaffer waves ( cyan ) disorganize ( from dashed line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 013 We then related the CA3som and Schaffer activities and over a large time-scale ( seconds ) , the two followed similar trends ( Figure 4—figure supplement 1B ) by displaying parallel gamma strings and pauses . However , clear differences were observed over shorter time-scales ( hundredths of milliseconds ) . Since we observed no differences between the sides and sites , we pooled the data from all shanks and animals where both activities could be recovered ( n = 19 ) . As expected from the flat baseline in the CA3som component , in all cases the spectral coherence between CA3som and Schaffer activities was only significant in the gamma band ( Figure 4B ) . Despite the tight temporal relationship between individual Schaffer and CA3som waves , the CC was only moderate ( 0 . 4 ± 0 . 02 , τmax of 4 . 72 ± 0 . 4 ms ) , in part due to the different duration of the gamma waves ( Schaffer waves are ~2 . 5 times longer than the CA3som ones and hence , the respective time courses are largely out-phased ) . We also applied the GC test to explore the relationship between the CA3som and Schaffer activities obtained at the same a-p site and side ( two shanks on each of six animals: n = 12 ) . In all cases , the GC test indicated statistically significant reciprocal relationship between the two activities that peaked in the gamma frequency band , and the CA3som to Schaffer link strongly dominated . We also evaluated the phase shift that yielded a highly significant peak at Δϕ = 1 . 46 ± 0 . 18 radians , the CA3som leading Schaffer activity . The estimate time lag was Δt = 6 . 18 ± 0 . 7 ms , which is in a good agreement with the value of the τmax obtained from CCs . With regards to the features of extracted waves , the most relevant finding was the poor amplitude co-variation between paired Schaffer and CA3som gamma waves recorded by the same shank ( ρA = 0 . 16 ± 0 . 02 ) . This observation may indicate a different topology , composition , or location of the CA3 PCs receiving a synchronous inhibitory volley and those setting the Schaffer waves in the CA1 . In addition , the time lag between paired Schaffer and CA3som waves was not significant ( 0 . 12 ± 0 . 3 ms ) , which should be interpreted in light of the different target cells and the reduced sampling ( 42 . 3 ± 2% of CA3som waves remained unpaired ) . Globally , these results conform to a strong but not necessarily causal lock of oscillatory somatic inhibition in CA3 to the outgoing gamma excitation , while the weak bilateral coupling of the former indicates that it operates mostly ipsilaterally . Schaffer gamma waves are excitatory inputs to CA1 units and they arrive synchronously with those from the contralateral side due to the tight coupling of CA3 gamma oscillators , which should have an impact on their output . We checked this inference by exploring the combined efficiency of bilateral gamma waves to bring CA1 units to fire . The excitatory nature of Schaffer waves was confirmed by the stronger response of CA1-PCs to bilateral as opposed to unilateral waves ( Figure 5A , Figure 5—source data 1 , and Figure 5—figure supplement 1 ) . The precise timing of firing was not solely determined by these inputs since spikes were relatively dispersed over the gamma cycle ( Figure 5B and Figure 5—source data 2 ) . Nevertheless , each PC had preferred intervals from the start of paired waves , indicating that they were cell-specific optimal intervals for bilateral summation ( Figure 5—figure supplement 1 ) . In turn , a large proportion of putative CA1 interneurons showed tight phase-locked firing shortly after the start of the bilateral gamma waves , and they also fired more frequently than PCs in response to unilateral waves on the same side ( Figure 5B and Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 16658 . 014Figure 5 . Left-Right synchrony of Schaffer gamma waves drives firing of CA1 PCs and putative interneurons distinctly . ( A ) Spike trains were correlated to the activity of the Schaffer generators in both sides , and the spikes were associated to unilateral ( only R , only L , R or L groups ) or bilateral ( i . e . synchronous ) gamma waves , or to none . Only statistically significant cells are represented . PCs fired preferentially upon convergence of bilateral gamma waves , while putative interneurons showed a marked unilateral drive . ( B ) Firing of representative cells in function of when the gamma waves started in the L and R sites ( additional data in Figure 5—figure supplement 1 ) . Spikes were sorted that occurred within 40 ms of the beginning of a gamma wave in both sides ( paired waves ) . The spike occurs at ( 0 , 0 ) , while the X and Y axes represent the start time of the R and L gamma waves , respectively . The calibration bar indicates the firing density upon removal of chance firing . Most firings approach the diagonal as a result of tight bilateral gamma coupling . However , PCs fired at varying cell-specific intervals , while firing of putative interneurons was phase-locked with a short lag from the start of the excitatory gamma waves . The relation of spike firing to LFP power is shown in Figure 5—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 01410 . 7554/eLife . 16658 . 015Figure 5—source data 1 . Spreadsheet containing the unitary results for the bilateral/unilateral test performed on CA1 units in Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 01510 . 7554/eLife . 16658 . 016Figure 5—source data 2 . Analysis of CA1 units in relation to the occurrence of gamma waves ( co-modulograms ) . The data pertain to Figure 5B and Figure 5—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 01610 . 7554/eLife . 16658 . 017Figure 5—figure supplement 1 . Relation of spike firing to bilateral gamma waves ( additional data ) . ( A ) Measurement of the temporal relation between the spike firing and start of excitatory gamma waves . ( B ) The time lag from the start of the waves in both sides is plotted as X-Y counts in time-to-time-density histograms ( firing comodulograms ) . The spike time occurs at ( 0 , 0 ) , and the X and Y axes mark the start time of the R and L gamma waves , respectively . The plotted time interval is ( −40 , +5 ) ms in both axes . ( C ) Additional examples of firing comodulograms as in Figure 2B of the main text . The two upper rows correspond to 10 different PCs and the two lower rows to 10 interneurons recorded in the CA1 strata indicated below . In the upper row of plots , from blue to red color indicate the raw spike density , whereas in the lower plots the color from black to yellow corresponds to statistically significant spike density . For PCs , a total of 1500 s were analyzed . Only for graphing purposes , the epoch was shortened discretionarily for some interneurons in order to avoid excessive filling and maintain visual discrimination . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 01710 . 7554/eLife . 16658 . 018Figure 5—figure supplement 2 . Distribution of PC firing induced by bi- or unilateral gamma waves over the power of the Schaffer activity . ( A ) Distribution of the Schaffer power ( gray curve ) and its piecewise-linear fit ( dashed curve ) . The distribution can be separated into two regions with different power laws . ( B ) Differential probability of PC spikes of different origin versus the instantaneous power in the left ( left subplot ) and right ( right subplot ) Schaffer generators . Red , blue and black curves correspond to spikes induced by R , L and bilateral waves , respectively . Gray areas mark the confidence intervals for bilateral spikes ( α = 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 018 Since unilateral Schaffer waves were slightly smaller than bilateral waves , we explored the possibility that increased association of PC firing to bilateral waves was due to stronger ipsilateral input rather than to the bilateral convergence of Schaffer and Commissural inputs . Given the sparseness of cell firing compared to the abundance and amplitude range of gamma waves we devised a global approach by estimating the probability with which spikes emitted by PCs deviate from spikes drawn from a random spike train over the instantaneous power distribution of the Schaffer generator . The typical distribution of the instantaneous power of the Schaffer generator ( Figure 5—figure supplement 2A ) has two regions that follow power laws , axk , with different scaling constants ( k1 = −1 . 22 for power < 1 , and k2 = −2 . 66 for power > 1; to compare between different experiments we scaled the generator power equal to 1 at the critical point ) . The first region corresponds to noisy dynamics , whereas the second one is associated with rarer but sufficiently strong gamma waves ( bi- and unilateral ) . The mean differential probabilities of PC spikes emitted during unilateral R and L waves ( blue and red curves , respectively ) or bilateral waves ( black curve; data obtained over n = 40 PCs ) were plotted ( Figure 5—figure supplement 2B ) . In both cases the differential probabilities were negative in the noisy regions ( power < 1 ) , i . e . : PC firing was unlikely in association with small noisy events . For strong gamma waves ( power > 1 ) bilateral spikes exhibited quite a flat distribution over the wave power . Moreover , the probability was smaller than that for spikes emitted during unilateral gamma waves on the same side . Thus , the coincidence of left-hand and right-hand waves ( i . e . : bilateral excitation ) plays crucial role in PC firing , while the associated power increase has a marginal effect . The presence of unilateral waves , frequent epochs of loose bilateral co-modulation , and the wide range of delays between bilateral waves appears to be incompatible with their joint initiation through the ipsi- and contralateral axonal branches of PCs in a CA3 assembly on the leading side . Thus , we explored the role of interhippocampal CA3↔CA3 connections by micro-injection of lidocaine in the VHC . Effective disruption was confirmed by the abolition of evoked fEPSPs in the CA1 of both hemispheres following stimulation through the injecting pipette ( Figure 6A , C ) . Notably , although Schaffer gamma strings were still observed bilaterally , frequent unilateral strings occurred that were not seen in the controls ( Figure 6B and Figure 6—figure supplement 1 ) . When they co-occurred we found that the overall power and wave characteristics on either side did not change ( Figure 6C , Figure 6—source data 1 , and Figure 6—figure supplement 1 ) , evidence that each CA3 can autonomously generate gamma in vivo , as reported earlier for chemically-induced gamma oscillations in vitro ( Oren et al . , 2010 ) . Coarse bilateral co-modulation was maintained , albeit reduced ( CC , 0 . 62 ± 0 . 03 vs 0 . 45 ± 0 . 05; p=0 . 007; mean of epochs lasting 100 s in n = 4 animals; Figure 6D and E ) . Moreover , bilateral gamma coherence was lost ( Figure 6F and Figure 6—source data 1 ) and consequently , the L-R synchrony of individual gamma waves disappeared while the amplitude variability of the waves remained high in both sides ( Figure 6C ) . This indicates that different gamma strings may circulate independently through L and R hippocampi . 10 . 7554/eLife . 16658 . 019Figure 6 . Disruption of the VHC uncouples CA3 gamma oscillators . ( A ) Experimental setup: lidocaine was injected through a pipette into the VHC , which also served to deliver electric pulses; ( B ) Gamma strings may appear on only one side after lidocaine injection ( ovals ) ( additional examples in Figure 6—figure supplement 1 ) . ( C ) Lidocaine in the VHC modifies L-R gamma synchronization without altering the gamma power on each side ( see Figure 6—figure supplement 1 ) . Note the tight co-variation and L-R phase synchronization of individual gamma waves in the controls , and their outphasing after lidocaine ( left tracings ) . Evoked fEPSPs on both sides were fully blocked after lidocaine microinjection . The autocorrelation ( AC ) of the Schaffer activity showed oscillatory gamma waves on both sides , with a similar power before and after lidocaine administration , whilst the cross-correlogram ( CC ) was drastically reduced ( D ) . ( E ) The CC between L and R Schaffer activities reduces significantly ( n = 4 experiments; t test ) . ( F ) L-R spectral coherence before and after lidocaine injection . Note the disappearance of significant bars ( in blue ) in the gamma band ( 40 Hz ) after lidocaine ( arrows ) . The effects of unilateral CA3 blockade on unit firing are in Figure 6—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 01910 . 7554/eLife . 16658 . 020Figure 6—source data 1 . Effect of disruption of the VHC by lidocaine on the features of gamma waes . The zip file contains the quantification of gamma power and the amplitude and duration of gamma waves before and after lidocaine in the VHC ( VHC lidocaine . xlsx ) , and the spectral coherences between left and right sides ( coherences . ppt ) . Data pertain to Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 02010 . 7554/eLife . 16658 . 021Figure 6—figure supplement 1 . Uncoupling of CA3 gamma oscillators ( additional data ) . Sample traces of Schaffer activity simultaneously obtained in L ( cyan ) and R ( black ) CA1 homotopic sites following lidocaine injection in the ventral hippocampal commissure . ( A ) Successive enlargement to give increasing temporal detail of gamma strings and waves . Despite the bilateral co-occurrence of gamma strings , the internal pace of gamma waves loses bilateral co-variation . As a result , paired waves could not be established and wave-wave comparisons are no longer possible . ( B ) Only following lidocaine injection could gamma strings ( red oval ) be observed unilaterally ( sample taken 10 s later than in A ) . ( C ) Normalized amplitude and band power of Schaffer gamma waves before and after lidocaine was micro-injected into the VHC . Values are normalized to the control in each experiment . Lidocaine does not modify the mean amplitude or duration of individual gamma waves nor the power of gamma ( n = 4; epochs lasted 50–100 s ) . None of the values were significant ( Student t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 02110 . 7554/eLife . 16658 . 022Figure 6—figure supplement 2 . Inactivation of the CA3 in one side alters individual but not population firing rates in the other . ( A ) Scheme of the experimental design . ( B ) Evoked potentials recorded on the left side upon stimulation on the left ( st1 ) and the right CA3 ( st2 ) before ( black traces ) and after lidocaine inactivation of the left CA3 ( red ) . ( C ) Features of the Schaffer activity and Schaffer waves in the right-hand side after contralateral CA3 inactivation . None of the parameters underwent significant changes ( n = 4; p>0 . 1 in all cases ) . ( D ) Raster plots of firing in representative units of a single experiment and population results ( E ) . Units were classified according to their subfield and subtype ( PC or putative interneuron ) . None of the groups showed significant changes . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 022 It might be expected that bilateral disruption affects the firing of units receiving ipsi and contralateral CA3 inputs . Given their low firing rate , PCs require long recording times and thus , we used an alternative experimental design that offers higher long-term stability , as the blockade of the CA3 itself on the left side . Units in the right CA3 and CA1 subregions were recorded for 700 s before and after lidocaine administration . A total of 33 PCs and 23 putative interneurons were recorded ( n = 4 experiments ) , and nearly all cells underwent changes in the mean firing rate following contralateral CA3 blockade ( Figure 6—figure supplement 2 ) . The firing of 10 out of 19 PCs in the CA1 , and of 4 out of 14 PCs ( all from the same experiment ) in the CA3 increased . As for interneurons , 7 out of 9 in the CA1 and 8 out 14 in the CA3 reduced firing . Neither the overall power nor the wave characteristics changed on the side contralateral to the lidocaine injection .
The GC test indicates that Schaffer activity on any side may predict activity on the other , and the R-to-L influence is more robust . However , the persisting finding among different experiments of a lower p for the R-to-L statistics versus L-to-R cannot be used for qualifying unequivocally a preferred directionality . Similar caution should be taken to interpret the lags yielded by the τmax from CCs and the phase-difference test , both pointing to a precedence of right-hand over left-hand activities . In a strict sense , the primacy of the right lobe is supported by evidence from the extracted waves , which can be considered discrete events independent from each other: ( a ) the mean lag between paired bilateral waves shows quantitative advance on the right-hand side ( around 0 . 6–0 . 8 ms ) ; ( b ) The mean amplitude of Schaffer waves was 30% larger on the right-hand side; ( c ) CA3som waves showed a similar advance on the right-hand side; ( d ) The cumulated lag in crossed comparisons , either summed ( Lp-Ra ) or balanced ( La-Rp ) , in consonance with the partial lags between a-p and R-L sites . The advantage of such asymmetry attains functional meaning when we consider that lateralized Schaffer gamma strings converge on the CA1 units of each side on a wave-to-wave basis . Examination of the relative timing of the waves at different sites sheds light on the mechanisms underlying the organization and dynamic behavior of natural CA3 assemblies ( for a summary see Figure 7 ) . Since the fate of a Schaffer wave is to excite CA1 units where it integrates with input from the contralateral side it is crucial to optimize the temporal overlap of the respective excitatory envelopes . Therefore , the short mean lag between paired bilateral waves indicates that the system is so finely adjusted ( see functional implications below ) . However , in terms of connectivity it appears too small lag to consider unilateral CA3 assemblies driving Schaffer waves on the opposite side to be a general mechanism . Yet , since waves may lead on either side , the mean bilateral lag is to a large extent balanced . Subgroups of longer left or right waves do present longer mean lags ( 1 . 5 and 3 . 2 ms ) . Nevertheless , the minimum time required for CA3 spikes to reach contralateral CA3 PCs is 2 . 5 ms , which when added to the time required for synaptic integration and assembly recruitment is insufficient to explain the lag in the majority of paired waves . Such view might change depending on how PC assemblies are recruited . For instance if PC assemblies are recruited through the so-called leader PC cells ( Fujisawa et al . , 2006; Wittner and Miles , 2007 ) , the driving factor between such cells may require shorter lags . On the other hand , a direct contralateral driving of gamma waves in the CA1 via commissural fibers is less likely for several reasons: ( a ) the minimum lag of commissural evoked potentials is already in the higher range of the lags between spontaneous bilateral waves; ( b ) the commissural pathway contributes little to field potentials due to anatomical and geometrical factors ( Martín-Vázquez et al . , 2015 ) , making gamma waves of such origin unlikely; ( c ) the presence of unilateral waves coincides on the same idea , i . e . , since a CA3 assembly may initiate a Schaffer wave that has no contralateral counterpart , when contralateral waves appear they most likely have an ipsilateral origin; ( d ) finally , the disruption of the VHC , or the unilateral inactivation of the CA3 would markedly reduce the presence of Schaffer waves and dampen the mean power of the Schaffer generator , which does not happen . Therefore , although unilateral CA3 assemblies driving contralateral ones cannot be ruled out entirely , the most likely origin is an independent ipsilateral driving . 10 . 7554/eLife . 16658 . 023Figure 7 . Scheme of ipsi and bilateral CA3 and CA1 relationships associated to the occurrence of Schaffer gamma waves . Cells , domains and pathways activated during an imaginary gamma cycle in a coronal representation of the two hippocampal lobes ( CA1 and CA3 subfields are in grey and beige , respectively ) . Only anatomical and functional relationships that are relevant to the present findings are included . Dashed squares depict lamellar-like CA1 domains . Spindle-like forms represent the CA1 dendritic domains activated by a CA3 assembly , i . e . : a Schaffer gamma wave . Dashed outlines in red represent spatial domains in the CA3 soma layer inhibited by CA3som waves . The CA3 assemblies elicit ipsilateral Schaffer waves that are matched by near synchronous waves on the other side at roughly homotopic sites ( i . e . : bilateral waves a-h , b-g ) . Excitation in the contralateral side does not produce field potentials ( Martín-Vázquez et al . , 2015 ) , as noted by the presence of unilateral waves ( e . g . , c-f ) . Schaffer waves may initiate in anterior or posterior sites of the CA1 indistinctly ( indicated by the wider end of the spindles ) , probably originated by assemblies in different CA3 zones . Very few assemblies projecting to a given CA1 domain overlap in the same cycle . For example , domain d receives two ipsilateral Schaffer waves ( in cyan and dark green ) and a contralateral ( unseen ) wave ( in orange ) . As part of the intrinsic gamma oscillator , different CA3 assemblies activate the same basket cells , which inhibit many more PCs , independently of whether they belong to the firing assembly or not ( amplitude mismatch between CA3som and Schaffer waves ) . The double-headed red arrow represents the overall R-to-L precedence , and the vertical arrows represent the overall anticipation of waves at anterior sites of the lamellar CA1 domains . DOI: http://dx . doi . org/10 . 7554/eLife . 16658 . 023 Schaffer waves may lead in anterior or posterior sites on the same side , and their mean lag increased far longer than expected for axonal conduction when longer waves led . Such long delays may indicate independent ignition of CA3 assemblies whose projection to CA1 may follow either lamellar or longitudinal topologies ( Li et al . , 1994 ) , while the lead on anterior or posterior sites appears to reproduce the different septotemporal CA1 projections of the CA3 subzones ( Ishizuka et al . , 1990 ) . We also reported previously that Schaffer coherence decreases rapidly over the septo-temporal axis ( Benito et al . , 2014 ) . The emerging view ( Figure 7 ) is that once every gamma cycle a collection of spatial modules of the CA1 dendritic space is activated over the septo-temporal axis by multiple independent CA3 assemblies with either a dominant lamellar or longitudinal projection , converging with others entering from the contralateral side . Such complex modular bilateral mating requires precise intra and interhippocampal control of their timings to achieve efficient global orchestration . There is insufficient data available to draw a more complete picture , and obtaining such information requires simultaneous recording over more CA1 and CA3 sites than was performed here , and additional techniques to monitor the spontaneous activity of other subpopulations not accessible from LFPs . Nevertheless , some roles of local inhibitory networks can be derived from present CA3som data and the available literature . Globally , the bilateral CA3som relationships replicate those of Schaffer activities , although some of the differences indicate that they are less relevant for bilateral entrainment . For instance , the bilateral coherence between right and left CA3som activities , and the covariation of paired waves , is very small . Instead , the CA3som gamma wavelets are tightly locked to ipsilateral Schaffer waves . The results obtained with the phase-difference test indicate a preferred CA3som to Schaffer direction , possibly distorted by the frequent mismatch of waves and shorter duration of CA3som waves . The lag between paired waves is a more realistic estimation , and it determines that the two initiate nearly simultaneously , as seen in evoked potentials . We should have in account that the Schaffer waves are detected in CA1 recordings ~2–3 ms after CA3 PCs fire , a lag that is fully compatible with CA3som waves arising from basket cell-mediated recurrent inhibition following the same PC firing ( Hájos et al . , 2004 ) . All these findings support a main participation of the CA3som activity in ipsilateral mechanisms , as also indicated by the large ipsilateral coherence and the nearly synchronous occurrence of waves at anterior and posterior sites . This latter feature is consistent with the reported coupling of activity in subnetworks of inhibitory cells ( Galarreta and Hestrin , 1999; Tamás et al . , 2000 ) , which provides modulation of CA3 output over extended domains . Since the marked gamma modulation indicates their participation in the CA3 gamma oscillator ( Fisahn et al . , 1998; Hájos et al . , 2004; Mann et al . , 2005; Gulyás et al . , 2010 ) , it would be important to determine the extent of septo-temporal coherence . On another line , one might think that a somatic inhibitory input plays a role in selecting the CA3 domain of PC cells from which the functional assembly forming a Schaffer wave arises . However , the strikingly poor covariation between CA3som and Schaffer waves does not appear to be in consonance with this view . Another possibility is that multiple CA3 assemblies share a common domain , since nearby PC cells rarely fire together ( Takahashi et al . , 2010; Matsumoto et al . , 2013 ) . Thus they would deliver Schaffer waves of similar spatial coverage in CA1 . It is also compatible with the finding that Schaffer and CA3som waves have a tight covariation in a longer time scale . A possible explanation of this is that both CA3 PCs and basket cells receive a common tonic input ( Glickfeld and Scanziani , 2006; Freund and Katona , 2007 ) that codes for the duration of the gamma strings and intermissions , possibly as an energy-saving mechanism to stop the intrinsic oscillators in absence of activity and when bilateral entrainment is unnecessary . The extreme amplitude variability of successive Schaffer waves and the degree of bilateral covariation is of particular interest . Although the functional significance of this is unknown , some possibilities arise within a global framework that includes the oscillatory nature of the activities and their cross-checking through the commissural system . Theoretical studies show that gamma oscillations form rapidly from small clusters of interconnected cells ( Börgers et al . , 2012 ) . The rhythmic succession of LFP events implies a certain loss of coding performance , although the variability in amplitude exhibited by Schaffer waves destroys the repetitive character of the transmission at the cellular level as it involves a different number of firing PCs , and even a different functional assembly ( Fernandez-Ruiz et al . , 2012a , 2012b ) . Indeed , coding is optimized by global oscillations within a population whose units fire sparsely ( Engel et al . , 2001; Fries , 2009; Chalk et al . , 2016 ) . From the side of target CA1 units , the gamma rhythmic CA3 input implies some sort of chunked information , as recently proposed in the cortex ( Hyafil et al . , 2015 ) . Indeed , the optimal use of these chunks would be to meet other inputs at precise times . One of these is the contralateral homonymous input with which it maintains tight covariation in some gamma strings but not in others , albeit always in phase . This observation already indicates that in some epochs the information conveyed is similar on both sides and it differs in others , which is also supported by the bilateral out-phasing of gamma waves and the occurrence of some unilateral gamma strings upon VHC disruption . Similarly , it is interesting that CA1 PC cells fire preferentially on bilateral waves and we ruled out the possibility that this was a mere response to larger ipsilateral inputs . Curiously , the blockade of the contralateral CA3 input modifies the firing rate of units but the changes were balanced at the population level , in agreement with previous findings ( Zemankovics et al . , 2013; Middleton and McHugh , 2016 ) . Similar population balance of unitary changes has been reported following long-term synaptic plasticity ( Martin and Shapiro , 2000; Dragoi et al . , 2003; Yun et al . , 2007; Fernández-Ruiz et al . , 2012b ) . In the context of bilateral integration , we may then infer that the two CA3-originated excitations summed in CA1 units do not operate as a simple integrator , but rather the output is defined on a single cell basis in concert with other inputs , such as that from inhibitory neurons that are also modulated by interhippocampal disruption . The rather disperse firing of PCs over the gamma cycle , in clear contrast to the phase-locked firing of putative interneurons , supports this concept . Such dispersion may arise from the slower dynamics of intracellular EPSPs compared to EPSCs ( reflected in extracellular field waves ) , and/or the variable influence of different interneuron subtypes with distinct phase coupling to gamma waves ( Hájos et al . , 2004; Tukker et al . , 2007; Vinck et al . , 2010; Varga et al . , 2014 ) . Indeed some interneurons can be envisaged as timing devices for different inputs to converge at optimal instants . In turn , the gamma-phase shifting of CA1-PCs is in contrast with the tight gamma coupling of CA3-PCs ( Hájos et al . , 2004; Fernandez-Ruiz et al . , 2012a ) , indicating a short-lived gamma pattern restricted to this segment of the hippocampal circuit . Indeed , we have not found a gamma pattern in the subicular projection of the CA1 population ( Makarova and Herreras , unpublished results ) . Theoretical studies indicate that connection asymmetry favors temporal association ( Sompolinsky and Kanter , 1986 ) and the generation of robust transients of sequential activation ( Rabinovich et al . , 2008 ) . The dominance of the right side of the CA3 may thus constitute a physiological mechanism to provide a fast dynamic trade-off between left and right CA3 gamma oscillators for the fast bilateral synchronization and stabilization of the entrainment of autonomous CA3 gamma oscillators despite of the strong amplitude variability of individual waves . Overall , the present observations are compatible with the hypothesis that the main functional role for CA3 gamma oscillations is their bilateral entrainment , rather than transmitting gamma activity to the cortex . It is important to differentiate between asymmetric connectivity and lateralized routing of information . The larger Schaffer waves in the right-hand side may reflect left-right asymmetry in the subunit composition of Glutamate receptors reported in the CA1 ( Kawakami et al . , 2003; Shinohara et al . , 2008 ) , although a larger size of CA3 assemblies on the right-hand side cannot be excluded . In addition , right dominance better fits a hardware-like role to promote the integration of lateralized streams of information , but we cannot rule the priority or the qualitative relevance of that running through the left and right circuits during specific tasks ( Klur et al . , 2009; Shinohara et al . , 2013; Shipton et al . , 2014 ) . Different hippocampal subregions are known to be differentially involved in the encoding and consolidation/retrieval of spatial information ( Jerman et al . , 2006; Hunsaker et al . , 2008 ) , or in perceived versus reconstructed scenes ( Zeidman et al . , 2015 ) . Also , spatial and non-spatial information converge in the hippocampus to form episodic memories ( Leutgeb et al . , 2004 ) , although functional difference between hemispheres has rarely been invoked . A significant finding is that both CA3 are required for short-term memory , yet only inactivation of the left CA3 impairs performance in an associative spatial long-term memory task and plasticity ( Shipton et al . , 2014 ) . Along with the present observations of independent gamma strings upon VHC blockade , it becomes clear that the left and right sides do not convey equivalent information . Whether lateralization is devoted to different sensory modalities , to features of a scene , or to the perceived/recalled nature remains unknown . The hippocampus is thought to represent behavioral episodes as sequences of events ( Eichenbaum , 2004 ) . How these events are coded at the cellular level is a matter of intense study , although experimental evidence that is largely based on unitary studies is hard to unite in a single framework . Based on the assembly code of CA3-transmitted activity , one interesting possibility is that groups of CA3 assemblies distributed over the septotemporal axis ( Benito et al . , 2014 ) sequentially code the relevant features of a scene in a gamma string , which are checked against each other bilaterally to form higher order features and to modulate parallel or lateralized output to the cortex ( Ciocchi et al . , 2015 ) ( Figure 7 ) . The simplicity of this mechanism makes likely that it subserves bilateral assemblage of lateralized activity in other brain structures .
The animals were anesthetized with urethane ( 1 . 2 g/kg , i . p . ) , fastened to a stereotaxic frame ( Narishige , mod . SR-6R ) and the body temperature was maintained at 37°C with a heating pad and feedback control . The long-lasting anesthetic bupivacaine ( 0 . 75% ) was applied at surgical wounds . In experiments aimed at exploring the inter- and intrahippocampal Schaffer gamma ( n = 7 animals ) one or two concentric bipolar stimulating electrodes were placed in the soma layer of the CA3b region in the left or in both hemispheres for orthodromic activation of the CA1 ( AP 2 . 9; L ± 2 . 6; V 3 . 4 mm from Bregma and cortical surface ) . Recordings were obtained from a total of 64 sites by two twin-shank linear probes ( 16 channels per shank , 100 μm intersite distance: Neuronexus , Ann Arbor , MI ) that were located at homotopic sites of the dorsal hippocampus across the CA1 region and that also spanned the DG/CA3 ( AP 4–4 . 5; L ± 2 . 6 mm ) ( Figure 1B ) . The shanks ( 0 . 5 mm apart ) were oriented parallel to the midline to maximize spatial coherence of Schaffer activity ( Benito et al . , 2014 ) . In a different group of experiments aiming to disrupt the commissural fibers ( n = 4 animals ) two single shank probes were used ( 32 channel each , 50 μm intersite distance ) ( AP 4 . 5; L ± 2 . 6 ) . CA1 units were obtained from the recordings in the first animal group as well as from an additional group of 3 animals using four-shank probes ( 150 μm apart ) , each with two tetrodes separated vertically by 200 μm . The probes were soaked in DiI ( Molecular Probes , Invitrogen , Carlsbad , CA ) before insertion for postmortem evaluation of their placement in histological sections . A silver chloride wire in the neck skin served as a reference for recordings . Signals were amplified and acquired using MultiChannel System ( Reutlingen , Germany ) recording hardware and software ( 50 kHz sampling rate ) . The commissural pathway was functionally blocked by local application of the Na-channel blocker lidocaine ( HCl 2%: Braun ) to the left side of the VHC ( AP 1 . 5; L 0 . 5 mm ) . We injected a microdrop of drug solution ( ~50 nl ) through a glass pipette ( 7–12 µm at the tip ) using pressure through a syringe connected by a plastic tube ( Mizumori et al . , 1989 ) ( Picospritzer , General Valve ) . The same pipette was also employed for electrical stimulation , and the bilaterally recorded CA1 evoked fEPSPs guided the placement of the linear probes , as well as the effectiveness of the drug . Placement stability of the VHC pipette was weak upon injection . Therefore , we only used experiments in which it remained in site after delivery of the drug and the evoked potentials were abolished for at least 2–3 min . The design described above is not sufficient to study the effects of interhippocampal disruption on the spontaneous unit firing of CA1 cells , which requires longer epochs . To this end in another group of experiments ( n = 4 ) we used the selective blockade of the CA3 itself on the left-hand side , sparing the circuitry on the right-hand side . Larger microdrops ( 0 . 1–0 . 2 µl ) of lidocaine were injected into the septal pole of the left CA3 through a 200 µm wide silica cannula inserted into the cannula of the concentric stimulating electrode along with the inner wire , connected through a plastic tube to a 5 µl Hamilton syringe . Admittedly , the disruption of the interhippocampal communication through the VHC may not be complete since fibers originating in dorso-caudal CA3 sites running into the fimbria ( Laurberg , 1979 ) are spared by the injection . The extension of the drug was monitored by the modulation of evoked potentials elicited by the ipsi- and/or contralateral CA3 , whose fibers pass above the injection site on their way from the fimbria to the CA1 . Typically 1 or 2 injections were sufficient to ensure complete blockade of ipsilateral CA1 fEPSPs that was stable for at least 10 min . For longer drug action , successive microdrops were injected at 5 min intervals , resulting in reasonable stability as witnessed by the selective steady effect on ipsilateral CA1 fEPSPs . Although evoked activity began to recover 15–30 min after injection , we noted that hippocampal LFP activity displayed abnormal interregional patterns for long periods ( >60–90 min ) . Hence , a recovery period could not be reasonably established for unitary analysis . At the end of the recording session the animals were sacrificed by anesthetic overdose , and their brain was removed and maintained in 4% paraformaldehyde in saline . Sagittal brain sections ( 100 μm ) were stained with bis-benzimide and the electrode positions assessed by fluorescence microscopy ( Figure 2—figure supplement 1 ) . | In humans and other backboned animals , the brain is divided into the left and right hemispheres , which are connected by several large bundles of nerve fibers . Thanks to these fiber tracts , sensory information from each side of the body can reach both sides of the brain . However , although many areas of the brain work with a counterpart on the opposite hemisphere to process this sensory information , they do not necessarily perform the same tasks , or perform them at the same time as their partner . The hippocampus is a brain region that helps to support navigation , to detect novelty , and to produce memories . In fact , our brains contain two hippocampi – one in each hemisphere . Previous studies of the hippocampus have tended to record from only one side of the brain . Benito , Martín-Vázquez , Makarova et al . now compare the activity of the left and right hippocampi , and consider how the two structures might work together . Recordings of the electrical activity of the hippocampi of anesthetized rats show that different groups of neurons fire in rhythmic sequence , forming waves called gamma waves . Successive waves have different amplitudes , and can be thought to form ‘strings’ . The recordings made by Benito et al . show that the two hippocampi produce parallel strings of waves , although the waves that originate in the right hemisphere are generally larger than those that originate in the left . Right-hemisphere waves also tend to begin slightly earlier than their left-hemisphere counterparts . Further experiments revealed that disrupting the fiber tracts between the hemispheres uncouples the waves that no longer occur at the same time , and the strings of waves may remain constrained to one side of the brain . In healthy animals , however , the right-hand dominance acts as a master-slave device , and makes the waves from the two hemispheres pair up and merge in the neurons that receive them both . Thus the information running in both hippocampi can be integrated or compared before sending to the cortex for task execution or storage . Overall , the findings reported by Benito et al . suggest that different types of information flow through the left and right hemispheres , and that the brain integrates these two streams using asymmetric connections . The next challenge is to identify how the information in the two streams differs: whether each stream reflects different sensory stimuli , different features of a scene , or the difference between recalled and perceived information . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"short",
"report",
"neuroscience"
] | 2016 | The right hippocampus leads the bilateral integration of gamma-parsed lateralized information |
Random search is a behavioral strategy used by organisms from bacteria to humans to locate food that is randomly distributed and undetectable at a distance . We investigated this behavior in the nematode Caenorhabditis elegans , an organism with a small , well-described nervous system . Here we formulate a mathematical model of random search abstracted from the C . elegans connectome and fit to a large-scale kinematic analysis of C . elegans behavior at submicron resolution . The model predicts behavioral effects of neuronal ablations and genetic perturbations , as well as unexpected aspects of wild type behavior . The predictive success of the model indicates that random search in C . elegans can be understood in terms of a neuronal flip-flop circuit involving reciprocal inhibition between two populations of stochastic neurons . Our findings establish a unified theoretical framework for understanding C . elegans locomotion and a testable neuronal model of random search that can be applied to other organisms .
Random search is an evolutionarily ancient set of foraging strategies that evolved as an adaptation to environments in which prey items are undetectable at a distance and occur at unpredictable locations . Rather than attempting to exhaustively search a region of interest , the organism samples the environment at randomly selected points . This is achieved by executing a series of straight-line movements , called 'runs , ' terminated at random intervals by sampling episodes during which the organism may or may not find prey . Sampling ends in a reorientation event , called a 'turn , ' such that the next run is usually in a different direction from the preceding one . In optimal random foraging strategies the probability distribution of run length is matched to the statistical distribution of isolated food patches or prey items ( Viswanathan , 2011 ) , with power law distributions predominating when resources are sparsely distributed and exponential distributions predominating when resources are densely distributed ( Humphries et al . , 2010; Sims et al . , 2012; Humphries et al . , 2012 ) . Random search has been documented in a wide range of species including microorganisms , nematodes , insects , mollusks , fish , birds , and mammals including humans ( Viswanathan , 2011; Berg and Brown , 1972; Pierce-Shimomura et al . , 1999 ) . In humans this strategy is observed in diverse contexts , from traditional hunter-gatherer societies ( Brown et al . , 2007; Humphries and Sims , 2014 ) to technologically enhanced fishing industries ( Bertrand et al . , 2007 ) . The formal similarities between random search across widely diverse phyla and spatial scales ( Viswanathan , 2011 ) may point to a common mechanism , even in organisms that are highly cognitive . Despite the universality of random search , little is known about its neuronal basis , in part because of the difficulty of recording and manipulating activity in the brain of an unrestrained animal while it explores a large region of space . The relatively small spatial scale of random search behavior in C . elegans , coupled with the simplicity of its nervous system , provides a unique opportunity to identify the neuronal basis of random search in this species . To the unaided eye , C . elegans search behavior consists of forward runs , each terminated after a variable interval by a briefer period of reverse locomotion , which is also variable in duration ( Pierce-Shimomura et al . , 1999; Zhao et al . , 2003; Wakabayashi et al . , 2004 ) , with apparently stochastic switching between these two behavioral states . Reversals are followed by resumption of forward movement that frequently begins with a deep body bend . These bends are highly variable in amplitude and lead to movement in a new direction . Thus , the sequence reverse–forward–deep bend , called a 'pirouette' ( Pierce-Shimomura et al . , 1999 ) is the fundamental turning event in C . elegans random search , with functional analogies to tumbles in bacterial chemotaxis ( Berg and Brown , 1972 ) . Careful inspection reveals a third state , called “pause , ” in which locomotion ceases for a fraction of a second or more ( Croll , 1975; Shingai , 2000; Stephens et al . , 2008; Rakowski et al . , 2013; Salvador et al . , 2014 ) . Thus , C . elegans locomotion consists of three main behavioral states – forward , reverse , and pause – together with the transitions between them . C . elegans subsists on a diet of bacteria that it finds mainly in rotting plant material ( Frézal and Félix , 2015 ) . In the laboratory , search behavior is studied in worms foraging on agar plates containing one or more dense bacterial lawns , analogous to food patches in the ethological literature . Like many other organisms , C . elegans can tune the spatial scale of random search to its physiological state , the availability of food ( Wakabayashi et al . , 2004; Gray et al . , 2005 ) , and prior knowledge of its distribution ( Calhoun et al . , 2014 ) . The lowest values of search scale are observed during “cropping , ” ( Jander , 1975 ) the exploitation of a dense food patch . In C . elegans , two substates of cropping have been described: "dwelling , " characterized by especially low crawling speed and frequent ( presumably short ) reversals , and "roaming , " characterized by somewhat higher speeds and less frequent reversals . Transitions between dwelling and roaming , like the transitions between forward and reverse locomotion , are stochastic ( Ben Arous et al . , 2009; Fujiwara et al . , 2002; Flavell et al . , 2013 ) . Intermediate values of search scale are observed during "local search" ( Wakabayashi et al . , 2004; Hills et al . , 2004 ) when , for example , the animal is suddenly transferred from a bacterial lawn to a foodless region of the plate . The highest values of search scale are observed during “ranging , ” when food is exhausted , starvation sets in , and the need to find a new food patch becomes urgent ( Wakabayashi et al . , 2004; Gray et al . , 2005 ) . Worms sometimes spontaneously leave a food patch well before it is exhausted , with leaving rate inversely related to food quality and food density ( Shtonda and Avery , 2006; Harvey , 2009 ) , which may reflect a trade-off between exploitation and exploration ( Bendesky et al . , 2011 ) . At the heart of the C . elegans locomotion circuit are five pairs of premotor 'command' interneurons organized into two functional groups that promote forward and reverse locomotion , respectively ( Chalfie et al . , 1985; Zheng et al . , 1999; Stirman et al . , 2010; Schmitt et al . , 2012 ) . The two groups are reciprocally connected , and make output synapses onto distinct , non-overlapping sets of motor neurons that control body-wall muscle . The locomotory state ( forward or reverse ) is believed to be determined mainly by whichever set of motor neurons is more highly activated by input from the command neurons ( Kawano et al . , 2011; Xie et al . , 2013; Gao et al . , 2015; Liu et al . , 2014 ) . Command neuron activation depends upon influences that are both external and intrinsic to the command neuron network , and appears to have a strong stochastic component that underlies switching between forward and reverse locomotion . Some command neurons are tightly linked both functionally and synaptically to upstream interneurons that also switch state stochastically in concert and counterpoint to them ( Gordus et al . , 2015 ) , providing a potential additional source of the stochasticity on which random search depends . At least nine classes of chemosensory neurons and twelve classes of upstream interneurons are required for normal regulation of the duration of forward locomotion ( Viswanathan , 2011; Gray et al . , 2005; Tsalik and Hobert , 2003; Fang-Yen et al . , 2015 ) . Input from these neurons onto the command neuron network modulates the mean run length and , thereby , the spatial scale of random search . Search scale also appears to be modulated by neurons that release biogenic amines ( serotonin , dopamine , and tyramine ) ( Flavell et al . , 2013; Hills et al . , 2004; Bendesky et al . , 2011 ) or peptides ( Ben Arous et al . , 2009; Flavell et al . , 2013; Gloria-Soria and Azevedo , 2008; Styer et al . , 2008; Reddy et al . , 2009; Bhattacharya et al . , 2014 ) . These diverse signaling pathways may provide the means by which the worm optimizes its search strategy in response to feeding history ( Gray et al . , 2005 ) , the quality , density and spatial distribution of food ( Shtonda and Avery , 2006; Calhoun et al . , 2015 ) , and other factors that constrain survival and reproduction ( Gloria-Soria and Azevedo , 2008; Pujol et al . , 2001; Pradel et al . , 2007; Lipton et al . , 2004 ) . Although the neural circuitry for local search has been described in considerable detail , our understanding of the system remains limited , partly for lack of key physiological data , but also for lack of a model in which to interpret the data . Common sense suggests that the forward and reverse command neurons should inhibit each other to minimize simultaneous occurrences of neuronal states for incompatible behaviors ( Zheng et al . , 1999 ) . A plausible anatomical substrate for such reciprocal inhibitory connections between command neurons exists in the C . elegans connectome ( White et al . , 1986 ) , but anatomical data do not specify the signs or strengths of synaptic connections . A quantitative model that incorporates physiological properties of the command neurons and their synaptic connections is needed to interpret experimental results , such as the unexpected observation that silencing some of the reverse command neurons causes a reduction in forward dwell time , and conversely for forward command neurons ( Rakowski et al . , 2013; Zheng et al . , 1999 ) . It is also needed to explain complex patterns of changes in dwell times observed across the three locomotory states caused by introducing or eliminating tonic membrane conductances in the command neurons , and to answer basic mechanistic questions about the control of C . elegans locomotion . At present , the experimental data are insufficient for creating a neuron-by-neuron model of the command network that incorporates biophysical details such as synaptic and membrane conductances without introducing a heavy load of unconstrained parameters ( Rakowski et al . , 2013 ) . Nor would such a mechanistically detailed model necessarily provide the appropriate level of abstraction in which to intuitively understand C . elegans search behaviors , including their strong stochastic component . Instead , we have kept the level of biological detail to the minimum needed to predict the statistical distributions of dwell times in forward , reverse and pause states , and other fundamental aspects of the behavior . Each of the model's three main assumptions remains within the bounds of widely accepted experimental results; our mathematical analysis simply shows what follows necessarily from these assumptions . To provide an empirical basis for the model we quantified C . elegans search behavior in terms of tangential velocity , defined as the speed and direction of worm's movement along its sinuous trajectory , which we recorded at higher resolution than previously possible . Behavioral data were then fit to a four-state hidden Markov model in which each state corresponds to a unique pattern of activation across the command neurons . Importantly , rate constants governing probabilistic transitions between states in the Markov model are expressed in terms of synaptic weights in an analytically tractable version of the model . We were therefore able to validate the model by showing that it correctly predicts phenomena on which it was not fit , such as reciprocal inhibition between forward and reverse command neurons in the biological network and the behavioral effects of perturbations introduced by laser ablations and genetic mutations . Although the model is inherently probabilistic , we found that it also makes accurate predictions concerning deterministic behaviors in C . elegans , indicating a potentially high level of generality . The present findings thus establish a simple theory of C . elegans locomotory control and provide a testable model of random search that can be applied to other organisms .
Figure 1A– D describes important features of search behavior obtained by regarding the worm as a point moving in an external reference frame ( allocentric coordinates ) without regard to the orientation of the body axis . The speed distribution was bimodal ( Figure 1B ) with a broad peak around 180 µm/s that includes both forward and reverse motion , and a narrower peak near zero that corresponds to pauses . The speed autocovariance function had multiple exponential components ( Figure 1C ) , suggesting at least three locomotory states . The average change in heading angle ( |Δφ|¯ ) , plotted as a function of the intervening time interval ( Figure 1D ) , showed that worms maintained a nearly constant heading for up to 10 s ( Stephens et al . , 2010; Peliti et al . , 2013 ) , but reoriented randomly within ~30 s , establishing the shortest time scale over which the behavior can be considered a Brownian random walk ( Figure 1—figure supplement 2 ) , the simplest form of random search . On shorter time scales the path takes on the character of a truncated Lévy flight ( Mantegna and Stanley , 1994 ) . For more detailed analysis we distinguished forward from reverse movement by visual inspection of the recorded videos , and defined velocity , V ( t ) , to be a signed scalar value that denotes the speed of movement along the worm’s track in the direction of the head ( + ) or tail ( - ) ( Figure 1E; see Materials and methods ) . The probability distribution of V ( t ) ( Figure 1F ) showed two broad peaks that correspond to forward and reverse movement , and a narrow third peak centered at zero that corresponds to pauses . For the initial analysis we defined pauses using a fixed speed threshold of 0 . 05 mm/sec ( Rakowski et al . , 2013 ) . Pauses occurred most frequently as transient interruptions of forward locomotion , causing the worm to stutter as it moves ( Figure 1E; Video 1 ) ; stuttering also occurred , albeit less frequently , during reverse locomotion ( Figure 1E; Video 2 ) . Distinct pauses were also observed during transitions from forward to reverse ( Figure 1G; Video 3 ) and from reverse to forward ( Figure 1H; Video 4 ) . Most pauses lasted longer than one video frame , indicating the presence of a locomotory state having a detectable dwell time; thus pauses were not merely zero crossings in plots of velocity versus time . We found that pauses during forward to reverse transitions were on average longer in duration than pauses during reverse to forward transitions ( Figure 1I; p<10–5 ; Mann-Whitney U-test ) . These findings are consistent with the predictions of the model presented below , which uses a probabilistic criterion rather than a fixed velocity threshold to identify pauses . Based on the results presented in Figure 1 and previous studies noted below , we propose a minimal model for the control of random search behavior that involves two opposing neuron-like “units” that can exist in four distinct states corresponding to forward locomotion , reverse locomotion , and two pause states . This model differs from a previous model that represents the worm as a point in “shape space” ( Stephens et al . , 2008 ) in that here velocity is measured directly by observing the motion of a point on the body surface relative to the substrate , rather than indirectly by the temporal progression of shape changes . It also differs from previous models ( Rakowski et al . , 2013; Zheng et al . , 1999; Wicks et al . , 1996; Kunert et al . , 2014 ) by representing changes in locomotory state as probabilistic transitions in a Markov process . Ablation of individual premotor interneurons ( Chalfie et al . , 1985 ) has led to the hypothesis that the direction of locomotion is controlled by a network comprising five pairs of premotor command interneurons organized into two functional groups that promote forward and reverse locomotion , respectively . Although the anatomical pattern of synaptic connectivity among these interneurons has been established ( White et al . , 1986 ) ( Figure 2A ) , this information does not yield an intuitive explanation of how the direction of locomotion is regulated . Nor , in our view , does the present state of the anatomical connectivity provide the basis for a neuron-by-neuron simulation of the network ( but see Rakowski et al . , 2013 ) , as neither signs nor physiological strengths ( weights ) of synapses in C . elegans can be inferred reliably from anatomical structure or neurotransmitter type , and almost nothing is known about the intrinsic membrane currents of these neurons or how they shape the input-output function of individual command neurons . To establish a mathematically tractable framework for understanding how the command network functions during search behavior , we created a minimal model based on three simplifying assumptions , each of which was biologically motivated . ( i ) Command neurons act like binary units ( Hopfield , 1982 ) . This assumption was based on voltage recordings from command neurons in which we regularly observed two stable membrane potentials with rapid transitions between them ( Figure 2B; also see Kato et al . , 2015 ) . It is also supported by the observation of a bimodal distribution of calcium activity in AVA neurons and their upstream partners AIB and RIM ( Gordus et al . , 2015 ) , and the report of distinct up and down states in voltage recordings from motor neruons ( Liu et al . , 2014 ) . ( ii ) Command neurons switch state stochastically . This assumption was based on the observation that C . elegans locomotory behavior has a strong stochastic component , with exponentially distributed dwell times in forward and reverse states ( Pierce-Shimomura et al . , 1999; Zhao et al . , 2003; Flavell et al . , 2013; Gordus et al . , 2015; Stephens et al . , 2011 ) . ( iii ) Command neurons within the forward pool are co-active , as are command neurons in the reverse pool . This assumption is based on simultaneous calcium imaging data from multiple command neurons in freely moving animals which suggest that the activity of neurons within the reversal pool is tightly correlated ( Schrödel et al . , 2013; Pokala et al . , 2014 ) . Additionally , neurons in opposing groups are likely to be reciprocally active , as indicated by simultaneous calcium imaging from AVA and AVB ( Pokala et al . , 2014; Faumont et al . , 2011 ) , as well as AVE and AVB ( Kawano et al . , 2011 ) . A fourth assumption , concerning the relationship between neuronal states and behavioral states , is introduced below . The three simplifying assumptions , together with the anatomical data ( White et al . , 1986 ) , lead to a model that has two binary stochastic elements , ℱ and ℛ , and six synaptic weights ( Figure 2C ) . Each type of weight has a specific interpretation . The cross-connections ( Wℱℛ , Wℛℱ ) represent mono- and polysynaptic connections between command neurons in different groups . The self-connections ( Wℱℱ , Wℛℛ ) represent connections between command neurons in the same group , including recurrent polysynaptic pathways involving neurons outside the command network . Self-connections also represent possible intrinsic voltage dependent currents within the command neurons , such as C . elegans plateau currents ( Mellem et al . , 2008 ) . The pair of connections originating from an ℱor ℛ unit can have either the same sign or different signs . Allowing a single unit to have opposing effects on different postsynaptic targets is justified by the fact that synaptic weights in the model represent polysynaptic pathways , the effects of which can be excitatory or inhibitory , and by the observation that some C . elegans neurons can monosynpatically excite some postsynaptic neurons while inhibiting others ( Chalasani et al . , 2007 ) . Two additional weights , hℱ and hℛ , represent inputs from sensory neurons , interneurons , neural modulators , and any other sources outside the network ( Gray et al . , 2005; Fry et al . , 2014 ) , plus intrinsic membrane conductances that produce sustained effects on membrane potential ( Zheng et al . , 1999; Gao et al . , 2015 ) . The summed synaptic inputs onto ℱ and ℛ are , respectively , Sℱ ( t ) =hℱ+Wℱℱbℱ ( t ) +Wℛℱbℛ ( t ) hℱ and Sℛ ( t ) =hℛ+Wℛℛbℛ ( t ) +Wℱℛbℱ ( t ) , where bℱ ( t ) and bℛ ( t ) are the states of ℱ and ℛ at time t ( 1 = ON , 0 = OFF ) . The quantities hℱ and hℛ were assumed to be constant during the 10 min observation period of local search behavior on a bare agar surface . State transitions of ℱ and ℛ were modeled as independent non-homogeneous Poisson processes in which the transition rates are exponential functions of the summed synaptic input to the units , as shown in Figure 2—figure supplement 1 . Changes of the state of ℱ and ℛ can be regarded as thermally-driven transitions across energy barriers of height proportional to Sℱ ( t ) and Sℛ ( t ) , respectively . Inhibitory synaptic input increased the height of the barrier for OFF→ON transitions while decreasing the height of the barrier for ON→OFF transitions by the same amount; excitatory synaptic inputs had the opposite effect . The variable A ( Materials and methods , Equations 26 , 27 ) represents the fundamental timescale of the system , defined as the rate at which units ℱ and ℛ change state when the summed synaptic input is zero . The present model is distinct from deterministic models of the command neuron network ( Rakowski et al . , 2013; Zheng et al . , 1999; Wicks et al . , 1996; Kunert et al . , 2014 ) in that it is inherently stochastic , like the behavior it is meant to predict . In particular , the synaptic input to a unit does not immediately determine its state , but instead modifies the transition rates between ON and OFF states . The two binary units of the model can exist in four states ( F , R , X , Y; Figure 2D ) , and provide the basis for a hidden Markov model having eight transitions in which a single unit changes state . The model was further constrained by the synaptic model , which allows the eight transition rate constants to be specified by only six synaptic weights as shown in Equations 31–35 ( Materials and methods ) . A Markov model was adopted to represent the biological system because dwell times in Markov states , like the observed dwell times in forward and reverse states ( Zhao et al . , 2003; Viswanathan , 2011 ) , are exponentially distributed . A hidden Markov model was required because , as noted above , states of command neurons cannot be observed directly in freely moving animals , even using optical recording methods . The fourth assumption is a particular mapping between the states of the two command units and behavioral states of the worm . The command units , ℱ and ℛ , are intended to represent the two pools of forward and reverse command neurons , respectively , such that the worm moves forward when ℱ is ON and ℛ is OFF ( state F ) , backwards when ℛ is ON and ℱ is OFF ( state R ) , and pauses when both ℱ and ℛ are OFF ( state X ) . These associations between states of the model and activation states of the command neurons are well supported by previous experimental evidence , including studies showing that genetic ablation or silencing of all command interneurons induces prolonged pauses ( Zheng et al . , 1999; Kawano et al . , 2011 ) , but they also assume the major simplification that all command neurons in a given pool act together as a unit . The model also permits a fourth state , in which ℱ and ℛ are simultaneously ON ( state Y ) . Whether the corresponding co-activation state of forward and reverse command neurons normally exists with any significant probability remains to be shown , but it has been observed that their downstream targets , the forward and reverse motor neurons , can be active simultaneously , causing the worm to pause ( Kawano et al . , 2011 ) . Given the existence of gap junction synapses between the main forward and reverse command neurons and their respective sets of forward and reverse motor neurons , it is reasonable to suppose that forward and reverse command neurons are co-active when their motor neurons are co-active . Thus , there is some evidence to designate state Y as a second pause state , which we consider to be a working hypothesis . Together , states X and Y comprise the phenomenological pause state denoted P . In what follows , we explore the logical consequences of the model’s four assumptions; it remains to be shown experimentally how closely the states of the model correspond to activity states of the command neurons . We used a maximum likelihood method ( Colquhoun and Hawkes , 1995 ) ( Materials and methods ) to estimate the set of transition rate constants that had the highest probability of generating the observed time series v ( t ) . Direct transitions between F and R , and between X and Y , were disallowed because the assumed statistical independence of the two command units implies that the probability of simultaneous transitions in ℱ and ℛ is vanishingly small . ( Note , however , that the model does allow transitions between any two states during the interval between successive video frames by making two or more non-simultaneous transitions; see Equation 21 ) . We first fit the velocity distribution for each cohort with three overlapping probability distributions corresponding to forward , reverse and pause states ( Figure 2—figure supplement 2 ) , then searched for the set of transition rate constants that maximized the likelihood of the observed v ( t ) given the velocity distributions . The resulting rate constants were used to compute the most likely sequence of states via the Viterbi algorithm ( Rabiner , 1989; Viterbi , 1967 ) . The agreement between observed velocity data and the sequence of states shown in Figure 2E was typical of the entire data set . The maximum likelihood rate constants for 5 wild-type cohorts , together with the predicted state probabilities and mean dwell times computed from them , are given in column A of Table 1 . The model’s predicted mean dwell time in the reverse state ( dR= 1 . 94 ± 0 . 04 s ) agreed with previously reported values ( Zheng et al . , 1999; Kawano et al . , 2011 ) . In contrast , the predicted mean dwell time in the forward state ( dF= 5 . 33 ± 0 . 25 s ) was smaller than previously reported when dwell time was measured by eye ( 13–35 sec ) ( Zhao et al . , 2003; Zheng et al . , 1999; Brockie et al . , 2001; Ryu and Samuel , 2002 ) or by velocity threshold crossings ( 9–16 sec ) ( Rakowski et al . , 2013; Stephens et al . , 2011 ) . To determine whether this difference arose because we used a hidden Markov model rather than a fixed velocity threshold , we also identified states based on a fixed velocity threshold of 0 . 05 mm/s , and calculated the resulting mean dwell times: dF , 0 . 005= 1 . 86 ± 0 . 03 s; dR , 0 . 005= 1 . 23 ± 0 . 02 s; dP , 0 . 005= 0 . 14 ± 0 . 001 s . We attribute the short mean dwell times in state F that we observed using either the hidden Markov model or a fixed velocity threshold to the fact that our tracking system is capable of revealing briefer visits to state P , which interrupt forward runs , than previous methods . Ignoring transient interruptions of forward locomotion ( i . e . , FPF transitions ) and using the fixed velocity threshold of 0 . 05 mm/s yielded longer a mean forward dwell time of 9 . 13 ± 0 . 15 s , which matches the value obtained by others using the same threshold ( 8 . 98 ± 0 . 57 s ) ( Rakowski et al . , 2013 ) . Predicted mean dwell times in the two pause states differed substantially from each other ( dX=0 . 44 ± 0 . 03 s , dY= 0 . 21 ± 0 . 02 ) . We assigned the long and short pause states to X and Y , respectively , based on the idea that the energetically expensive state in which both units are on should be relatively short-lived . In previous work , transitions between locomotory states in C . elegans have been analyzed by choosing a speed threshold to distinguish pause states from the movement states ( Rakowski et al . , 2013; Salvador et al . , 2014; Stephens et al . , 2011 ) . The choice of threshold is important because it affects the measured dwell times , yet is necessarily arbitrary because the velocity distributions of the states overlap ( Figure 1F ) . The hidden Markov model used here replaces arbitrary thresholds with empirically determined state transition rates ( i . e . , the set of rates that maximizes the probability of the observed velocity time series ) , from which one can determine the sequence of states that is most likely to have generated the data ( the Viterbi algorithm ) . The hidden Markov model thus offers two advantages: ( 1 ) it provides a statistical criterion for selecting the best parameter values and ( 2 ) it takes into account the uncertainties in identifying the state of the system from velocity data . Under the assumptions of the hidden Markov model the state of the system cannot be observed directly because the velocity distributions overlap , making it impossible to test directly whether the predicted state probabilities agree with the observed velocity data . Nevertheless , an important test of the model can be obtained using the Viterbi algorithm to identify the most likely sequence of states given the observed velocity data , from which the histogram of dwell times in each state can be computed and compared to the exponential distribution predicted by the Markov model ( Figure 2—figure supplement 3 ) . The degree of agreement between the distributions shows that our model provides a good description of the system . The initial rationale for including two pause states in the hidden Markov model came from our model-independent analysis of the tracking data ( Figure 1I ) , which showed different dwell time distributions for pauses at FPR and RPF transitions . To test whether having two pause states yielded a statistically significant improvement in the ability of the model to fit the data , we eliminated one of the pause states and asked whether the resulting reduction in likelihood was greater than could be attributed to the reduction in the number of free parameters ( see Table 1 ) . For this comparison we constrained the transition rates into state Y to be extremely small ( aFY=aRY= 10-10 s-1 ) , effectively eliminating state Y and reducing the number of free parameters from six to four . The reduction in likelihood caused by eliminating one of the pause states was highly significant , and cannot be attributed simply to the elimination of two parameters ( p<10-100; likelihood ratio test ) . Separately , we considered the most general one-pause state model , which allows direct transitions between states F and R and has no constraints on the 6 transition rates other than that they are all ≥0 . The fit of this model ( Table 1 column C ) converged to nearly the same set of transition rates as the one-state model described above ( Model B ) . These comparisons show that our model with two pause states and six free parameters ( the six synaptic weights ) provides a much better fit to the data than models with only one pause state . We conclude that the tracking data contain a statistically significant signature of two distinct pause states . The model explains the observation that the pause dwell times during FPR transitions are longer than during RPF transitions ( Figure 1I ) in terms of the different dwell times in states X and Y ( dX>dY ) , and the strong tendency to cycle clockwise through state space , exiting from state F to state X and from state R to state Y as shown by the fate diagram ( Figure 3 ) . It has been reported that pauses in C . elegans locomotion occur at specific points in “shape space” ( Stephens et al . , 2011 ) , suggesting the worm pauses in preferred postures . To investigate this possibility , we analyzed worm tracks before and after pauses , inferring posture from the path of the tracking spot . This inference is justified by the fact that on an agar surface the worm moves without slipping , such that each segment of the body traces the trajectory of the one before it . Thus , the path of the tracking spot leading up to the pause reveals the worm’s posture posterior to the spot during forward locomotion , and anterior to the spot during reverse locomotion ( Figure 4 ) . Plotting mean curvature versus distance along the track ( Figure 4A ) reveals only a weak tendency to stop in a particular posture in state X ( r = 0 . 14; Figure 4B ) . Nearly all of the transitions into state X were either stutters during forward locomotion ( FXF transitions ) or reversals ( FXR transitions ) ; when these were analyzed separately , similarly weak postural preferences were found at FXF transitions ( r = 0 . 14 ) and FXR transitions ( r = 0 . 14 ) . A nearly identical result ( r = 0 . 14 ) was obtained using a fixed velocity threshold of 0 . 05 mm/s rather than the hidden Markov model to determine state . For the latter case , in which there is only one pause state , we analyzed the posture at all FP transitions , which almost always correspond to FX transitions in the hidden Markov model because FY transitions are extremely rare ( see Figure 3 ) . To test whether the failure to find a strong postural preference at FX transitions was due to including very short pauses in the analysis , we repeated the analysis after reclassifying all pauses shorter than a minimum duration as a continuation of the previous state , and obtained the same result; we found no strong postural preference at FX transitions for minimum pause durations up to 2 s ( r = 0 . 16 , 0 . 19 , 0 . 23 , 0 . 3 for X dwell times > 0 . 33 s , 0 . 67 s , 1 s , and 2 s , respectively ) ; longer dwells in state X were too rare to analyze . Thus , FX transitions can occur at any locomotory phase and do not occur preferentially at a particular posture ( Figure 4D ) ; in the case of FXR transitions the worm generally retreats along the same track . In contrast , we found a strong tendency to stop in a particular posture in state Y ( Figure 4A , C , E; r = 0 . 71 ) . Almost all entries into state Y were RYF transitions and these were associated with a ventral bend in the middle of the worm ( Figure 4E ) . These results suggest fundamental differences between the control of forward and reverse locomotion . In our model , forward locomotion terminates when forward command neurons turn OFF , and this can happen at any phase , whereas reverse locomotion terminates when forward neurons turn ON , and this is most likely to happen at a particular phase . The latter could be explained by phasic feedback from the locomotory pattern generator to the forward neurons ( Li et al . , 2006 ) . To determine the contributions of individual command neurons to the overall function of the command network , we separately ablated the pair of neurons that comprises each command neuron class , then tracked ablated and sham operated animals during local search . Mean velocities in F and R , if significantly changed , were reduced ( Pokala et al . , 2014 ) ( Figure 5A; ⋆⋆ ) , as was the frequency of undulations during forward and reverse locomotion ( Table 3 ) . In many organisms , the frequency of rhythmic behaviors is regulated by the amplitude of tonic excitatory drive to the associated pattern generator ( Weeks and Kristan , 1978; Satterlie and Norekian , 2001; Böhm and Schildberger , 1992; Deliagina et al . , 2000; Dembrow et al . , 2003; Hedwig , 2000; Sirota et al . , 2000 ) . To explain our results we propose that ablation of the locomotory command neurons reduces tonic drive to the presumptive locomotory pattern generator ( Xie et al . , 2013; Gao et al . , 2015 ) . A previous study found that ablating a subset of the reverse command neurons ( AVAL and AVAR ) reduces dwell time in the reverse state but also paradoxically reduces dwell time in the forward state ( Zheng et al . , 1999 ) . Similarly paradoxical effects have been reported following ablation of the reciprocally connected brain stem nuclei that regulate sleep and wakefulness ( Saper et al . , 2010 ) . The stochastic switch model predicts and explains such effects . In principle , the ablation of a subset of neurons in a pool of co-active neurons can have widespread effects on the pool’s overall input and output connectivity . Widespread effects can be expected because ablation removes not only the outgoing synaptic connections from the ablated neurons , but also the targets of incoming synaptic connections . In the C . elegans command neuron network , ablating a reverse command neuron such as AVA potentially reduces four of the six weights in the network: hℛ , Wℛℛ , Wℛℱ , and Wℱℛ . Thus , a single ablation can move the system a considerable distance in weight space toward the uncoupled state in which all weights are zero . In the limiting case of a fully uncoupled network , all dwell times approach a value of 1/2A , where A is the intrinsic switching time of the stochastic units ( see Materials and methods , equations 31–34 ) ; henceforth we will use d0 to denote the uncoupled dwell time . Dwell times that in intact animals are greater than d0 will be reduced by ablation , whereas dwell times that are less than d0 will be increased . In particular , if dF and dR are both greater than d0 , ablation of a reverse command neuron is expected to reduce both dwell times; the same is true for ablation of a forward command neuron . Thus the observed paradoxical effects of ablations are to be expected if d0 is below dF and dR . To determine the actual relationship between d0 and dwell times in the forward and reverse state , we estimated the rate constants in ablated animals and sham operated controls , and computed the corresponding dwell times ( Figure 5B; Table 4 ) . Dwell times in F and R , if significantly altered by the ablation ( ⋆⋆ ) , were reduced , indicating that d0 is indeed below dF and dR . Additionally , dwell times in the pause states dX and dY were increased , with one exception ( dY , AVB ) . Thus , the observed pattern of dwell time changes is consistent , overall , with a value of d0 that is between the dwell times of the movement states and the dwell times of the pause states . This finding allowed us to place bounds on d0 . Specifically , d0 must be less than or equal to the lowest post-ablation value of dR , and greater than or equal to the largest post-ablation value of dX; thus , 0 . 58 ≤ d0 ≤ 1 . 24 sec . Furthermore , because A=1/2d0 , we can infer that 0 . 40 Hz ≤ A ≤ 0 . 86 Hz . This inequality provides an estimate of the fundamental time scale of stochastic switching in C . elegans locomotion . For subsequent analysis , we defined Amin= 0 . 40 Hz and Amax= 0 . 86 Hz . Having placed bounds on A , we were able to compute synaptic weights in the model ( Table 2 ) . This was done by deriving expressions for the weights in terms of the rate constants ( Materials and methods , Equations 36–38 ) and substituting into these equations our estimates of rate constants together with the values of Amin and Amax . We found that input weights , hℱ and hℛ , are small and positive , suggesting that these inputs may provide modest but steady excitation to the system ( Figure 6A ) . The self-connections Wℱℱ and Wℛℛ are also mainly positive , indicating that the ON states may be stabilized by intrinsic or extrinsic positive feedback . The cross-connections Wℱℛ and Wℛℱ are negative , indicating reciprocal inhibition , as expected for neurons that activate opposing behavioral states . Furthermore , the magnitude of Wℱℛ is greater than the magnitude of Wℛℱ , suggesting that the animal spends more time in the forward state than the reverse state in part because the forward neurons inhibit the reverse neurons more strongly than the reverse neurons inhibit the forward neurons . Synaptic weights in an abstract network model such as this one , where neuronal state is activation rather than voltage , are not generally interpretable as synaptic conductances . Rather , they represent the functional effects of one neuron on another , such as the degree of excitation or inhibition produced by a unit change in activation . Thus , synaptic weights in the Stochastic Switch model cannot be said to predict the magnitude of synaptic conductances , but they can be said to predict aspects of functional connectivity in certain cases . For example , as command neurons AVA and AVB are behaviorally much more important than the others ( Chalfie et al . , 1985 ) ( see also Figure 5A , B ) , it is reasonable to assume that the signs of their functional synaptic connections match the signs of the net functional connections in the biological network . Thus , the model predicts reciprocal inhibition ( Qi et al . , 2012 ) between AVA and AVB under this assumption . We tested this prediction by photoactivating either AVA or AVB with channelRhodopsin-2 and recording electrophysiologically from AVB or AVA , respectively ( Figure 6B , C ) . We found that the reversal potential of optically induced synaptic currents in AVA and AVB was more negative than the zero-current potential in these neurons ( Figure 6B , C , D ) , indicating synaptic inhibition as predicted by the model . This inhibition is likely to be monosynaptic as C . elegans command neurons are cholinergic and express inhibitory postsynaptic receptors that respond to acetylcholine ( Pereira et al . , 2015 ) . Additionally , the connection from AVB to AVA appeared to be stronger thaCroninn the connection from AVA to AVB ( Figure 6E ) , measured in terms of the amplitude of the synaptic current at a holding potential approximately equal to the membrane potential when command neurons are in their depolarized state ( Figure 2B ) . However , we do not exclude the possibility that AVB was more strongly activated than AVA as a result of differential expression of the photoprobe . These findings demonstrate the feasibility of using the worm’s velocity , v ( t ) , a simple behavioral measure , to predict functional synaptic connections between populations of neurons in a biological neural network , at least under certain assumptions concerning the relationship between model network weights and physiological synaptic strengths . Two classes of ion channel mutants that affect membrane conductances in the command neurons are also known to alter locomotory behavior in systematic ways , thus providing key insights into command neuron function ( Zheng et al . , 1999 ) . The hyperpolarizing class ( “HYP” ) comprises three genotypes in which release of the excitatory neurotransmitter glutamate , presumed to be tonic , is disrupted by mutations that affect either presynaptic ( eat-4 ( ad572 ) , eat-4 ( ky5 ) ) or postsynaptic mechanisms ( glr-1 ( n2461 ) ) . These mutations are hypothesized to cause chronic hyperpolarization of the command neurons by reducing depolarizing currents . The depolarizing class ( “DEP” ) comprises two genotypes in which a constitutively activated glutamate receptor is expressed in the command neurons ( glr-1::glr-1 ( A/T ) , nmr-1::glr-1 ( A/T ) ) . These mutants are hypothesized to chronically depolarize the command neurons . We found that the frequency of locomotory undulations was decreased in HYP mutants and increased in DEP mutants compared to wild-type controls ( Table 5 ) , consistent with the likely effects of respectively increasing and decreasing tonic drive to the presumptive pattern generator for locomotion . Importantly , however , it is possible that both classes of mutation also alter the input resistance of the command neurons . The closure or removal of glutamate receptors in HYP mutants should increase input resistance whereas the introduction of constitutively active glutamate receptors in DEP mutants should decrease it . Thus , the previously observed effects of these mutations on locomotory state transitions ( Zheng et al . , 1999 ) could be the result of changes in membrane potential ( ΔV ) , input resistance ( Δr ) , or both . Changes in membrane potential and input resistance can both be represented in the stochastic switch model by changes in synaptic weights . We modeled the effects of ΔV by adding an increment Δh ( −1≤Δh≤ 1 ) to wild type h values , with negative Δh for HYP mutations and positive Δh for DEP mutations . We modeled the effect of Δr as a change in the magnitude of synaptic weights ( h and w quantities ) . This representation of Δr is appropriate because changes in input resistance alter the magnitude of the voltage change that would be produced by a fixed postsynaptic current . All weights were scaled by a common factor z ( 1 < z < 2 for HYP mutants; 0 < z < 1 for DEP mutants ) . Here we consider the effects of ΔV and Δr on dwell times in the stochastic switch model to enable direct comparison with the original study of HYP and DEP strains ( Zheng et al . , 1999 ) . Dwell times can be written as functions of weights: ( 1 ) dX=aXF+aXR-1=[A exp ( hℱ ) +A exp ( hℛ ) ]−1 ( 2 ) dF=aFX+aFY-1=[A exp ( −hℱ−wℱℱ ) +A exp ( hℛ+wℱℛ ) ]−1 ( 3 ) dR=aRX+aRY-1=[A exp ( −hℛ−wℛℛ ) +A exp ( hℱ+wℛℱ ) ]−1 ( 4 ) dY=aYR+aYF-1=[A exp ( −hℱ−wℱℱ−wℛℱ ) +A exp ( −hℛ−wℛℛ−wℱℛ ) ]−1 These equations show that the ΔV and Δr hypotheses make qualitatively distinct predictions . The simplest case is dwell dX , which depends only on hℱ and hℛ . Equation 1 shows that dX rises and falls as h terms are made more negative or positive , respectively . Thus , under the ΔV hypothesis , dX should rise in HYP mutants and fall in DEP mutants ( Figure 7A , row 4 ) . In contrast , under the Δr hypothesis , in which weight magnitudes ( |W| and |h| ) decrease in DEP mutants and increase in HYP mutants , dX should rise in DEP mutants and fall in HYP . To distinguish between these hypotheses , we measured dwell times in mutants and wild type animals during local search . The pattern of observed changes in dX matched the pattern predicted by the ΔV hypothesis but not the Δr hypothesis ( Figure 7C , row 4 ) . Thus , the effects of membrane potential appear to dominate the effects of changes in synaptic strength in the case of mutant values of dX . In contrast to dX , dF and dR depend on w terms as well as h terms . Under the ΔV hypothesis , the h terms but not the w terms would be affected by the mutations . Positive and negative increments in h have the effects shown in Figure 7A , rows 1 and 2; dF and dR are predicted to shift in opposite directions . Changes in dF are dominated by the effects of hℱ on the first term in Equation 2 ( which represents aFX ) because the second term in the equation ( which represents aFY ) remains close to zero in the mutants . Analogously , changes in dR are dominated by the effects of hℱ on the second term in Equation 3 ( aRY ) because the first term in the equation ( aRX ) remains close to zero in the mutants . The Δr hypothesis makes a distinctly different prediction . In this version of the model , w terms and h terms would both be affected by the mutations . Now , the predicted pattern of dwell time changes across both dF and dR is such the both dwell times shift in the same direction ( Figure 7B , rows 1 and 2 ) ; specifically , dwell times in DEP and HYP mutants move toward or away from their uncoupled dwell times , respectively . Taken together , the pattern of observed changes in dF and dR matched the pattern predicted by the Δr hypothesis ( Figure 7C , rows 1 and 2 ) but not the ΔV hypothesis . We conclude that changes in synaptic strength may dominate the effects of changes in membrane potential on mutant values of dF and dR . Neither hypothesis predicts the observed changes in dY ( Figure 7C , row 5 ) which resembled the pattern of changes in dX ( Figure 7C , row 4 ) . However , the ΔV hypothesis correctly predicts observed dwell times in the overall pause state dp ( Figure 7C row 3 ) . This is because dp is dominated by dX and changes in dX are well-predicted by the ΔV model as noted above . Overall , our analysis of the effects of HYP and DEP mutations in terms of the Stochastic Switch Model points to a role for changes in both membrane potential and input resistance in regulating dwell times . The Stochastic Switch Model immediately suggests a family of models for the regulation of the spatial scale of random search in response to the availability of food and the worm’s physiological state . The scale of random search is determined primarily by mF , the mean distance traveled during a forward run . In C . elegans , a run begins with a transition from state R ( via P ) into state F and continues until the next transition into state R . Any run may include one or more visits to state P , but FPF transitions are not usually associated with changes in heading . In terms of the Stochastic Switch Model , mF=V¯FpF/fRPF , where V¯F is the average velocity in state F , pF is the probability of being in state F , and fRPF is the frequency of RPF transitions ( Materials and methods , Equation 39 ) , which coincide with random reorientations . Importantly , under the approximation aFY≅0 ( Table 1 , column A ) , mF is can be expressed as a function of just three of the six weights in the network: ( 5 ) mF≅vFA¯·exp ( hℱ ) +exp ( hℛ ) exp ( hℛ−hℱ−wℱℱ ) We refer to these weights as potential control points in the network . In a minimal model of search scale regulation , mF could be controlled by sensory inputs represented by hℱ and hℛ ( Figure 8A ) . Search scale ( mF ) together with the frequency of reversals ( FPR transitions ) , have been used to define the three search modes commonly recognized in C . elegans: cropping , local search , and ranging . To find minimal models for regulation of search mode , we performed exhaustive searches of subregions of network’s six-dimensional weight space . Subspaces , defined by one , two , or three weights , were scanned across a wide range of values ( -6 ≤ w ≤ 6 ) while other weights remained fixed at their wild type levels ( Figure 7B–H ) . The performance of each configuration of the network was scored according to whether it matched the range of mF magnitudes and reversal frequencies characteristic of each mode ( see Materials and methods ) . Another consideration was the number of distinct search modes available; accordingly , we also noted the density with which the plane defined by reversal frequency and mF was covered in the scan ( Figure . 8B-H , gray symbols ) . All three search modes were available in the subspace defined by the control points ( hℱ , hℛ , Wℱℱ ) ( Figure 8B , Figure 8—figure supplement 1 ) . However , only cropping and local search were available in the complementary subspace ( Wℛℛ , Wℛℱ , Wℱℛ ) ( Figure 8C ) ; thus , to achieve the full set of search modes , at least one of the weights in equation 5 must be free to change . None of the control-point weights was sufficient on its own to produce all three search modes ( Figure 8D–F ) . Scanning the subspaces ( hℱ , Wℱℱ ) and ( hℛ , Wℱℛ ) showed these pairs of weights to be sufficient for all modes ( Figure 8G , H ) , but a three-dimensional subspace containing at least one of the control-point weights was a necessary condition for both dense coverage of this plane and the presence of all three search modes ( Table 7 ) . We suggest that these three-weight subspaces constitute the most likely minimal models for the regulation of search in C . elegans . They could be tested by chronic manipulation of control-point weights utilizing a variety of approaches , such as chemical or optical probes that alter tonic inputs to the command network from sensory neurons and interneurons represented by the parameters hℱ and hℛ . Mean forward run length is also modulated during biased random walks , increasing or decreasing when the animal is moving in a favorable or unfavorable direction , respectively ( Pierce-Shimomura et al . , 1999; Block et al . , 1982; Iino and Yoshida , 2009; Luo et al . , 2014 ) . When C . elegans is engaged in chemotaxis toward an attractive substance , the direction of motion relative to the gradient is represented by specialized chemosensory neurons that respond either to increases ( ON cells ) or decreases in concentration ( OFF cells ) ( Thiele et al . , 2009 ) ; moreover , interventions that activate ON cells or OFF cells promote runs and pirouettes , respectively ( Suzuki et al . , 2008 ) . Thus , in one simple model of random-walk chemotaxis , ON cells increase hℱ and decrease hℛ , whereas OFF cells do the opposite . Simulations show that this model is sufficient to generate realistic chemotaxis in a point model of search behavior in C . elegans ( Figure 8—figure supplement 2 ) when the worm is below the target concentration of attractant . Similar circuitry can explain biased random walks in response to other physical gradients ( Lockery , 2011 ) . In addition to random search , the command neurons in C . elegans are required for a variety of escape responses ( Hart et al . , 1995 ) that are deterministic in that pR closely approaches unity for strong stimuli ( Wittenburg and Baumeister , 1999; Tobin et al . , 2002; Liu et al . , 2012; Mohammadi et al . , 2013 ) . C . elegans escape responses can be produced by two pathways , one that requires the reverse command neurons ( Chalfie et al . , 1985 ) and one that does not ( Piggott et al . , 2011 ) . Three distinct circuit motifs for the functional connectivity underlying escape responses requiring reverse command neurons are conceivable ( Figure 9A ) . In the Push motif , nociceptive neurons excite reverse command neurons via hℛ thereby increasing the rate constants for transitions in which ℛ turns ON ( aXR and aFY ) , and decreasing the rate constants for transitions in which ℛ turns OFF ( aXR and aFY ) . In the limit where hℛ→ ∞ , both aXR and aFY→ ∞ , whereas aRX and aYF→ 0 ( Figure 9B ) . The system now inhabits only states R and Y , and pR=aYR/ ( aYR+aRY ) . In the Pull motif , nociceptive neurons inhibit the forward command neurons via hℱ . In the limit where hℱ→ -∞ , the system switches only between states R and X and pR=aXR/ ( aXR+aRX ) . In the third motif , in which Push and Pull are combined , R becomes an absorbing state ( pR=1 ) . Using the rate constants shown in column A of Table 1 to compute limiting values of pR in each motif , we found that the Pull and Push-Pull motifs are sufficient for deterministic escape , whereas the Push motif is not ( Figure 9B ) . Thus , inhibition of forward command neurons is required for deterministic escape , predicting that nociceptive neurons functionally inhibit these neurons . To test this prediction we examined the ASH neurons , a pair of nociceptive sensory neurons required for the majority of escape responses in C . elegans . ASH neurons have anatomically defined monosynaptic and polysynaptic connections to both the behaviorally dominant command neurons AVB and AVA ( Chalfie et al . , 1985; White et al . , 1986 ) . We have previously shown that the functional connection from ASH to AVA is excitatory ( Lindsay et al . , 2011 ) . To test whether the functional connection from ASH to AVB is inhibitory , we photoactivated ASH neurons while recording from AVB ( Figure 9C , D ) . The reversal potential of this connection was more negative than the zero current potential , indicating inhibition as predicted by the model . Thus , ASH-mediated escape may be controlled by a Push-Pull motif , further demonstrating the feasibility of using behavioral data to predict population-level synaptic connectivity . The source of the AVB inhibition could be the inhibitory connection from AVA , polysynaptic pathways from ASH to AVB , or both . Notably , the Pull and Push-Pull motifs are equally effective in driving pR to 1 . 0 ( Figure 9B ) . Nevertheless , computation of the expected latency to the first reversal event when a forward moving animal suddenly encounters a strong nociceptive stimulus indicates a 2 . 3-fold reduction in latency for the Push-Pull motif ( Figure 9B , parenthetical values ) . We conclude that the ASH mediated escape circuit in C . elegans may be specialized for short latency escape responses .
The Stochastic Switch Model is cast at a level of biological detail that is minimally sufficient to capture the stochastic dynamics of C . elegans locomotion in neuronal terms . Despite its simplicity , the model predicts the unexpected effects of neuronal ablations and genetic manipulations . It also predicts the sign and strengths of key synaptic connections , which were confirmed by combining optogenetics with electrophysiology . The model is immediately extensible to random search at a variety of spatial scales , biased random walks such as chemotaxis , and deterministic escape behaviors . The predictive success of the model indicates that random search in C . elegans can be understood in terms of a neuronal flip-flop circuit involving reciprocal inhibition between two populations of stochastic neurons . Two likely sources of stochastic state transitions are quantal synaptic transmission and ion channel gating . Both of these sources derive their randomness from thermal fluctuations at the molecular level , a phenomenon that is common to all nervous systems . The stochasticity underlying search behavior in C . elegans could be intrinsic to the command neurons , their presynaptic neurons ( Gordus et al . , 2015 ) , or both . The simplifying assumptions of the model introduce several limitations worth noting . ( i ) By representing the ten command neurons as only two functional units , the model ignores possible functional differences between individual neurons within each group . ( ii ) By design , the model predicts exponentially distributed dwell times , but Figure 2—figure supplement 3 shows that this relationship is only approximate . ( iii ) The model also has no provision to explain the strong correlation between locomotory phase and entry into state Y ( Figure 4 ) , although this could be added by modeling feedback from the pattern generator as a time-varying component of hℱ and hℛ . ( iv ) The model does not take into account temporal correlations in velocity , but instead uses only the present velocity , along with the present state , to compute transition probabilities . For example , the fact that locomotion gradually slows before the worm enters the pause state ( Figure 1G , H ) suggests that transition probabilities might be more reliably calculated by including the recent velocity history , rather than just the present velocity . ( v ) Finally , the model does not attempt to explain the observation that the number of command neurons that are present and the degree of command neuron activation has an effect on velocity and undulation frequency ( Figure 5A , Table 3 , Table 5 ) . Velocity modulation could be incorporated by relaxing the assumptions that command neurons within pools are co-active and have a single non-zero level of activation . Although the model correctly predicts several unexpected and even paradoxical observations at the behavioral and electrophysiological levels , it would be premature to conclude that the biological system functions as assumed . This caution extends to all of the model's assumptions , including the mapping relationship between pause states X and Y and their behavioral correlates . We view the pause states as theoretical constructs having an epistemological status akin to theoretical constructs in many widely accepted models , such as the gating particles that were proposed in the Hodgkin-Huxley model of the squid action potential to explain the voltage sensitivity of ion channels . An altogether different method for analyzing locomotory states in C . elegans also proposed the existence of two pause states ( Stephens et al . , 2008 ) . In that work , each pause state was associated with a particular locomotory phase . In contrast , we found that only state Y occurred in association with a particular posture ( a ventral bend in the middle of the body ) , whereas state X occurred with essentially no postural preference . The reason for this discrepancy may be that pauses are identified in different ways in the two studies . Here pauses are identified in terms of tangential velocity . In Stephens et al . ( Stephens et al . , 2008; 2011; 2010 ) , however , pauses are identified in the phase space defined by the amplitudes of first two principle components of the worm's instantaneous shape . For the two approaches to yield the same result , minima in the magnitudes of tangential and phase velocity would have to be coincident at all times . We believe this outcome is unlikely because the third and fourth principle shape components , which account for approximately 30% of the shape variance ( Stephens et al . , 2008 ) , meet the necessary and sufficient conditions for generating tangential thrust: a gradient of curvature along the worm’s centerline ( Gray , 1946; Gray , 1953; Gray and Lissmann , 1964 ) ; this is one way thrust is believed to be generated during omega turns ( Stephens et al . , 2008 ) . Thus , the worm can be moving with respect to the substrate even when phase velocity is zero . Overall , we speculate that pauses in phase velocity are a subset of pauses in tangential velocity . The extent to which this is true could be determined by performing spot tracking and shape analysis on the same individual worms . It will be interesting to test several additional predictions of Stochastic Switch Model: Like the Stochastic Switch Model , a previous model of the command neuron circuit by Rakowski et al . ( Rakowski et al . , 2013 ) predicts reciprocal inhibition between command neurons . Although the two models analyze locomotion behavior in terms of the same three behavioral states – forward , reverse , and pause – the models have essentially no points of mathematical contact . In the Rakowski model , neurons are deterministic electrical compartments and only the long-term average state probabilities of the network are computed . In the Stochastic Switch Model , by contrast , neurons are inherently stochastic and instantaneous state is computed . These disparities are significant because only the Stochastic Switch Model can predict temporal phenomena including such fundamental quantities as transition rates and mean dwell times . The fact that the both models predict reciprocal inhibition may reflect that fact that the behavioral signal of reciprocal inhibition is strong enough to transcend large differences between models . Mammalian sleep , like C . elegans locomotion , is composed of numerous abrupt alternations between opposing behavioral states . Sleep is punctuated frequently by brief periods of wakefulness , and dwell time distributions in sleep and wake states indicate that switching between them is a stochastic process ( Lo et al . , 2004 ) . Sleep and wakefulness are controlled by mutually inhibitory brain-stem nuclei , implying a reciprocal inhibition motif . In a significant parallel to the effects of command neuron ablations on dwell times in C . elegans locomotion ( Figure 5B ) , lesions of sleep-related nuclei simultaneously reduce the dwell times in both sleep and wake states , as do lesions of wakefulness nuclei ( Saper et al . , 2010 ) . Thus the relationship between synaptic uncoupling of the circuit and changes in dwell times may be a general principle of reciprocal inhibition in stochastic neuronal networks . Further study of invertebrate models of this circuit motif would be a productive means of identifying the genetic and physiological underpinnings of such circuits . The debut of the essentially complete wiring diagram of the C . elegans nervous system raised the prospect of the first account of the entire behavioral repertoire of an organism at single-neuron resolution ( White et al . , 1986; Varshney et al . , 2011 ) . To date , the repertoire of behaviors commonly recognized in C . elegans can be divided into three main functional categories , subsuming 23 different elementary actions ( Faumont et al . , 2012 ) . Because the command neurons considered here are required for almost half of this repertoire , the Stochastic Switch Model is a significant step toward a comprehensive understanding of the neuronal basis of behavior in this animal , bringing us closer to the goal of computing the behavior of an entire organism . Though abstract by design in its representation of individual neurons and synapses , the model accommodates not only random search at multiple spatial scales ( Figure 8 ) , but also biased random walks ( Figure 8—figure supplement 1 ) and deterministic escape behaviors ( Figure 9 ) . We propose , therefore , that the Stochastic Switch Model could serve as a multipurpose module for computing C . elegans behavior . Combining this mathematically tractable module with others representing sensory inputs , modulatory states , and the presumptive pattern generators for forward and reverse locomotion , could lead to essentially complete models of the C . elegans nervous system that are at once predictive and intuitively comprehensible ( Abbott , 2008 ) .
All strains were cultivated at 22 . 5°C on low-density NGM ( nematode growth medium ) agar plates seeded with the E . coli bacteria ( OP50 ) as described by Brenner ( Brenner , 1974 ) . Transgenic lines were made using standard protocols ( Mello and Fire , 1995 ) . External saline for electrophysiology ( mM ) : 5 KCl , 10 HEPES , 8 CaCl2 , 143 NaCl , 30 glucose , pH 7 . 2 ( NaOH ) ; internal saline for electrophysiology ( mM ) : 125K-gluconate , 1 CaCl2 , 18 KCl , 4 NaCl , 1 MgCl2 , 10 HEPES , 10 EGTA , pH 7 . 2 ( KOH ) . Medium for behavioral assays ( mM ) : NH4Cl 2 , CaCl2 1 , MgSO4 1 , and KPO4 25 , pH 6 . 5; M9 Buffer ( grams ) : 3 KH2PO4 , 6 Na2HPO4 , 5 NaCl , 1 ml 1 M MgSO4 , H2O to 1 liter . Prior to each assay , an individual adult hermaphrodite was picked to a bacteria-free agar transfer plate by means of a platinum-wire pick . The worm was then washed in M9 to remove excess bacteria , then transferred in a pipette filled with assay medium to a 10 cm petri plate containing 1 . 7% agarose in assay medium . A black dot approximately 40 microns in diameter was applied to the center of the body as shown in Figure 1A ( see Spotting procedure ) . The worm was allowed to recover from transfer and handling for 2 min . , then recorded for 10 min . The assay plate rested on a motorized microscope stage ( Applied Scientific Instrumentation MS-2000 , Eugene , OR USA ) fitted with position encoders ( Gurely Precision Instruments LE-1800 , Troy , NY USA ) having a resolution of 0 . 5 μm . Behavior was recorded using an analog video camera ( CCD Sony XC-ST70 , 29 . 97 frames per second ) fitted with a 12× zoom lens ( Navitar 50486D , Rochester , NY USA ) . For tracking purposes , video was analyzed in real time by custom software to calculate the eccentricity of the ink spot relative to the center of the field of view , and to compute the stage movements required to re-center the spot . Motion blur was minimized by making stage speed during corrective movements an increasing exponential function of target eccentricity such that small corrections were made more slowly than large corrections . Position encoders were read in synchrony with the video stream and this information was stored for off-line analysis . The overall trajectory of the worm was computed by combining the location of the spot in the field of view with stage position in each video frame . The direction of movement ( forward or reverse ) at the start of each recording was keyed by the observer and subsequent assignments were made automatically by computer . Each recording was spot-checked for correct assignments at four or more points during the recording . In experiments involving neuronal ablations or genetic mutations , recordings of sham operated controls or wild type worms , respectively , were interleaved with worms in each treatment group . The animal was immobilized by a stream of humidified CO2 emitted by a 1 . 5 mm diameter glass pipette positioned near the worm . The spotting ink was comprised of petroleum jelly ( 1 ml ) , mineral oil ( 1 ml ) , and black iron oxide ( 3 g ) . Ink was applied by means of 1 . 5 mm diameter glass rod that had been pulled to a fine point , fire polished to produce a bulbous tip , and dipped in the ink . The rod was positioned by means of a micromanipulator . To control for the effects of the spotting procedure , we compared the speed of locomotion of worms that had been immobilized , or immobilized and spotted , to untreated worms . There were no significant differences between these three groups . Worms were glued to an agarose coated coverslip using cyanoacrylate adhesive as previously described ( Lindsay et al . , 2011 ) . The coverslip formed the bottom of the recording chamber , which was filled with external saline . The cell body of the neuron to be recorded was exposed by making a small slit in the cuticle using a finely drawn glass rod . Recording pipettes had resistances of 10–20 MΩ when filled with internal saline . Voltage- and current-clamp recordings were made with a modified Axopatch 200A amplifier ( Lockery and Goodman , 1998 ) . In reversal potential measurements , recordings of photostimulation-evoked synaptic currents were filtered at 2 kHz and sampled at 10 kHz . Postsynaptic neurons ( AVA , AVB ) were identified using a combination of fluorescent markers and distinctive voltage clamp currents as described ( Lindsay et al . , 2011 ) . Presynaptic neurons ( AVA , AVB , and ASH ) were activated by expression of ChannelRhodopsin-2 expressed under the control of neuron-specific promoters as described ( see “Strains” ) . Worms were photostimulated in electrophysiological experiments using the blue channel ( 470 nm ) of a dual-wavelength LED module ( Rapp OptoElectronic , Wedel , Germany ) that was focused by a 63× , 1 . 4 NA oil immersion objective lens ( Zeiss , part number 440762–9904 ) . Irradiance ( 12 . 5 mW/mm2 ) was determined by measuring the power emitted from the objective using an optical power meter placed above the front lens of the objective and dividing by the area of the field of illumination at the focal plane of the preparation . Neurons were ablated using a laser as described previously ( Bargmann and Avery , 1995 ) . L1 larvae were mounted on 2 . 5% agarose pads containing 5–7 mM of the immobilizing agent NaN3 . AVA and AVB neurons were ablated in N2 animals and identified by position . AVD , AVE and PVC were ablated in animals expressing nmr-1::GFP and identified by a combination of position and GFP expression . To limit potential behavioral differences in the two strains , we outcrossed ( 4× ) the nmr-1::GFP strain to the N2 strain used for AVA and AVB ablations . All animals were remounted 1–3 hr after surgery to confirm the ablation; those with collateral damage were discarded . Sham-operated animals were treated in the same manner except that the laser was not fired . Statistical significance for the results shown in Figures 4B and 6C , and in Tables 4 and 6 were obtained using the likelihood ratio test ( see Table 1 and 4 legends ) . Otherwise , two-tailed t-tests or 2-tailed Mann-Whitney U tests were used . The worm’s position in video frame k is represented as the row vector: ( 6 ) R ( tk ) =xtk , ytk k=1 , 2 , … , Nwhere X ( tk ) and Y ( tk ) are the coordinates of centroid of the tracking spot in the frame of reference of the agar plate , tk=kΔt , ∆t=33 ms , and N≅18000 is the number of video frames analyzed in a continuous recording of one worm . We made the following definitions: To analyze locomotory states we converted the velocity vector , V ( t ) , into a signed scalar quantity v ( t ) that represents the component of velocity in the direction of the worm’s track , with positive values indicating forward movement . We first smoothed x ( t ) and y ( t ) using an 11 frame window , assigned a direction to the smoothed track with respect to the head/tail orientation of the worm , and projected V ( t ) onto the smoothed track to obtain v ( t ) . For each cohort of worms we collected all v ( t ) values into a single velocity distribution g ( v ) . The central peak of g ( v ) was fit by a Cauchy distribution with median 0 and half-width b = 18 µm/s ( Figure 2—figure supplement 2 ) , which we used to approximate the pause velocity distribution for states X and Y for all worms: ( 17 ) gX ( v ) =gY ( v ) =gP ( v ) =bπ ( b2+v2 ) We used a Cauchy distribution because it has long tails that describe the pause velocity distribution better than a Gaussian distribution ( i . e . , the worm does not stop instantaneously when it switches from forward or reverse locomotion into one of the pause states ) . We estimated the forward and reverse velocity distributions gF ( v ) and gR ( v ) by scaling gP ( v ) to fit the peak at v=0 , subtracting it from the overall distribution and splitting the remaining distribution into gF ( v ) for v>0 and gR ( v ) for v<0 . Velocity distributions were scaled to be probability densities ( area =1 ) and collected into a row vector: ( 18 ) G ( v ) = [gF ( v ) , gR ( v ) , gX ( v ) , gY ( v ) ]where gi ( v ) is the estimated probability density that worms move at velocity v when in state i . The goal of the maximum likelihood fitting procedure is to find the set of state transition rates {aXF , aFX , aXR , aRX , aFY , aYF , aRY , aYR}that maximize the probability of the observed velocity time series v ( t ) given the velocity distribution G ( v ) . All transition rates were constrained to be ≥0 , and usually were additionally constrained to correspond to valid synaptic weights as described below . The likelihood is most conveniently calculated using matrix notation as follows; see Colquhoun and Hawkes , 1995 for a more complete explanation of these computations . Let: ( 19 ) Q=− ( aFX+aFY ) 0aXFaYF0 − ( aRX+aRY ) aXRaYRaFXaRX − ( aXF+aXR ) 0aFYaRY0 − ( aYF+aYR ) Element qij ( i≠j ) of matrix Q is the transition rate from state i to state j ( i . e . , the instantaneous probability per unit time that the system in state i will make a transition to state j , and element qii is the negative of the total transition rate out of state i , which is related to the mean dwell time in state i by: ( 20 ) di=−1/qii Matrix Q is composed of instantaneous transition rates , which can be converted into the matrix of transition probabilities during a brief time interval of duration ε by multiplying Q by ε and adding 1 to each diagonal element ( i . e . , by calculating ε·Q+I , where I is the 4×4 identity matrix ) . If ε is sufficiently small that multiple state transitions can be ignored , then element ij of matrix ε·Q+I is the probability that the system is in state j at the end of a time interval of duration ε given that it was in state i at the beginning of the interval . For longer time intervals during which multiple state transitions may occur , transition probabilities can be calculated by repeatedly multiplying matrix ε·Q+I by itself . Thus , if ( 21 ) M= ( ε·Q+I ) Kthen M is the matrix of transition probabilities during a time interval of duration Kε . If K and ε are chosen such that ∆t=Kε , then element ij of matrix M is the transition probability from state i to state j during one video frame of duration ∆t . We chose K=230 and let ε=∆t/K=30 . 7 picoseconds , a time interval during which multiple state transitions can safely be ignored . Since K was chosen to be a power of 2 , M could be rapidly and accurately calculated by 30 serial multiplications using 64-bit floating point arithmetic . Let P ( t ) be the row vector of history-dependent state probabilities: ( 22 ) P ( t ) = [ pF ( t ) , pR ( t ) , pX ( t ) , pY ( t ) ]where pi ( t ) is the probability of being in state i at time t given v ( u ) for all u up to and including the present time ( u≤t ) . The matrix product P ( t ) ·M is the state probability vector at time t+∆t prior to accounting for the observed velocity at time t+∆t . To account for v ( t+∆t ) we used the information contained in G ( v ( t+∆t ) ) and applied Bayes theorem: ( 23 ) P ( t+∆t ) =l·P ( t ) ·M·diagG ( v ( t+∆t ) ) where diagG ( v ( t+∆t ) ) is the 4×4 matrix with the elements of G ( v ( t+∆t ) ) along the diagonal , and l is the scalar multiplicative factor required for the sum of the four elements of P ( t+∆t ) to equal 1 ( i . e . , P ( t+∆t ) is a vector of probabilities ) . Initially ( t=0 ) we set P ( 0 ) equal to the steady-state probability vector P∞ , which is given by: ( 24 ) P∞·Q=0 ⇒ P∞=U4· ( Qa·Qat ) −1where U4 is the 1×4 row vector of ones and Qa is the 4×5 matrix constructed by appending a column of ones to Q . To break the symmetry between the behaviorally indistinguishable states X and Y , we identified X as the state with higher steady-state probability . We then calculated the log-likelihood , summed over all worms in the cohort: ( 25 ) lnL=∑t , w ln ( Pw ( t ) ·Gt ( vw ( t ) ) ) where VW ( t ) is the velocity and PW ( t ) is the history-dependent state probability vector of worm w at time t . We used a random optimization algorithm to find the set of transition rates that maximized lnL . Initial guesses for 6 of the 8 rates were chosen independently from log uniform distribution between 0 . 01 Hz and 10 Hz . The remaining 2 rates were calculated to satisfy the constraints needed to generate valid synaptic weights ( see below ) . At each iteration , each of the 6 independently chosen rates was altered by adding a random number chosen from a Cauchy distribution with median 0 and width brandom ( initially brandom= 0 . 01 Hz ) , and the remaining 2 rates were recalculated . To avoid getting trapped in local likelihood maxima , the new rates were rejected and another set was calculated if any of the new rates were <0 . 01 Hz . If the new rates generated an increased likelihood , the new rates were accepted and brandom was increased by 3% . Otherwise the old rates were retained and brandom was decreased by 0 . 5% . The procedure was iterated until brandom< 0 . 001 Hz . The random optimization procedure was replicated 10 times for each cohort using different randomly chosen initial guesses . In 71% of the replicates the procedure converged on a set of transition rates in which none of the transition rates differed from the best set by more than 5% . The best set of transition rates was then refined by applying the optimization procedure using a success criterion of brandom< 10-5 and constraining transition rates to be ≥ 10–4 Hz . The likelihood calculations described above use only past and present velocity observations to calculate P ( t ) , but once the optimal transition rates were determined , the Forward-Backward algorithm ( Rabiner and Juang , 1986 ) can be used to yield a better estimate of the state probabilities based on past , present and future velocity observations , and the Viterbi Algorithm can be used to find the sequence of states with the highest probability of producing the observed velocities ( Figure 2E ) . We expressed the effect of synaptic inputs to command units ℱ and ℛ by equations of the form: ( 26 ) aON=A·eS ( 27 ) aOFF=A·e-Swhere aOFF is the transition rate from ON to OFF , aON is the transition rate from OFF to ON , and S is the total synaptic input to the unit . We do not attach any mechanistic significance to these equations , but note that they are analogous to the Arrhenius Equation ( Stiller , 1989 ) an approximation commonly used to describe the rates of chemical reactions in terms of an activation energy , E: ( 28 ) a=A·e- EkBTwhere a is the reaction rate constant , A is an empirically determined constant , KB is the Boltzmann constant , and T is the absolute temperature . Under this interpretation , S is analogous to activation energy expressed in units of KBT . Thus , ℱ and ℛ are assumed to be symmetrical bi-stable units that change state at rate A when S=0 . Deviations from this baseline condition are modelled as external synaptic inputs hℱ and hℛ . We represented the total synaptic input onto units ℱ and ℛ , respectively , by: ( 29 ) Sℱ=hℱ+bℱtwℱℱ+bℛtwℛℱ ( 30 ) Sℛ=hℛ+bℛtwℛℛ+bℱtwℱℛwhere bℱ ( t ) and bℛ ( t ) are the states of ℱ and ℛ ( 1 = ON , 0 = OFF ) , Wℛℱ and Wℱℛ are the synaptic weights from ℛ onto ℱ and from ℱ onto ℛ , respectively , and Wℱℱ and Wℛℛ represent synaptic interactions among command neurons of the same class , plus any intrinsic membrane properties that may promote bistability . Applying these definitions to the rate constants in Figure 2C gives: ( 31 ) aXF=A exp ( hℱ ) aXR=A exp ( hℛ ) ( 32 ) aFX=A exp ( −hℱ−wℱℱ ) aRX=A exp ( −hℛ−wℛℛ ) ( 33 ) aRY=A exp ( hℱ+wℛℱ ) aFY=A exp ( hℛ+wℱℛ ) ( 34 ) aYR=A exp ( −hℱ−wℱℱ−wℛℱ ) aYF=A exp ( −hℛ−wℛℛ−wℱℛ ) In these experiments the sensory environment was kept constant ( e . g . , no chemical or temperature gradients ) . Therefore hℱ and hℛ were assumed to be constant . For simulations of chemotaxis hℱ and hℛ varied with position in the chemical gradient . Equations 31–34 express the 8 transition rates in terms of 6 parameters and yield the following two constraints on the transition rates: ( 35 ) aFX aXF = aRY aYR aFY aYF=aRX aXR The inverse relations between transition rates and synaptic parameters are: ( 36 ) hℱ=ln ( aXF ) −lnA hℛ =ln ( aXR ) −lnA ( 37 ) wℛℱ=ln ( aRY/aXF ) wℱℛ=ln ( aFY/aXR ) ( 38 ) wℱℱ=−ln ( aXF aFX ) +2·lnA wℛℛ=−ln ( aXR aRX ) +2·lnA The time series of the worm’s locomotory states can be divided into forward runs , during which the worm is in either the F or P state , and reverse runs , during which the worm is in either the R or P state . Forward runs always begin with an RPF transition and end with the next FPR transition , which marks the beginning of a reverse run . Thus forward runs and reverse runs occur in strict alternation , such that the number of forward runs equals the number of reverse runs . Let mF denote the mean distance traveled during a single forward run , assuming that forward runs are straight . The value of mF is most easily calculated by dividing time into non-overlapping epochs , each of which begins with an RPF transition and ends immediately before the next RPF transition . Each epoch thus contains exactly one forward run , which includes all visits to state F during the epoch . Therefore , mF is also equal to the mean distance travelled while in the forward state during a single epoch: ( 39 ) mF=vF¯pFfRPFwhere V¯F is the mean velocity in the forward state and fRPF is the frequency of RPF transitions . Since FPR and RPF transitions occur in strict alternation they must occur in equal numbers: fRPF=fFPR . Thus , eq . 39 can also be written with fFPR in the denominator , which is more useful for the calculation that follows , although the form shown above is more directly interpreted in terms of the frequency of random reorientations , which occur at the RPF transitions . It is straightforward to calculate fFPR given pF , aFX , aFY , and the probabilities that the transitions out of states X and Y will be into state R: ( 40 ) prob ( X→R ) =aXR/ ( aXF+aXR ) ( 41 ) prob ( Y→R ) =aYR/ ( aYF+aYR ) ( 42 ) fFPR=pF ( aFXaXRaXF+aXR+aFYaYRaYF+aYR ) Combining eqns . 39 and 42 yields: ( 43 ) mF=vF¯ ( ( aXF+aXR ) ( aYF+aYR ) aFXaXR ( aYF+aYR ) +aFYaYR ( aXF+aXR ) ) An approximation to mF in terms of synaptic weights is obtained by noting that transitions from F to Y were extremely rare ( aFY= 0 . 007 s-1; Table 1 ) . Setting aFY≅0 yields: ( 44 ) mF≅vF___ ( aXF+aXRaFXaXR ) =vF___A ( exp ( hℱ ) +exp ( hℛ ) exp ( hℛ−hℱ−wℱℱ ) ) In Figure 8—figure supplement 1 and 2 , the worm was represented as a point that moved forward or backward at speeds of 0 . 2 and 0 . 3 mm/sec , respectively , and was stationary during the pause state . Rate constants were calculated according to equations 31–34 based on the weights that pertain under random search or chemotaxis , using either Amin= or Amax= . Weights were used to compute the state transition matrix M . At each time step ( ∆t = 33 ms ) , the next state was selected randomly according to the state probabilities given by M . When an RPF transition occurred , a new direction of movement ( heading ) was selected from a uniform distribution . The random component of the heading was modeled as Gaussian noise having a standard deviation of 0 . 001 degrees . In the case of chemotaxis simulations , the values of hℱ and hℛ were updated at every time step according to the direction in which the worm was heading , leading to an updated set of weights and a new M matrix . To date , these behaviors have been defined mainly in operational terms . Following the terminology of Jander ( 1975 ) : ( i ) cropping is the locomotory behavior exhibited by well-fed worms on plates with densely populated patches of bacteria; ( ii ) local search ( also “area restricted search” ( Hills et al . , 2004 ) or “pivoting” ( Wakabayashi et al . , 2004 ) is exhibited by well-fed worms within about 10 min after being transferred to a foodless plate; and ( iii ) ranging ( “dispersal” ( Gray et al . , 2005 ) or “traveling” ( Wakabayashi et al . , 2004 ) is exhibited by well-fed worms tens of minutes after being transferred to a foodless plate . Each mode can be associated with approximate ranges of three parameters: mean forward run length ( mF ) , mean frequency of reversals ( fFPR ) , and mean reverse run length ( mR ) . Local search serves as a useful reference point . During cropping , mF is greatly reduced , fFPR is greatly increased , and mR is also reduced , being limited to “short reversals” ( the distance traveled in one or two head sweeps , or about 0 . 5 mm ( Gray et al . , 2005 ) ; during local search , reverse runs are almost always “long” ( the distance traveled in at least three head sweeps ) . During ranging , mF is greatly increased , fRPF is reduced , and reversals are long . Cutoff values for search modes , inferred from behavioral data ( Wakabayashi et al . , 2004; Gray et al . , 2005; Fujiwara et al . , 2002; Hills et al . , 2004 ) were: Dwelling – short forward run length ( mF < 0 . 5 mm ) , high reversal frequency ( fFPR > 6 . 0/min ) , short reversals ( mR < 0 . 5 mm ) ; Local search – moderate forward run length ( 0 . 5 mm ≤ mF < 5 . 0 mm ) , moderate reversal frequency ( 2 . 0/min ≤ fFPR < 6 . 0/min ) , non-short reversals ( mR ≥ 0 . 5 mm ) ; Ranging – long forward run length ( mF ≥ 5 . 0 mm ) , low reversal frequency ( fFPR< 2/min ) , non-short reversals ( mR ≥ 0 . 5 mm ) . All data and the analysis program are publicly available at doi:10 . 5061/dryad . 35qv6 . | An animal’s ability to rapidly and efficiently locate new sources of food in its environment can mean the difference between life and death . As a result , animals have evolved foraging strategies that are adapted to the distribution and detectability of food sources . Organisms ranging from bacteria to humans use one such strategy , called random search , to locate food that cannot be detected at a distance and that is randomly distributed in their surroundings . The biological mechanisms that underpin random search are relatively well understood in single-cell organisms such as bacteria , but this information tells us little about the mechanisms that are used by animals , which use their nervous system to control their foraging behavior . Roberts et al . have now investigated the biological basis for random search behavior in a tiny roundworm called Caenorhabditis elegans . This worm forages for pockets of bacteria in decaying plant matter and has a simple and well-understood nervous system . Roberts et al . used information on how the cells in this worm’s nervous system connect together into so-called “neural circuits” to generate a mathematical model of random searching . The model revealed that the worm’s neural circuitry for random searching can be understood in terms of two groups of neuron-like components that switch randomly between “ON” and “OFF” states . While one group promotes forward movement , the other promotes backward movement , which is associated with a change in search direction . These two groups inhibit each other so that only one group usually is active at a given time . By adjusting this model to reproduce the behavioral records of real worms searching for food , Roberts et al . could predict the key neuronal connections involved . These predictions were then confirmed by taking electrical recordings from neurons . The model could also account for the unexpected behavioral effects that are seen when a neuron in one of these groups was destroyed or altered by a genetic mutation . These findings thus reveal a biological mechanism for random search behavior in worms that might operate in other animals as well . The findings might also provide future insight into the neural circuits involved in sleep and wakefulness in mammals , which is organized in a similar way . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"neuroscience"
] | 2016 | A stochastic neuronal model predicts random search behaviors at multiple spatial scales in C. elegans |
The flight muscles , dorsal air sacs , wing blades , and thoracic cuticle of the Drosophila adult function in concert , and their progenitor cells develop together in the wing imaginal disc . The wing disc orchestrates dorsal air sac development by producing decapentaplegic and fibroblast growth factor that travel via specific cytonemes in order to signal to the air sac primordium ( ASP ) . Here , we report that cytonemes also link flight muscle progenitors ( myoblasts ) to disc cells and to the ASP , enabling myoblasts to relay signaling between the disc and the ASP . Frizzled ( Fz ) -containing myoblast cytonemes take up Wingless ( Wg ) from the disc , and Delta ( Dl ) -containing myoblast cytonemes contribute to Notch activation in the ASP . Wg signaling negatively regulates Dl expression in the myoblasts . These results reveal an essential role for cytonemes in Wg and Notch signaling and for a signal relay system in the myoblasts .
Flight muscles of the Drosophila adult drive the coordinated movements of the wings and thoracic cuticle to power flight , and many thin tubes ( tracheoles ) that emanate from the thoracic dorsal air sacs penetrate the muscles to oxygenate them . Thus , the functions of the muscles , wings , thoracic cuticle , and trachea are linked , and the physical associations are intimate . The progenitor cells that produce these tissues develop together in the wing imaginal disc . Previous studies from this lab showed that the air sac primordium ( ASP ) , which is the progenitor of the dorsal air sacs , depends on Branchless/FGF ( FGF ) and Dpp signaling proteins that the wing disc produces ( Sato and Kornberg , 2002; Roy et al . , 2014 ) . Here , we describe two other signaling systems that coordinate the progenitors of the flight muscles with the wing disc and trachea . The wing disc can be described as a flattened sac that juxtaposes the apical surfaces of two connected epithelial sheets across a narrow lumen . One of the sheets , called the columnar epithelium because its cells are highly elongated along their apical/basal axis , generates the wing blade and most of the notum , the dorsal cuticle of the thorax . The wing disc is encapsulated by a basement membrane , but a branch of the tracheal system ( the transverse connective ) penetrates the basement membrane at several sites in the dorsal region of the disc ( Guha et al . , 2009 ) . Transverse connective that is within the basement membrane lies adjacent to the basal surface of the columnar epithelium , and during the third instar ( L3 ) , this segment of the transverse connective sprouts a tubular outgrowth—the ASP—in response to FGF expressed by a group of nearby columnar epithelial cells ( Sato and Kornberg , 2002 ) . Myoblasts that are the progenitors of the flight muscles are also at the basal surface of the columnar epithelium , underneath the basement membrane , and in the vicinity of the tracheal branches . They proliferate during L3 to extend over most of the dorsal part of the disc where the cells that will produce the notum cuticle grow ( Sudarsan et al . , 2001; Gunage et al . , 2014 ) . Signaling proteins that contribute to the growth and diversification of the cells of the wing disc have been extensively characterized . Three that are relevant to the ASP and myoblasts are Notch , Dpp , and Wg ( Couso et al . , 1995; Ng et al . , 1996; Brennan et al . , 1999; Steneberg et al . , 1999; Sudarsan et al . , 2001; Baena-Lopez et al . , 2003; Giraldez and Cohen , 2003; Marois et al . , 2006; Herranz et al . , 2008; Gunage et al . , 2014 ) . Notch signaling has essential roles at both the dorsal/ventral and anterior/posterior compartment borders of the disc , and although it has been shown to specify fusion cell fate and branch identity during formation of tracheal system in the embryo , a role in larval trachea has not been reported . Studies in several other contexts indicate that Notch signaling may be mediated by cytonemes that make direct contacts between signaling cells ( Renaud and Simpson , 2001; Cohen et al . , 2010 ) . Dpp-expressing cells line the anterior side of the anterior/posterior compartment border at all stages of L3 discs , and Dpp that is produced near the ASP activates Dpp signal transduction in the ASP that is necessary for its morphogenesis . ASP cells express the Dpp receptor but do not express Dpp . The mechanism by which Dpp signals from disc cells to the ASP involves exchange of Dpp between producing and receiving cells at synapses that form where cytonemes link ASP cells to Dpp-producing disc cells ( Roy et al . , 2014 ) . ASP cytonemes that contain the Dpp receptor have been observed extending as far as 40 μm , crossing over approximately 15–20 disc cells to reach sources of Dpp . These cytonemes transport Dpp from producing cells to the ASP , and signal transduction is dependent on the contacts they make with the disc cells . Comparably long ASP cytonemes containing the FGF receptor have been observed reaching FGF-expressing disc cells , and in the wing disc , Hh dispersion is effected by a similar mechanism ( Callejo et al . , 2011; Bischoff et al . , 2013 ) . In these contexts , the evidence that Dpp , FGF , and Hh paracrine signaling are mediated by cytonemes is strong . Expression patterns of Wg change throughout the L3 . In the wing blade primordium , Wg is expressed broadly in early L3 discs , but in late L3 discs , it is expressed in well-delineated bands both at the dorsal/ventral compartment border and around the periphery ( Phillips and Whittle , 1993; Couso et al . , 1995; Strigini and Cohen , 2000; Alexandre et al . , 2014 ) . Wg-expressing cells at the dorsal/ventral border may function as a signaling center for the growth and diversification of the wing cells . Wg is also expressed in the notum primordium where it has a role in myoblast proliferation and diversification ( Sudarsan et al . , 2001; Gunage et al . , 2014 ) . The mechanism by which Wg signals from the disc cells to myoblasts has not been studied . In other contexts , it has been assumed that Wg is secreted and released by producing cells , and that it reaches target cells by passive diffusion , and models that describe its movement posit that its path may be restricted by structures it encounters , but that its hypothetical journey is similar to a ‘drunken sailor’ ( Muller et al . , 2013 ) . In wing discs , a mutant form of Wg that is tethered to the plasma membrane by a heterologous transmembrane domain is active ( Alexandre et al . , 2014 ) , but it is not known how signaling by this tethered form is related to the normal processes that present wild type Wg to receiving cells . There is no experimental evidence that shows directly that Wg disperses by passive diffusion or that it signals in vivo as a soluble , free protein . The study reported here investigated signaling between disc cells , myoblasts , and the ASP . The Wg-Fz and Notch-Dl signaling systems were identified as important to ASP development , and two types of myoblast cytonemes were characterized that mediate exchange and transport of Wg and signaling by Dl .
To investigate the relative proximity of the myoblasts and trachea as they grow and develop during L3 , we simultaneously expressed membrane-tethered mCherry in the trachea ( with btl-LHG lexO-mCherry:CAAX , a tracheal-specific driver [Shiga et al . , 1996] ) and membrane-tethered GFP in myoblasts ( with 1151-Gal4 UAS-CD8:GFP; a myoblast-specific driver [Roy and VijayRaghavan , 1997 , 1998] ) , and monitored both fluorescent proteins in disc preparations . During L3 , the myoblasts increase in number from approximately 250 to 2500 ( Gunage et al . , 2014 ) , and the ASP buds from the transverse connective to grow posteriorly across the anterior compartment . Images of the mid-dorsal region of early , mid , and late L3 discs revealed the expansion of the myoblasts and growth of the ASP as well as the close proximity of the trachea and myoblasts ( Figure 1A–C ) . These figures also show that the distal ASP extended beyond the myoblast domain; cytonemes that emanate from the distal tip and that take up FGF from wing disc ( Sato and Kornberg , 2002; Roy et al . , 2011 , 2014 ) were also visible ( arrows ) . The close proximity of the ASP and myoblasts is also apparent in a sagittal view ( Figure 1D ) and in an electron micrograph image ( Figure 1E ) . The drawing in Figure 1F depicts the ASP , myoblasts , and disc cells at the late L3 . 10 . 7554/eLife . 06114 . 003Figure 1 . The close proximity of wing disc myoblasts and the ASP . ( A–C ) Confocal images show the trachea and myoblasts at early ( A ) , mid ( B ) , and late ( C ) L3 stages . ( Genotype: 1151-Gal4/+;btl-LHG/+;UAS-CD8GFP/lexO-mCherry-CAAX ) . Scale bar: 50 μm . ( D ) Sagittal cross-section shows myoblasts adjacent to the proximal portion of the ASP . Genotype as in ( A–C ) . Scale bar: 50 μm . ( E ) Electron microscopic image shows a sagittal view of the wing disc columnar epithelium , associated myoblasts ( white arrows ) , and the ASP . Scale bar: 5 μm . ( F ) Cartoon showing a cross-section of the ASP ( red ) and wing disc epithelium ( blue ) . In the left drawing , dotted lines represent the approximate positions of upper and lower layers of ASP and the vertical dashed line corresponds to the location of the transverse optical section on the right and in Figure 2B . Vertical dashed line in transverse section corresponds to plane imaged in Figure 2B′ . DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 003 To investigate whether the ASP , myoblasts , and disc cells contact each other directly , we applied the GFP reconstitution across synaptic partners ( GRASP ) technique ( Feinberg et al . , 2008 ) . This method monitors the interaction of two non-fluorescent fragments of GFP ( GFP1–10 and GFP11 ) that are expressed separately as external , membrane-tethered proteins on different cells; reconstitution of GFP by fragments present on cells that juxtapose at <100 nm generates green fluorescence . We co-expressed mCherry-CAAX ( to label ASP membranes ) and CD4:GFP11 in the ASP , and expressed CD4:GFP1–10 in myoblasts . As shown in panels A , B , and B′ in Figure 2 , GFP fluorescence was present at the periphery of the ASP . Most of the GRASP fluorescence was at proximal and medial regions of the ASP , and little was near the distal tip , consistent with the observed distribution of myoblasts ( Figure 1D–F ) . ASP cytonemes marked with mCherry were visible that extended as much as 25 μm ( approximately 5 myoblast cell diameters ) from the ASP periphery . Whereas most of the GFP fluorescence on the dorsal side ( downward orientation in Figure 2A ) was associated with these cytonemes , most of the ventral fluorescence was close to the ventral edge of the ASP . 10 . 7554/eLife . 06114 . 004Figure 2 . Wing disc myoblasts contact the lower layer of the ASP . ( A ) Medial optical section , ( B ) sagittal plane and ( B′ ) transverse plane of the ASP ( red , Cherry fluorescence ) with green fluorescence indicating reconstituted GFP at regions of contact between myoblasts that expressed GFP1–10 and the ASP that expressed GFP11 . Cytonemes ( arrows in [A] ) extend from the ASP; green fluorescence marks cytonemes that contact myoblasts . ( Genotype: 1151-Gal4/+;btl-LHG , lexO-mCherry-CAAX/lexO-CD4-GFP11;UAS-CD4-GFP1–10/+ ) . ( C ) Medial optical section and ( D ) sagittal plane of a preparation similar to ( A and B ) , but with the myoblasts marked with Cherry fluorescence . Myoblast cytonemes extend to the ASP; green fluorescence marks cytonemes that contact the ASP . ( Genotype: 1151-Gal4/+;UAS-CD8-mCherry/btl-LHG;UAS-CD4-GFP1–10 lexO-CD4-GFP11/+ ) . ( E–H′ ) Upper ( left panels ) and lower ( right panels ) optical sections ( see Figure 1F ) show GFP reconstitution across synaptic partners ( GRASP ) fluorescence ( E and E′ ) , staining with α-Hnt antibody ( F–G′ ) , and staining with α-ß-galactosidase antibody ( H and H′ ) . GRASP fluorescence , Hnt levels and Su ( H ) lacZ levels were higher in the lower than in the upper layer of the ASP . Scale bars: 25 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 004 A ‘reciprocal’ experiment imaged ASP-myoblast GRASP in the context of myoblasts that were marked with mCherry , and this genotype generated similar patterns of GFP fluorescence . Both GRASP contexts illuminated contacts between the ASP and myoblasts that can be seen in frontal optical sections ( Figure 2A , C ) and in sagittal projections of optical sections at the middle of the ASP ( Figure 2B , D; see Figure 1F ) . Transverse projections of optical sections at the middle of the ASP ( Figure 2B′; see Figure 1F ) show that the ASP lays in a depression of the disc's basal surface and that ASP cells at the ‘lowest level’ ( ‘6 o'clock’ in transverse sections as in Figures 1F , 2B′ ) contacted myoblasts . The transverse projection ( Figure 2B′ ) also reveals contacts between myoblasts and cells at lateral positions around the ASP circumference . GRASP fluorescence in frontal sections ( Figure 2E , E′ ) at the lower and upper layers of the ASP ( as defined in Figure 1F ) confirms this pattern of contacts . In addition , mCherry fluorescence marked myoblast cytonemes , revealing that cytonemes that link myoblasts to the ASP extended from both types of cells ( Figure 2A , C ) . Previous studies showed that Dpp signal transduction is elevated in cells in the ‘lower layer’ of the ASP and is reduced in the ‘upper layer’ , a pattern that correlates with GRASP-marked contacts between ASP cytonemes and Dpp-expressing disc cells ( Roy et al . , 2014 ) . In order to identify the signaling system that correlates with the contacts between myoblasts and cells in the ‘lower layer’ of the ASP , we examined the expression patterns of several candidate genes . We did not detect expression of Senseless , Vestigial , Cut , or Distal-less in the ASP , and although we detected expression of Escargot , Trachealess , and Pointed , there was no apparent difference in expression levels between the ‘upper’ and ‘lower layers’ . In contrast , we found that both the intensity of staining with α-Hindsight ( Hnt ) antibody and the expression of Su ( H ) GBE-lacZ ( Su ( H ) lacZ ) were significantly greater in the ‘lower layer’ ( Figure 2F–H′ ) . Hnt and Su ( H ) lacZ are targets of Notch signaling ( Furriols and Bray , 2001; Bray , 2006; Sun and Deng , 2007 ) , suggesting that Notch signaling may be activated in the ASP and may be associated with contacts between myoblasts and the ASP . To determine if Notch signaling has a role in tracheal development , we altered Notch signaling in several ways , including expressing dominant-negative Notch ( NotchDN ) , constitutive active Notch ( NotchCA ) , and Dl-RNAi . In comparison with controls ( Figure 3A , D ) , the ASP was reduced in size by tracheal expression of NotchDN ( Figure 3B ) and was misshapen and expanded following tracheal expression of NotchCA ( Figure 3C ) . These effects suggest that Notch activity is necessary for ASP development . To determine if Notch signaling in the ASP involves the disc-associated myoblasts , we over-expressed the pro-apoptotic cell death gene reaper ( rpr ) in myoblasts to reduce their number ( Figure 3M , N ) . ASP growth was impaired by the rpr over-expression ( Figure 3E ) . We then investigated a role for Delta ( Dl ) because it is a Notch ligand that is highly expressed in myoblasts ( Gildor et al . , 2012 ) . To test whether the myoblasts produce Delta that signals to activate Notch in the trachea , Delta was depleted by expression of Dl-RNAi in myoblasts; abnormally small and misshapen ASPs were observed ( Figure 3F ) . In contrast , expression of Dl-RNAi in the disc had no apparent effect on the ASP ( Figure 3O ) . 10 . 7554/eLife . 06114 . 005Figure 3 . Notch signaling in the ASP depends upon myoblast-produced Delta . ( A–F ) Confocal images show that compared to controls ( A and D ) , the ASP morphology was abnormal under conditions that either knockdown ( B ) or stimulate ( C ) Notch signaling , or that perturb myoblasts by expression of Rpr ( E ) or DlRNAi ( F ) . Animals were reared at 18°C for 4 days ( to L2 stage ) and shifted to 29°C for >48 hr ( to late L3 stage ) . ( Genotypes: ( A ) btl-Gal4 UAS-CD8:GFP/+;tub-Gal80ts/+; ( B ) btl-Gal4 , UAS-CD8:GFP/UAS-NotchDN;tub-Gal80ts/+; ( C ) btl-Gal4 , UAS-CD8:GFP/UAS-NotchCA;tub-Gal80ts/+; ( D ) 1151-Gal4/+;btl-LHG lexO-Cherry:CAAX/+; ( E ) 1151-Gal4/+;btl-LHG lexO-Cherry:CAAX/UAS-rpr; and ( F ) 1151-Gal4/+;btl-LHG lexO-Cherry:CAAX/+;UAS-DlRNAi/+ ) . ( G–I ) Compared to controls ( G and J ) , levels of Hnt and Su ( H ) lacZ expression were lower under conditions that reduce Notch activity in ASP cells ( H and K ) or reduce Dl expression in myoblasts ( I and L ) . Staining was with α-Hnt ( G–I ) and α-ß-galactosidase ( J–L ) antibodies and Alexa Fluor 555 secondary antibodies . ASPs are outlined by dotted lines . ( M and N ) The number of myoblasts ( marked with CD8:GFP ) and intensity of GFP fluorescence were reduced by expression of Rpr ( N ) relative to control ( M ) . ( O ) Expression of DlRNAi in the wing disc had no apparent effect on ASP development . ( Genotypes: ( G ) 1151-Gal4/+; ( H ) btl-Gal4/UAS-NotchDN;tub-Gal80ts/+; ( I ) 1151-Gal4/+;UAS-DlRNAi/+; ( J ) Su ( H ) lacZ/+;btl-Gal4/+ , ( K ) Su ( H ) lacZ/+;UAS-NotchDN/+;btl-Gal4/tub-Gal80ts; ( L ) Su ( H ) lacZ/1151-Gal4;UAS-DlRNAi/+; ( M ) 1151-Gal4/+;UAS-CD8:GFP/+; ( N ) 1151-Gal4/+;UAS-rpr/+;UAS-CD8:GFP/+; ( O ) ap-Gal4/+;btl-LHG lexO-CD8:GFP/DlRNAi ) . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 005 Notch signaling in the ASP was monitored by assaying the Notch reporters Hnt and Su ( H ) lacZ . Compared to control ASPs , levels of Hnt and Su ( H ) lacZ in the lower layer of the ASP ( where Notch is activated [Figure 2F–H′] ) were reduced by expression of NotchDN in trachea and by expression of Dl-RNAi in the myoblasts ( Figure 3G–L ) . These data suggest that Dl expressed in myoblasts activates Notch signaling in the ASP , and that this signaling system is necessary for ASP development . We investigated whether the myoblast cytonemes that contact the ASP ( Figure 2C ) may mediate signaling by myoblast-produced Dl . To track myoblast-produced Dl , we expressed Dl:RFP together with a fluorescent membrane protein marker ( CD4:IFP2 . 0-HO1 ) in the myoblasts ( Yu et al . , 2014 ) , and in the same animals expressed membrane-tethered GFP in the ASP . In the plane of focus captured in the images in Figure 4 , cytonemes with GFP fluorescence are visible at the surface of the ASP , and cytonemes with IFP fluorescence are visible that emanated from the myoblasts . Many Dl:RFP-containing puncta are visible in the myoblasts , and many of these fluorescent puncta colocalized with and moved along IFP-marked cytonemes . The motile puncta indicated in Figure 5A–C and visible in Video 1 moved at 0 . 33 μm/s , a speed that is consistent with rates of myosin motors . 10 . 7554/eLife . 06114 . 006Figure 4 . Delta localizes to myoblast cytonemes . Confocal image of a late L3 ASP marked with CD2:GFP ( btl-LHG lexO-CD2:GFP ) and myoblasts marked with CD4:IFP ( 1151-Gal4 UAS-CD4:IFP2 . 0-HO1 ) and that express Dl:RFP ( 1151-Gal4 UAS-Dl:RFP ) . Dl:RFP puncta in IFP-marked myoblast cytonemes and CD2:GFP-containing ASP cytonemes are visible extending across the basal surface of the lower layer of the ASP . Scale bar: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 00610 . 7554/eLife . 06114 . 007Figure 5 . Delta is motile in myoblast cytonemes and activates Notch signaling in the ASP . ( A–C ) Confocal images of a late L3 ASP marked with CD2:GFP ( btl-LHG lexO-CD2:GFP ) and myoblasts marked with CD4:IFP ( 1151-Gal4 UAS-CD4:IFP2 . 0-HO1 ) and that express Dl:RFP ( 1151-Gal4 UAS-Dl:RFP ) . Dl:RFP puncta are visible in myoblasts and myoblast cytonemes . Images taken 5 s apart captured the motion of Dl-containing puncta ( arrowheads in A , B , C ) . Scale bar: 10 μm . ( D–D′ ) Projection images in control ( 1151-Gal4/+;UAS-CD8:GFP/+ ) and diaRNAi ( 1151-Gal4/+;UAS-CD8:GFP/UAS-diaRNAi ) flies show that myoblasts with reduced diaRNA had fewer cytonemes . ( E–F′ ) ASPs with the same genotypes as ( D and D′ ) stained with α-Hnt ( E and E′ ) and α-ß-galactosidase antibodies ( F and F′ ) and with Alexa Fluor 555 secondary antibodies show that both Hnt and Su ( H ) lacZ were reduced under conditions of diaRNAi . ( G and H ) Bar graphs showing the dependence on dia and nrg function of myoblast cytoneme numbers per unit length of the imaged circumference of the ASP ( G ) , and ( I ) Hnt and Su ( H ) lacZ levels ( measured as fluorescence intensity in arbitrary units ) . p values , ( G ) 7 . 17E-12 and 8 . 475E-08 for diaRNAi and nrgRNAi , respectively; ( H ) 3 . 48E-05 and 1 . 79E-03 for diaRNAi and nrgRNAi levels of Hnt , respectively , and 5 . 9E-04 for diaRNAi levels of Su ( H ) lacZ; error bars: standard deviation . ( I–L ) Myoblast cytonemes ( marked with CD8:GFP ) that extend across the ASP were reduced in the presence of nrgRNAi ( I and J ) ; and in the ASP , the level of α-Hnt staining ( a readout of Delta-Notch signaling ) was also reduced ( K and L ) . Genotypes: control ( 1151-Gal4/+;UAS-CD8:GFP/+ ) and nrgRNAi ( 1151-Gal4/+;UAS-nrgRNAi/+;UAS-CD8:GFP/+ ) . ( M and M′ ) Sagittal views of ASPs ( red , Cherry fluorescence ) show that contacts between myoblasts and the ASP marked by GRASP fluorescence were reduced in dia-RNAi ( M′; 1151-Gal4/+;UAS-CD4:GFP1–10 lexO-CD4:GFP11/btl-LHG , lexO-mCherry-CAAX;UAS-diaRNAi/+ ) flies compared to controls ( M; 1151-Gal4/+;UAS-CD4:GFP1–10 lexO-CD4:GFP11/btl-LHG lexO-Cherry:CAAX ) . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 00710 . 7554/eLife . 06114 . 008Video 1 . Motile Delta-RFP in myoblast cytonemes . Dl:RFP expressed in myoblasts was detected in fluorescent puncta that moved along dynamic myoblast cytonemes ( labeled by CD4:IFP2 . 0–HO1 ) ; the ASP was labeled by CD2:GFP . ( Genotype: 1151-Gal4/+;btl-LHG lexO-CD2:GFP/+; UAS-Dl:RFP UAS-CD4:IFP2 . 0-HO1/+ . ) DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 008 To investigate whether the IFP-marked , Dl-containing cytonemes function in Notch signaling , we genetically ablated myoblast cytonemes by inactivating diaphanous ( dia ) and neuroglian ( nrg ) . The Dia protein is a formin whose activated form localizes to tips of filopodia ( Homem and Peifer , 2008; Rousso et al . , 2013 ) and cytonemes ( Roy et al . , 2014 ) , and Nrg is an L1-type cell adhesion protein that has been implicated in the development and stability of neuronal synapses ( Enneking et al . , 2013 ) . Loss-of-function phenotypes are dependent on context and degree of knockdown and negative results are uninformative , but it has been reported that Dia and Nrg are necessary for cytoneme-mediated Dpp signaling in the ASP ( Roy et al . , 2014 ) . We examined the effects of diaRNAi and nrgRNAi expression on myoblast cytonemes and on Notch signaling in the ASP and found that both the relative number of myoblast cytonemes and Notch signal transduction in the ASP were reduced ( Figure 5D–L ) . Notch signaling was monitored by expression of Hnt ( for diaRNAi and nrgRNAi ) and Su ( H ) lacZ ( for diaRNAi ) . Expression of diaRNAi reduced contacts between myoblasts and the ASP ( as indicated by reduced GRASP fluorescence; Figure 5M , M′ ) . These results suggest that the Dl-containing myoblast cytonemes present Delta to the ASP and make contacts that activate Notch signaling in the ASP . Notch signal transduction involves internalization of activating ligands by ligand-producing cells ( Seugnet et al . , 1997; Fischer et al . , 2006; Windler and Bilder , 2010 ) . To test whether endocytosis by myoblasts is required for Notch signaling in the ASP , we disrupted endocytosis in myoblasts by expressing dominant-negative forms of the Rab5 GTPase ( Vaccari et al . , 2008; Windler and Bilder , 2010 ) and the dynamin ortholog Shibire ( Shi ) ( Moline et al . , 1999 ) with the myoblast-specific 1151-Gal4 driver . Myoblasts were not obviously affected by these conditions , but stunted ASPs were observed in animals that expressed Rab5DN or ShiDN ( Figure 6A–F ) . We also examined Notch signaling in animals with mutant myoblasts by monitoring Hnt and Su ( H ) lacZ expression . Both were reduced ( Figure 6G–L ) . These results suggest that myoblast-produced Dl and the Dl-containing myoblast cytonemes activate Notch by the normal pathway of Notch signal transduction . 10 . 7554/eLife . 06114 . 009Figure 6 . Myoblast functions necessary for Notch signaling in the ASP . Compared to controls ( A , D , G , J ) , reduction of Rab5 function ( B , E , H , K ) or of Dynamin function ( C , F , I , L ) perturbed ASP development ( B , C , E , F ) and reduced Hnt ( H and I ) and Su ( H ) lacZ ( K and L ) expression . ( G–L ) Staining was with α-Hnt ( G–I ) and α-ß-galactosidase antibodies ( J–L ) and Alexa Fluor 555 secondary antibodies . ( Genotypes: ( A ) 1151-Gal4/+;btl-LHG , lexO-mCherry-CAAX/+; ( B ) 1151-Gal4/+;btl-LHG , lexO-Cherry:CAAX/+;UAS-Rab5DN/+; ( C ) 1151-Gal4/+;btl-LHG , lexO-Cherry:CAAX/+;UAS-ShiDN/+; ( D ) 1151-Gal4/+;UAS-CD8:GFP/+; ( E ) 1151-Gal4/+;UAS-CD8:GFP/UAS-Rab5DN; ( F ) 1151-Gal4/+;UAS-CD8:GFP/UAS-Shi5DN; ( G ) 1151-Gal4/+; ( H ) 1151-Gal4/+;UAS-Rab5DN/+; ( I ) 1151-Gal4/+;UAS-ShiDN/+; ( J ) Su ( H ) lacZ/1151-Gal4; ( K ) Su ( H ) lacZ/1151-Gal4;UAS-Rab5DN/+; ( L ) Su ( H ) lacZ/1151-Gal4;UAS-ShiDN/+ ) . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 009 The sensitivity of Dl-dependent Notch signaling in the ASP implies that Dl expression in the myoblasts is regulated . To investigate how the abundance of myoblast Dl might be controlled , we carried out screens to identify genes whose mis-expression perturbed Notch signaling in the ASP . To search for genes whose expression in the wing disc affects the ASP , we expressed candidates in the dorsal disc cells ( with ap-Gal4 ) while simultaneously marking the ASP ( with btl-LHG lexO-mCherry-CAAX ) and monitoring ASP growth and morphogenesis . Ectopic over-expression of Wg was found to severely reduce the ASP ( Figure 7A , B , J , K ) and to phenocopy Notch loss-of-function ( Figure 3B ) . Conversely , reducing Wg below its normal levels in the disc by expressing wg-RNAi with a wg-Gal4 driver generated abnormally large and malformed growths ( Figure 7C ) . These growths appeared similar to those produced by btl > NotchCA flies ( Figure 3C ) . Moreover , ASP expression of the Hnt and Su ( H ) lacZ Notch targets was decreased in wing discs with elevated levels of Wg ( Figure 7D , E , G , H ) and was increased in discs with decreased levels of Wg ( Figure 7D , F , G , I ) . These results suggest that wing disc-produced Wg negatively regulates Notch signaling in the ASP . 10 . 7554/eLife . 06114 . 010Figure 7 . Wg expression in the wing disc affects ASP development and Notch signaling . Compared to controls ( A , D , G ) , ectopic Wg expression ( B , E , H ) and wgRNAi expression ( C , F , I ) in the wing disc perturbed ASP development ( B and C ) . Ectopic Wg expression reduced levels of Hnt ( E ) and Su ( H ) lacZ ( H ) ; wgRNAi increased Hnt and Su ( H ) lacZ ( F and I ) . ( D–I ) Staining was with α-Hnt ( D–F ) and α-ß-galactosidase antibodies ( G–I ) and Alexa Fluor 555 secondary antibodies . ( J ) Drawings showing ( left ) areas of sal expression ( red ) in the wing disc ( Grieder et al . , 2009 ) and ( right ) Tr2 tracheal branches ( Rao et al . , 2015 ) . Dashed white line indicates approximate location of sagittal sections shown in Figure 11E–H . ( K ) Over-expression of Wg in the sal domain reduced ASP development; ( L ) Expression of TCFDN in trachea had no apparent effect on ASP development . ( Genotypes: ( A ) btl-LHG , lexO-Cherry:CAAX/+; ( B ) btl-LHG , lexO-Cherry:CAAX/ap-Gal4; UAS-wg/tub-Gal80ts; ( C ) wg-Gal4/UAS-wgRNAi; btl-LHG , lexO-Cherry:CAAX/UAS-wgRNAi; ( D ) wg-Gal4/+; ( E ) ap-Gal4/+;UAS-wg/tub-Gal80ts; ( F ) wg-Gal4/UAS-wgRNAi;UAS-wgRNAi/+; ( G ) Su ( H ) lacZ/+; ( H ) Su ( H ) lacZ/+;ap-Gal4/tub-Gal80ts;UAS-wg/+; ( I ) Su ( H ) lacZ/+;wgGal4/UAS-wgRNAi;UAS-wgRNAi/+; ( K ) sal-Gal4/+;UAS-wg/tub-Gal80ts; ( L ) btl-Gal4 UAS-CD8:GFP/UAS-TCFDN ) . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 010 To determine whether Wg produced by the wing disc signals directly or indirectly to the ASP , we expressed a dominant-negative form of the transcription factor Pangolin ( TCF ) in tracheal cells ( with the btl-Gal4 driver ) . TCFDN inhibits Wg signaling , but its expression in the ASP had no apparent effect ( Figure 7L ) . This result suggests that Wg signaling is not activated in the ASP and that Wg does not signal directly to the ASP . To evaluate Wg signaling in the myoblasts , we generated clones of mutant myoblasts that were incapacitated for Wg signal transduction . Dishevelled ( dsh ) is an essential component of the Wg signal transduction pathway ( Habas and Dawid , 2005 ) , and although null dsh myoblasts did not activate Wg signal transduction , they grew to generate large clones ( Figure 8A , B and Figure 8—figure supplement 1 ) . However , their levels of Dl were elevated relative to dsh+ neighbors ( Figure 8C ) . Ubiquitous expression of TCFDN in myoblasts ( with the 1151-Gal4 driver ) decreases the number of myoblasts ( Sudarsan et al . , 2001 ) , but ASPs that developed in discs with these compromised myoblasts were large and deformed ( Figure 8D , E ) . Increased Hnt and Su ( H ) lacZ expression in the abnormal ASPs ( Figure 8G , H , J , K ) indicates that Notch signal transduction was increased . These results suggest that Notch signaling in the ASP is sensitive to levels of Dl in myoblasts and that Dl expression in myoblasts is negatively regulated by Wg signaling . Genetic linkage between Wg signaling in myoblasts and Notch signaling in the ASP was demonstrated by simultaneously expressing dTCFDN and Dl-RNAi in myoblasts ( with the 1151-Gal4 driver ) . ASPs in these double mutant discs were neither stunted nor misshapen ( Figure 8D , F ) and expression of Hnt and Su ( H ) lacZ was similar to controls ( Figure 8G , I , J , L ) . These results imply that Wg signaling affects the ASP indirectly—that although Wg signaling may not be activated in the ASP , Wg signaling nevertheless plays an important role by down-regulating Dl expression in ASP-associated myoblasts that activate Notch signaling in the ASP . Wg-dependent down-regulation of Notch signaling has also been shown to be important for the myoblasts that generate the flight muscles of the adult fly ( Gunage et al . , 2014 ) . 10 . 7554/eLife . 06114 . 011Figure 8 . Wg signaling regulates the abundance of Delta in wing disc myoblasts . ( A–C ) A MARCM clone ( Lee et al . , 2000 ) of dishevelled mutant cells ( outlined with dashed white line ) expressed GFP ( B ) and up-regulated Dl ( C ) . α-Apontic antibody staining identified myoblasts ( A ) . Clone was induced by 1 hr heat shock 3–4 days after egg laying . ( D–F ) The ASP developed abnormally in the presence of TCFDN in wing disc myoblasts ( E ) compared to control ( D ) ; the phenotype was suppressed by expression of DeltaRNAi in the myoblasts . ( G–L ) TCFDN expression in the wing disc myoblasts increased Hnt ( H ) and Su ( H ) lacZ ( K ) relative to controls ( G and J ) ; DlRNAi reduced Hnt and Su ( H ) lacZ expression to control levels ( I and L ) . ( Genotypes: ( A–C ) dsh3 FRT19A/hsFLP tub-Gal80ts , FRT19A;act-Gal4 UAS-GFP/+; ( D ) 1151-Gal4/+;btl-LHG , lexO-Cherry:CAAX/+ , ( E ) 1151-Gal4/+;btl-LHG lexO-Cherry:CAAX/UAS-TCFDN and ( F ) 1151-Gal4/+;btl-LHG , lexO-Cherry:CAAX/UAS-TCFDN;UAS-DeltaRNAi/+ ) . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 01110 . 7554/eLife . 06114 . 012Figure 8—figure supplement 1 . disheveled mutant clones up-regulate Delta expression . MARCM clones mutant for dsh3 in myoblasts ( identified by expression of Apontic , left panels ) that were induced 3–4 days after egg laying expressed GFP ( middle panels ) and increased Dl expression ( right panels ) . Staining was with α-Apontic and α-Hnt antibodies . ( Genotype: dsh3 FRT19A/hsFLP tub-Gal80ts , FRT19A; act-Gal4 UAS-GFP/+ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 012 To investigate how Wg that is produced in the wing disc signals to the myoblasts and how Dl produced in myoblasts signals to the ASP , we first sought to understand where Wg-expressing disc cells are relative to myoblasts and the ASP . The number of myoblasts in early stage L3 discs is small , and we observed few of them directly overlayed Wg-expressing cells ( Figure 9A ) . During L3 , the myoblasts proliferate ( Gunage et al . , 2014 ) to cover most of the dorsal disc by late L3 , and we observed that the domain of Wg-expressing cells in the dorsal disc also expanded ( Figure 9B , C ) . However , the relative proximity of these Wg-expressing disc cells to the ASP decreased during the L3 period: although the most ventral Wg-expressing cells were within 5–10 μm of the ASP in early L3 discs , there were no Wg-expressing cells within 60–70 μm in late L3 discs ( Figure 9D–F ) . In order to determine whether myoblasts contact Wg-expressing cells and where such contacts are relative to the ASP , we applied the GRASP technique . The fluorescence of the reconstituted GFP revealed extensive contacts between myoblasts and Wg-expressing disc cells throughout L3 , but these contacting myoblasts were adjacent to the ASP only at the early L3 stage ( Figure 9G–I ) . Wg-expressing disc cells did not contact the ASP or nearby tracheal cells at mid or late L3 ( Figure 10A ) . These results suggest that Wg signaling from the disc to the ASP is indirect and that in the myoblasts , Wg signaling that is relevant to the ASP may be specific to the early L3 period . 10 . 7554/eLife . 06114 . 013Figure 9 . Proximity of Wg-expressing cells , myoblasts and ASP in the wing disc . ( A–C ) Images show the presumptive notum region of the wing disc during the third instar; the Wg domain ( red ) labeled by α-Wg antibody staining and myoblasts by GFP fluorescence ( Genotype: 1151-Gal4 UAS-CD8:GFP ) . ( D–F ) Images of the presumptive notum region of the wing disc show the ASP ( red ) and Wg-expressing disc cells ( green ) in early ( D ) , mid ( E ) and late ( F ) L3 stage discs . ( Genotype: btl-LHG/wg-Gal4 UAS-CD8:GFP; lexO-Cherry:CAAX/+ ) . ( G–I ) Images show contacts ( green fluorescence ) between Wg-expressing disc cells and myoblasts generated by reconstituted GFP ( GRASP ) . Auto-fluorescence of air-filled tracheal lumen was detected at 405 nm; perimeter of ASP is indicated by dashed white lines . ( Genotype: 15B03-lexA/wg-Gal4;UAS-CD4-GFP1–10 lexO-CD4-GFP11/+ ) . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 01310 . 7554/eLife . 06114 . 014Figure 10 . ASP development depends on Wg expression at the early third instar stage . ( A ) Image shows ASP ( highlighted by α-Discs large antibody staining and outlined with dashed white line ) from a GRASP experiment in which the fragments of GFP were expressed in the Wg domain of the disc and in the trachea . The absence of GFP fluorescence indicates that Wg-expressing disc cells did not directly contact the ASP . ( Genotype: wg-Gal4/lexO-CD4-GFP11;btl-LHG/UAS-CD4-GFP1–10 ) . ( B–I ) Expression of wgRNAi in the wing disc during the early L3 was induced as depicted in the line drawings above the images—either as a 24 hr temperature pulse at the non-permissive temperature for Gal80ts ( 29°C ) 7 days after egg laying ( early L3 , B and D ) , continuously at 29°C after 7 days ( C and E ) , 8 . 5 days ( mid L3 , F and H ) or after 10 days ( late L3 , G and I ) . wgRNAi perturbed ASP development ( D and E ) and increased Hnt expression ( B and C ) when expressed at early L3 , but had no apparent effect on either Hnt expression ( F and G ) or ASP development ( H and I ) when expressed only at mid or late L3 . ( F , G , H , I ) Staining with α-Hnt antibody was detected with Alexa Fluor 555 secondary antibody . ( J ) Quantitation of Hnt staining in ( B , C , F , G ) normalized to control ( wg > wgRNAi , Gal80ts at 18°C ) . N = 5 for each time point and control; error bars: standard deviation . ( Genotypes: ( B , C , F , G ) wg-Gal4/UAS-wgRNAi;UAS-wgRNAi/tub-Gal80ts; ( D , E , H , I ) btl-LHG lexO-Cherry:CAAX/wg-Gal4;UAS-wgRNAi/tub-Gal80ts ) . ( K–P ) Expression of TCFDN in myoblasts ( with the 1151-Gal4 driver ) was induced as depicted in the line drawings above the images—either at the non-permissive temperature for Gal80ts ( 29°C ) 7 days after egg laying ( early L3 ) , 8 . 5 days ( mid L3 ) or 10 days ( late L3 ) . ASP development was abnormal and Hnt staining increased with expression of TCFDN at early L3 , but had no apparent effect on either Hnt or ASP development when expressed only at mid or late L3 . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 014 To determine when the ASP requires Wg signaling in the myoblasts , we expressed wg-RNAi in the wing disc ( with a wg-Gal4 driver ) at various times during L3 , and the ASP was analyzed for Hnt expression ( to monitor Notch signaling ) and morphology ( Figure 10B–I ) . We used the Gal80ts repressor to conditionally express the RNAi . At 18°C , the permissive temperature for Gal80ts , animals developed normally . However , expression of wg-RNAi at early L3 , ( 7 days after egg laying at 18°C ) affected both ASP morphology and Hnt levels . wgRNAi , expressed either for a pulse of 24 hr ( between 7 and 8 days ) , or continuously ( for 40 hr after 7 days ) , increased Hnt levels and resulted in ASPs that were blunted and enlarged ( Figure 10B–E ) . In contrast , expression of wg-RNAi at mid ( 8 . 5 days after egg laying ) and late ( 10 days after egg laying ) L3 had no apparent effect on either Hnt levels or morphology ( Figure 10F–J ) . Similar results were observed after ectopic expression of dTCFDN in early , mid , and late L3 stages ( Figure 10K–P ) . These results suggest that Wg signaling in myoblasts is required at the early L3 stage to control the level of Delta-dependent Notch signaling in the ASP and that the consequences of the early stage Wg signaling perdure to later stages . We designed two types of experiments to investigate how Wg moves from the disc cells to the myoblasts . We first marked the membranes of myoblasts with tethered GFP ( using the 15B03-lexA driver ) in animals whose wg-expressing cells were marked with membrane-tethered RFP ( using the wg-Gal4 driver ) . In early L3 discs , we observed that myoblasts extended cytonemes both toward wg-expressing cells and the ASP ( Figure 11A , B ) . Myoblasts that had both types of cytonemes were detected ( Figure 11A ) . We next investigated whether the myoblast cytonemes contain Fz and Wg . We analyzed preparations from larvae that expressed a Fz:Cherry protein fusion in myoblasts and whose cytonemes were marked with membrane-tethered GFP . Fz:Cherry was present in motile puncta , many of which localized to myoblast cytonemes and to cytoneme tips ( Figure 11C and Video 2 ) . We also analyzed preparations from larvae whose wg-expressing cells expressed a Wg:Cherry protein fusion and whose myoblasts expressed membrane-tethered GFP ( driven by 1151-Gal4 ) . Many Wg:Cherry puncta were observed , some of which co-localized with and moved along GFP-marked myoblast cytonemes ( Figure 11D and Video 3 ) . To examine the role of the myoblast cytonemes , diaRNAi was expressed in the myoblasts , and Delta and β-catenin were monitored . In the presence of diaRNAi , levels of Delta increased ( ∼3×; Figure 11E , F ) and levels of β-catenin decreased ( ∼0 . 6×; Figure 11G , H ) . Neuralized expression was also detected coincident with Delta ( Figure 11–figure supplement 1 ) . These results support the conclusion that myoblasts extend cytonemes that take up Wg directly from Wg-expressing disc cells . 10 . 7554/eLife . 06114 . 015Figure 11 . Delta and Wg localize to myoblast cytonemes . ( A and B ) Myoblast cytonemes ( marked with CD8:GFP ) at early L3 . Images show Wg-expressing disc cells ( red ) , the ASP ( outlined by white dashed line in ( A ) ) , trachea ( autofluorescence at 405 nm in ( A ) ) , and myoblasts ( green , and indicated with black stars in ( A ) ) . Myoblast cytonemes extended toward the ASP ( white arrows ) and toward Wg-expressing cells ( white arrowheads ) . Yellow arrow in ( A ) indicates a myoblast with both types of cytonemes . ( C ) Fz:Cherry puncta ( white arrowhead points to one ) visualized in myoblasts and myoblast cytonemes ( marked with CD8:GFP ) . ( D ) Images obtained at 5 s intervals visualizing Wg:Cherry expressed in the Wg domain of the wing disc . Fluorescent puncta associated with disc cells ( red background area ) , with myoblasts ( green area ) and with myoblast cytonemes ( white arrowhead indicates a motile puncta ) . ( E–H ) Sagittal images showing wing disc and myoblasts ( green; encircled by dotted lines ) and stained with α-Delta ( E and F ) and α-β-catenin ( G and H ) antibodies ( red ) in the absence ( ( E and G ) or presence ( F and H ) of diaRNAi expressed in myoblasts . Fluorescence intensity in the indicated boxes was measured using ImageJ and the levels in experimental samples relative to controls were calculated by comparing the ratio of myoblast to disc fluorescence . ( Genotypes: ( A and B ) UAS-CD8:RFP lexO-CD8:GFP/+;wg-Gal4 , 15B03-LexA/+; ( C ) 15B03-LexA lexO-Fz:Cherry/lexO-CD2:GFP; ( D ) 1151-Gal4/+;wg:Cherry/+; UAS-CD8:GFP/+; ( E and G ) 1151-Gal4/+;UAS-CD8:GFP/+; ( F and H ) 1151-Gal4/+;UAS-CD8:GFP/UAS-diaRNAi ) . Scale bars: 30 μm ( A–D ) , 25 μm ( E–H ) . ) DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 01510 . 7554/eLife . 06114 . 016Figure 11—figure supplement 1 . Expression of Neuralized in the wing disc and associated myoblasts . Staining with α-Neuralized antibody ( red ) detected elevated levels in the myoblasts relative to the disc epithelium . DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 01610 . 7554/eLife . 06114 . 017Video 2 . Motile Frizzled receptors in the cytonemes . Fz:Cherry expressed in myoblasts was detected in fluorescent puncta that move along dynamic myoblast cytonemes ( labeled with CD2:GFP ) . ( Gentoype: 15B03-LexA lexO-Fz:Cherry/lexO-CD2:GFP . ) DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 01710 . 7554/eLife . 06114 . 018Video 3 . Wg puncta is transported in myoblast cytonemes . Wg:Cherry expressed in the wg domain of the wing disc was detected in fluorescent puncta that move along myoblast cytonemes ( labeled with CD8:GFP ) . ( Genotype: 1151-Gal4/+;wg–wg:Cherry/+;UAS-CD8:GFP/+ . ) DOI: http://dx . doi . org/10 . 7554/eLife . 06114 . 018
These histological and genetic studies identified regulatory interactions and physical contacts between the wing disc epithelium , wing disc-associated myoblasts and the ASP that presage the functional relationships of the adult tissues they produce . Whereas some interactions between the disc and ASP are direct , such as the signaling by disc-generated Dpp and FGF that activate signal transduction in the ASP ( Roy et al . , 2014 ) , our results now show that disc-dependent regulation of Notch signaling in the ASP is indirect . We found that the myoblasts are intermediaries that relay Wg signaling from the disc to control Notch signal transduction in the ASP . There are precedents for intermediaries that link signaling pathways in different cells . In the wing disc , for example , Hh produced in the posterior compartment induces Dpp expression in a spatially limited group of anterior compartment cells that are situated just across the anterior/posterior compartment border . These cells function as the anterior/posterior organizer , dispersing Dpp protein that induces the expression of various target genes by cells across the disc . The Dpp-expressing cells may thus be considered to be intermediaries that relay Hh signaling . This relay system transforms a pattern of binary expression ( Hh on in the posterior compartment , off in the anterior compartment ) to the more complex nested patterns of expression of Dpp target genes . The function of the myoblast relay system is not clear , and in contrast to the Hh-Dpp system , it may not be related to spatial patterns of gene expression . Key differences between the systems that may be relevant are timing and physical proximity . Hh induces Dpp expression at the anterior/posterior compartment border in the embryo and appears to do so continuously until the late L3 . Moreover , the spatial relationship between the Hh and Dpp-expressing cells remains essentially unchanged as the disc grows and matures . In contrast , Wg signaling in the myoblasts is required only in early L3 ( Figure 10 ) when the ASP and the Wg-expressing disc cells are relatively close ( Figure 9 ) . Myoblasts in this region facilitate Wg signaling by extending Fz-containing cytonemes that take up Wg . At later L3 stages , the patterns of Wg expression in the disc are quite different and the distance between Wg-expressing disc cells and the ASP is much greater ( Figure 9 ) . At these stages , the myoblasts that contact the ASP with Dl-containing cytonemes are juxtaposed to the ASP , far from the myoblasts that contact the Wg-expressing cells ( Figures 2 , 9 , 10 ) . The ASP has no apparent requirement for Wg signaling at these stages . The late L3 ASP is a tubular structure that has a lumen and morphologically distinct stalk , medial , and tip regions that define a proximal/distal axis ( Figure 1C ) . An orthogonal dorsal/ventral axis in the ASP is defined by cells that have contrasting levels of Dpp ( Roy et al . , 2014 ) and Notch ( Figure 2 ) signal transduction in the ‘upper’ and ‘lower’ layers . These regional specializations materialize as the ASP grows during the mid and late L3 stages ( Guha and Kornberg , 2005; Guha et al . , 2009 ) , but in early L3 when Wg signaling is needed , the ASP appears to be simply a small outgrowth with no discernable lumen or regional specializations . Our studies suggest that the myoblasts that activate Notch signaling in the lower layer cells of the late L3 ASP are defined by their prior exposure to Wg . We suggest that the myoblast relay system may accommodate the changing spatial relationship between the disc cells that express Wg and the ASP cells that require Notch signaling . The framework for understanding the functions of homeotic genes in Drosophila has come from seminal studies that analyzed null mutants , temperature sensitive mutants , and somatic clones of mutant cells . The consistent finding that mutant cells respond similarly in the embryo and in the larval stages led to the general concept that homeotic genes are required continuously during development ( Lawrence and Morata , 1994 ) . Our observation here , that ASP requires Wg signaling only in early L3 , is unusual in this context , but it is consistent with two prior reports . One described somatic clones in the wing disc that ectopically activated the Dpp pathway , and analyzed the expression of the Dpp target gene optomotor-blind ( omb ) ( Lecuit et al . , 1996 ) . Differences between omb expression in the clones and in normal discs led to the proposal that some omb-expressing cells in late L3 discs are not active for Dpp signal transduction but express omb because of their exposure to Dpp at an earlier stage . The second study analyzed the expression of Wg target genes in wing discs that expressed a form of Wg that is fused to a transmembrane domain of an unrelated protein ( Alexandre et al . , 2014 ) . It reported that Wg is normally expressed in a pattern of well-defined stripes in late L3 discs and more broadly in younger discs , and to reconcile the apparent discrepancy in late L3 discs between the restricted distribution of the membrane-tethered Wg ( detected by immunofluorescence ) and the broader expression domains of the Wg targets ( frizzled3 and Distal-less ) , it proposed that target genes that are induced in cells at early stages continue to be expressed by descendants of these cells that are not active for Wg signal transduction . Neither study directly examined the temporal dependence of signaling on target gene expression or characterized thresholds for the identified phenotypes . Wg and its family of related Wnt proteins have inductive signaling activities in many contexts ( Willert and Nusse , 2012 ) , and the long-standing assumption has been that it is released into the extracellular aqueous environment in order to travel from producing to target cells . A requirement for cofactors that enhance solubility in vitro and promote long range signaling in vivo is consistent with this idea ( Mulligan et al . , 2012 ) . However , the form of Wg that moves between pre- and post-synaptic cells at the Drosophila neuromuscular junction suggests a different mechanism . EM micrographs show that motoneurons secrete Wg in a vesicular form and that postsynaptic cells take up exosome-like vesicles that contain Wg ( Korkut et al . , 2009 ) . We have previously reported that Dpp and FGF signaling in the ASP is mediated by cytonemes that synapse with signal-producing wing disc cells , and that several genes that are required at neuronal synapses also have essential roles at cytoneme synapses ( Roy et al . , 2014 ) . We have also suggested that signaling at neuronal and cytoneme synapses is conceptually and structurally similar ( Kornberg and Roy , 2014 ) . Although cytoneme synapses that link Wg-producing wing disc cells and myoblasts have not been imaged with EM resolution , the most parsimonious model posits that cytoneme-dependent Wg signaling between disc cells and myoblasts is similar to Wg signaling at the neuromuscular junction and is also vesicle-mediated . The idea is that synaptic transfer and transit in exosome-like vesicles may be universal to Wg signaling . This reasoning has implications for signaling by membrane-tethered Wg . The discovery that flies that depend on membrane-tethered Wg are viable and morphologically normal was interpreted as evidence that juxtacrine signaling by the tethered Wg is sufficient ( Alexandre et al . , 2014 ) . An alternative possibility is that membrane-tethered Wg localizes to exosome-like vesicles that traffic along cytonemes and activates signal transduction while membrane-bound . Previous studies indicate that Dpp , FGF , EGF , Hh , and Notch may be cytoneme-mediated ( Hsiung et al . , 2005; Cohen et al . , 2010; Callejo et al . , 2011; Roy et al . , 2011; Bischoff et al . , 2013; Sanders et al . , 2013; Chen and Kornberg , 2014; Roy et al . , 2014 ) . The results reported here add Wg to this list and provide evidence that Delta-Notch signaling is cytoneme-dependent . Exchange of signals at synapses may be a universal mechanism of paracrine signaling .
Flies were reared on standard cornmeal and agar medium at 22–25°C , unless otherwise stated . btl-Gal4 , btl-LHG , lexO-CD2-GFP , and UAS-CD8:GFP were described ( Roy et al . , 2014 ) . lexO-CD4-GFP11 and UAS-CD4-GFP1–10 , from K Scott; lexO-mCherry-CAAX , from K Basler; UAS-CD4:IFP2 . 0-T2A-HO1 ( Yu et al . , 2014 ) ; Wg-Cherry knock-in , from JP Vincent; UAS-NotchDN , from N Perrimon ( Micchelli and Perrimon , 2006 ) , and UAS-NotchCA , from E Rulifson; wg-Gal4 , from T Adachi-Yamada ( Takemura and Adachi-Yamada , 2011 ) ; wg-RNAi ( II and III ) , from the Vienna Drosophila RNAi Center stock center; Su ( H ) lacZ , from S Bray ( Furriols and Bray , 2001 ) ; 1151-Gal4 , from K VijayRaghavan ( Roy and VijayRaghavan , 1997 ) ; sal-Gal4 , UAS-GFP/CyO; Gal80ts/TM6b ( Makhijani et al . , 2011 ) . Larvae were dissected in cold phosphate-buffered saline ( PBS ) . Wing imaginal discs together with Tr2 trachea were fixed in 4% formaldehyde . After several washes , the discs were permeablized with TritonX-100 and blocked in 10% donkey serum . The following primary antibodies were used: α-Apontic ( Eulenberg and Schuh , 1997 ) , α-Hindsight , α-Wg , α-β-catenin , and α-β-galactosidase ( Developmental Studies Hybridoma Bank , Iowa City , IA ) . Secondary antibodies ( Jackson ImmunoResearch , West Grove , PA ) were fluorescence-conjugated . Samples were mounted in Vectashield and imaged with a Leica TCS SPE confocal microscope . Wing discs were dissected and placed in a drop of PBS underneath a coverslip using the ‘hanging drop’ method ( Roy et al . , 2014 ) . Samples were imaged with a Leica TCS SPE or TCS SP2 confocal microscope with LAS-AF software . For time-lapse imaging , larvae were dissected in PBS and mounted in a slot formed between two strips of double sided tape , with the columnar layer facing the coverslip . Wing discs were incubated in Schneider's Drosophila medium containing fly extract , insulin , and penicillin-streptomycin . Images and videos were taken with a Nikon spinning-disc confocal microscope with 405 nm , 488 nm , 561 nm or 640 nm wavelength lasers . Videos were processed with NIS-Elements . Larvae were dissected and fixed in 0 . 12 M Na-cacodylate buffer , pH 7 . 4 for 1 hr on ice . The wing discs were post-fixed with 1% OsO4 in 0 . 12 M Na-cacodylate buffer , rinsed with distilled water several times and incubated overnight in cold 2% uranyl acetate . Dehydration was with an ascending series of ethanol , infiltrated with Durcupan ACM resin and embedded in Durcupan resin . After polymerization , the resin blocks were sectioned by microtome . Images were captured with a FEI Tecnai12 TEM . Cytonemes that extend to the ASP were scored . For each genotype , z-section stacks of confocal images from five ASPs were analyzed . The ratios in Figure 5H represent the number of cytonemes per unit length along the circumference of the ASP . To quantify Hnt staining ( Figures 5I , 10J ) , the mean intensity of 555 nm fluorescence was measured in an area ( containing approximately 20 cells ) of the lower layer of the ASP . The value ( with background fluorescence subtracted ) was normalized with respect to Hnt levels in an area containing approximately 10 cells at the dorsoventral boundary of the wing disc . For each experiment , comparisons were made to control genotypes that were prepared and analyzed together with experimental genotypes in order to control for differences in staining and changes to laser intensity . Statistical significance was calculated by t-test . The Frizzled coding sequence was amplified from a cDNA clone ( from P Adler ) with primers: Forward: 5′-TCGAGAATTCCCAAAATGTGGCGTCAAATCC-3′ and Reverse: 5′-CGTCGAATTCAAGACGTACGCCTGCGCC-3′ . Insert and lexO-Cherry vector were digested by EcoRI . | Fruit fly larvae undergo a remarkable physical transformation to become an adult fly . During this transformation , the tissues in the larvae change into the structures found in the adult . For example , the adult wings , flight muscles , and other structures needed for coordinated flight form from a pair of disc-like tissues called the wing imaginal discs . For these structures to develop correctly , the cells in the wing imaginal discs need to receive coordinated instructions about what types of cells they need to become . Within the wing discs , finger-like projections called cytonemes link specific cells together to allow signal molecules to move between the cells; this controls the development of the wing disc itself as well as structures called dorsal air sacs , which supply oxygen to the flight muscles in the adult fly . However , it is not known if cytonemes allow the exchange of signal molecules between cells involved in the formation of other structures needed for flight . Here , Huang and Kornberg investigated the role of cytonemes in the development of the flight muscles in fruit flies . The experiments reveal that cells called myoblasts—which will later become the flight muscle cells—form two sets of cytonemes with other cells . One set connects the myoblasts to cells in the developing air sac , which allows a signal protein called Delta to signal from the myoblasts into the air sac cells . The other set of cytonemes connects the myoblasts to wing disc cells . This enables another signal molecule called Wingless , which is produced in wing disc cells , to move into the myoblasts and block the production of Delta . Huang and Kornberg's findings reveal a new role for cytonemes in coordinating the development of the flight muscles and the dorsal air sacs . A future challenge will be to understand how individual cytonemes are able to connect to specific cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2015 | Myoblast cytonemes mediate Wg signaling from the wing imaginal disc and Delta-Notch signaling to the air sac primordium |
In most organisms , clamp loaders catalyze both the loading of sliding clamps onto DNA and their removal . How these opposing activities are regulated during assembly of the DNA polymerase holoenzyme remains unknown . By utilizing FRET to monitor protein-DNA interactions , we examined assembly of the human holoenzyme . The results indicate that assembly proceeds in a stepwise manner . The clamp loader ( RFC ) loads a sliding clamp ( PCNA ) onto a primer/template junction but remains transiently bound to the DNA . Unable to slide away , PCNA re-engages with RFC and is unloaded . In the presence of polymerase ( polδ ) , loaded PCNA is captured from DNA-bound RFC which subsequently dissociates , leaving behind the holoenzyme . These studies suggest that the unloading activity of RFC maximizes the utilization of PCNA by inhibiting the build-up of free PCNA on DNA in the absence of polymerase and recycling limited PCNA to keep up with ongoing replication .
Replicative DNA polymerases ( pols ) alone are distributive , synthesizing very few nucleotides of complementary DNA before disengaging from an elongating primer strand . To achieve the high processivity required for efficient DNA replication , replicative pols anchor to ring-shaped sliding clamps , forming the holoenzyme . Sliding clamps are dimers or trimers of identical subunits aligned head-to-tail , forming a highly-conserved , toroidal structure with two structurally distinct faces and a central cavity large enough to encircle double-stranded DNA and slide freely along it . The ‘C-terminal face' contains all C-termini and serves as a platform for interaction with replicative pols ( Hedglin et al . , in press; Yao and O'Donnell , 2012 ) . In solution , most sliding clamps are closed , requiring an enzyme-catalyzed mechanism which; ( 1 ) disrupts an interface within the sliding clamp ring and holds it open for assembly; ( 2 ) targets it to primer/template ( P/T ) junctions where DNA synthesis is initiated; ( 3 ) orients it correctly for interaction with pols; and ( 4 ) closes it around DNA ( Yao et al . , 1996; Schurtenberger et al . , 1998; Matsumiya et al . , 2003; Zhuang et al . , 2006; Paschall et al . , 2011 ) . Such feats are achieved by heteropentameric complexes referred to as clamp loaders , which utilize ATP to catalyze site-directed loading of sliding clamps onto DNA . Clamp loaders are composed of two to five distinct subunits that assemble into a tight heteropentameric complex with an overall structure that is conserved throughout evolution . Each subunit belongs to the AAA+ superfamily of ATPases that contain characteristic ATP-binding/hydrolysis motifs that convert the energy from ATP binding and hydrolysis to mechanical force . Over the years , information gathered from detailed mechanistic and structural studies has converged on a similar sequential mechanism for clamp loader-mediated assembly of the DNA pol holoenzyme ( Hedglin et al . , in press; Yao and O'Donnell , 2012 ) . In the presence of ATP , clamp loaders bind the C-terminal face of their respective clamp and open it for assembly . Once formed , the open clamp•clamp loader complex binds a P/T junction , adopting a ‘notched screw cap arrangement' that matches the helical geometry of the DNA duplex ( Simonetta et al . , 2009; Kelch et al . , 2011 ) . In this orientation , ATP hydrolysis is optimized and the clamp loader reverts to a low-affinity DNA-binding state upon hydrolysis of ATP and ejects from the P/T junction , leaving behind the loaded clamp . Concurrent or subsequent to clamp loader ejection , the open sliding clamp ring contracts around the P/T junction and is ready to anchor an incoming pol to the 3′ end of the primer strand , forming the holoenzyme ( Hedglin et al . , in press; Yao and O'Donnell , 2012 ) . Support for the generic model described above was primarily garnered from studies focused on holoenzyme assembly within E . coli , T4 bacteriophage , and S . cerevisiae while parallel human studies are gravely lacking . Also , distinguishing subtleties are present within each of these model systems such that a defined model for assembly of the human pol holoenzyme cannot be inferred . For instance , the fate of the clamp loader upon loading the clamp onto DNA is quite distinct for each model system . In T4 bacteriophage , the clamp loader chaperones the incoming pol onto the correct face of loaded clamp prior to dissociating from the clamp•DNA complex ( Trakselis et al . , 2003; Perumal et al . , 2012 ) . In E . coli , the clamp loader disengages from the P/T junction but remains at the replication fork through association with the replicative helicase and single-stranded DNA binding protein and chaperones one of the three pols bound to its own subunits to the newly loaded clamp ( Downey and McHenry , 2010 ) . In S . cerevisiae , a recent report suggests that the clamp loader does not actually eject from the P/T junction but rather remains part of the pol holoenzyme through association with one of the sliding clamp subunits ( Kumar et al . , 2010 ) . Furthermore , the few studies on assembly of the human pol holoenzyme failed to account for the clamp unloading activity of the clamp loader , replication factor C ( RFC ) . In a replicating cell , the number of P/T junctions are in excess of sliding clamps , necessitating an efficient unloading mechanism for recycling during S-phase ( Yao et al . , 1996; Leu et al . , 2000 ) . The human sliding clamp , proliferating cell nuclear antigen ( PCNA ) , is very stable on DNA by itself and is catalytically unloaded by RFC in an ATP-dependent manner . Such an activity is also present in archaea and bacteria while it seems to be missing from S . cerevisiae ( Yao et al . , 1996; Cai et al . , 1997; Cann et al . , 2001; Matsumiya et al . , 2003; Kumar et al . , 2010 ) . In T4 bacteriophage , enzyme-catalyzed clamp removal is unnecessary since the gp45 sliding clamp is unstable on DNA in the absence of polymerase ( Kaboord and Benkovic , 1996; Soumillion et al . , 1998; Perumal et al . , 2012 ) . Currently , holoenzyme assembly within humans has only been inferred from studies focused on the clamp loading-unloading pathway . Such studies addressed only one direction of the pathway at a time . The forward direction ( clamp loading ) was monitored beginning with only free components ( DNA , PCNA , RFC ) while the reverse direction ( clamp unloading ) was monitored by subjecting isolated PCNA•DNA complexes to RFC . Thus , it is assumed that RFC ejects immediately upon loading sliding clamps onto DNA , before the arrival of the polymerase , and then rebinds to unload them only after the polymerase disengages . However , without monitoring clamp loading and unloading simultaneously , the temporal correlation between the opposing activities of RFC and how they are regulated during assembly of the DNA polymerase holoenzyme remains unknown . Here we report studies on the assembly of the human replicative polδ holoenzyme . By utilizing FRET signals from fluorescently labeled P/T DNA and PCNA to monitor protein-DNA interactions , we were able to observe both activities of RFC in real time and directly monitor assembly of the DNA pol holoenzyme . We obtained data detailing the step-wise manner in which the human polδ holoenzyme is assembled; ( 1 ) RFC loads a stoichiometric amount of PCNA onto a P/T junction in an ATP-dependent manner; ( 2 ) upon hydrolysis of ATP , RFC releases the closed clamp ring onto DNA but remains transiently bound near the P/T junction; ( 3 ) unable to ‘slide away , ' PCNA re-engages with bound RFC in the absence of polδ and together RFC and all of the loaded PCNA dissociate from the DNA; ( 4 ) in the presence of polδ , DNA-bound PCNA is captured and physically occluded from the unloading activity of RFC and then; ( 5 ) RFC dissociates from the P/T junction , leaving behind the functional holoenzyme consisting of PCNA and polδ . Thus , human RFC does not immediately eject from P/T DNA in between PCNA loading and unloading as has been assumed for so long and the replicative pols play a paramount role by capturing loaded PCNA from DNA-bound RFC to inhibit unloading of PCNA and complete assembly of the holoenzyme . Upon dissociation of polδ from PCNA and DNA , RFC may then unload the clamp for recycling . These studies provide the first comprehensive analysis of assembly of the human replicative pol holoenzyme . The outcomes of these studies suggest that the unloading activity of the human clamp loader serves a dual purpose; it increases the efficiency of holoenzyme assembly by inhibiting the build-up of free clamps on DNA in the absence of replication and recycles scarce clamps to keep up with ongoing DNA replication .
In order to study holoenzyme assembly within humans , we utilized FRET to monitor protein-DNA interactions . A forked DNA substrate in agreement with the minimal requirements for assembly of human RFC and PCNA onto DNA was labeled with a Cy3 dye that served as the FRET donor ( Tsurimoto and Stillman , 1991 ) . This substrate , referred to herein as Cy3-P/T DNA , resembles the in vivo replication fork and also carries a 3′-biotin label ( Figure 1A and Table 1 ) . Together with the flap , the 3′-biotin label in complex with Neutravidin prevents loaded clamp from sliding off the DNA ( Yao et al . , 2000 ) . The position of PCNA residue 107 within the clamp ring is structurally conserved among eukaryotes ( Figure 1B ) and has been previously utilized for labeling as a FRET acceptor through mutation to cysteine ( Gulbis et al . , 1996; Zhuang et al . , 2006; Kumar et al . , 2010 ) . A mutant PCNA with the surface-exposed cysteines mutated and asparagine 107 changed to cysteine ( N107C ) was labeled with Cy5-maleimide ( Figure 1—figure supplement 1 ) . The labeled mutant PCNA ( Cy5-PCNA ) stimulated the ATPase activity of RFC equal to that for wild-type PCNA , indicating a fully functional protein ( Figure 1—figure supplement 2 ) . A truncated form of RFC ( hRFCp140ΔN555 ) was used in these studies in which the N-terminal 555 amino acids of the large subunit ( RFC1 , p140 ) were deleted to remove the nonspecific DNA-binding affinity of RFC ( Figure 1—figure supplement 1 ) . This region of the human clamp loader is conserved among eukaryotes and is dispensable for PCNA loading ( Uhlmann et al . , 1997; Podust et al . , 1998; Gomes et al . , 2000 ) . Herein , the truncated form of the clamp loader ( hRFCp140ΔN555 ) will be referred to as simply RFC . 10 . 7554/eLife . 00278 . 003Figure 1 . RFC-mediated loading of PCNA onto DNA . ( A ) Schematic representation of the Cy3-P/T DNA substrate used in this study . The primer had a Cy3 dye at the 5′ end and biotin was attached to the 3′ end of the template . Sequences for the primer , template , and flap constructs of all substrates used in this study are shown in Table 1 . The recombinant human proteins used in this study are shown in Figure 1—figure supplement 1 . ( B ) Model of human PCNA generated using Pymol from PDB code 1AXC ( Gulbis et al . , 1996 ) . PCNA subunits are shown in ribbon form in green , orange , and blue . The asparagine 107 residue , shown in red for one PCNA monomer in space-filling form , was mutated to cysteine for dye labeling . On average , each PCNA trimer has at least one labeled monomer . The mutations nor the labeling of PCNA had any effect on its ability to interact of RFC ( Figure 1—figure supplement 2 ) . Frontal and side views are shown . ( C ) Schematic representation of RFC-catalyzed loading of PCNA onto DNA . The N107C residue of PCNA is located on the face opposite that which interacts with RFC and faces the Cy3 FRET donor on the P/T DNA . ( D ) Fluorescence emission spectra obtained by exciting the Cy3-P/T DNA with a 514-nm light source . Cy5-PCNA can be excited through FRET from Cy3 only when the two dyes are in close proximity ( <∼10 nm ) . Cy5 fluorescence intensity peaks at 665 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 00310 . 7554/eLife . 00278 . 004Figure 1—figure supplement 1 . SDS–PAGE analysis of recombinant human proteins used in this study . ( Left ) RFCp140ΔN555 ( Lane 1 , 12 . 5 pmol ) , wtPCNA ( Lane 2 , 25 pmol ) , mutPCNA ( Lane 3 , 25 pmol ) , Cy5-PCNA ( Lane 4 , 25 pmol ) , and polδ ( Lane 5 , 12 . 5 pmol ) were loaded on a SDS 10% polyacrylamide gel and stained with Coomassie Blue . ( Right ) wtPCNA ( Lane 1 , 25 pmol ) , mutPCNA ( Lane 2 , 25 pmol ) , and Cy5-PCNA ( Lanes 3–4 , 25 pmol ) , were loaded on a SDS 10% polyacrylamide gel and imaged with UV trans illumination to visualize Cy5-label ( Lane 4 ) prior to staining with Coomassie Blue . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 00410 . 7554/eLife . 00278 . 005Figure 1—figure supplement 2 . Activities of labeled PCNA mutants in stimulating human RFC ATPase activity . The ATPase activity was determined at 25°C in an assay solution containing 125 nM RFC , 125 nM wtPCNA , mutPCNA , or Cy5PCNA , 125 nM P/T DNA ( 500 nM Neutravidin ) , 1 mM ATP , 3 mM phosphoenolpyruvate , 200 mM NADH , and 6–8 units of phosphoenolpyruvate kinase-lactate dehydrogenase mix . For each condition , the initial rates of ATP hydrolysis are reported as an average of three independent experiments ±SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 00510 . 7554/eLife . 00278 . 006Table 1 . DNA constructs for forked DNA substrates used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 006SubstrateTemplatePrimerFlapCy3-P/T DNACGG ACT GCA CGT GCC GCG TGG GCA TTC GTC GCG CAG GCT CAG CGT CCA TCG CGA GAC CAG /3Bio//5Cy3/CTG GTC TCG CGA TGG ACG CTG AGC CTG CGCCGT GGT GGT AGG TGA GGG CGG CAC GTG CAG TCC GCy3-P/T DNA-no flapCGG ACT GCA CGT GCC GCG TGG GCA TTC GTC GCG CAG GCT CAG CGT CCA TCG CGA GAC CAG /3Bio//5Cy3/CTG GTC TCG CGA TGG ACG CTG AGC CTG CGCGCG GCA CGT GCA GTC CGCy3-P/T DNA-25 bpCGG ACT GCA CGT GCC GCG TGG GCA TTC GTC GCG CAG GCT CAG CGT CCA TCG CGA G/3Bio//5Cy3/CT CGC GAT GGA CGC TGA GCC TGC GCCGT GGT GGT AGG TGA GGG CGG CAC GTG CAG TCC G The Cy3-P/T DNA substrate was tested for PCNA loading by monitoring the steady-state FRET signal between the Cy3 and Cy5 dyes by fluorescence spectroscopy ( Figure 1C ) . In the presence of RFC and ATP ( or ATPγS ) , a distinct FRET signal was observed at 665 nm , indicating that loading of Cy5-PCNA onto Cy3-P/T DNA is favored at equilibrium ( Figure 1D ) . Both RFC and ATP ( or ATPγS ) are required for PCNA loading , confirming that the ATP-dependent PCNA-loading activity of RFC is responsible for the observed steady-state FRET signal . In order to study the kinetics of PCNA loading in real time , we pre-assembled the RFC•Cy5-PCNA complex in the presence of ATP and mixed it with Cy3-P/T DNA in a stopped-flow instrument ( Figure 2A ) . In the absence of RFC , the signal remains flat over the entire time course ( Figure 2B , black trace ) again demonstrating that RFC is responsible for the observed FRET signal . In the presence of RFC , an extended time course of the FRET trace displayed three distinct phases ( Figure 2B , blue trace ) . A very rapid increase in FRET occurred within the ‘dead time' of the instrument and was not observed in the trace ( please see ‘Materials and methods' ) . This is most likely binding of the RFC•ATP•Cy5-PCNA complex to DNA . This was followed by a FRET increase with a rate constant of kinc = 6 . 4 ± 0 . 28 s−1 and a FRET decrease with a rate constant of kdec , 1 = 1 . 4 ± 0 . 038 s−1 . After ∼5 s , the FRET signal stabilized and remained flat for up to 1 min ( Figure 2B , Inset ) . 10 . 7554/eLife . 00278 . 007Figure 2 . RFC loads PCNA with rates independent of RFC concentration . ( A ) Schematic representation of experimental procedure for Figure 2B . ( B ) Cy5-PCNA ( 200 nM ) was incubated with RFC ( 200 nM ) in the presence of 1 mM ATP . This preformed RFC•Cy5-PCNA•ATP complex was mixed with Cy3-P/T DNA ( 200 nM ) in a stopped-flow instrument and the FRET signal was followed ( Blue trace ) . The loading traces were fit to a double-exponential equation . If RFC was omitted , no FRET signal was observed ( Black trace ) . An extended time course of 60 s is shown in the inset . ( C ) The experiments depicted in Figure 2A were also performed with varying concentrations of either RFC ( 0–300 nM ) or ATP ( 0–5 mM ) . Shown in panel C is the RFC titration of the FRET signal . The initial 2 . 5 s of the FRET traces is shown in the inset . The loading traces were fit to double-exponential equations and the respective rate constants for the fitted FRET increases and decreases are presented as a function of [RFC] in Figure 2—figure supplement 1 . The ATP titration of the FRET signal is presented in Figure 2—figure supplement 2 . ( D ) The overall amplitude of the signal ( sum of all amplitudes ) from Figure 2C plotted against [RFC] . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 00710 . 7554/eLife . 00278 . 008Figure 2—figure supplement 1 . RFC loads PCNA with rates independent of RFC concentration . The respective rate constants of the fitted FRET increase ( kinc ) and decrease ( kdec , 1 ) from Figure 2C as a function of [RFC] . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 00810 . 7554/eLife . 00278 . 009Figure 2—figure supplement 2 . RFC-catalyzed loading of PCNA is dependent upon ATP . ( A ) and ( B ) Experiments were performed as depicted in Figure 2A with varying concentrations of ATP ( 0–5 mM ) . Loading traces were fit to double-exponential equations and the rate constants for the fitted FRET increase , kinc ( shown in panel A ) , and FRET decrease , kdec , 1 ( shown in panel B ) , were plotted against [ATP] . In the absence of ATP , no FRET signal was observed ( data not shown ) . ( C ) Experiments were performed as depicted in Figure 2A in the presence of 1 mM ATPγS . The loading traces ( black ) were overlaid on corresponding traces from Figure 2B performed at 1 mM ATP ( grey ) . The loading traces were fit to a double-exponential equation with rate constants of 0 . 16 ± 1 . 3 × 10−2 s−1 and 1 . 1 × 10−2 ± 6 . 2 × 10−4 s−1 for the fitted fast and slow phases . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 009 The rate constants for the two observed phases ( kinc , kdec , 1 ) were both independent of RFC concentration ( Figure 2—figure supplement 1 ) . This suggests that each observed phase represents a first-order kinetic process , that is , a conformational change . The bimolecular association of the RFC•ATP•Cy5-PCNA complex with Cy3-P/T DNA is apparently fast as it was not observed at any RFC concentration on the time scale of our experiments . Interestingly , the overall amplitude ( AT ) of the signal ( i . e . , the sum of the amplitudes of all phases in the FRET trace , please see ‘Materials and methods' ) increases linearly throughout the entire range of RFC concentrations and does not saturate at a concentration where the RFC•Cy5-PCNA•Cy3-P/T DNA complex is stoichiometric ( Figure 2D ) . In the presence of ATP , binding of human RFC to PCNA is very tight with a KD ∼ 0 . 2 nM ( Cai et al . , 1997; Shiomi et al . , 2000 ) . Thus , at 200 nM Cy5-PCNA in Figure 2B–C , the RFC•Cy5-PCNA complex is saturated . This suggests that the FRET value at infinite time does not reflect a static endpoint but rather the amount of PCNA loaded onto DNA at equilibrium . Only the rate constant for the observed FRET increase ( kinc ) was independent of [ATP] ( Figure 2—figure supplement 2A ) . Even at the lowest [ATP] ( 62 . 5 μM ) , kinc was maximal , demonstrating that RFC has moderately high affinity for ATP . In the presence of 1 mM ATPγS , kinc splits into two phases , both of which are more than 30-fold slower than kinc observed at 1 mM ATP ( Figure 2—figure supplement 2C ) . Due to the differing kinetic characteristics of the FRET increases observed in the presence of ATP and ATPγS , it cannot be concluded that each represents an approach to the same FRET state . Furthermore , the presence of multiple phases within the observed FRET increase with ATPγS may suggest either multistep loading or the presence of more than one population of the ternary complex . Nonetheless , these studies demonstrate that only ATP binding , not hydrolysis , is required for opening of the clamp ring and assembly of the RFC•PCNA•DNA complex , in agreement with previous reports on the human system ( Lee and Hurwitz , 1990; Mossi et al . , 1997; Shiomi et al . , 2000 ) . The rate constant ( Figure 2—figure supplement 2B ) as well as the amplitude ( data not shown ) for the observed FRET decrease ( kdec , 1 ) was inhibited at concentrations of ATP greater than 1 mM . Perhaps at [ATP] > 1 mM , nucleotide exchange begins to compete with ATP hydrolysis and/or ADP binding within one or more of the RFC subunits , returning the RFC•ADP complex to the ATP-bound form . In contrast to kinc , these results suggest that ATP hydrolysis is required for the observed FRET decrease , kdec , 1 . Indeed , a FRET decrease is not observed in the presence of 1 mM ATPγS ( Figure 2—figure supplement 2C ) , a non-hydrolysable analog of ATP . Collectively , the studies in this section show that PCNA is loaded onto P/T DNA in a process that requires both ATP and RFC . Furthermore , it appears that PCNA loading proceeds through two observed phases; a conformational change that requires only ATP binding followed by a conformational change that requires ATP hydrolysis . The bimolecular association of the RFC•ATP•Cy5-PCNA complex with Cy3-P/T DNA is rapid and was not observed on the timescale of our experiments . The presence of multiple FRET changes within the traces presented in Figure 2B–C may suggest multistep loading . However , under these conditions , it cannot be concluded that each step occurs within the initial binding encounter of RFC•ATP•PCNA and P/T DNA . Furthermore , even though PCNA and RFC•ATP form a tight complex ( KD ∼ 0 . 2 nM ) , the alternative possibility that multiple reaction species are present cannot be ruled out ( Cai et al . , 1997; Shiomi et al . , 2000 ) . In order to address this , we carried out pulse-chase experiments as described in Figure 3A . A pre-formed RFC•ATP•Cy5-PCNA complex was briefly incubated with Cy3-P/T DNA ( ‘Pulse' ) and then mixed with a large excess unlabeled PCNA ( ‘Chase' ) in a stopped-flow instrument . Under these conditions , only Cy5-PCNA bound to RFC prior to the addition of Cy3-P/T DNA and chase should generate a FRET signal , thus eliminating free RFC and free Cy5-PCNA . In the absence of chase , the loading traces appear as in Figure 2 with rate constants of 7 . 6 ± 0 . 20 s−1 and 1 . 50 ± 0 . 083 s−1 for the fitted FRET increase ( kinc ) and decrease ( kdec , 1 ) , respectively ( Figure 3B , blue trace ) . If unlabeled PCNA chase is added prior to Cy5-PCNA , the FRET signal remains flat ( Figure 3B , red trace ) . This represents the zero FRET state and demonstrates that unlabeled PCNA is an efficient trap . When the experiment was carried out in the presence of chase , a FRET increase was observed with a rate constant of kinc = 6 . 1 ± 0 . 59 s−1 , followed by a unique FRET decrease with three distinct exponential decay phases instead of one; kdec , 1 = 1 . 6 ± 0 . 33 s−1 , kdec , 2 = 0 . 43 ± 0 . 070 s−1 , and kdec , 3 = 0 . 058 ± 5 . 7 × 10−4 s−1 ( Figure 3B , orange trace ) . The rate constants for the FRET increase ( kinc ) and the initial phase of the FRET decrease ( kdec , 1 ) agree very well with that observed in the absence of chase , demonstrating that these two observed phases occur within the initial binding encounter of Cy5-PCNA and Cy3-P/T DNA . Furthermore , these two observed phases exactly align in the presence and absence of chase ( Figure 3B , inset ) , suggesting that the FRET signal was generated from a single reaction species and that all of the pre-formed RFC•ATP•Cy5-PCNA complex commits through the loading pathway . Thus , dissociation of the RFC•ATP•Cy5-PCNA complex from the Cy3-P/T DNA is much slower ( ≥10-fold ) than the first conformational change ( kinc ) . The two additional phases within the FRET decrease only appeared when rebinding of Cy5-PCNA to Cy3-P/T DNA was inhibited by excess unlabeled PCNA chase and decreased the FRET signal to zero where all Cy3-P/T DNA is devoid of Cy5-PCNA . This suggests that these previously unobserved phases represent dissociation of Cy5-PCNA from Cy3-P/T DNA and that the final amplitude observed in the absence of chase does not reflect a static endpoint but rather the amount of PCNA loaded onto DNA at equilibrium where the rates for PCNA loading and dissociation are equal . Collectively , these studies demonstrate that a single turnover of RFC-catalyzed loading of PCNA actually proceeds through four observed phases , the last two representing dissociation of Cy5-PCNA from Cy3-P/T DNA . 10 . 7554/eLife . 00278 . 010Figure 3 . Monitoring a single binding encounter between PCNA and P/T DNA . ( A ) Schematic representation of pulse-chase experiment for Figure 3B . ( B ) Cy5-labeled PCNA ( 100 nM ) was incubated with RFC ( 100 nM ) and ATP ( 1 mM ) . This preformed RFC•Cy5-PCNA•ATP complex was mixed with Cy3-P/T DNA ( 100 nM ) for 37 ± 2 . 9 ms ( Pulse ) prior to mixing with unlabeled PCNA at 0 or 2 μM concentration ( Chase ) in a stopped-flow instrument and the FRET signal was followed . In the absence of chase ( ‘−Chase , ' blue trace ) , the traces appear biphasic as in Figure 2 . In the presence of chase ( ‘+Chase , ' orange trace ) , two additional phases appear in the FRET decrease . No FRET signal appears if unlabeled trap PCNA ( 2 μM ) is added prior to Cy5-PCNA ( ‘Chase Control , ' red trace ) . The initial 2 . 25 s of the FRET traces are shown in inset . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 010 Isolated human PCNA•DNA complexes are incredibly stable in the absence of free DNA ends , with reported half-lives of 22–80 min ( Jonsson et al . , 1995; Podust et al . , 1995; Yao et al . , 1996 ) . However , the results presented in Figure 3 suggest that DNA-bound PCNA is very unstable and rapidly dissociates into solution with a minimal half-life of t1/2 = ln2/kdec , 3 = 12 s . Thus , the appearance of two additional phases ( kdec , 2 and kdec , 3 ) within the FRET decrease in the presence of chase ( Figure 3B ) are quite profound as they raise the possibility that upon loading PCNA onto a P/T junction , RFC unloads PCNA without first dissociating into solution . Thus , the two opposing activities of RFC may occur within the same binding encounter with DNA . In order to address this , we directly monitored dissociation of PCNA from P/T DNA . As depicted in Figure 4A , the RFC•Cy5-PCNA•Cy3-P/T DNA complex was pre-assembled in the presence of ATP and then mixed with excess unlabeled PCNA chase . In the absence of chase , the FRET signal does not change over time , again signifying that the loading reaction has reached equilibrium where PCNA loading and dissociation are equal and a net change in the FRET signal is no longer observed ( Figure 4B , black trace ) . In the presence of chase , reloading of free Cy5-PCNA is inhibited and the FRET signal decreases with two distinct exponential decay phases ( Figure 4B , orange trace ) with rate constants of 0 . 40 ± 0 . 083 s−1 for the fast phase and 0 . 039 ± 0 . 012 s−1 for the slow phase with relative amplitudes of 83 ± 1 . 1% and 17 ± 1 . 1% respectively ( Table 2 ) . These rate constants , along with their relative amplitudes , are independent of the unlabeled PCNA trap concentration ( Figure 4—figure supplement 1 ) demonstrating that unlabeled PCNA serves as a passive trap . Furthermore , these values are in excellent agreement with data from Figure 3B obtained in the presence of chase . Together , this demonstrates that PCNA indeed rapidly dissociates from P/T DNA in a biphasic manner and suggests that at least two populations of the ternary complex are formed . 10 . 7554/eLife . 00278 . 011Figure 4 . polδ inhibits RFC-catalyzed unloading of PCNA to form the holoenzyme . ( A ) Schematic representation of experimental procedure for Figure 4B . ( B ) Cy5-labeled PCNA ( 100 nM ) loaded onto DNA ( 200 nM ) by RFC ( 100 nM ) in the presence of 1 mM ATP was mixed with 0 ( −Chase , black trace ) or 2 µM unlabeled PCNA chase ( +Chase , orange trace ) and the FRET signal was followed . The unloading traces were fit to a double-exponential equation and the rate constants and their relative amplitudes are reported in Table 2 . These rate constants , along with their relative amplitudes , are independent of the unlabeled PCNA chase concentration ( Figure 4—figure supplement 1 ) . Only the faster of the two rate phases is dependent upon the concentration of ATP ( Figure 4—figure supplement 2B ) . ( C ) Schematic representation of experimental procedure for Figure 4D . ( D ) Unlabeled PCNA ( 100 nM ) loaded onto Cy3-P/T DNA ( 200 nM ) by RFC ( 100 nM ) in the presence of 1 mM ATP was mixed with Cy5- PCNA ( 100 nM ) in the absence ( black trace ) or presence of polδ ( blue trace ) and the FRET signal was followed . The loading traces were fit to a double-exponential equation and the rate constants and their relative amplitudes are reported in Table 3 . These experiments were also carried out with either varying concentrations of Cy5-PCNA in the absence of polδ ( Figure 4—figure supplement 3 ) or with varying concentrations of polδ at a constant [Cy5-PCNA] ( Figure 4—figure supplement 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 01110 . 7554/eLife . 00278 . 012Figure 4—figure supplement 1 . Unlabeled PCNA serves as a passive trap . ( A ) Experiments were performed as depicted in Figure 4A with unlabeled PCNA trap at a concentration of either 2 ( black trace ) or 5 μM ( grey trace ) . Representative loading traces were overlaid for comparison . At 2 μM trap concentration , the FRET signal decreases with biphasic behavior with rate constants of 0 . 32 ± 0 . 022 s−1 for the fast phase and 0 . 026 ± 0 . 014 s−1 for the slow phase with amplitudes of 0 . 35 ± 0 . 013 ( 81 ± 1 . 9% ) and 0 . 080 ± 0 . 013 ( 19 ± 19% ) , respectively . At 5 μM trap concentration , the FRET signal decreases with biphasic behavior with rate constants of 0 . 35 ± 0 . 030 s−1 for the fast phase and 0 . 029 ± 0 . 010 s−1 for the slow phase with amplitudes of 0 . 39 ± 0 . 0010 ( 82 ± 0 . 90% ) and 0 . 084 ± 0 . 0053 ( 18 ± 3 . 1% ) , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 01210 . 7554/eLife . 00278 . 013Figure 4—figure supplement 2 . Unloading of PCNA by RFC is dependent upon ATP . ( A ) Experiments were carried out as depicted in Figure 4A in the presence of 1 mM ATPγS . The unloading trace ( black ) fits to a single exponential equation with rate constant of 6 . 2 × 10−3 ± 3 . 2 × 10−3 s−1 . The unloading trace is overlaid on corresponding trace in presence of 1 mM ATP from Figure 4B ( grey trace ) for comparison . ( B ) Experiments were carried out as depicted in Figure 4A in the presence of varying concentrations of ATP ( 0–5 mM ) . The unloading traces were fit to double-exponential equations and the rate constants were plotted versus ATP concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 01310 . 7554/eLife . 00278 . 014Figure 4—figure supplement 3 . All PCNA loaded onto P/T DNA dissociates with RFC . Experiments carried out as depicted in Figure 4C with various concentrations of Cy5-PCNA ( 33 , 100 , or 300 nM ) . For each [Cy5-PCNA] , the overall amplitude of the signal ( AT , Reload ) was divided by the overall amplitude for PCNA unloading ( AT , Unload ) determined in Figure 4B . Results were plotted against the fraction of Cy5-PCNA and fit to a linear regression . Each data point represents the average ± SEM of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 01410 . 7554/eLife . 00278 . 015Figure 4—figure supplement 4 . The single-stranded DNA flap has no effect on the dissociation of RFC or PCNA from Cy3-P/T DNA . ( A ) Experiments carried out as depicted in Figure 4A on the Cy3-P/T DNA substrate containing ( ‘+Flap , ' black trace ) or lacking ( ‘−Flap , ' grey trace ) the 17 nt single-stranded DNA flap . Representative traces were overlaid for comparison . All traces were fit to double-exponential equations and the rate constants and relative amplitudes for the fitted fast ( kdec , 2 ) and slow ( kdec , 3 ) are reported in Table 1 in the main text . When the flap was present , the FRET signal decreased with biphasic behavior with amplitudes of 0 . 32 ± 0 . 0055 for the fast phase and 0 . 067 ± 0 . 0038 for the slow phase ( AT = 0 . 39 ± 0 . 0049 ) . When the flap was removed , the FRET signal decreased with biphasic behavior with amplitudes of 0 . 34 ± 0 . 0076 for the fast phase and 0 . 068 ± 0 . 0021 for the slow phase ( AT = 0 . 41 ± 0 . 0024 ) . ( B ) All experiments were carried out as depicted in Figure 4C in the absence ( black trace ) or presence ( blue trace ) of polδ on a Cy3-labeled P/T DNA substrate lacking the single-stranded DNA flap . FRET traces were fit to double exponential equations and the calculated rate constants and relative amplitudes for the fitted fast ( k1 ) and slow ( k2 ) phases as well as AT , Reload/AT , Unload are reported in Table 2 in the main text . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 01510 . 7554/eLife . 00278 . 016Figure 4—figure supplement 5 . polδ captures a stoichiometric amount of loaded PCNA from DNA-bound RFC to form the holoenzyme . Experiments carried out as depicted in Figure 4C with various concentrations of polδ ( 0–133 nM ) . For each [polδ] , overall amplitude of the signal ( AT , Reload ) was divided by the overall amplitude for PCNA unloading ( AT , Unload ) determined in Figure 4B . Results were plotted against [polδ] and showed saturation at a concentration equivalent to the concentrations of RFC ( 100 nM ) and PCNA ( 100 nM ) used in the reaction . The rate constants for each phase as well as their relative amplitudes are independent of polδ concentration ( Figure 4—figure supplement 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 01610 . 7554/eLife . 00278 . 017Figure 4—figure supplement 6 . polδ passively captures loaded PCNA from DNA-bound RFC to form the holoenzyme and RFC dissociates from P/T DNA independently of polδ and PCNA . polδ passively captures loaded PCNA from DNA-bound RFC to form the holoenzyme and RFC dissociates from P/T DNA independently of polδ and PCNA . The respective rate constants ( A ) and relative amplitudes ( B ) of the fitted fast ( k1 ) and slow ( k2 ) phases from Figure 4—figure supplement 5 as a function of [polδ] . Each data point represents the average ±SEM of at least two independent measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 01710 . 7554/eLife . 00278 . 018Figure 4—figure supplement 7 . RFC releases PCNA onto P/T DNA . ( A ) Experiment performed as depicted in Figure 4A in the presence ( blue trace ) or absence ( red trace ) of Neutravidin . Representative unloading traces were overlaid for comparison . In the presence of Neutravidin , the FRET signal decreases with biphasic behavior with rate constants of 0 . 44 ± 0 . 033 s−1 for the fast phase and 0 . 042 ± 0 . 010 s−1 for the slow phase with amplitudes of 0 . 24 ± 0 . 024 ( 83 ± 0 . 97% ) and 0 . 051± 0 . 0067 ( 17 ± 0 . 97% ) , respectively . In the presence of Neutravidin , the FRET signal decreases with biphasic behavior with rate constants of 0 . 65 ± 0 . 054 s−1 for the fast phase and 0 . 051 ± 0 . 015 s−1 for the slow phase with amplitudes of 0 . 11 ± 0 . 011 ( 85 ± 2 . 9% ) and 0 . 018 ± 0 . 0031 ( 15 ± 2 . 9% ) , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 01810 . 7554/eLife . 00278 . 019Figure 4—figure supplement 8 . Removing 5 bp from the double-stranded region of the P/T junction has no effect on RFC-catalyzed unloading of PCNA from Cy3-P/T DNA or formation of the polδ holoenzyme . Removing 5 bp from the double-stranded region of the P/T junction has no effect on RFC-catalyzed unloading of PCNA from Cy3-P/T DNA or formation of the polδ holoenzyme . All experiments were performed on a Cy3-P/T DNA substrate in which 5 bp within the double-stranded region of the P/T junction was removed ( Cy3-P/T DNA-25 bp ) . ( A ) Experiments were carried out as depicted in Figure 4A . All FRET traces were fit to double-exponential equations and the rate constants and relative amplitudes for the fitted fast ( kdec , 2 ) and slow ( kdec , 3 ) are reported in Table 1 in the main text . ( B ) Experiments were carried out as depicted in Figure 4C in the absence ( black trace ) or presence ( blue trace ) of polδ . All FRET traces were fit to double exponential equations and the calculated rate constants and relative amplitudes for the fitted fast ( k1 ) and slow ( k2 ) phases as well AT , Reload/AT , Unload are reported in Table 2 in the main text . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 01910 . 7554/eLife . 00278 . 020Table 2 . Rate constants and relative amplitudes for the dissociation of Cy5-PCNA from Cy3-P/T DNADOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 020kdec , 2kdec , 3SubstrateRate constant , s−1% ATRate constant , s−1% ATCy3-P/T DNA0 . 40 ± 0 . 08383 ± 1 . 10 . 039 + 0 . 01217 ± 1 . 1Cy3-P/T DNA-No Flap0 . 40 ± 0 . 06884 ± 1 . 50 . 031 + 0 . 009316 ± 1 . 5Cy3-P/T DNA-25 bp0 . 23 ± 0 . 008083 ± 0 . 900 . 023 ± 0 . 002917 ± 0 . 90 Only the faster of the two phases ( kdec , 2 ) observed in the dissociation of Cy5-PCNA from Cy3-P/T DNA is dependent upon the concentration of ATP ( Figure 4—figure supplement 2B ) . Up to 1 mM ATP , kdec , 2 remains constant at 0 . 54 ± 0 . 080 s−1 and then decreases at concentrations greater than 1 mM ATP . This dependence is reminiscent of that observed for kdec , 1 ( Figure 2—figure supplement 2 ) . Furthermore , at each [ATP] , kdec , 2 was 1 . 5- to 3 . 4-fold slower than kdec , 1 , demonstrating that each is a distinct ATP hydrolysis-dependent step in the clamp loading-unloading pathway . The slower of the two phases in Figure 4B ( kdec , 3 ) remains constant at 0 . 057 ± 0 . 0044 s−1 across the entire range of [ATP] ( Figure 4—figure supplement 2B ) . Interestingly , spontaneous disassembly of PCNA from P/T DNA through dissociation of the PCNA trimer into its subunits ( 6 . 30 × 10−3 ± 1 . 7 × 10−3 s−1 , Senthil Perumal , manuscript in preparation ) was never observed in the presence of ATP . This process , referred to as subunit exchange , sets the maximum timescale for disassembly of sliding clamps from DNA in the absence of an enzymatic unloading activity and free DNA ends . In T4 bacteriophage , this process is the sole mechanism for disassembly of the gp45 clamp from DNA ( Kaboord and Benkovic , 1996; Yao et al . , 1996; Soumillion et al . , 1998 ) and also contributes substantially to disassembly of PCNA from DNA within S . cerevisiae ( Kumar et al . , 2010 ) . However , within the human system , this process is ∼6 . 7- and 67-fold slower than the two phases observed in Figure 4B and is only observed in ternary complexes formed in the presence of ATPγS ( Figure 4—figure supplement 2A ) . Assuming that the two kinetic phases observed in the dissociation of PCNA from P/T DNA represent defined FRET complexes , these results suggest that one or both populations of loaded PCNA may be actively unloaded by ( and , hence , dissociate with ) RFC . In order to address this , we sought to establish whether a population of RFC dissociates along with either PCNA species and whether or not any PCNA is left behind on P/T DNA . We carried out the trapping experiments depicted in Figure 4C for direct comparison to the PCNA dissociation experiments described in Figure 4A above . First , the RFC•ATP•PCNA•Cy3-P/T DNA complex was pre-assembled in the presence of excess Cy3-P/T DNA to ensure all RFC was engaged with PCNA and Cy3-P/T DNA . The PCNA in this complex is unlabeled so a FRET signal is not observed prior to mixing . This complex was then mixed with Cy5-PCNA at a 1:1 ratio of labeled to unlabeled PCNA . Under these conditions , the only source of RFC is from the pre-assembled RFC•ATP•PCNA•Cy3-P/T DNA complex . Thus , if all PCNA is actively unloaded from Cy3-P/T DNA by RFC and rapidly exchanges with Cy5-PCNA , the rate of Cy5-PCNA loading will be entirely rate-limited by dissociation of the RFC•PCNA complex from Cy3-P/T DNA . Hence , the rate of appearance of the FRET signal should be the same as the rates observed for the disappearance of the FRET signal in Figure 4B . Furthermore , the overall amplitude of the FRET increase ( AT , Reload ) should be exactly one-half the overall amplitude for the FRET decrease in Figure 4B ( AT , Unload ) if all PCNA is unloaded from Cy3-P/T DNA by RFC , as follows . Upon dissociation of RFC from the pre-assembled complex , the probability that RFC will exchange for and load Cy5-PCNA will be equal to the fraction of Cy5-PCNA in solution . At a 1:1 ratio of PCNA:Cy5-PCNA , the concentrations of unlabeled and labeled PCNA in solution will be equal if RFC unloads all loaded PCNA from the Cy3-P/T DNA and RFC will have an equal probability of loading either onto Cy3-P/T DNA . Thus , the overall amplitude of the FRET increase ( AT , Reload ) will be exactly half the overall amplitude for the FRET decrease ( AT , Unload ) observed in Figure 4B . In other words , AT , Reload/AT , Unload will be equal to 0 . 5 , the fraction of labeled PCNA in the reaction . As presented in Figure 4D , the FRET time trace was essentially the mirror image of Figure 4B , with two phases with rate constants of 0 . 45 ± 0 . 080 s−1 and 0 . 040 ± 0 . 0093 s−1 and relative amplitudes of 86 ± 2 . 1% and 14 ± 2 . 1% ( Table 3 ) that are within error of those reported for Figure 4B ( Table 2 ) . Thus , dissociation of RFC and PCNA from P/T DNA is intimately coordinated such that all RFC ejects from the P/T junction simultaneously with PCNA through dissociation of both populations of the RFC•PCNA•P/T DNA complex . This is definitive proof that the clamp loading-unloading pathway reaches equilibrium after the initial loading of PCNA onto P/T DNA because subsequent loading events occur with rate constants equal to the rate constants for dissociation of RFC•PCNA from DNA . Furthermore , AT , reload/AT , unload ( 0 . 51 ± 0 . 055 ) is equal to the fraction of labeled PCNA ( 0 . 5 ) in the reaction presented in Figure 4D ( Table 3 ) . This relationship holds true for other ratios of Cy5-PCNA:PCNA as well and a plot of AT , Reload/AT , Unload vs the fraction of Cy5-PCNA in the reaction yielded a straight line with a Y-intercept of ∼0 . 0 and a slope of ∼1 . 0 ( Figure 4—figure supplement 3 ) , indicating that no PCNA is left behind on P/T DNA upon ejection of RFC . Rather , all PCNA loaded onto P/T DNA by RFC is unloaded back into solution by RFC . 10 . 7554/eLife . 00278 . 021Table 3 . Rate constants and relative amplitudes for the dissociation of RFC from Cy3-P/T DNADOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 021k1k2SubstratepolδRate constant , s−1% ATRate constant , s−1% ATAT , Reload/AT , UnloadCy3-P/T DNA−0 . 45 ± 0 . 08086 ± 2 . 10 . 040 ± 0 . 009314 ± 2 . 10 . 51 ± 0 . 055+0 . 48 + 0 . 06580 ± 2 . 20 . 042 ± 0 . 01020 ± 2 . 21 . 0 ± 0 . 0063Cy3-P/T DNA-No Flap−0 . 48 ± 0 . 02691 ± 1 . 40 . 081 ± 0 . 0189 . 0 ± 1 . 40 . 51 ± 0 . 028+0 . 52 ± 0 . 05383 ± 1 . 50 . 037 ± 0 . 001317 ± 1 . 50 . 99 ± 0 . 060Cy3-P/T DNA-25 bp−0 . 25 ± 0 . 03885 ± 0 . 640 . 040 ± 0 . 0008415 ± 0 . 640 . 51 ± 0 . 046+0 . 27 ± 0 . 01270 ± 0 . 490 . 026 ± 0 . 001730 ± 0 . 491 . 0 ± 0 . 051 The experiments depicted in Figure 4 were also carried out on a Cy3-P/T DNA substrate lacking the single-stranded DNA flap ( Cy3-P/T DNA-No Flap , Table 1 ) and yielded the same results ( Figure 4—figure supplement 4 and Tables 2 and 3 ) . This demonstrates that all PCNA dissociates from the Cy3-P/T DNA substrate via the same pathway and independently of the single-stranded DNA flap and rules out the possibility that the flap traps RFC on the DNA and favors clamp unloading by keeping RFC in close proximity to the P/T junction upon loading PCNA . Indeed , human RFC is highly specific for P/T junctions and has very low affinity for purely single-stranded DNA ( Tsurimoto and Stillman , 1990 ) . Together , this suggests that upon loading PCNA onto a P/T junction RFC remains at or near the P/T junction by anchoring to the adjacent single-stranded DNA of the template strand , not the single-stranded DNA flap . Subsequently , RFC dissociates back into solution taking all loaded PCNA with it . Thus , the opposing activities of RFC , clamp loading and unloading , both occur within the same binding encounter with a given P/T junction . Like other systems , the human replicative pols δ and ε share common binding sites on PCNA with RFC ( Zhang et al . , 1999 ) . The results presented above imply that incoming pols must capture loaded PCNA rings from DNA-bound RFC to block unloading and complete assembly of the pol holoenzyme . In order to gain insight into how this is achieved , we repeated the trapping experiments depicted in Figure 4C in the presence of polδ . Under these conditions , the overall amplitude of the FRET increase ( AT , reload ) will report on the amount of both unlabeled PCNA and RFC in solution as before . At a ratio of 1:1:1 PCNA:Cy5-PCNA:polδ , if polδ captures all loaded PCNA from DNA-bound RFC and all RFC ejects into solution , there will be no unlabeled PCNA in solution and RFC can only load Cy5-PCNA onto Cy3-P/T DNA . Thus , the overall amplitude of the FRET increase ( AT , Reload ) should be equal to the overall amplitude of the FRET decrease ( AT , Unload ) in Figure 4B . Furthermore , the rate of appearance of the FRET signal should be faster in the presence of polδ if polδ actively displaces RFC . As presented in Figure 4D , only the overall amplitude ( AT , Reload ) increases in the presence of polδ . Two phases are present within the FRET time trace with rate constants and relative amplitudes that are within error of those reported in the absence of polδ ( Table 3 ) as well as those reported for the dissociation of PCNA from DNA ( Table 2 ) . Furthermore , these values are independent of polδ concentration ( Figure 4—figure supplement 6 ) . Thus , polδ does not actively displace RFC from the P/T junction . Rather , RFC dissociates from the P/T DNA in a biphasic manner independently of both PCNA and polδ . Thus , the rate constants for the disappearance of the FRET signal in Figure 4B and re-appearance of the FRET signal in Figure 4D all reflect dissociation of RFC from the P/T DNA and suggest that at least two forms of RFC are present in the ternary complex . AT , Reload/AT , Unload increased from 0 . 51 ± 0 . 055 in the absence of polδ to 1 . 0 ± 0 . 0063 in the presence of a stoichiometric amount of polδ ( Table 3 ) , suggesting that polδ captured all loaded PCNA from DNA-bound RFC . Importantly , this same behavior was also observed when the single-stranded flap was removed from the Cy3-P/T DNA substrate ( Figure 4—figure supplement 4B and Table 2 ) , again suggesting that RFC remains at or near P/T junction upon loading PCNA and does not retreat to the single-stranded DNA flap to allow the holoenzyme to form . When the concentration of polδ was varied , AT , Reload/AT , Unload increased linearly to the level of the unlabeled PCNA and RFC concentrations ( 100 nM ) and plateaued thereafter ( Figure 4—figure supplement 5 ) . At the break point where AT , reload/AT , unload = 1 . 0 , the concentrations of unlabeled PCNA , RFC , and polδ were all equivalent indicating that polδ stabilized all loaded PCNA on the Cy3-P/T DNA . Furthermore , this indicates that all PCNA loaded onto DNA by RFC is in the same form and competent for holoenzyme formation , that is the PCNA ring is closed around the P/T DNA in the correct orientation . Taken together , this suggests that polδ captures loaded PCNA rings from DNA-bound RFC to complete assembly of the pol holoenzyme . This effectively inhibits the unloading activity of DNA-bound RFC by physical occlusion and RFC subsequently dissociates from the single-stranded region of the template strand adjacent to the P/T junction , leaving behind the functional holoenzyme consisting of PCNA and polδ .
Previous studies on the human clamp loading-unloading pathway provided valuable insights into how this process may unfold and laid the groundwork for subsequent investigations . However , such studies were often indirect and qualitative , failing to monitor clamp loading-unloading in real time . Thus , several points along the reaction pathway have remained contested and a description of the complete reaction cycle , that is beginning and ending with RFC and PCNA free in solution , is currently lacking . By utilizing FRET experiments to monitor protein-DNA interactions , we were able to monitor the clamp loading-unloading process in real time using recombinant human proteins . Taken in a historical context , our results provide the first description of the complete human clamp loading-unloading pathway ( Figure 5 ) . 10 . 7554/eLife . 00278 . 022Figure 5 . Stepwise assembly of the human DNA polymerase holoenzyme . ( 1 ) RFC•ATP binds PCNA and opens it for assembly onto DNA . ( 2 ) The open PCNA•RFC•ATP complex binds to a P/T junction and ( 3 ) adopts the notched screw cap arrangement . ( 4 ) RFC hydrolyzes ATP , closing the PCNA ring and releasing it onto DNA . ( 5 ) In the absence of polymerase , loaded PCNA is unable to ‘escape' from DNA-bound RFC and is unloaded back into solution by RFC . ( 6 ) RFC subsequently releases PCNA , exchanges ADP for ATP , and the cycle repeats . ( 7 ) In the presence of polymerase , loaded PCNA is ‘captured' from DNA-bound RFC by an incoming polymerase , blocking the unloading activity of DNA-bound RFC by physical occlusion . ( 8 ) RFC subsequently dissociates , leaving behind the functional holoenzyme consisting of polymerase and PCNA . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 02210 . 7554/eLife . 00278 . 023Figure 5—figure supplement 1 . The human notched screw cap complex . The DNA footprint of the human RFC•ATPγS•PCNA complex ( Tsurimoto and Stillman , 1991 ) overlaid on the Cy3-P/T DNA substrate . Shown in magenta is the region protected by human RFC . Shown in cyan is the region protected by human PCNA . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 02310 . 7554/eLife . 00278 . 024Figure 5—figure supplement 2 . The clamp loader gp44/62 of T4 bacteriophage retracts towards the P/T junction upon hydrolysis of ATP and closure of the gp45 clamp ring around DNA . The open clamp:clamp loader:DNA complex ( PDB code 3U60 ) and the closed clamp:clamp loader:DNA complex ( PDB code 3U61 ) are shown in the same orientation ( via alignment of the AAA+ module of the C subunit ) . Images were generated using Pymol . P/T DNA and gp45 ( gray ) are shown in cartoon form . The surface of gp44/62 ( multicolored ) is shown . The clamp loader subunits ( A–E ) , gp45 as well as the primer ( orange ) and template ( yellow ) strands of the duplex DNA are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 00278 . 024 In the presence of ATP , RFC binds the C-terminal face of PCNA and opens the PCNA ring for assembly on P/T DNA [step 1 , ( Zhang et al . , 1999; Shiomi et al . , 2000 ) ] . Although ring opening has yet to be directly monitored within humans , previous studies suggest that it requires only ATP binding and not hydrolysis ( Lee and Hurwitz , 1990; Shiomi et al . , 2000 ) . Indeed , our pre-steady-state and steady-state FRET experiments demonstrated that assembly of the RFC•ATP•PCNA•P/T DNA complex , a process which ultimately requires PCNA ring opening , will occur in the presence of ATPγS , a non-hydrolysable analog of ATP . Once formed , the open RFC•ATP•PCNA complex specifically recognizes and binds a P/T junction [step 2 , ( Tsurimoto and Stillman , 1990; Lee et al . , 1991 ) ] . Our pre-steady state FRET experiments demonstrate that the bimolecular association of RFC•ATP•PCNA with DNA is very rapid , most likely diffusion-limited ( Berg and von Hippel , 1985 ) . Upon binding to DNA , ATP-bound clamp loaders adopt a conserved , right-handed spiral conformation referred to as the ‘notched screw cap' that closely mimics the helical geometry of the bound DNA ( Bowman et al . , 2004; Simonetta et al . , 2009 ) . Through extensive interactions with the C-terminal face of open clamps , clamp loaders pull the clamp subunits out-of plane into a complementary conformation ( Miyata et al . , 2005; Kelch et al . , 2011 ) . Although the structure of such complexes have yet to be determined for the human system , the asymmetric DNA footprints of human RFC alone and the RFC•PCNA complex exactly agree with that predicted by crystal structures of the notched screw cap arrangement from other organisms ( Tsurimoto and Stillman , 1991; Bowman et al . , 2004; Kelch et al . , 2011 ) . Furthermore , the DNA footprints could only be mapped in the presence ATPγS , demonstrating that the notched screw cap arrangement occurs prior to ATP hydrolysis within the RFC ATPase sites . Our pre-steady state FRET experiments show that after binding P/T DNA , the RFC•ATP•PCNA•P/T DNA complex undergoes a conformational change ( kinc ) that requires only the presence of ATP . Specifically , kinc is independent of [ATP] up to as high as 5 mM where it is assumed that each ATPase site within RFC is saturated with ATP . Furthermore , kinc is observed in the presence of ATPγS ( Figure 2—figure supplement 2 ) . Together , this suggests that kinc represents the RFC•ATP•PCNA complex adopting the notched screw cap arrangement on P/T DNA ( step 3 ) in which all RFC ATPase sites are occupied by ATP . After adopting the notched screw cap arrangement , the RFC•ATP•PCNA•P/T DNA complex undergoes a second conformational change ( kdec , 1 ) which requires ATP hydrolysis . It should be noted that these studies cannot discern whether the observed rate constant describes the actual conformational change , a rate-limiting ATP hydrolysis , or both . In the notched screw cap arrangement , the ATPase sites of clamp loaders are aligned properly for catalysis and ATP hydrolysis is optimized ( Simonetta et al . , 2009; Kelch et al . , 2011 ) . Furthermore , ATP hydrolysis is essential for assembly of a PCNA clamp that can slide along DNA and associate with polδ , both processes that require closure of the PCNA ring around DNA ( Lee and Hurwitz , 1990; Tinker et al . , 1994 ) . Thus , we speculate that the conformational change described by kdec , 1 reflects closing of the PCNA ring around DNA upon hydrolysis of ATP by RFC ( step 4 ) . Indeed , this conformational change is not observed in the presence of ATPγS ( Figure 2—figure supplement 2 ) . However , these studies cannot discern how many ATP molecules are hydrolyzed or in which ATPase sites within RFC such events occur ( discussed further below ) . Upon hydrolysis of ATP , RFC does not immediately eject from the PCNA•P/T DNA complex . Rather , RFC remains transiently bound near the P/T junction , dissociating slowly in a biphasic manner independently of both PCNA and polδ ( Figures 3 , 4 , and Figure 4—figure supplement 6 ) . Furthermore , in the absence of polδ , all PCNA loaded onto DNA by RFC is unloaded by RFC back into solution ( Figure 4 , Table 3 , and Figure 4—figure supplement 3 ) . Thus , PCNA and RFC are always present together on a given P/T DNA in the absence of polδ . However , this does not imply that RFC and PCNA remain engaged at the P/T junction while both are bound to a given DNA . In numerous reports , PCNA•DNA complexes devoid of RFC have been isolated in vitro using recombinant human proteins . Within these studies , the circular DNA substrates did not contain any blocks to PCNA sliding ( Cai et al . , 1996 , 1997; Yao et al . , 1996; Uhlmann et al . , 1997; Zhang et al . , 1999 ) . Thus , upon hydrolysis of ATP by RFC and closure of the PCNA ring , at least some fraction of loaded PCNA must slide away from the P/T junction within these substrates , escaping the unloading activity of RFC . Indeed , RFC's affinity for PCNA decreases ∼3 orders of magnitude in the presence of ADP alone ( Zhang et al . , 1999; Shiomi et al . , 2000 ) and if Neutravidin is omitted from the Cy3-P/T DNA substrate , allowing PCNA to slide off the primer end of the Cy3-P/T DNA substrate , the amount of Cy5-PCNA unloaded by RFC decreases by 2 . 3 ± 0 . 27-fold ( Figure 4—figure supplement 7 ) . Thus , we propose that upon hydrolysis of ATP ( s ) and closure of the PCNA clamp ring , RFC releases PCNA onto DNA ( step 4 ) . This would also be most consistent with the passive model for the capture of loaded PCNA by polδ ( discussed later ) . In the absence of polδ , all loaded PCNA re-engages with DNA-bound RFC and the RFC•PCNA complex together ejects from the P/T DNA into solution ( step 5 ) . This implies that RFC must re-open the closed PCNA ring prior to unloading . Based on the requirements for PCNA ring opening in solution , it is presumed that only ATP binding , not hydrolysis , is required for opening PCNA on DNA . We are currently testing this hypothesis and thus it cannot be ruled out that RFC requires the energy of ATP hydrolysis to pry open closed PCNA rings on DNA that are reinforced by electrostatic interactions between the positively-charged inner surface of PCNA and the negatively-charged DNA backbone . Regardless , unloading of PCNA from DNA is not limited by re-opening of the PCNA ring . Rather , this process is entirely rate-limited by RFC dissociation from P/T DNA as the rate constants for dissociation of RFC•PCNA and RFC alone are equivalent ( Figure 4 and Table 3 ) . Dissociation of RFC from P/T DNA is biphasic , with ∼80% dissociating fast and the remaining 20% dissociating ∼10-fold slower . This suggests that at least two forms of RFC are present with only the faster of the two phases being dependent upon ATP hydrolysis . As discussed above , clamp loaders revert to a low-affinity DNA-binding state upon hydrolysis of ATP and eject from the P/T junction . Perhaps as the concentration of ATP is increased , nucleotide exchange begins to compete with ATP hydrolysis and/or ADP binding such that dissociation of most RFC from the P/T junction is inhibited . This suggests that the minor population of RFC represents that which dissociates from the P/T junction independently of its nucleotide-bound state . Once free in solution , RFC releases bound PCNA , exchanges ADP for ATP ( step 6 ) , and the cycle repeats . The proposed model suggests that there are at least two distinct ATP hydrolysis-dependent steps that occur within the initial binding encounter of RFC with a P/T junction . This implies that ATP hydrolysis occurs ‘all or none' or sequentially . In the former model , each ATP-hydrolysis dependent event is catalyzed by ATP hydrolysis occurring within all ATPase sites of RFC at once . Thus , in our PCNA loading-unloading model , nucleotide exchange would have to occur while RFC is still bound to P/T DNA , in between closure of the clamp ring and unloading of PCNA into solution . In the latter model , there is a division of labor in which the ATPase sites of certain RFC subunits ‘fire' at defined moments along the reaction coordinate in order to catalyze distinct steps ( i . e . , ring closure , ejection of the clamp loader , etc . ) . Although our studies on human RFC cannot discern between these models , previous reports on clamp loaders from E . coli ( Snyder et al . , 2004 ) , the hyperthermophilic euryarchaeon Archaeoglobus fulgidus ( Seybert and Wigley , 2004 ) , and S . cerevisiae ( Johnson et al . , 2006 ) have provided experimental evidence for the latter model and recent crystallographic studies on the T4 bacteriophage clamp loader have provided the first structural insights . In brief , after the clamp loader•clamp complex binds to a P/T junction , ATP is hydrolyzed in one of the clamp loader subunits causing the open clamp to close around the DNA . The conformational changes within the hydrolyzing subunit are propagated to the remaining subunits of the clamp loader , presumably inducing sequential ATP hydrolysis ( Kelch et al . , 2011 ) . Thus , the state of bound nucleotide within the clamp loader progressively changes along the reaction coordinate , temporally defining the catalytic events . Future studies with human proteins will discern whether such a model is conserved . As discussed above , in the absence of a replicative pol , RFC loads a stoichiometric amount of PCNA onto a P/T junction and then disengages , taking all loaded PCNA along with it . The results presented in Figure 4 and Table 3 demonstrate that polδ captures loaded PCNA from DNA-bound RFC , inhibiting RFC's unloading activity by physical occlusion and completing assembly of the pol holoenzyme . Furthermore , polδ stabilizes a stoichiometric amount of loaded PCNA on DNA demonstrating that all loaded PCNA is closed around the P/T DNA and competent for holoenzyme assembly ( Figure 4—figure supplement 5 ) . This high efficiency is in contrast to holoenzyme assembly in S . cerevisiae where only one of the two populations of loaded PCNA is competent for polδ binding ( Kumar et al . , 2010 ) . The human replicative pols ( δ and ε ) share common binding sites on PCNA with RFC ( Zhang et al . , 1999 ) . This suggests that assembly of the human holoenzyme may occur by one of two models . Incoming pols may either actively displace RFC from an RFC•PCNA complex present at the P/T junction or passively capture a closed PCNA clamp while it is transiently disengaged from the DNA-bound RFC . As presented in Table 3 , and Figure 4—figure supplement 6 , dissociation of RFC from the P/T DNA is independent of polδ , consistent with a passive capture of loaded PCNA . This agrees with the data presented here and elsewhere that suggest that PCNA is released onto DNA upon hydrolysis of ATP by RFC ( Cai et al . , 1996 , 1997; Yao et al . , 1996; Uhlmann et al . , 1997; Zhang et al . , 1999 ) . However , it should be noted that this does not rule out that DNA-bound RFC may ‘chaperone' in replicative pols to the P/T junction as suggested to occur during nucleotide excision repair ( Overmeer et al . , 2010 ) . Indeed , our studies suggest that RFC remains transiently bound near the P/T DNA upon binding of polδ to the loaded PCNA clamp ( discussed further below ) and human RFC has been shown to interact with polδ ( Yuzhakov et al . , 1999 ) . In the notched screw cap complex , human RFC anchors to the single-stranded region of the P/T junction , covering 12 nucleotides of the template strand , and extends into the double-stranded region , covering 15 and 8 nucleotides within the primer and template strands , respectively ( Tsurimoto and Stillman , 1991; Gulbis et al . , 1996 ) . The open PCNA clamp expands the protected region on the double-stranded side of the P/T junction out to ∼25 bp ( Figure 5—figure supplement 1 ) . This arrangement of the clamp loader•ATP•clamp complex is highly conserved in all domains of life as well as T4 bacteriophage ( Hedglin et al . , in press ) as discussed further below . The closed ring of human PCNA has a width ( 30 Å ) that is equivalent to ∼10 bp of B-form DNA ( Gulbis et al . , 1996 ) . Thus , due to the minimal double-stranded region of the P/T junction ( 30 bp ) and the Neutravidin block within the Cy3-P/T DNA substrate , loaded PCNA can slide at most 20 bp away from the P/T junction to allow an incoming polymerase to bind to its C-terminal face ( Figure 1A ) . We believe this best represents the situation in vivo where PCNA is loaded onto ∼30 nt RNA-DNA hybrid primers that are blocked on both sides by single-stranded DNA binding protein ( replication protein A , RPA , in humans ) that restrict the loaded PCNA to the P/T junction ( Burgers , 2009 ) . However , the experiments performed on the Cy3-P/T DNA substrate lacking the single-stranded DNA flap ( Figure 4—figure supplement 4 and Table 3 ) demonstrate that human RFC remains at or near the P/T junction upon loading PCNA onto DNA by anchoring to the adjacent single-stranded DNA of the template strand , implying that RFC must also retract towards the P/T junction upon closure and release of PCNA onto DNA to allow for an incoming DNA polymerase . Recent , ground-breaking crystallographic studies from the Kuriyan laboratory suggest this may be the case . In their 2011 publication , Kelch et al . crystallized the clamp loader•clamp•DNA complex from T4 bacteriophage in two forms ( Figure 5—figure supplement 2 ) . In the ATP bound form , the clamp loader•clamp complex has adopted a right-handed spiral conformation ( notched screw cap ) that closely mimics the helical geometry of the bound DNA . In another form , the clamp is closed around DNA and the B subunit of the clamp loader has hydrolyzed ATP , moved away from the adjacent C subunit , and partially disengaged from the template strand and the clamp . The new conformation of the B subunit is incompatible with the symmetric spiral of the other clamp loader subunits . This partial collapse of the clamp loader spiral retracts the clamp loader , particularly the N-terminal domain of the A subunit , towards the P/T junction . The authors suggest that as ATP hydrolysis continues around the clamp loader spiral , the matching symmetry between the clamp loader and the DNA•clamp is progressively broken ( Kelch et al . , 2011 ) . Perhaps as ATP hydrolysis propagates around the human clamp loader after closing of the clamp ring on DNA , RFC retracts towards the 3′ end of the primer strand as the clamp loader spiral collapses , revealing more of the double-stranded region of the P/T junction . In addition to the closed PCNA ring sliding away from the P/T junction , such a conformational change would allow sufficient room for an incoming polymerase to form the holoenzyme . Indeed , limiting the double-stranded region of the P/T junction to the DNA footprint ( 25 bp ) of the human notched screw cap complex ( Cy3-P/T DNA-25 bp in Table 1 ) had no effect on the amount of PCNA unloaded in the absence of polδ or the amount of polδ holoenzyme formed ( Figure 4—figure supplement 8 and Tables 2 and 3 ) , suggesting that DNA-bound RFC retracts towards the P/T junction upon closure and release of PCNA onto DNA ( Figure 5 , step 4 ) . Once loaded PCNA is captured by polδ ( Figure 5 , step 7 ) , RFC subsequently dissociates , leaving behind the functional holoenzyme consisting only of PCNA and polδ ( Figure 5 , step 8 ) . The amount of PCNA trimers present in the nucleus of human cells during S-phase is estimated to be ∼2 × 105/cell ( Morris and Mathews , 1989 ) . Based on the size of the human genome and the number of P/T junctions ( Okazaki fragments and origins of replication ) , it is suggested that PCNA trimers are in demand during S-phase and must be re-used hundreds to thousands of times per cell cycle . As discussed above , human PCNA is incredibly stable on DNA in the absence of free DNA ends and an enzymatic unloading activity . Thus , an efficient unloading mechanism is required for recycling during S-phase ( Yao et al . , 1996; Leu et al . , 2000 ) . The results presented here demonstrate that RFC loads PCNA onto a P/T junction and , in the absence of polδ , catalytically unloads all loaded PCNA from the P/T DNA without first dissociating into solution . This may serve to maximize the utilization of limited PCNA by inhibiting the build-up of free PCNA on DNA in the absence of polymerase . Presumably , once polδ disengages from the P/T junction and PCNA , RFC may unload the PCNA from DNA , recycling limited PCNA to keep up with ongoing replication . However , in replicating eukaryotic cells , PCNA left behind on DNA serves as a cell signal during S-phase by marking replicated DNA for chromatin assembly ( Shibahara and Stillman , 1999 ) and sister chromatid cohesion ( Moldovan et al . , 2006 ) . Furthermore , replication-blocking lesions encountered during S-phase are also marked by PCNA for replication by alternative DNA polymerases that can accommodate lesions . Such a process , referred to as translesion ( TLS ) synthesis , is initiated by ubiquitylation of PCNA at the site of the lesion and may occur immediately at the replication fork or at a later time ( Sale et al . , 2012 ) . Thus , some PCNA rings must remain on DNA after dissociation of replicative pols . This suggests that the PCNA unloading activity of RFC is tightly coordinated with the aforementioned activities , perhaps through post-translational modifications of PCNA , RFC , or both . Indeed , SUMO ( small ubiquitin-like modifier ) is attached to PCNA during S-phase in S . cerevisiae . This serves to prevent homologous recombination by recruiting the Srs2 helicase but also leads to accumulation of SUMO-PCNA on chromatin in the absence of an alternative clamp loader , ELG1-RFC , suggesting that this modification may also inhibit the PCNA unloading activity of native RFC ( Parnas et al . , 2010 ) . Attachment of SUMO to PCNA has only recently been established in humans and it remains to be seen what effect , if any , this modification has on the stability of the human PCNA clamp on DNA ( Gali et al . , 2012; Moldovan et al . , 2012 ) . Furthermore , it was recently shown that human RFC is ubiquitylated in a DNA damage-dependent manner in vivo ( Tomida et al . , 2008 ) . Although it is currently unknown what effect this modification has on the catalytic activities of RFC , its temporal correlation with ubiquitylation of PCNA and TLS tempts one to speculate that ubiquitin conjugation to RFC may serve to inhibit its PCNA unloading activity during TLS .
DNA constructs were synthesized by Integrated DNA Technologies ( Coralville , IA ) and purified on denaturing polyacrylamide gels . Concentrations were determined from the absorbance at 260 nm using the calculated extinction coefficients . For annealing of all substrates , the Cy3-labeled primer strand was mixed with a 1 . 1-fold excess each of Biotin-labeled template and flap in 1× Annealing Buffer ( 10 mM TrisHCl , pH 8 . 0 , 100 mM NaCl , 1 mM EDTA ) , heated to 95°C for 5 min , and allowed to slowly cool to room temperature . The sequences of the primer , template , and flap constructs for each forked DNA substrate are shown in Table 1 . In all experiments described in this study , DNA was bound to a fourfold excess of Neutravidin ( Sigma-Aldrich ) unless otherwise stated . The plasmid for expression of wild-type human RFC was a generous gift from Dr . Paul Modrich ( Duke University School of Medicine , Durham , NC ) . Details of the truncated human RFC expression vectors and purification procedures will be described elsewhere ( manuscript in preparation ) . Wild-type human polymerase δ ( polδ ) was expressed and purified from E . coli as described previously ( Masuda et al . , 2007 ) . RFC and polδ concentrations were determined via Bradford Assay using BSA as a standard . The plasmid for expression of poly ( His ) -tagged wild-type PCNA was a generous gift from Prof . Ulrich Hubscher ( University of Zurich Institute of Veterinary Biochemistry and Molecular Biology , Zurich , Switzerland ) . Details of the human PCNA mutant ( mutPCNA ) expression vector will be described elsewhere ( manuscript in preparation ) . Wild-type and mutant PCNA were expressed in E . coli and purified by a published protocol ( Jonsson et al . , 1995 ) . PCNA concentrations were determined from the absorbance at 280 nm using the calculated extinction coefficients . The mutant human PCNA was labeled with Cy5-maleimide at N107C and checked for activity as described previously for S . cerevisae PCNA ( Kumar et al . , 2010 ) . The extent of cysteine-labeling was determined by calculating the molar concentrations of the dye ( ε650 = 250 , 000 M−1cm−1 ) and the protein ( ε280 = 13 , 670 M−1cm−1 ) by absorbance and correcting for the absorbance of the dye at 280 nm ( ∼5% of the absorbance at 650 nm ) . PCNA monomer labeling efficiency was 38 . 3% , indicating that each PCNA trimer has at least one labeled monomer on average . ATPase activity of RFC ( hRFCp140ΔN555 ) was assayed spectrophotometrically via an NADH oxidation enzyme-coupled assay as previously described ( Norby , 1988; Kumar et al . , 2010 ) . The ATPase activity was determined at 25°C in an assay solution containing 1× Replication Buffer ( 25 mM TrisOAc , pH 7 . 5 , 10 mM Mg ( OAc ) 2 , 125 mM KOAc ) , 0 . 1 mg/ml BSA , 1 mM DTT , 1 mM ATP , 125 nM RFC , 125 nM forked DNA ( 500 nM Neutravidin ) , and 125 nM of either wild-type PCNA ( wtPCNA ) , mutant PCNA ( mutPCNA ) , or Cy5-labeled mutant PCNA ( Cy5-PCNA ) . The initial rates of ATP hydrolysis are reported . Steady-state fluorescence spectroscopy was carried out in a Jobin Yvon FluoroMax-4 fluorimeter . The buffer solution contained 1× Replication Buffer , 0 . 1 mg/ml BSA , and 1 mM DTT . The assay solution contained 250 nM forked DNA , 1 μM Neutravidin , and 5 mM of either ATP or ATPγS equilibrated at 25°C . To this solution , 250 nM Cy5-labeled PCNA mutant and 250 nM RFC were sequentially added and fluorescence emission spectra recorded from 530 to 725 nm after 514 nm excitation . Stopped-flow studies were performed on an Applied Photophysics SX20 stopped-flow machine equipped with a fluorescence detector . PCNA loading experiments were performed by mixing RFC•Cy5-PCNA•ATP in one syringe with Cy3-P/T DNA in the other syringe , unless stated otherwise . PCNA unloading experiments were performed by mixing RFC•Cy5-PCNA•ATP•Cy3-P/T DNA in one syringe with unlabeled mutant PCNA in the other syringe , unless stated otherwise . Loading and unloading was monitored by exciting the donor ( Cy3 ) at 514 nm and following the resulting FRET signal using a 645 nm cutoff filter ( Andover Corporation , Salem , NH ) . The FRET traces for the PCNA loading experiments were recorded over 60 s by collecting 8000 total time points; 5000 time points over the initial 10 s of the reaction and 3000 time points over the remaining 50 s . The FRET traces for the PCNA unloading experiments were recorded over 60–180 s by collecting 1000 time points per 60 s . All traces were analyzed using Applied Photophysics ProDataTM software . Conditions for each experiment are detailed in the figure captions . For the PCNA loading experiments presented in Figure 2 , the FRET signal remains flat in the absence of RFC and represents the zero FRET state ( i . e . , fluorescence value at zero time , y0 ) . However , in the presence of RFC , an initial FRET increase occurs within the dead time of the instrument such that the observed FRET value is elevated at zero time . Although this phase cannot be fit to a rate equation , its amplitude ( Adead ) is reflected in the fluorescence value at infinite time ( c ) and can be deduced by modifying the double exponential equation as follows . For a given condition ( 1 ) y=Aince∧ ( −kinct ) +Adec , 1e∧ ( −kdec , 1t ) +c′ , wherec′=y0+Adead+Ainc+Adec , 1 , where y represents the fluorescence value at time t , Adead the amplitude of the initial FRET increase occurring within the dead time of the instrument , Ainc and kinc the amplitude and rate constant for the observed FRET increase , Adec , 1 and kdec , 1 the amplitude and rate constant for the observed FRET decrease , and c′ the fluorescence value at infinite time . y0 represents the fluorescence value at time 0 and was obtained at 0 nM RFC ( Figure 2C ) where the FRET signal remains flat over the entire time course . Thus , for a given condition , ( 2 ) Adead=c′−y0−Ainc−Adec , 1 . | All living organisms use enzymes called replicative DNA polymerases to produce copies of their genome during cell division . These enzymes promote the formation of new DNA strands by catalyzing the polymerization of deoxyribonucleotides , the single units that make up DNA . In humans these polymerases contain multiple protein subunits , as well as a specific cofactor—a magnesium ion—that is required for the enzyme to be active . DNA polymerases are able to add several nucleotides at once because they are anchored to ring-shaped protein complexes called sliding clamps that encircle the DNA template . This structure , known as the holoenzyme , is able to slide freely along the DNA template , which allows the polymerase to promote the addition of nucleotides in a highly efficient manner . Protein complexes called clamp loaders are responsible for attaching the holoenzyme to the DNA template , and also for detaching it . Studies of model organisms , including bacteria , viruses and yeast , have provided insights into the assembly of the holoenzyme in humans , but the exact mechanism behind this process has remained unknown . Now , Hedglin et al . use fluorescence resonance energy transfer ( FRET ) , a powerful microscopy technique that can monitor interactions between proteins , and also between proteins and DNA , to study the assembly of the holoenzyme . Whenever a sliding clamp is loaded onto a DNA template in the absence of polymerase , the clamp loaders quickly remove it . Whenever a polymerase is present , however , it captures the sliding clamps; the clamp loaders then dissociate from the newly assembled holoenzyme and DNA replication begins . By revealing that clamp loaders recycle scarce sliding clamps , and that they boost the efficiency of holoenzyme assembly by preventing clamps from accumulating on DNA in the absence of polymerase , Hedglin et al . have redefined our understanding of human holoenzyme assembly . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology"
] | 2013 | Stepwise assembly of the human replicative polymerase holoenzyme |
Flowering plants utilize small RNA ( sRNA ) molecules to guide DNA methyltransferases to genomic sequences . This RNA-directed DNA methylation ( RdDM ) pathway preferentially targets euchromatic transposable elements . However , RdDM is thought to be recruited by methylation of histone H3 at lysine 9 ( H3K9me ) , a hallmark of heterochromatin . How RdDM is targeted to euchromatin despite an affinity for H3K9me is unclear . Here , we show that loss of histone H1 enhances heterochromatic RdDM , preferentially at nucleosome linker DNA . Surprisingly , this does not require SHH1 , the RdDM component that binds H3K9me . Furthermore , H3K9me is dispensable for RdDM , as is CG DNA methylation . Instead , we find that non-CG methylation is specifically associated with sRNA biogenesis , and without H1 sRNA production quantitatively expands to non-CG-methylated loci . Our results demonstrate that H1 enforces the separation of euchromatic and heterochromatic DNA methylation pathways by excluding the sRNA-generating branch of RdDM from non-CG-methylated heterochromatin .
Transposable elements ( TEs ) and their remnants comprise a substantial fraction of eukaryotic genomes and generally must be kept silent to ensure genome integrity and function ( Bourque et al . , 2018 ) . TE silencing is achieved despite the disruption caused by each cell division , whereby half of the genome and histone proteins are made anew . Robust cellular memory of the inactive state is achieved by feedback interactions that reinforce and concentrate chromatin features and factors that contribute to transcriptional silencing and exclude activating factors ( Allshire and Madhani , 2018; Zhang et al . , 2018b ) . However , silent chromatin domains are not homogenous . Flowering plants have two major types of TE-associated silent chromatin: GC-rich coding regions of autonomous TEs , and AT-rich chromatin comprised of gene-proximal TE remnants , short nonautonomous TEs , and edges of autonomous TEs ( Sequeira-Mendes et al . , 2014; To et al . , 2020; Zemach et al . , 2013; Zhong et al . , 2012 ) . Although both are comprised of TEs , these chromatin types have distinct features ( Sequeira-Mendes et al . , 2014; Zemach et al . , 2013 ) . How two types of silent TE chromatin are distinguished and kept separate within the nucleus is a major open question . Both types of TE chromatin feature extensive cytosine methylation in the CG context catalyzed by MET1 ( plant homolog of Dnmt1 ) ( Cokus et al . , 2008; Lister et al . , 2008; Zemach et al . , 2013 ) , and are also methylated at non-CG ( CHG and CHH , where H is A , T , or C ) cytosines ( Stroud et al . , 2014; Zemach et al . , 2013 ) . GC-rich TE sequences have high levels of histone modifications associated with heterochromatin , including methylation of lysine nine of histone H3 ( H3K9me ) , and are therefore known as heterochromatic TEs ( Sequeira-Mendes et al . , 2014; Zemach et al . , 2013 ) . Non-CG methylation ( mCH ) at heterochromatic TEs is catalyzed primarily by chromomethylases ( CMTs; CMT3 for CHG methylation and CMT2 for CHH ) , which are recruited to H3K9 dimethylated ( H3K9me2 ) nucleosomes by histone-tail-interacting domains ( Du et al . , 2012; Stoddard et al . , 2019; Stroud et al . , 2014; Zemach et al . , 2013 ) . SUVH family H3K9 methyltransferases are in turn recruited to methylated DNA via SRA domains , forming a self-reinforcing loop ( Du et al . , 2014; Johnson et al . , 2007; Rajakumara et al . , 2011 ) . Arabidopsis thaliana plants lacking functional chromomethylases ( cmt2cmt3 mutants ) almost completely lack mCH at heterochromatic TEs , and their H3K9 methylation is greatly reduced ( Stroud et al . , 2014 ) . AT-rich TE sequences are low in H3K9me and other heterochromatic histone modifications , and are therefore known as euchromatic TEs ( Sequeira-Mendes et al . , 2014; Zemach et al . , 2013 ) . In contrast to the SUVH/CMT feedback loop that predominates in heterochromatin , RNA-directed DNA methylation ( RdDM ) catalyzes cytosine methylation within euchromatic TEs ( Zemach et al . , 2013; Zhong et al . , 2012 ) . RdDM loci are transcribed by a methylation-tolerant RNA polymerase II derivative ( Pol IV ) that couples cotranscriptionally with RNA-dependent RNA polymerase 2 ( RDR2 ) to make double stranded RNA , which is processed into 23/24-nt fragments by Dicer-like 3 ( DCL3 ) ( Singh and Pikaard , 2019 ) . These 24-nt small RNAs ( sRNA ) are subsequently denatured and loaded into Argonaute ( AGO ) protein complexes . AGO–sRNA complexes associate with another Pol II family enzyme , Pol V , to recruit Domains Rearranged Methylases ( DRMs; primarily DRM2 in Arabidopsis ) ( Erdmann and Picard , 2020; Matzke and Mosher , 2014; Raju et al . , 2019; Wendte and Pikaard , 2017 ) . Like the SUVH/CMT pathway , RdDM comprises positive feedback loops . Pol V is recruited to methylated DNA , effectively seeking its own product ( Liu et al . , 2014; Wongpalee et al . , 2019; Zhong et al . , 2012 ) . A more paradoxical feedback loop is thought to involve recruitment of Pol IV to H3K9me ( Erdmann and Picard , 2020; Matzke and Mosher , 2014; Raju et al . , 2019; Wendte and Pikaard , 2017 ) . This hypothesis emerged from the observation that Pol IV-mediated sRNA production at many loci requires SHH1/DTF1 , a protein that binds H3K9me2 and monomethylated H3K9me ( H3K9me1 ) in vitro ( Law et al . , 2013; Zhang et al . , 2013 ) . This model of Pol IV recruitment necessitates explaining how RdDM in general , and Pol IV specifically , is excluded from heterochromatic TEs with high H3K9me and targeted to euchromatic TEs with low H3K9me . Reliance of Pol IV on H3K9me also poses two theoretical questions . First , why would RdDM depend on a core component of the SUVH/CMT feedback loop ( H3K9me2 ) , when the two DNA methylation systems have largely nonoverlapping primary targets ( Stroud et al . , 2014 ) , and RdDM targets are H3K9me depleted ? Second , the euchromatic TEs targeted by RdDM are often comprised of just one or two nucleosomes ( Zemach et al . , 2013 ) . Maintenance of histone modifications is expected to be unstable at such short sequences due to the random partitioning of nucleosomes to sister chromatids following DNA replication ( Angel et al . , 2011; Berry and Dean , 2015; Lövkvist and Howard , 2021; Ramachandran and Henikoff , 2015; Zilberman and Henikoff , 2004 ) . Why would RdDM , a pathway capable of almost nucleotide-level resolution ( Blevins et al . , 2015; Zhai et al . , 2015 ) and specialized for silencing short TEs , be tied to a histone modification that requires longer sequences for stable propagation ? Here , we show that Pol IV activity is recruited to sequences with non-CG DNA methylation regardless of H3K9me , so that both the Pol IV and Pol V branches form positive feedback loops with the ultimate product of RdDM . We also show that linker histone H1 impedes RdDM activity in GC-rich heterochromatin , thereby restricting RdDM to AT-rich euchromatic TEs . We propose that without H1 , RdDM would be diluted into and effectively incapacitated by the vast stretches of non-CG-methylated heterochromatin common in plant genomes ( Feng et al . , 2010; Niederhuth et al . , 2016; Ritter and Niederhuth , 2021; Zemach et al . , 2010 ) . The affinity of H1 for GC-rich heterochromatin ( Choi et al . , 2020 ) focuses RdDM activity on short , AT-rich euchromatic TEs that RdDM is uniquely suited to silence .
To understand how the CMT and RdDM pathways are separated , we categorized Arabidopsis TEs by the dependence of their CHH methylation ( mCHH ) either on CMT2 ( CMT TEs ) or DRM2 ( DRM TEs ) . Among 18784 TEs with more than 2% mCHH in wild-type ( wt ) plants , 4486 TEs were demethylated in cmt2 plants and 3039 TEs lost mCHH in drm2 ( mCHH in the mutants <0 . 02 , Fisher’s exact test p < 0 . 01 , TEs longer than 200 bp; Figure 1—figure supplement 1A and Figure 1—source data 1 ) . Only 80 TEs had mCHH diminished below 2% in both mutants ( Figure 1—source data 1 ) , consistent with the largely separate sets of primary DRM and CMT targets ( Sigman and Slotkin , 2016; Stroud et al . , 2014 ) . Next , we used random forest classification ( Breiman , 2001; Ishwaran et al . , 2012 ) to identify predictors of DRM or CMT targets ( Figure 1A ) . We included genetic and epigenetic features known to be associated with RdDM or CMT activity , as well as linker histone H1 . H1 is specifically enriched in heterochromatic TEs , and its loss leads to increased DNA methylation at heterochromatic TEs and decreased methylation at euchromatic ones ( Bourguet et al . , 2021; Lyons and Zilberman , 2017; Papareddy et al . , 2020; Rutowicz et al . , 2015; Zemach et al . , 2013 ) . As expected , sRNA abundance can distinguish CMT and DRM TEs ( Figure 1A ) . H3K9me1 is also a good classifier ( Figure 1A ) . However , the best classifier turned out to be H1 ( Figure 1A ) . Using all variables in Figure 1A , we could predict CMT and DRM TEs with an error rate of 2 . 15% ( Figure 1B ) . With just H3K9me1 and H1 , the prediction is almost as accurate ( 5 . 42% error; Figure 1B ) . Remarkably , H1 alone successfully identifies CMT and DRM TEs ( 12 . 17% error; Figure 1B ) , suggesting that H1 is fundamental to separating these silencing pathways . To understand how H1 regulates the CMT and DRM pathways , we analyzed 24-nt sRNA expression , DNA methylation , and H3K9me2 in h1 plants that have inactivating mutations in both of the canonical Arabidopsis H1 genes ( Zemach et al . , 2013 ) . Consistent with published results ( Bourguet et al . , 2021; Lyons and Zilberman , 2017; Papareddy et al . , 2020; Rutowicz et al . , 2015; Zemach et al . , 2013 ) , we found an elevation of CHG methylation ( mCHG ) , H3K9me2 and mCHH at CMT TEs ( Figure 1C , D ) . CMT TEs are depleted of sRNAs in wt leaves , but sRNA expression increases 5 . 6-fold in h1 plants ( Figure 1D , Figure 1—figure supplement 1B , C ) . sRNA expression in h1 positively correlates with that in wt ( Figure 1—figure supplement 1B ) , indicating that loss of H1 amplifies sRNA expression at RdDM-capable CMT TEs rather than creating de novo RdDM targets . In contrast to the hypermethylation of CMT TEs , DRM TEs lose H3K9me2 , mCHG , mCHH , and sRNA expression in h1 plants ( Figure 1C , D and Figure 1—figure supplement 1D ) . Despite the loss of sRNA at DRM TEs , global 24-nt sRNA abundance is not altered in h1 plants ( Figure 1—figure supplement 1E ) , indicating the reallocation of RdDM activity from DRM to CMT TEs . This phenomenon can be observed within individual TEs , with sRNA biogenesis and mCHH relocating from the AT-rich edges in wt to the GC-rich internal sequences in h1 ( Figure 1E ) . The relocation of sRNA production and mCHH into TE interiors in h1 plants is also apparent in aggregate at TEs that retain substantial mCHH in drm2 and cmt2 mutants ( intermediate TEs that are not classed either as DRM or CMT TEs; Figure 1—figure supplement 1A and F , G ) . CMT TE mCHH increases to the same relative extent in h1 plants devoid of CMT2 ( h1c2; Figure 1F and Figure 1—figure supplement 1H ) , indicating that mCHH hypermethylation at CMT TEs in h1 mutants is caused by RdDM . These results indicate that RdDM relocates into heterochromatin in the absence of H1 and are consistent with recently published work ( Bourguet et al . , 2021; Papareddy et al . , 2020 ) . Absence of H1 in Arabidopsis causes a preferential increase of heterochromatic TE DNA methylation within linker DNA , the regions between nucleosomes ( Lyons and Zilberman , 2017 ) . The average distance between heterochromatic nucleosomes is also reduced from ~180 to 167 bp ( Choi et al . , 2020 ) . Given the relative promiscuity of RNA Pol IV initiation ( Zhai et al . , 2015 ) and the increased sRNA abundance at CMT TEs in h1 ( Figure 1D , Figure 1—figure supplement 1B , C ) , we asked whether patterns of sRNA production with respect to nucleosomes are altered in h1 . As expected , overall levels of sRNA are increased around nucleosomes of CMT TEs and decreased at DRM TEs ( Figure 2A–C and Figure 2—figure supplement 1 ) . An overt sRNA linker bias is apparent in both h1 and wt around the best-positioned nucleosomes ( Figure 2A–C and Figure 2—figure supplement 1 ) . This pattern becomes less obvious at less-well-positioned loci until it disappears completely ( Figure 2B and Figure 2—figure supplement 1 ) , as illustrated by measuring the correlation of the sRNA signal to itself ( Figure 2D ) . The shortening h1 sRNA autocorrelation around better positioned nucleosomes ( Figure 2D ) demonstrates how the linker histone dictates sites of sRNA production directly through linker occlusion and indirectly through nucleosome positioning . Because H3K9me is thought to recruit Pol IV activity ( Law et al . , 2013; Zhang et al . , 2013 ) , we investigated how sRNA distribution changes in relation to H3K9me1/2 in h1 plants . In wt , sRNA expression increases as H3K9me1 and H3K9me2 levels rise , but this trend reverses at TEs with more H3K9me and H1 ( Figure 3A , B ) . In contrast , sRNA expression shows a relatively simple , direct relationship with H3K9me1 and H3K9me2 in h1 plants ( Figure 3A , B ) , suggesting that H1 prevents Pol IV from following the H3K9me gradient . Unlike TEs , gene bodies normally have low levels of H3K9me , mCH , and sRNA ( Figure 3C; Zhang et al . , 2018b ) . However , many genes gain H3K9me and mCH ( especially mCHG ) in plants lacking the H3K9 demethylase IBM1 ( Miura et al . , 2009 ) . Although this hypermethylation does not require RdDM ( Inagaki et al . , 2010; Saze et al . , 2008 ) , recruitment of Pol IV by H3K9me would predict sRNA biogenesis in ibm1 genes . Indeed , we find increased sRNA and mCHH levels in ibm1 genes associated with high H3K9me2 and mCHG ( Figure 3C , D ) . Hence , the presence of H3K9me or mCH may be sufficient to trigger 24-nt sRNA production . The only H3K9me-binding factor implicated in Pol IV recruitment is SHH1 ( Law et al . , 2013; Zhang et al . , 2013; Zhou et al . , 2018 ) . Therefore , we tested whether CMT TE hypermethylation in h1 plants requires SHH1 . CMT TEs remain hypermethylated in h1cmt2shh1 plants to about the same extent as in h1cmt2 plants ( Figure 4A ) , demonstrating that in the absence of H1 , Pol IV is recruited to CMT TEs independently of SHH1 . Pol IV activity depends on a family of four CLSY putative chromatin remodeling proteins ( Greenberg et al . , 2013; Smith et al . , 2007; Zhou et al . , 2018 ) . Simultaneous loss of CLSY1 and CLSY2 has the same effect as loss of SHH1 , whereas CLSY3 and CLSY4 mediate RdDM at a largely distinct set of loci ( Yang et al . , 2018; Zhou et al . , 2018 ) . Mutations of SHH1 and CLSY1/2 preferentially reduce mCHH and sRNA at DRM TEs and increase mCHH at CMT TEs ( Figure 4A ) . In contrast , clsy3/4 mutant plants have reduced mCHH and sRNA at CMT TEs and increased mCHH and sRNA at DRM TEs ( Figure 4A ) , suggesting that SHH1 and CLSY1/2 preferentially mediate RdDM at DRM TEs , whereas CLSY3/4 preferentially recruit Pol IV to CMT TEs . Consistently , TEs hypermethylated in h1cmt2 and h1cmt2shh1 show a strong overlap with published CLSY3/4-dependent sRNA clusters and little overlap with CLSY1/2-dependent clusters ( Figure 4B and Figure 4—figure supplement 1 ) , suggesting that Pol IV relocation into heterochromatin involves CLSY3/4 . However , our results do not rule out the possibility that some of the RdDM expansion in h1 plants is mediated by CLSY1/2 or is independent of CLSY activity . Also , please note that the wt sRNA patterns in Figures 1D and 4A are distinct because the former is from leaves and the latter from inflorescences . Leaf sRNA levels are lower at CMT TEs and CLSY3/4 clusters compared to flowers ( Figure 4—figure supplement 2 ) , presumably due to higher expression of CLSY3/4 in reproductive tissues ( Long et al . , 2021; Zhou et al . , 2021 ) . Overall , our results indicate that SHH1 is relatively unimportant for RdDM activity at H3K9me-rich CMT TEs with or without H1 . The entry of Pol IV into H1-depleted heterochromatin must either involve a different H3K9me-interacting factor , or a chromatin feature other than H3K9me . Our results suggest that sRNA biogenesis at CMT TEs in h1 mutants is mediated by CLSY3/4 Pol IV complexes . Recruitment of these complexes has been proposed to involve mCG ( Zhou et al . , 2018 ) . Therefore , we examined sRNA levels and DNA methylation in h1met1 mutants ( Choi et al . , 2020 ) . Although MET1 is a CG methyltransferase , its loss also perturbs mCH and H3K9me2 at some CMT TEs ( Figure 5A and Figure 5—figure supplement 1A; Choi et al . , 2020; Deleris et al . , 2012; Zabet et al . , 2017; Zhang et al . , 2018a ) . To understand how these changes impact sRNA production , we differentiate between two groups of CMT TEs in met1 plants . MET1-independent CMT TEs keep mCH and H3K9me2 in met1 ( Figure 5—figure supplement 1A; Choi et al . , 2020 ) and accordingly maintain sRNA expression ( Figure 5B ) . These CMT TEs gain sRNA expression and mCHH in h1met1 relative to met1 and wt ( Figure 5B ) , demonstrating that mCG is not required for RdDM expansion into heterochromatin . In contrast , MET1-dependent CMT TEs , which lose mCH and H3K9me in met1 ( Figure 5—figure supplement 1A; Choi et al . , 2020 ) , lose sRNA in met1 and do not recover sRNA expression or mCHH in h1met1 ( Figure 5C ) , suggesting that mCH or H3K9me is necessary for sRNA biogenesis . To test the above hypothesis , we grouped CLSY3/4 targets by mCHH level in met1 ( mCHH ≥0 . 05 in wt and met1; mCHH ≥0 . 05 in wt and <0 . 05 in met1 ) . Even though all CLSY3/4 targets lose mCG in met1 , sRNA expression is reduced only when mCH and H3K9me2 are reduced ( Figure 5D and Figure 5—figure supplement 1B ) , implying that the presence of mCH and/or H3K9me is sufficient to maintain CLSY3/4-dependent sRNA biogenesis . In h1met1 , sRNA levels increase at CLSY3/4 targets where mCH is maintained: among 1565 CLSY3/4 clusters with wt mCH ( >0 . 01% ) , 72% keep mCH in met1 and gain sRNA expression in h1met1 ( met1 mCH >0 . 01 ) , whereas 15% effectively lose all mCH in met1 and have similarly low sRNA levels in met1 and h1met1 ( met1 mCH <0 . 005 , Figure 5E and Figure 5—figure supplement 1C ) . These results indicate that neither CLSY3/4 Pol IV activity , nor the RdDM expansion triggered by loss of H1 , depend on mCG . Our results so far indicate that H1 prevents RdDM from following a gradient of either H3K9me or mCH into heterochromatin . However , heterochromatin is structurally complex and contains many factors ( Feng and Michaels , 2015 ) . To understand the overall importance of heterochromatin integrity , we tested the effects of H1 on sRNA distribution in plants with a mutation in the Swi/Snf2 chromatin remodeler DDM1 , which have severely compromised heterochromatin ( Kim and Zilberman , 2014; Sigman and Slotkin , 2016 ) . The ddm1 mutation greatly reduces heterochromatic DNA and H3K9 methylation ( Ito et al . , 2015; Lyons and Zilberman , 2017; Osakabe et al . , 2021; Teixeira et al . , 2009; Zemach et al . , 2013 ) , activates TE expression ( Lippman et al . , 2004; Osakabe et al . , 2021; Panda et al . , 2016; Panda and Slotkin , 2020; Rougée et al . , 2021 ) , and disperses nuclear heterochromatic foci ( Rougée et al . , 2021; Soppe et al . , 2002; Figure 6A , B and Figure 6—figure supplement 1A ) . However , 24-nt sRNA expression in ddm1 is broadly similar to wt ( Figure 6C , D and Figure 6—figure supplement 1B ) . Simultaneous lack of H1 and DDM1 in h1ddm1 mutants ( Lyons and Zilberman , 2017; Zemach et al . , 2013 ) causes relocation of sRNA biogenesis into CMT and intermediate TEs that mirrors that in h1 plants ( Figure 6C , D and Figure 6—figure supplement 1B ) , indicating that overall heterochromatin integrity is not required for this process . Furthermore , RdDM expansion into heterochromatin occurs in h1ddm1 despite strong H3K9me reduction compared to wt and h1 ( Figure 6A , B and Figure 6—figure supplement 1A ) . This does not rule out the possibility that H3K9me promotes Pol IV activity , because the H3K9me remaining in h1ddm1 may be sufficient . However , the observation that sRNA production at CMT TEs is largely unaffected by a bulk H3K9me reduction argues against a primary role for H3K9me in Pol IV recruitment . H3K9me and mCH are closely associated in heterochromatin due to the feedback loop between CMT2/3 and the SUVH4/5/6 H3K9 methyltransferases ( Du et al . , 2012; Stoddard et al . , 2019; Stroud et al . , 2014 ) . To isolate the effects of these features on sRNA biogenesis , we examined DNA methylation , H3K9me and sRNA levels in c2c3 and h1c2c3 plants . While mCG is largely unaffected , mCH is specifically abolished at CMT TEs in these plants ( Figure 7—figure supplement 1A ) , consistent with previously published c2c3 results ( Stroud et al . , 2014 ) . As expected , H3K9me is also greatly reduced ( Figure 7—figure supplement 1A ) , but some H3K9me1 and H3K9me2 remains in heterochromatin . Specifically , 875 CMT TEs maintain H3K9me1 and 1126 maintain H3K9me2 in c2c3 , while in h1c2c3 we identified 2434 H3K9me1-enriched CMT TEs and 1443 H3K9me2-enriched CMT TEs ( Figure 7A , B ) . Principal component analysis shows that H3K9me in these mutants associates with mCG , followed by CG and CCG density ( which contribute to mCG density; Figure 7C and Figure 7—figure supplement 1B ) , suggesting that SUVH4/5/6 are recruited to mCG in the absence of mCH . This conclusion is supported by a complementary pattern of H3K9 methylation changes in h1c2c3 vs . met1 . TEs that lose H3K9me2 in met1 , suggesting H3K9me dependence on mCG , maintain H3K9me in the absence of mCH in h1c2c3 ( Figure 7D ) . Conversely , TEs that lose H3K9me in h1c2c3 , suggesting H3K9me dependence on mCH , retain H3K9me2 in met1 ( Figure 7E ) . This indicates that H3K9me at mCG-dense CMT TEs is partially dependent on mCG , leading to considerable H3K9me retention in c2c3 , and especially h1c2c3 plants . The ability of mCG to recruit H3K9me is consistent with published work , including studies that show RdDM-independent initiation of the CMT-SUVH feedback loop specifically at CG-methylated sequences ( Miura et al . , 2009; To et al . , 2020; Zabet et al . , 2017 ) and the observed affinity of SUVH histone methyltransferase SRA domains for mCG in vitro ( Johnson et al . , 2007; Li et al . , 2018; Rajakumara et al . , 2011 ) . The decoupling of H3K9me and mCH in h1c2c3 plants allowed us to determine how each feature is associated with sRNA biogenesis . In h1 plants , H3K9me2 , DNA methylation in every context , and sRNA expression together increase in direct relation to wt H1 prevalence , as loss of H1 increases accessibility of previously H1-rich TEs ( Figure 8A and Figure 8—figure supplement 1A; Bourguet et al . , 2021; Lyons and Zilberman , 2017; Papareddy et al . , 2020; Zemach et al . , 2013 ) . H3K9me1/2 , DNA methylation , and sRNA levels are also all positively correlated in h1 plants , though the correlation between H3K9me2 and sRNA is weak ( Figure 8B and Figure 8—figure supplement 1B ) . In contrast , the coupling of H3K9me with DNA methylation and sRNA levels nearly disappears when comparing h1c2c3 to c2c3 ( Figure 8C , D and Figure 8—figure supplement 1C , D ) . Relative H3K9me1/2 abundance increases with wt H1 levels , whereas DNA methylation and sRNA changes show at best a very weak relationship with wt H1 enrichment ( Figure 8C and Figure 8—figure supplement 1C ) . Two correlated groups remain in h1c2c3: H3K9me1/2 with mCG , and sRNA with mCHG/mCHH ( Figure 8D and Figure 8—figure supplement 1D ) . The linear correlations between sRNA and either H3K9me1 or mCG observed in h1 ( Figure 8E ) become kinked in h1c2c3 ( Figure 8F ) , resembling the association between sRNA and H3K9me1 in wt ( Figure 3B ) . The overall pattern of h1c2c3 sRNA at CMT and intermediate TEs resembles wt far more than h1 ( Figure 8G and Figure 8—figure supplement 1E ) . The patterns and levels of sRNA and mCHH at DRM TEs are also similar between h1c2c3 and wt ( Figure 8—figure supplement 1F , G ) . Only the association between mCH and sRNA remains linear in h1c2c3 ( Figure 8E , F ) . This dynamic can be observed at an individual array of CMT TEs ( Figure 8H ) . 24-nt sRNA expression is confined to the edges of the CMT TE array in wt , but follows H3K9me and DNA methylation throughout the array in h1 plants ( Figure 8H ) . In h1c2c3 , mCH within the array is strongly reduced , but H3K9me is maintained , and sRNA expression exhibits a broadly wt pattern associated with remaining mCHH but not with H3K9me ( Figure 8H ) . It is important to note that in plants lacking CMT2/3 , all mCHH should be catalyzed by RdDM , and a correlation between sRNA ( product of the Pol IV pathway ) and mCHH ( product of the Pol V pathway ) is therefore expected regardless of how Pol IV is recruited . The key observations are that loss of CMT2/3 in h1c2c3 plants ( and the associated loss of mCHG/mCHH ) largely abrogates the relocation of Pol IV activity into heterochromatin ( Figure 8G , H and Figure 8—figure supplement 1E , G ) , and the remaining heterochromatic sRNA biogenesis is not associated with H3K9me or mCG ( Figure 8D–F ) . These results do not support the hypothesis that Pol IV is recruited by H3K9me , and offer mCH as the most likely alternative . Our data suggest the hypothesis that without H1 , mCH catalyzed by CMT2/3 pulls Pol IV into heterochromatin , and loss of CMT2/3 allows Pol IV to return to its mostly euchromatic wt targets . 24-nt sRNA expression is globally associated with mCH rather than H3K9me in h1c2c3 , but these correlations are primarily driven by heterochromatic regions with low wt RdDM . To determine if this trend translates to euchromatic TEs where SHH1 is required for RdDM , we analyzed associations between H3K9me , DNA methylation , and sRNA expression in published CLSY1/2 sRNA clusters in wt plants ( Figure 9A; Zhou et al . , 2018 ) . In clusters grouped by H3K9me and mCHH , sRNA expression is associated with high mCHH , but not with high H3K9me ( Figure 9A ) , supporting the idea that mCH dictates Pol IV localization ( with the caveat that mCH is a product of RdDM ) . As a further test of our hypothesis , we analyzed published data from plants lacking the three H3K9 methyltransferases implicated in the CMT/SUVH positive feedback loop . In these suvh4/5/6 mutants , H3K9me2 and mCH are strongly diminished and sRNA expression of CLSY1/2 clusters is decreased ( Stroud et al . , 2014; Zhou et al . , 2018 ) . If H3K9me2 recruits Pol IV via SHH1 , the limited remaining H3K9me would be expected to correlate with sRNA . Instead , we find sRNA expression in suvh4/5/6 follows mCHH but not H3K9me2 ( Figure 9B , C , compare left and right elements in Figure 9C ) , consistent with our observations in heterochromatin . 24-nt sRNA correlates much more strongly with mCH than with H3K9me2 in suvh4/5/6 plants ( Figure 9D ) , highlighting the limited importance of H3K9me for sRNA biogenesis . Finally , we assayed CLSY1/2 clusters with low wt H3K9me2 but high wt sRNA and mCHH ( LH CLSY1/2 clusters ) in polv mutants to determine whether mCH is required to maintain sRNA expression . RNA Pol V is not directly involved in sRNA production , but is an essential RdDM component required for DNA methylation because it recruits DRM2 ( Erdmann and Picard , 2020; Matzke and Mosher , 2014; Raju et al . , 2019; Wendte and Pikaard , 2017 ) . Therefore , polv mutants allow us to differentiate mCH as a cause vs . a consequence of Pol IV activity . 90% of the 662 LH CLSY1/2 clusters lose mCHH in polv plants ( mCHH <0 . 05 , Figure 9E ) , and the overall mCH of LH CLSY1/2 clusters is greatly reduced without Pol V ( Figure 9F ) . In suv4/5/6 mutants , LH CLSY1/2 clusters maintain sRNA expression , whereas sRNA expression in polv mutants is greatly reduced ( Figure 9G ) . Furthermore , mCG at LH CLSY1/2 clusters is higher in polv than in suvh4/5/6 plants ( Figure 9H ) . Therefore , sRNA biogenesis is not sensitive to the loss of either H3K9me2 or mCG and specifically requires mCH .
We have examined intertwined chromatin features – sRNA production , DNA methylation , and H3K9 methylation – to understand how the genomic sites of Pol IV activity are specified . We find that two main factors are involved . First , linker histone H1 prevents sRNA production in heterochromatin ( Figure 10 ) . Without H1 , RdDM relocates from its usual euchromatic targets into heterochromatic TEs ( Figure 1 and Figure 1—figure supplement 1 ) , as has been recently observed by an independent study ( Papareddy et al . , 2020 ) . Another heterochromatic protein , the histone variant H2A . W , may also contribute to the exclusion of RdDM from heterochromatin , but this effect is modest and only observed when H1 is absent ( Bourguet et al . , 2021 ) . In the presence of H1 , lack of H2A . W instead strengthens the exclusion of RdDM from heterochromatin , potentially due to enhanced heterochromatic H1 accumulation ( Bourguet et al . , 2021 ) . Overall , the available evidence indicates that H1 is the major factor excluding Pol IV from heterochromatin . Second , we find that mCH promotes Pol IV activity ( Figure 10 ) , contrary to the well-established view that Pol IV is recruited by H3K9me ( Erdmann and Picard , 2020; Law et al . , 2013; Raju et al . , 2019; Wendte and Pikaard , 2017; Zhang et al . , 2013 ) , and the more recent proposal that mCG may be involved ( Zhou et al . , 2018 ) . The hypothesis that mCH recruits Pol IV has a long history ( Herr et al . , 2005; Li et al . , 2020; Zemach et al . , 2013 ) , but testing it has been challenging because mCH is associated with other epigenetic and chromatin features , including mCG and H3K9me ( Law and Jacobsen , 2010; Xu and Jiang , 2020; Zhang et al . , 2018b ) . The link with H3K9me has been particularly difficult to break because of the CMT-SUVH feedback loop ( Du et al . , 2012; Johnson et al . , 2007; Li et al . , 2018; Stoddard et al . , 2019 ) . However , we have used h1c2c3 , suvh4/5/6 and polv mutants to disentangle H3K9me and mCH . In all three backgrounds , sRNA biogenesis follows mCH instead of H3K9me ( Figures 8 and 9 and Figure 8—figure supplement 1 ) . The h1c2c3 line has been particularly informative due to the many TEs that maintain H3K9me but lack mCH ( Figure 8 and Figure 8—figure supplement 1 ) . H3K9me may be substantially retained in h1c2c3 heterochromatin because lack of H1 allows SUVH methyltransferases easier access , so that the weak affinity of their SRA domains for mCG suffices for effective recruitment ( Johnson et al . , 2007; Li et al . , 2018; Rajakumara et al . , 2011 ) . Whatever the mechanism , the strong linear association between sRNA biogenesis and mCH , and the lack of such an association with H3K9me and mCG ( Figures 8 and 9 ) , provide strong support for the hypothesis that mCH recruits Pol IV ( Figure 10 ) . Our data linking 24-nt biogenesis with mCH do not mean that such methylation is absolutely required for Pol IV recruitment . Indeed , there is residual 24-nt biogenesis in ddcc mutants that lack mCH ( Stroud et al . , 2014 ) . One possibility is that the factor or factors recruiting Pol IV to mCH have weak affinity for mCG , which could recruit Pol IV in the absence of mCH , analogous to our proposed mode of SUVH4/5/6 recruitment in plants lacking CMT2/3 . Other chromatin features may also recruit or facilitate Pol IV activity . However , our results indicate that mCH is the major Pol IV recruiting genomic feature under normal conditions . The linking of Pol IV activity to mCH instead of H3K9me resolves several thorny issues . First , the observation that SHH1 – the proposed H3K9me reader – is preferentially required for RdDM where H3K9me is low ( Zhou et al . , 2018 ) can be easily accommodated if H3K9me is not directly involved in RdDM . Similarly , the finding that severe loss of H3K9me in suvh4/5/6 mutants is accompanied by only a modest reduction of sRNA levels ( Zhou et al . , 2018 ) is no longer mysterious . At a more fundamental level , this hypothesis ties RdDM in a feedback loop with its product and unties it from a histone modification produced by the distinct CMT-SUVH pathway and depleted from RdDM target sequences . Breaking RdDM from dependence on any histone modification is also conceptually important because a core theoretical strength of RdDM is the ability to maintain methylation at much shorter sequences than those where stable histone-based epigenetic inheritance is possible ( Angel et al . , 2011; Lövkvist and Howard , 2021; Ramachandran and Henikoff , 2015; Zilberman and Henikoff , 2004 ) . Long TEs that can be effectively silenced by the histone-dependent CMT-SUVH pathway tend to be relatively GC-rich because they contain coding sequences ( Sequeira-Mendes et al . , 2014; To et al . , 2020; Zemach et al . , 2013 ) . In contrast , short nonautonomous TEs and TE remnants tend to lack coding sequences and are thus AT-rich . In this context , the GC sequence preference of Arabidopsis H1 ( Choi et al . , 2020 ) may be key . GC bias is far from a H1 universal , with most animal H1 variants preferring AT-rich DNA ( Cao et al . , 2013; Izaurralde et al . , 1989; Tomaszewski and Jerzmanowski , 1997 ) . The preferences of plant H1 may have evolved , at least in part , to target it to coding sequences , including those of autonomous heterochromatic TEs . This would allow H1 to exclude RdDM from such sequences , which can cover vast tracts of plant genomes ( Michael , 2014; Suzuki and Bird , 2008 ) , and focus RdDM on the short TEs it is specialized to silence . The interplay of H1 and mCH can thus produce the preferential activity of RdDM at short , AT-rich TEs observed throughout flowering plants ( Gouil and Baulcombe , 2016; Numa et al . , 2015; Tan et al . , 2018 ) .
cmt2 and cmt2cmt3 ( Stroud et al . , 2014; Zemach et al . , 2013 ) plants were crossed to h1 . 1h1 . 2 ( Zemach et al . , 2013 ) plants to generate h1cmt2 and h1cmt2cmt3 plants . To establish the h1cmt2shh1 mutant line , we crossed h1 +/- cmt2 plants with shh1 ( SALK_074540C ) plants , then isolated h1cmt2shh1 homozygous siblings . met1 , h1met1 , ddm1 , and h1ddm1 plants were described previously ( Choi et al . , 2020; Lyons and Zilberman , 2017 ) . Arabidopsis thaliana seedlings were germinated and grown for 4–5 weeks on soil at 20–25℃ in growth chambers ( 16 hr day/8 hr night ) for all the experiments performed except for met1 , h1met1 , and corresponding wt seedling sRNA libraries . These seedlings were germinated and grown for 2 weeks in half-strength Gamborg’s B-5 liquid media ( Caisson Labs , cat . no . GBP07 ) at 22–25℃ under continuous light with shaking at 125 rpm . Bisulfite sequencing ( BS-seq ) libraries were constructed using genomic DNA ( gDNA ) extracted from rosette leaves of 4–5-week-old plants . 500 ng total gDNA was sheared to 100–1000 bp using Bioruptor Pico ( Diagenode ) , then purified with 1 . 2× volume of SPRI beads ( Beckman Coulter , cat . no . A63881 ) . Fragmented gDNA was ligated to NEBNext Adaptor for Illumina using NEBNext Ultra II DNA library prep kit for Illumina ( New England Biolabs , cat . no . E7645 ) . We performed bisulfite conversion twice with ligated libraries ( QIAGEN , cat . no . 59104 ) to prevent incomplete conversion ( <99% conversion ) of unmethylated cytosines . Converted libraries were subjected to SPRI bead purification with 0 . 8× volume of beads . We amplified bisulfite-converted libraries with NEB next indexing primers ( New England Biolabs Inc , cat . no . E7335S ) . To isolate sRNA , we extracted total RNA from rosette leaves of 4–5-week-old plants using Trizol ( Invitrogen , cat . no . 15596026 ) according to the manufacturer’s manual . To remove DNA from samples , 5 μg of RNA was treated with DNA-free DNA removal kit ( Thermo , cat . no . AM1907 ) . 1 μg of DNA-free total RNA was subjected to sRNA library construction according to the manufacturer’s protocol ( Illumina , cat . no . RS-200-0012 and RS-200-0024 ) . MNase digestion of native chromatin was carried out on 0 . 5 g of 4-week-old Arabidopsis rosette leaves as described previously ( Lyons and Zilberman , 2017 ) . Digestion was stopped with EGTA and chromatin was rotated at 4℃ for 30 min . The preparation was then centrifuged for 10 min at 2000 rpm and solubilized chromatin fragments were isolated by aspirating supernatant immediately . Chromatin was then diluted to 1 ml in wash buffer A ( 50 mM Tris–HCl pH 7 . 5 , 50 mM NaCl , 10 mM EDTA ) and antibody added at 1 μl per 0 . 1 g of total starting material ( Millipore , cat . no . 07-450 for H3K9me1 , Abcam , cat . no . ab1220 for H3K9me2 ) . Dilute Tween-20 was added to a final concentration of 0 . 1% , and the mixture was rotated overnight at 4℃ . All buffers were supplemented with PMSF and protease inhibitor ( Roche [Merck] , cat . no . 11873580001 ) . A standard immunoprecipitation procedure was used the following day . Briefly , preblocked Protein-A and -G dynabeads ( Invitrogen , cat . no . 10 , 001D and 10 , 003D ) were incubated with the chromatin preparation for 3 hr . rotating at 4℃ , and the beads/chromatin mixture was then washed on ice in Tris–EDTA buffer with increasing concentrations of NaCl , starting at 50 mM and ending at 150 mM . DNA was eluted from beads by shaking in 1% SDS and 1% NaHCO3 for 10 min at 55℃ , and DNA was purified with phenol–chloroform extraction . Input and ChIP DNA was converted into sequencing libraries using Celero DNA reagents ( Tecan , cat . no . 3460-24 ) following the manufacturer’s instructions . Sequencing was performed at the John Innes Centre with the NextSeq 500 ( Ilumina ) , except for sRNA libraries from seedlings ( wt , met1 , and h1met1 ) . These seedling libraries were sequenced at the Vincent J . Coates Genomic Sequencing Laboratory at the University of California , Berkeley with the HiSeq 4000 ( Illumina ) . For sRNA-seq libraries , adapter sequences were removed from reads using cutadapt ( Martin , 2011 ) . 18–28 bp , 21 nt , and 24 nt fragments were isolated using the following cutadapt options: -m 18 M 28 , -m 21 M 21 , -m 24 M 24 . Reads were mapped with Bowtie ( Langmead et al . , 2009 ) allowing up to one mismatch and up to 10 multimapped reads . Aligned 21-nt or 24-nt read counts were normalized by reads per kilobase per million mapped reads ( rpkm ) of 18–28 bp fragments . ChIP-seq libraries were mapped with Bowtie ( Langmead et al . , 2009 ) allowing up to 2 mismatches and up to 10 multimapped reads . To calculate enrichment , ChIP samples were divided by input samples and transformed into log2 ratio values using deepTools2 bamCompare ( Ramírez et al . , 2016 ) . For H3K9me1 and H3K9me2 from WT , h1 , ddm1 , h1ddm1 , c2c3 , and h1c2c3 , we used a random subset of input reads equivalent to 25% of the total uniquely mapped reads of the corresponding IP for input into bamCompare . For BS-seq libraries , reads were mapped with the bs-sequel pipeline ( https://zilbermanlab . net/tools/ ) . ‘Transposable elements’ include transposon annotation from Panda and Slotkin , 2020 . Araport11 TE genes and pseudogenes , and genomic regions with TE-like DNA methylation ( Cheng et al . , 2017; Choi et al . , 2020; Panda and Slotkin , 2020; Shahzad et al . , 2021 ) . We filtered out elements shorter than 250 bp . Previously , we merged overlapping TE annotations into single TE unit , then defined heterochromatic TEs and euchromatic TEs as transposons that have more than 0 or less than 0 H3K9me2 ( log2 ChIP/Input ) in wt plants ( Choi et al . , 2020 ) . Both CMT and DRMs target these merged , long TEs , as the edges of TEs are methylated by DRMs and the bodies of TEs are methylated by CMTs . Therefore , to isolate TEs with mCH dependent on CMTs or DRMs , we did not merge TE annotations here . Among TEs with mCHH methylation ( mCHH >0 . 02 ) , CMT-dependent TEs were defined as the TEs that lost mCHH methylation in cmt2 plants ( mCHH <0 . 02 in cmt2 ) . DRM-dependent TEs were defined as the TEs that lost mCHH methylation in drm2 plants ( mCHH <0 . 02 in drm2 ) . sRNA cluster annotation is from Zhou et al . , 2018 . We previously defined MET1-dependent TEs as the TEs that lost H3K9me2 in met1 plants ( Choi et al . , 2020 ) . In this study , to evaluate how DNA methylation affects CLSY3/4-dependent sRNA expression , we defined MET1-dependent TEs as the TEs that lost mCHH methylation in met1 ( mCHH in wt ≧0 . 05 , mCHH in met1 <0 . 02 ) , and MET1-independent TEs as ones that keep mCHH methylation in met1 ( mCHH in wt ≧0 . 05 , mCHH in met1 ≧0 . 05 ) . To measure the importance of each genetic and epigenetic marker to classify DRM and CMT TEs , we first calculated average enrichment of various histone modifications , histone H1 , average sRNA expression , and DNA methylation level at each TE using window_by_annotation . pl Perl script ( https://zilbermanlab . net/tools/ ) . We also included density of various cytosine sequence contexts . The importance of each variable was evaluated using ‘randomForest’ and ‘measure_importance’ function in RandomForestExplainer R package ( Ishwaran et al . , 2012 ) . The importance matrices were visualized by ‘plot_multi_way_importance’ function of the same package . To evaluate the predictive power of each variable , we randomly divided TEs into training and validation sets . The random forest classifier was built using TEs in the training set with indicated variables and the classification of each TE ( DRM or CMT ) . The trained model was used to predict the category of TEs in the validation set , and the error rate was calculated by comparing the predicted classification and its actual classification . We used ‘randomforest’ and ‘predict’ function in randomForest R package . Enrichment scores of various genomic and epigenomic features were generated by window_by_annotation . pl Perl scripts ( https://zilbermanlab . net/tools/ ) . For scatter plots and heatscatter plots in Figure 1 , the enrichment scores were imported to R ( Davey et al . , 1997 ) and visualized by ggplot2 R package ( Wickham , 2009 ) or ‘heatscatter’ function in LSD R package ( Venables and Ripley , 2002 ) . For scatter plots and heatscatter plots in other figures , TEs were sorted by their GC content , then average feature enrichments of 100 TEs were calculated to reduce the variability of data . DNA methylation , H3K9 methylation , and sRNA distribution around TEs were generated with ends_analysis . pl and average_ends_new . pl Perl scripts ( https://zilbermanlab . net/tools/ ) . For sRNA distribution , we removed bins with higher than 200 rpkm to prevent outliers skewing the average . For proportional Venn diagram , TE ID lists in each group were uploaded to BioVenn ( Hulsen et al . , 2008 ) . To visualize the relationship among genetic , epigenetic features and sRNA expression in c2c3 and h1c2c3 plants , principal component analysis was applied to arrays of features using Gene Cluster 3 . 0 ( de Hoon et al . , 2004; Figure 6C ) . For Pearson’s correlation coefficient plots , the DNA methylation , H3K9 methylation , and sRNA expression level matrices were imported to R and visualized using corrplot R package ( Friendly , 2002; Murdoch and Chow , 1996; Figures 7 and 8 ) . Screenshots of Arabidopsis genomic loci were taken in IGV ( Robinson et al . , 2011; Thorvaldsdóttir et al . , 2013 ) . Treeview was used to generate heatmaps ( de Hoon et al . , 2004 ) . For sRNA plots around nucleosomes ( Figure 2 ) , previously published nucleosome dyad coordinates were used ( Lyons and Zilberman , 2017 ) as anchors around which 10 bp bins of 24-nt sRNA were averaged and plotted . Autocorrelation estimates were generated on these averages using the built-in R ‘acf’ function . DNA methylation data of wt , drm2 , c2c3 , ddcc , and ibm1 plants ( Stroud et al . , 2014; Zemach et al . , 2013 ) , DNA methylation and sRNA data of clsy1/2 , clsy3/4 , and shh1 plants ( Zhou et al . , 2018 ) , DNA methylation , MNase , well-positioned nucleosome loci data of wt and h1 plants ( Lyons and Zilberman , 2017 ) , DNA methylation , H1 and H3K9me data of wt , met1 , and h1met1 plants ( Choi et al . , 2020 ) , H3K9me2 and sRNA expression data of wt and ibm1 plants ( Fan et al . , 2012; Lai et al . , 2020 ) , DNA methylation , H3K9me2 , and sRNA expression data of suvh4/5/6 plants ( Papareddy et al . , 2020; Stroud et al . , 2014 ) , and DNA methylation and sRNA data of polv plants ( Johnson et al . , 2014; Zhong et al . , 2012 ) were obtained through GEO ( GEO accessions: GSE51304 , GSE41302 , GSE99694 , GSE122394 , GSE108487 , GSE32284 , GSE152971 , GSE52041 , and GSE39247 ) . | Cells adapt to different roles by turning different groups of genes on and off . One way cells control which genes are on or off is by creating regions of active and inactive DNA , which are created and maintained by different groups of proteins . Genes in active DNA regions can be turned on , while genes in inactive regions are switched off or silenced . Silenced DNA regions also turn off ‘transposable elements’: pieces of DNA that can copy themselves and move to other regions of the genome if they become active . Transposons can be dangerous if they are activated , because they can disrupt genes or regulatory sequences when they move . There are different types of active and inactive DNA , but it is not always clear why these differences exist , or how they are maintained over time . In plants , such as the commonly-studied weed Arabidopsis thaliana , there are two types of inactive DNA , called E and H , that can silence transposons . In both types , DNA has small chemicals called methyl groups attached to it , which help inactivate the DNA . Type E DNA is methylated by a process called RNA-directed DNA methylation ( RdDM ) , but RdDM is rarely seen in type H DNA . Choi , Lyons and Zilberman showed that RdDM is attracted to E and H regions by previously existing methylated DNA . However , in the H regions , a protein called histone H1 blocks RdDM from attaching methyl groups . This helps focus RdDM onto E regions where it is most needed , because E regions contain the types of transposons RdDM is best suited to silence . When Choi , Lyons and Zilberman examined genetically modified A . thaliana plants that do not produce histone H1 , they found that RdDM happened in both E and H regions . There are many more H regions than E regions , so stretching RdDM across both made it less effective at silencing DNA . This work shows how different DNA silencing processes are focused onto specific genetic regions , helping explain why there are different types of active and inactive DNA within cells . RdDM has been studied as a way to affect crop growth and yield by altering DNA methylation . These results may help such studies by explaining how RdDM is naturally targeted . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"plant",
"biology"
] | 2021 | Histone H1 prevents non-CG methylation-mediated small RNA biogenesis in Arabidopsis heterochromatin |
Fibrosis of organs is observed in systemic autoimmune disease . Using a scleroderma mouse , we show that transplantation of MHC compatible , minor antigen mismatched bone marrow stromal/stem cells ( BMSCs ) play a role in the pathogenesis of fibrosis . Removal of donor BMSCs rescued mice from disease . Freshly isolated PDGFRα+ Sca-1+ BMSCs expressed MHC class II following transplantation and activated host T cells . A decrease in FOXP3+ CD25+ Treg population was observed . T cells proliferated and secreted IL-6 when stimulated with mismatched BMSCs in vitro . Donor T cells were not involved in fibrosis because transplanting T cell-deficient RAG2 knock out mice bone marrow still caused disease . Once initially triggered by mismatched BMSCs , the autoimmune phenotype was not donor BMSC dependent as the phenotype was observed after effector T cells were adoptively transferred into naïve syngeneic mice . Our data suggest that minor antigen mismatched BMSCs trigger systemic fibrosis in this autoimmune scleroderma model .
Systemic fibrosis is a feature of autoimmune disease such as systemic sclerosis ( SSc ) or Sjögren’s syndrome involving exocrine glands ( Ferrara et al . , 2009; Filipovich et al . , 2005 ) . A mouse model for human SSc reported by Zhang et . al . involves transplantation of B10 . D2 bone marrow into MHC matched , minor antigen mismatched BALB/c host ( Zhang et al . , 2002 ) . This model of SSc occurs spontaneously without the use of artificial agents such as bleomycin ( Yamamoto and Nishioka , 2004 ) , and exhibits characteristics of human SSc including fibrosis , inflammation , and autoimmunity . Animal models are effective in screening for therapeutic interventions such as anti IL-6 ( Le Huu et al . , 2012 ) and angiotensin II type-1 receptor antagonists ( Yaguchi et al . , 2013 ) . However , such a spontaneous model is also a valuable tool for investigating the pathogenesis of SSc , which is still largely unknown . In order to shed light onto the mechanisms leading to fibrosis in this SSc mouse model , it is necessary to isolate the different cellular fractions within the B10 . D2 donor bone marrow , namely , hematopoietic stem cells ( HSCs ) and bone marrow stromal/stem cells ( BMSCs ) . Multipotent BMSCs in the bone marrow differentiate into several mesenchymal lineages including fibroblasts , adipocytes , osteocytes , and chondrocytes ( Pittenger et al . , 1999; Prockop , 1997 ) . However , due to the lack of specific markers , a crucial step involving in vitro expansion was required to isolate BMSCs , which may modify their phenotype and function ( Banfi et al . , 2000 ) . Most current information on BMSCs comes from such in vitro studies of adherent cells referred to as fibroblast CFUs ( CFU-Fs ) ( Conget and Minguell , 1999; Friedenstein et al . , 1974; Pittenger et al . , 1999; Prockop , 1997 ) , which are a heterogeneous population of cells at best . Therefore , the in vivo dynamics of BMSCs after whole bone marrow transplantation ( WBMT ) are still unknown , and the establishment of a solid experimental system to trace the fate of BMSCs following transplantation was required . In order to establish an animal model with traceable donor BMSCs and HSCs , we applied our previously reported method for prospectively isolating murine BMSCs based on their expression of PDGF receptor α and Sca-1 ( PDGFRα+/ Sca-1+ ( PαS ) cells ) ( Morikawa et al . , 2009a ) . Selectively isolated PαS-BMSCs without in vitro expansion represents highly clonogenic and multi-potent population of cells including hematopoietic niche cells , osteoblasts , and adipocytes after systemic in vivo transplantation ( Morikawa et al . , 2009a; 2009b ) . Our model allows for the first time both whole bone marrow transplantation , as well as the selective transplantation of freshly isolated BMSCs and/or HSCs into recipient mice . By applying this modified SSc model using prospectively isolated BMSCs and HSCs , we sought to identify the role of donor HSCs and BMSCs in the pathogenesis of the autoimmune-related fibrosis in SSc . Here , we show how mismatched donor BMSCs not only contribute to fibrosis in various organs , but also trigger the onset of autoimmune disease by activating host T cells .
A known bone marrow transplantation ( BMT ) model using 8-week-old donor B10 . D2 ( H-2d ) mice and recipient BALB/c mice ( H-2d ) , is a MHC-compatible , minor histocompatibility antigen ( miHA ) –incompatible model of systemic fibrosis ( Figure 1—figure supplement 1A ) ( Zhang et al . , 2002 ) . Signs of fibrosis appear by 3 weeks after BMT and progresses to a full-blown disease by 8 weeks characterized by excessive fibrosis of the lacrimal gland , conjunctiva , salivary gland , skin , lung , liver , and intestine ( Figure 1—figure supplement 1B , C ) . In order to elucidate the source and role of donor-derived fibroblasts , we first modified the SSc model by co-transplanting prospectively isolated HSCs with prospectively isolated BMSCs . PαS-BMSCs and side population ( SP-HSCs ) from B10 . D2 or BALB/c were individually isolated ( Figure 1A and Figure 1—figure supplement 2 ) . We confirmed that PαS BMSCs cells expressed CD29 , CD90 , and CD106 , known as established markers of BMSCs . However , CD31 and CD133 , markers of endothelial cells , consisted of only a small portion of the PαS BMSCs fraction ( Figure 1—figure supplement 2C ) . These PαS-BMSCs and side population ( SP-HSCs ) from B10 . D2 or BALB/c were systemically co-transplanted into BALB/c recipients as defined combinations ( Figure 1Bi–iv ) . Unlike the original model using WBM , the graft does not include donor type mature hematopoietic and non-hematopoietic cells , and the cells were provided from each lineage of stem cell after bone marrow engraftment . 10 . 7554/eLife . 09394 . 003Figure 1 . Modified SSc model by co-transplanting isolated HSCs and BMSCs . ( A , B ) Transplantation model with different combinations of syngeneic ( white ) and mismatched ( black ) HSCs and BMSCs co-transplanted into BALB/c mice . ( B , i–v ) ( i ) Negative control , ( ii ) mismatched BMSC transplantation , ( iii ) syngeneic BMSC transplantation , ( iv ) positive control , and ( v ) mismatched MSC-depleted WBMT . Excessive fibrosis ( deep blue , and ∗ ) in various organs was observed in mismatched BMSC transplanted mice after 3 weeks . Double arrows indicate epidermal and dermal thickness . The fibrotic areas were assessed as the ratio of the blue-stained area per field . D , duct . Minimal Mallory staining areas ( ▲ ) in ( iii ) and ( v ) are physiological changes and necessary to support the structure of ducts and intestinal walls . Scale bar , 100 μm ( 50 μm in liver ) . ( C ) HSP47+ fibroblasts in the lacrimal glands were significantly higher following mismatched BMSC transplantation ( red ) compared to syngeneic BMSC ( blue ) , and BMSC-depleted WBM transplantation ( green ) . The number of HSP47+ cells in negative control ( i ) ( blue ) , and syngeneic MSC transplantation ( iii ) ( blue ) , and mismatched MSC transplantation ( ii ) ( red ) , and positive control ( iv ) ( red ) , and mismatched MSC-depleted WBMT ( green ) . Data are shown as mean ± SD , #p<0 . 05 , *p<0 . 01 , **p<0 . 001 . ( D ) Tear volume in the same groups described in ( C ) . Data are shown as mean ± SD , n = 2-5 per group , *p<0 . 05 . ( E , F ) The degree of fibrosis ( blue ) in the lacrimal glands of mismatched BMSC recipients was dose dependent . Excessive fibrotic areas are shown in deep blue ( ∗ ) . Data are shown as mean ± SD , n = 3 . Scale bar , 100 μm . *p<0 . 001 . BMSCs , bone marrow stromal/stem cells; HSCs , hematopoietic stem cells; SD , standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 00310 . 7554/eLife . 09394 . 004Figure 1—source data 1 . HSP47+ cells/ field in the lacrimal glands . Number of HSP47+ cells per field ( x200 ) from 5 to 13 areas from the lacrimal glands 3 and 8 weeks following transplantation . Source data for ( C ) . HSP , heat-shock protein . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 00410 . 7554/eLife . 09394 . 005Figure 1—source data 2 . Tear volume following transplantation . Source data for graph shown in ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 00510 . 7554/eLife . 09394 . 006Figure 1—source data 3 . The number of HSP47+ cells/field in the lacrimal glands after transplantation of 500 , 2000 , 8000 , and 16 , 000 syngeneic or mismatched BMSCs shown in ( F ) . HSP , heat-shock protein . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 00610 . 7554/eLife . 09394 . 007Figure 1—source data 4 . Number of HSP47+ cells/ field from in various target organs after whole bone marrow transplantation . Number of HSP47+ cells per field from 4 to 5 areas in each organ 3 and 8 weeks following WBM for Figure 1—figure supplement 1 ( C ) . HSP , heat-shock protein . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 00710 . 7554/eLife . 09394 . 008Figure 1—source data 5 . HSP47+ cells per field in the salivary gland , skin , lung , liver , and intestine shown in Figure 1—figure supplement 3B . Data for each organ are displayed on separate sheets . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 00810 . 7554/eLife . 09394 . 009Figure 1—figure supplement 1 . Autoimmune-associated fibrosis following whole bone marrow transplantation in mouse model . ( A ) Schematic diagram of MHC-compatible , multiple minor antigen ( miHA ) –incompatible model of scleroderma . BMMNCs: bone marrow mononuclear cells . ( B ) Histology of BALB/c recipients transplanted with syngeneic ( BALB/c ) or mismatched ( B10 . D2 ) WBM at 3 and 8 weeks after transplantation . Fibrosis shown by blue Mallory staining is more prominent in target organs of fibrosis . Representative data from 10 independent experiments ( n = 4–5 per group ) are shown . Scale bar , 100 μm ( liver , 50 μm ) . Excessive fibrotic areas are shown in deep blue ( ＊ ) . ( C ) HSP47+ fibroblasts in the lacrimal glands , conjuntiva , salivary glands , skin , lung , and intestine were significantly higher following mismatched whole bone marrow transplantation ( red ) compared to syngeneic whole bone marrow transplantation ( blue ) . Data are shown as mean ± SD . #p<0 . 05 , *p<0 . 01 . HSP , heat-shock protein; SD , standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 00910 . 7554/eLife . 09394 . 010Figure 1—figure supplement 2 . Flow cytometry protocol for Isolating BMSCs and HSCs . ( A , B ) PαS-BMSCs ( A ) and SP-HSCs ( B ) were isolated from BMMNCs by flowcytometry as shown . ( C ) Characterization of PαS BMSCs by other BMSCs marker CD29 , CD90 , and CD106 and endothelial marker , CD31 and CD133 by flowcytometry . BMSCs , bone marrow stromal/stem cells; BMMNC , bone marrow mononuclear cells; HSCs , hematopoietic stem cells . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 01010 . 7554/eLife . 09394 . 011Figure 1—figure supplement 3 . Modified SSc model by co-transplanting isolated HSCs and BMSCs in other target organs . ( A ) Mallory staining of the salivary gland , skin , lung , liver , and intestine tissue sections of BALB/c mice that received mismatched ( ii ) or syngeneic ( iii ) BMSC transplantation and negative ( i ) or positive control ( iv ) at 3 weeks after transplantation . Mismatched BMSCs depletion from WBM transplantation ( v ) . Excessive fibrotic areas are shown in ( * ) . ( B ) The number of HSP47+ fibroblasts in negative control ( i ) ( blue ) , and syngeneic BMSC transplantation ( iii ) ( blue ) , and mismatched BMSC transplantation ( ii ) ( red ) , and positive control ( iv ) ( red ) , and mismatched BMSC-depleted WBMT ( green ) . Double arrows in skin indicate the thickness of epidermal region leading to the loss of adipose tissue in the mismatched BMSC group . Error bars indicate SD . #p< 0 . 05 , *p< 0 . 01 , **p< 0 . 001 . Scale bar , 100 μm . BMSCs , bone marrow stromal/stem cells; HSCs , hematopoietic stem cells; HSP , heat-shock protein; SSc , systemic sclerosis . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 01110 . 7554/eLife . 09394 . 012Figure 1—figure supplement 4 . Mallory staining of normal organs . Minimally blue stained areas by Mallory staining ( ▲ ) in normal tissues including the lacrimal gland , conjunctiva , salivary gland , skin , lung , liver , and intestine . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 012 Notably , progressive fibrosis was only observed in mice receiving mismatched BMSCs ( Figure 1Bii , iv and Figure 1—figure supplement 3Aii , iv ) . The recipients of mismatched HSCs combined with syngeneic BMSCs were indistinguishable from syngeneic HSC and syngeneic BMSC recipients . Physicological connective tissues are also stained in blue in the lacrimal gland , salivary gland , and intestine when BALB/c mice received mismatched HSCs and syngeneic BMSCs ( Figure 1Bi , iii and Figure 1—figure supplement 3Ai , iii ) . Since heat-shock protein 47 ( HSP47 ) , a molecular chaperon specific to collagen-secreting cells , has been reported as a useful marker for fibroblasts in paraffin-embedded tissue sections ( Kuroda and Tajima , 2004 ) , we determined the state of fibrosis by ( 1 ) semi-quantitative analysis to count the number of HSP47+ fibroblasts per field and ( 2 ) Mallory staining , a specific marker of connective tissue ( Figure 1B , C and Figure 1—figure supplement 3A , B ) ( Brack et al . , 2007 ) . Fibrotic lesions in mismatched BMSC recipients had significantly higher HSP47+ fibroblasts ( Figure 1C ) . Lacrimal gland function was also reduced only in mismatched BMSC recipients ( Figure 1D ) . In addition , all these observations were canceled in recipients of BMSC-depleted WBM transplantation ( Figure 1Bv , Cv , Dv and Figure 1—figure supplement 3Av , Bv ) . Minimal Mallory staining areas are physiological changes and necessary to support the structure of ducts and intestinal walls ( Figure 1—figure supplement 4 ) . The severity of fibrosis triggered by mismatched BMSCs was dose dependent on the number of transplanted BMSCs ( Figure 1E , F ) . These results suggest that mismatched donor BMSCs or their progeny triggered the onset of disease . Cells double positive for HSP47 and enhanced green fluorescent protein ( EGFP ) were detected in fibrotic tissue when mismatched EGFP+B10 . D2 PαS-BMSCs were transplanted with wild-type B10 . D2 SP cells ( Figure 2A ) . Approximately 42 . 9 ± 7 . 3% of the HSP47+ cells in fibrotic tissue were GFP+ , indicating that the majority of fibroblasts were derived from donor PαS-BMSCs ( Figure 2A ) . On the other hand , a notable number of cells derived from mismatched donor EGF+SP-HSCs also migrated to fibrotic lesions , but none of the GFP+ cells expressed HSP47 ( Figure 2B ) . The lineage exclusivity of this transplantation excludes the possibility that HSP47+ fibroblasts were derived from SP-HSCs as shown in our previous report ( Koide et al . , 2007 ) . The accumulation of donor-derived fibroblasts was not observed in syngeneic transplantation , where fibrosis does not occur ( Figure 2C ) . HSP47 expression is clearly seen especially in activated fibroblasts ( Figure 2A ) , but very faint expression is detected in quiescent fibroblasts in the syngeneic BMSCs transplanted recipients’ target organs ( Figure 2C ) . These results suggest that mismatched BMSCs migrate into the target organs and proliferate under pathological microenvironment . 10 . 7554/eLife . 09394 . 013Figure 2 . Mismatched donor derived-fibroblasts engrafted in target tissues of fibrosis . ( A , B ) Transplantation scheme of EGFP+-labeled B10 . D2 PαS-BMSCs ( A ) or SP-HSCs ( B ) . Each was co-transplanted into wild-type BALB/c mice along with unlabeled SP-HSCs ( A ) or PαS-BMSCs ( B ) , respectively . Arrows indicate colocalized cells in yellow ( GFP-labeled BMSCs expressed HSP47 ) in the lacrimal gland and intestine . ( C ) Syngeneic wild-type B10 . D2 SP-HSCs and EGFP+-labeled B10 . D2 PαS-BMSC co-transplantation into wild B10 . D2 mice . HSP47+ ( red ) fibroblasts observed in mismatched BALB/c recipients were BMSC-derived , and not HSC cells . Nuclei were counter stained with DAPI ( blue ) . The data in Figure 2A–C from two replicate experiments ( n = 3 per group ) . Scale bar , 20 μm . ( A–C ) D , duct . ( D ) Cultured HSP47+ lacrimal gland fibroblasts after mismatched EGFP+ HSC transplantation were EGFP- ( left ) , while the majority of fibroblasts after mismatched EGFP+ BMSC transplantation were EGFP+ fibroblasts ( right , positive cells in yellow ) indicating the donor BMSC origin of fibroblasts . ( E ) Donor–derived EGFP+ cells were observed in the spleen 3 weeks after syngeneic EGFP+ WBMT ( left ) , while engraftment of donor cells were sparse following mismatched WBMT ( right ) , indicating a number of residual host cells remained after mismatched WBMT . Figure 2D , E from representative data of two replicate experiments ( n = 2 or 3 per group ) . Scale bar , D = 50 μm , E = 20 μm . BMSC , bone marrow stromal/stem cells; HSP , heat-shock protein; HSCs , hematopoietic stem cellsDOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 01310 . 7554/eLife . 09394 . 014Figure 2—source data 1 . Percentage of donor–derived EGFP+ cells in the spleen 3 weeks after EGFP+ WBMT . Source data for graph in right panel of Figure 2E . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 01410 . 7554/eLife . 09394 . 015Figure 2—figure supplement 1 . Mismatched PαS-BMSCs do not induce sclerodermatous fibrosis in recipient nude mice . ( A ) Transplantation scheme of B10D . 2 ( Mismatched ) PαS-BMSCs and BALB/c-nu/nu ( Syngeneic ) SP-HSCs into nu/nu or wild type BALB/c . ( B ) Mallory staining of the lacrimal gland , salivary gland , and skin in BALB/c-nu/nu ( top ) and wild-type ( bottom ) recipient mice lacrimal gland and salivary gland at 3 weeks and skin at 8 weeks after transplantation . Data collected from two replicate experiments ( n = 3 per group ) . Scale bar , 50 μm . * , areas of lymphocytic infiltration and excessive fibrosis in deep blue , double arrow; thickness of epidermis . BMSC , bone marrow stromal/stem cells; HSC , hematopoietic stem cellsDOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 015 In vitro culture also confirmed that the majority of fibroblasts from lacrimal gland cultures were derived from mismatched EGFP+ BMSC ( Figure 2D , right ) , whereas no EGFP+ fibroblasts were observed after mismatched EGFP+ HSC transplantation ( Figure 2D , left ) . This suggests that the migration of donor-derived fibroblasts to peripheral organs is one of the characteristic phenomena associated with pathogenesis or progression of this disease . The progression of fibrosis in this mouse model is T-cell-dependent as evidenced by the fact that fibrosis was not observed when BALB/c-nu/nu mice were used as recipients of mismatched B10 . D2 BMSCs transplantation ( Figure 2—figure supplement 1 ) . The next question was whether these pathogenic T cells were donor-derived T cells , or residual host T cells that lost self-tolerance . Both host-derived T cells and donor HSC-derived T cells are believed to exist in this mouse model , because mature T cells often escape radiation damage and remain in the peripheral blood of recipients ( Anderson et al . , 2004 ) . Interestingly , almost all spleen cells were donor-derived in matched WBMT , whereas the remaining host-derived cells after mismatched WBMT were higher in mismatched WBMT ( Figure 2E ) . To determine the origin of T cells , we used BALB/c-RAG2KO mice as recipients or donors of HSCs . In the former lacking recipient T cells ( Figure 3A ) , the fibrosis was not observed ( Figure 3A , C ) . However , donor RAG2KO HSCs lacking T cells combined with mismatched BMSCs in the latter ( Figure 3B ) induced fibrosis with increased HSP47+ fibroblasts in wild-type BALB/c recipients ( Figure 3B , C ) . In addition , we found that recipient-derived Th 17 cells were in proximity of donor-derived BMSCs ( Figure 4A , left ) and that donor BMSCs produce IL-6 ( Figure 4B , left ) around the ducts of the lacrimal gland in BALB/c recipient mice at 8 weeks after mismatched BMSC transplantation . In comparison , mismatched donor HSC did not produce Th 17 cells ( Figure 4A , right ) or IL-6 ( Figure 4B , right ) . 10 . 7554/eLife . 09394 . 016Figure 3 . Host T cells are required for the progression of fibrosis . ( A ) Mallory staining of recipient RAG2KO organs shows that transplantation of mismatched BMSCs did not induce fibrosis in the absence of recipient T cells . Scale bar , 100 μm . ( B ) Tissue inflammation and excessive fibrosis in deep blue ( * ) was observed in BALB/c recipient mice after B10 . D2 BMSC and RAG2KO HSC transplantation , despite the lack of donor T cells . Data collected from two replicate experiments ( n = 3 per group ) . Scale bar , 100 μm . ( A , B ) D , duct; Ac , Acinus . ( C ) : Significantly higher number of HSP47+ fibroblasts was observed in BALB/c recipients after B10 BMSC + RAG2KO HSC transplantation ( red ) compared to RAG2KO recipients after B10 BMSC + BALB/c HSC transplantation ( blue ) . Data from five different fields in two replicate experiments . Data are shown as mean ± SD . *p<0 . 01 , **p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 01610 . 7554/eLife . 09394 . 017Figure 3—source data-1 . Number of HSP47+ cells per field from the lacrimal gland , salivary gland , liver , and intestine . Source data for graphs in ( C ) . HSP , heat-shock protein . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 01710 . 7554/eLife . 09394 . 018Figure 4 . Donor BMSCs interact with recipient Th 17 cells and produce IL-6 . ( A , B ) Th 17 cells ( pink in A ) and IL-6-producing cells ( yellow in left panel of Figure 4B , arrows ) were observed in the lacrimal gland of mismatched BMSC transplanted mice 8 weeks after transplantation . Yellow cells in ( B ) are due to co-localization of IL-6 ( red ) and donor BMSCs ( GFP ) , resulting in yellow . Mismatched HSC transplanted mice did not show co-localization of donor cells ( green ) with IL-6 nor IL-17-producing cells . Representative images from two replicate experiments ( n = 3 per group ) . Scale bar , 20 μm . ( C ) CD4+IL-17+ cells comprised more than 50% of splenic cells from B10 . D2 BMSC + RAG2KO HSCs transplanted BALB/c recipients ( left ) , while the ratio was low in spleens from B10 . D2 BMSC + BALB/c HSC transplanted RAG2KO recipients ( right ) . BMSC , bone marrow stromal/stem cells; HSC , hematopoietic stem cells . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 01810 . 7554/eLife . 09394 . 019Figure 4—figure supplement 1 . GFP donor MSCs produce IL-6 . Split images of IL-6 immunostaining and GFP+ donor MSC/ HSC into BALB/c Mismatched MSC-transplanted recipient lacrimal gland ( left ) and mismatched HSC-transplanted recipient lacrimal gland ( right ) from Figure 4B . D , Duct . Scale bar , 20 μm . HSC , hematopoietic stem cellsDOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 019 In spleen cells , flow cytometric analysis revealed a significant increase of recipient-derived CD4+Th 17 cells when donor BMSCs with RAG2KO HSCs were transplanted into BALB/c recipients ( Figure 4C , Left ) . In contrast , CD4+Th 17 cells were scarcely detected when donor BMSCs with wild-type BALB/c HSCs were transplanted into RAG2KO recipient ( Figure 4C , Right ) . To further confirm the function of mismatched PαS-BMSC-transplanted recipient-derived T cells , we adoptively transferred splenic T cells from mismatched BMSC-transplanted recipient into naive nude mice ( BALB/c background ) . We found the primary inflammation and following fibrosis occurred in the target organs including the lacrimal glands , salivary glands , skin , lung , liver , and intestine ( Figure 5A , Figure 5—figure supplement 1 ) . The number of HSP47+ fibroblasts per field was significantly increased in the adoptively transferred nude mice ( Figure 5B ) compared with the wild-type control . Splenic PBMC of the adoptive transferred recipients revealed that CD4+Th 17 cells were markedly elevated ( Figure 5C , Left ) compared to the wild-type BALB/c background nude mice ( Figure 5C , Right ) . We conducted IL-17 ELISA using the serum samples from the same mice of the experiments shown in Figure 4C . IL-17 concentration significantly increased in the serum from the adoptive transferred recipients group compared with the control group ( Figure 5D ) . 10 . 7554/eLife . 09394 . 020Figure 5 . Adoptive transfer of recipient T cells from mismatched BMSC- transplanted recipients into BALB/c background Nude mice induce disease . ( A ) Adoptive transfer of BALB/c T cells from mismatched BMSC-transplanted mice induced excessive fibrosis accompanied by numerous inflammatory cells in the lacrimal gland of naive nude mice , as shown by Mallory staining ( excessive fibrotic area in deep blue [*] ) . Scale bar , 100 μm . Data collected from two replicate experiments ( n = 4 per group ) . ( B ) The number of HSP47+ fibroblasts was significantly higher in various target organs following adoptive transfer of BALB/c T cells from mismatched BMSC recipients into nude mice ( red ) , compared to wild-type ( WT ) nude mice ( blue ) . Data are shown as mean ± SD . *p<0 . 005 , **p<0 . 001 . ( C ) CD4+ Th 17+ splenic cells were markedly elevated from adoptively transferred nude mice ( left ) , compared to WT BALB/c background nude mice ( right ) ( n = 4 each ) . ( D ) 1L-17 concentration in the serum from the same mice of the experiments shown in Figure 4C ( n = 4 each ) . Data from one of two independent experiments ( A , D ) . BMSC , bone marrow stromal/stem cells; HSP , heat-shock protein . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 02010 . 7554/eLife . 09394 . 021Figure 5—source data 1 . Number of HSP47+ cells in various target organs following adoptive transfer of BALB/c T cells from mismatched BMSC recipients into nude mice . Data from the lacrimal gland , conjunctiva , salivary gland , lung , skin , liver , and intestine as shown in ( B ) . BMSC , bone marrow stromal/stem cells; HSP , heat-shock protein . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 02110 . 7554/eLife . 09394 . 022Figure 5—source data 2 . 1L-17 concentration in the serum from adoptively transferred nude mice , compared to WT BALB/c background nude mice . Source data for graph in ( D ) . WT , wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 02210 . 7554/eLife . 09394 . 023Figure 5—figure supplement 1 . Adoptive transfer of recipient T cells from mismatched BMSC-transplanted recipients into BALB/c background Nude mice induce disease in target organs other than the lacrimal glands . ( A ) Mallory staining of the conjunctiva , salivary gland , skin , lung , liver , and intestine tissue sections of BALB/c mice that received adoptive transfer and control . Excessive fibrotic areas are shown in deep blue ( ∗ ) . BMSC , bone marrow stromal/stem cellsDOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 023 These results strongly suggest that recipient mature T cells act as an effector for mismatched PαS-BMSCs during the onset of fibrosis . Since EGFP+ BMSC-derived cells were not observed in the cortical region of the thymus within this time frame ( data not shown ) , it is unlikely that escape of donor-derived T cells from negative selection by the host thymus is involved in the recognition of host tissue as non-self . To confirm that mature host T cells react to mismatched BMSCs , Thy-1+ T cells from various recipients ( mismatched BMSC-transplanted , syngeneic BMSC-transplanted , untreated B10 . D2 or untreated BALB/c mice ) were co-cultured with freshly isolated PαS-BMSCs or splenic dendritic cells ( DCs ) from BALB/c or B10 . D2 mice . T cells derived from mismatched BMSC-transplanted recipients proliferated and produced IL-6 in response to PαS-BMSCs but not DCs ( Figure 6A , B ) . Interestingly , T cells also responded to matched ( BALB/c ) BMSCs ( Figure 6A , B ) , suggesting that recipient T cells acquired the auto-reactive nature via antigen spreading ( Shlomchik , 2007 ) . In addition , either strain-derived BMSCs induced slight IL-6 secretion when reacted with naive T cells from wild-type mice ( Figure 6B ) . Both PαS-BMSCs and T cells from mismatched BMSC-transplanted recipients produce IL-6 ( Figure 6C ) . 10 . 7554/eLife . 09394 . 024Figure 6 . T cells following mismatched BMT are activated by PαS-BMSCs . ( A , B ) T cells isolated from mismatched BMSC-transplanted recipients proliferated when co-cultured with donor BMSCs ( A ) , which was significantly blocked by anti-MHC class II antibody treatment ( D ) . ( B ) Increased IL-6 production was observed following co-culture of T cells from mismatched BMSC-transplanted recipients with donor PαS-BMSCs , but not with splenic dendritic cells ( DCs ) . Color bars indicate source of T cells . Results are from triplicate cultures of two independent experiments in both ( A ) and ( B ) and from quintupulicate of two independent experiments in ( D ) . Data are shown as mean ± SD . *p<0 . 05 , **p<0 . 01 . ( C ) Both CD3+ T cells and Sca-1+ BMSCs produced IL-6 following co-culture described in ( A ) . Dot plots of mismatched BMSCs-transplanted recipient samples are shown in black , and isotype control in light grey . ( E ) The increase in T cell proliferation under co-culture with mismatched BMSCs ( red ) was due to the activation of CD4+ and not CD8+ T cells . Data collected from triplicate cultures of two replicate experiments ( n = 2 per group ) . Data are shown as mean ± SD . **p<0 . 01 . ( F ) MHC class II expression was upregulated in BMSCs after co-culture with T cells from mismatched BMSC-transplanted recipients . BMT , bone marrow transplantation; BMSC , bone marrow stromal/stem cells; SD , standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 02410 . 7554/eLife . 09394 . 025Figure 6—source data 1 . T cell proliferation after co-culturing of donor or recipient BMSCs and splenic dendritic cells ( DC ) . Sheet 1 shows the OD source values for each group in ( A ) . Sheet 2 shows collective data and SD for graph in ( A ) . BMSC , bone marrow stromal/stem cells . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 02510 . 7554/eLife . 09394 . 026Figure 6—source data 2 . IL-6 production following co-culture of T cells from various sources with donor or recipient BMSCs and splenic dendritic cells ( DCs ) . Sheet 1 shows the concentration of IL-6 in each group shown in ( B ) . Sheet 2 shows raw OD values prior to conversion to concentrateon . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 02610 . 7554/eLife . 09394 . 027Figure 6—source data 3 . T cells proliferation blocked by anti-MHC class II antibody treatment . Source data for graph in ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 02710 . 7554/eLife . 09394 . 028Figure 6—source data 4 . CD4+ T cells and CD8+T cells proliferation under co-culture with syngeneic or mismatched BMSCs . Source data for graph in ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 028 Since the reaction was significantly suppressed by anti-MHC-class II antibody ( Figure 6D ) and CD4+ T cells were predominant over CD8+ T cells ( Figure 6E ) , it was suggested that PαS-BMSCs themselves act as antigen-presenting cells via MHC-class II . Flowcytometry also revealed that MHC-class II molecule was expressed in only a rare subset of PαS cells , but the frequency of MHC-class II expressing cells was significantly increased after co-culture with recipient T cells ( Figure 6F ) . These data suggest that mismatched BMSCs and host T cells stimulate each other via unknown but specific antigens expressed in mismatched BMSCs , but not in DCs , presented by MHC class II molecules . We further examined the in vivodynamics of T cells and BMSCs . We found that serum levels of IL-6 increased in recipient mice transplanted with mismatched BMSCs starting at 3 weeks after transplantation ( Figure 7A ) . This coincided with the appearance of PαS-BMSC-derived cells in the peripheral blood of recipients starting at 3 weeks after transplantation and gradually increasing up to 7 weeks ( Figure 7B ) . Interestingly , significant number of the PαS-BMSC-derived cells in the peripheral blood expressed MHC class II antigens , whereas the vast majority of them were negative prior to transplantation ( Figure 6F , 7B ) . 10 . 7554/eLife . 09394 . 029Figure 7 . Autoimmune phenotype following mismatched BMSC transplantation . ( A ) Serum IL-6 concentration increased after mismatched BMSC transplantation ( red ) compared to syngeneic control ( blue ) . Duplicate experiments . Data are shown as mean ± SD , n = 2 . ( B ) GFP+ donor BMSCs appear in peripheral blood mononuclear cells after mismatched BMSCs transplantation , and peaks at approximately 7 weeks . The percentage of BMSCs expressing MHC class II antigen increases following transplantation . ( C ) CD4+ CD25+ Foxp3+ Tregs were suppressed in the spleen after mismatched WBMT transplantation ( right ) compared to syngeneic control ( left ) . ( D ) CD4+CD25+Foxp3+ Tregs were suppressed in both mismatched WBMT ( yellow ) and mismatched BMSC transplantation ( red ) compared to syngeneic WBMT ( green ) or syngeneic BMSCs transplantation ( blue ) . Data are shown as mean ± SD , triplicate experiments , n = 3–5 , **p<0 . 01 . ( E , F ) The ratio of CD4+ IL-17+ T cells in the spleen was significantly higher following mismatched WBMT ( yellow ) or mismatched BMSC transplantation ( red ) compared to syngeneic control ( green and blue ) . Data are shown as mean ± SD , triplicate experiments , n = 3–5 , *p<0 . 05 . BMSCs , bone marrow stromal/stem cells; SD , standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 02910 . 7554/eLife . 09394 . 030Figure 7—source data 1 . Serum IL-6 concentration after mismatched BMSC transplantation compared to syngeneic BMSC transplantation . Data are from 2 , 3 , and 4 weeks after mismatched and syngeneic BMSC transplantation shown in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 03010 . 7554/eLife . 09394 . 031Figure 7—source data 2 . Serial changes of CD4+CD25+Foxp3+ Tregs in spleen cells . Raw data and average values for statistical analysis use in ( D ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 03110 . 7554/eLife . 09394 . 032Figure 7—source data 3 . The ratio of CD4+ IL-17+ T cells in the spleen cells . Raw data and average values for statistical analysis used in ( E ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 09394 . 032 Increased IL-6 during an autoimmune process leads to the induction of Th17 cells and a decrease in regulatory T cells ( Tregs ) ( Bettelli et al . , 2006 ) . We found that CD4+CD25+Foxp3+ Tregs in peripheral blood transiently increased following lethal irradiation after syngeneic transplantation of BMSCs or WBM , which recovered to basal levels by 8 weeks . This transient increase in Tregs , reaching statistical significance at 3 weeks , was not observed after mismatched BMSC or WBM transplantation ( Figure 7C , D ) . This was followed by an increase of Th17 cells in both the mismatched WBMT and BMSC groups reaching statistical significance compared to syngeneic controls at 3 weeks in the WBMT group , and 8 weeks in the BMSC mismatched group . Mismatched BMT more rapidly induces the appearance of CD4+/Th17+ cells compared with mismatched PαS cells . ( Figure 7E , F ) . These results suggest that the interaction of mismatched BMSCs and recipient T cells induce IL-6 production and lead to pathological changes in major organs similar to autoimmune disease , with inflammation due to Th17 cells via suppression of transient increase in Tregs .
We have reported the role of donor BMSCs in the pathogenesis of fibrosis associated with autoimmune SSc in a MHC-matched , minor antigen mismatched mouse model . Our data show that depletion of BMSCs from donor whole bone marrow significantly reduced fibrosis in all organs examined , and rescued mice from lacrimal gland dysfunction associated with the disease . These findings suggest the possible role of donor BMSCs in the initiation of the autoimmune process , as the number of BMSCs in graft correlated with severity of fibrosis ( Figure 1E , F ) . Furthermore , the onset of true autoimmune disease in these mice was shown by the repression of Treg induction , followed by an increase in Th17 effector cells of host origin . This shows that migrating donor PαS-BMSCs are the initial trigger of events leading to increased levels of circulating IL-6 , followed by a decrease in Tregs , and conversely an increase in host-derived Th17 cells . Activation or maturation of BMSC-derived progeny is probably also involved in the progression of disease because accumulation of donor-derived fibroblasts was not observed in syngeneic BMSC transplantation , where fibrosis does not occur . In the mixed lymphocyte reaction experiments , Thy-1+ T-cells isolated from host mice after mismatched BMT were activated by PαS-BMSCs in vitro , as shown by enhanced proliferation and IL-6 secretion . In addition , the T-cell proliferation was blocked by anti-MHC antibody . Finally , adoptively transferred T cells from mismatched BMSCs recipients into nude-mice-induced autoimmune-like inflammation and fibrosis in targets organs , indicating that autoreactive recipient T cells were activated by antigens presented by MHC class II molecules in minor antigen mismatched donor BMSCs . Adoptive transferred T cells react recipient BMSCs leading to recipient’ fibrosis ( Figure 5 ) probably because adoptively transferred T cells has already acquired the autoreactive nature to activate the recipient derived BMSCs as shown in vitro analysis from Figure 6A and Figure 6B . Although the auto-antigen responsible for the autoimmune type reaction still needs to be identified , a common antigen specifically expressed in BMSCs of B10 . D2 and BALB/c with different isoforms or SNIPs , is one of the most probable candidates . Since naive T cells responded to PαS-BMSCs in vitro at low levels , we speculate that auto-reactive T cells that are only present in basal levels under normal conditions ( Sakaguchi et al . , 2008 ) can recognize minor differences between B10 . D2 and BALB/c BMSCs . The lack of fibrosis following mismatched HSC transplantation suggests that hematopoietic lineage cells , including DCs , did not express such a molecule . The identical phenotype was observed when host and recipient were reversed ( i . e . BALB/c BMSCs transplanted to B10 . D2 recipients , data not shown ) . This implies that both B10 . D2 and BALB/c BMSCs can be a primary inducer against mismatched recipient T cells , showing that the autoimmune-related fibrosis in this model was due to difference in strain , and not a specific reaction of B10 . D2 BMSCs and BALB/c T cells . The novelty of our study is the use of prospectively isolated BMSCs combined with prospectively isolated HSCs . Previous studies have used culture-isolated BMSCs , which do not reflect the physiological role of these cells in vivo . Functional differences between in vitro and in vivo observations may explain discrepancies in the anti-inflammatory and pro-inflammatory effects of BMSCs reported in the literature ( See review by Bernardo et . al . ( Bernardo and Fibbe , 2013 ) . Our study has also demonstrated that donor BMSCs can contribute to the fibroblast population observed in fibrotic lesions of the host . Our findings suggest that once mismatched BMSCs migrate into the target organs via certain homing signals , mismatched BMSCs encounter T cells and proliferate and activate under the pathological microenvironment ( Figure 2A ) . This phenomenon may not be present in syngeneic BMSCs transplantation where an allogeneic response does not occur . As for HSP47 expression , it is clearly seen in activated fibroblasts , but only very faint expression is detected in quiescent fibroblasts in the syngeneic BMSC-transplanted recipients’ target organs . Interestingly , a fraction of BMSCs mobilized in the peripheral blood expressed MHC class II molecules ( Figure 6B ) . Preliminary data show that these cells also express CD45 and type I collagen ( data not shown ) , which corresponds with the phenotype of fibrocytes reported in the literature ( Abe et al . , 2001; Chesney et al . , 1997; Mielcarek et al . , 2003; Phillips et al . , 2004; Wang et al . , 2007; Yang et al . , 2002 ) . Further studies are required to elucidate the association of PαS-BMSCs and fibrocytes . Interestingly , this SSc mouse model has also been used as a model of chronic graft-versus-host disease ( cGVHD ) ( Kaplan et al . , 2004; Kim et al . , 2007 ) . The phenotype of cGVHD characterized by systemic fibrosis and severe dry eye is very similar to this SSc mouse phenotype . Unlike conventional cGVHD that is believed to be caused by donor T cells , our results show that radio-resistant residual recipient T cells , but not donor-HSC derived de novo T cells , were activated following mismatched BMSC transplantation . Although we do not believe that our findings in this mouse model could be directly applied to human cGVHD cases , it is worth noting this phenomenon . In fact , several previous reports have shown that residual recipient CD4+ T cells regulate cGVHD ( Anderson et al . , 2004; Blazar et al . , 2000; Jaffee and Claman , 1983 ) . In addition , cGVHD has occurred in a surprisingly high fraction of nonmyeloablative stem cell transplant recipients ( Anderson et al . , 2004; Mielcarek et al . , 2003; Schetelig et al . , 2002 ) since residual host T cells remain in nonmyeloablative transplantation . Notably , BMSC depletion from grafts significantly reduced fibrosis in all organs that we examined . The use of CD34+ selected HSCs as an approach to reducing the risk of cGVHD was suggested in early studies of CD34+ selected peripheral blood HSCs . ( Martı́nez et al . , 1999; Urbano-Ispizua et al . , 1997; 2001 ) CD34+ selected HSC transplantation depletes not only mature T cells but also donor BMSCs because human BMSCs are negative for CD34 expression ( Mabuchi et al . , 2013 ) . In addition , recent reports showed that the incidence of cGVHD was less following cord blood transplantation ( CBT ) compared to peripheral blood stem-cell transplantation ( PBSCT ) ( Takahashi et al . , 2007; Uchino et al . , 2012 ) . We have found that BMSCs are rarely detected in cord blood but are abundant in G-CSF-mobilized PBL ( Mabuchi et al . , 2013 ) , which may indicate that the onset of cGVHD may correlate with the number of BMSCs transplanted . Taken all , the reduced risk of cGVHD may have been due to the removal or absence of donor BMSCs as we showed in our BMSC-depleted model . Further studies are required for monitoring the frequency of donor BMSCs in grafts and residual host T cells in human cGVHD patients to elucidate the possible role of donor BMSCs in the pathogenesis . In summary , prospective transplantation of freshly purified BMSCs and HSCs into BALB/c-RAG2KO suggests that transplantation of minor antigen-mismatched MHC-compatible BMSCs interact with residual host T cells to induce the autoimmune phenotype observed in fibrosis associated with the SSc mouse model . While the responsible antigen remains to be elucidated , our data suggest that accidental recognition of self-minor antigens on MHC class II+ BMSCs may be involved in systemic fibrosis observed in autoimmune disease .
B10 . D2 , BALB/c , and BALB/c background Nude mice 8-10 week of age were purchased from Sankyo Laboratory , Inc . ( Tokyo , Japan ) . B10 . D2 GFP mice were obtained by backcrossing of B10 . D2 mice with C57BL/6 GFP mice ( Japan SLC Ltd ) . The 10th generation of backcrossed B10 . D2 GFP progenies was used for experiments . RAG2KO on BALB/c background were kindly provided from Dr . Shigeo Koyasu ( Keio University ) . All experimental procedures and protocols were approved by the ethics committee of Keio University and were in accordance with the Guide for the Care and Use of Laboratory Animals ( # 09152 ) . We followed the ARRIVE guidelines for reporting in vivo experiments in animal research ( Kilkenny et al . , 2010 ) . Bone marrow cells , suspended at 1x106 cells/ml in calcium- and magnesium-free Hanks balanced salt solution ( HBSS ) supplemented with 2% fetal calf serum , 10 mM HEPES , and 1% penicillin/streptomycin ( HBSS+ ) were incubated with 5 μg/ml Hoechst 33342 ( Sigma Aldrich , St . Louis , MO ) for 60 min at 37°C . The side population ( SP ) was sorted as described previously ( Matsuzaki et al . , 2004 ) . ( Figure 1—figure supplement 2 ) Purified BMSCs were isolated by flowcytometry as described previously ( Morikawa et al . , 2009a; 2009b ) . Briefly , femurs and tibias were dissected and crushed with a pestle . The bone marrow was gently washed in HBSS+ . Collagenase ( 0 . 2% ) was used to digest the minced tibia and femur in DMEM without fetal bovine serum at 37°C for 1 hr . After red blood cell lysis , the residual cells were stained with FITC-labeled anti-Sca-1 ( Ly6A/E ) , APC-labeled anti-PDGFR-α ( APA-5 ) , PE-labeled anti-CD45 ( 30-F-11 ) , and PE-labeled anti-TER119 ( TER-119 ) ( all from e-Bioscience , San Diego , CA ) . Analysis and sorting were performed on a triple-laser MoFlo flow cytometer ( Beckman Coulter , Brea , CA ) . CD45- TER119- PDGFRα+ Sca-1+ cells were routinely prepared at 99% purity by this method ( Figure 1—figure supplement 2 ) ( Morikawa et al . , 2009a ) . CD45- TER119- PDGFRα+ Sca-1+ cells were further characterized by using FITC-labeled anti-Sca-1 ( E13-161 . 7 ) , PE- or APC-labeled anti-PDGFR-α ( APA-5 ) , PE-Cy7-labeled anti-CD45 ( 30-F-11 ) , and PE-Cy7 labeled anti-TER119 ( TER-119 ) ( all from BD Pharmingen , San Jose , CA ) and then Sca-1 and PDGFR-α double positive fraction were further analyzed with PE-labeled anti-CD31 ( MEC13 . 3: BD Pharmingen ) , PE-labeled anti-CD133 ( 13A4: e-Bioscience ) , PE-labeled anti-CD 29 ( Ha2/5: BD Pharmingen ) , APC-labeled anti-CD90 ( 30-H12: BioLegend , San Diego , CA ) , and APC-labeled CD106 ( 429: BioLegend ) , respectively . For PE-labeled isotype antibody , rat IgG2b , κ for CD45 and TER119 ( e-Bioscience ) , and rat IgG2a , κ for CD31 ( BD phamingen ) , rat IgG1 , κ for CD133 ( BD phamingen ) , and Armenian Hamster IgM , κ for CD29 ( BD pharmingen ) were used . For APC-labeled isotype antibody , rat IgG2a , κ for , rat IgG2b , κ for CD90 ( BD phamingen ) and rat IgG2a , κ for PDGFRα+ ( e-Bioscience ) and CD106 ( BD phamingen ) were used . FITC-labeled rat anti-mouse IgG2a , κ for Sca-1 ( e-Bioscience ) was used . For co-transplantation experiments , 1x104 PαS-BMSCs and 1x103 SP-HSCs ( see 'Methods' and Figure 1—figure supplement 2 ) were intravenously injected into the tail vein ( 200 μl/ per mouse ) of recipient mice that had been lethally irradiated with a dose of 7 . 0 Gy as indicated in our previous studies ( Matsuzaki et al . , 2004; Morikawa et al . , 2009a ) . This is approximately equivalent to 50 CFU-F in 1x104 PDGFRα+/Sca1+ cells ( Morikawa et al . , 2009a; 2009b ) while approximately 10 to 50 HSCs are found in 1000 SP KSL cells ( Okada et al . , 1992 ) . Therefore , while the ratio of PαS-BMSC to SP KSL was 10:1 in the donor , the actual number of BMSCs to HSCs was approximately 1:1 ( Okada et al . , 1992 ) . Spleens were obtained from mice that received B10 . D2 BMSC or BALB/c BMSC transplants . T cells and DCs were purified from the spleens by anti-CD90 . 2 monoclonal antibody ( mAb ) -conjugated microbeads or anti-CD11c mAb-conjugated microbeads ( Miltenyi Biotic , Bergisch Gladbach , Germany ) , respectively , according to the manufacturer’s instructions . The purity was consistently > 98% . T cells were cultured alone or co-cultured with PαS-BMSCs at a ratio of 10:1 ( T cells: BMSCs ) . Adoptive transfer was performed in accordance with previous publication ( Arakaki et al . , 2003; Niederkorn et al . , 2006 ) . Briefly , splenic T cells were isolated from mismatched PαS-BMSC transplanted BALB/c recipients as described above , and 3x106 cells were transferred into BALB/c background nude mice without any pre-treatment . The animals were sacrificed 45 days after transfer . Analyses of tissue samples from target organs were performed as shown in the section 'Immunohistochemistry and immunofluorescence' . Splenic cells were collected and prepared for flowcytometry and stained with CD4and IL-17 antibody as shown in the section 'Flowcytometry analysis for IL-17' . Purified B10 . D2 PαS-BMSCs and BALB/c PαS-BMSCs were plated at 1x104 cells/ well into 96-well plates in triplicate , which were irradiated at 52 Gy after adherence . 1x105 purified mouse T cells were added to each well . On the fourth day , 5-bromo-2’-deoxyuridine ( BrdU ) was added . Twenty-four hours later , BrdU uptake was quantified by cell proliferation enzyme-linked immunosorbent assay ( ELISA ) using a BrdU Kit ( Roche Applied Science , Penzberg , Germany ) ( Guo et al . , 2009 ) . In some experiments , purified B10 . D2 PαS-BMSCs were co-cultured with T cells isolated from mismatched PαS-BMSC-transplanted recipient mice . Cells were treated with either blocking anti-MHC class II antibody ( M5/114 . 15 . 2 , Biolegend ) or isotype control . The supernatant of the co-cultures was subjected to IL-6 ELISA using commercial kit ( BD Biosciences ) . For Foxp3 staining , whole blood samples were co-stained with anti-CD4-FITC and anti-CD25-PE ( PC61 . 5 ) . After fixation and permealization , cells were stained with APC-labeled anti-Foxp3 mAb ( FJK-16s ) ( all mAbs were from e-Bioscience ) as described ( Chen et al . , 2007 ) . For IL-6 , and IL-17 staining , 2 x 106 spleen cells were stimulated with 10 ng/ml phorbol myristate acetate ( PMA ) ( Sigma ) and 10 ng/ml ionomycin ( Sigma ) in the presence of the Golgi inhibitor Brefeldin ( Sigma ) ( 10 μg/ml ) for 4 hr . The cells were then stained with FITC-labeled anti-CD4 ( GK 1 . 5 , e-Bioscience ) and PE-labeled anti-IL-6 ( MP5-20F3 , BD Pharmingen ) or APC-labeled anti-IL-17 ( eBio17B7 , e-Bioscience ) as described ( Kappel et al . , 2009 ) . FITC-labeled anti-CD8 ( 53-6 . 7 ) and MHC class II ( M5/114 . 15 . 2 , e-Bioscience ) were used for FACS analysis on T cells and BMSCs . PE-labeled IgG2a , FITC-labeled IgG1 , APC-labeled IgG2a and IgG2a , κ were used as isotype controls ( all from e-Bioscience ) . Cells were analyzed on a FACScan with Cellquest software ( Beckton Dickinson ) . For the co-cultures with T cells and BMSCs , T cells were stimulated with Brefeldin ( 10 μg/ml ) alone for 4 hr . Bonferroni/Dunn test ( SPSS19 . 0 for Windows , SPSS Japan Inc . , Tokyo , Japan ) and two tailed Stutent’s t-test was used to analyze the number of HSP47+ fibroblasts per field . Two-tailed Student’s t-test ( SPSS19 . 0 for Windows ) was used to analyze tear volume , cytokine serum levels , and cell proliferation and cytokine production in co-cultures . Differences were considered significant when p<0 . 05 . Data presented as mean ± SD . | Systemic scleroderma is an autoimmune disease caused by the immune system attacking the body’s connective tissues , which provide the body with structural support . Immune cells called T cells accumulate in connective tissue , which leads to the hardening of the skin and may also damage the heart , lungs and other internal organs . However , it is not clear what prompts the T cells to accumulate in the connective tissues of these individuals . Autoimmune diseases develop when the immune system mistakenly identifies host cells as being a threat to the body . Normally , the immune system recognizes healthy body cells by the presence of particular proteins on the surface of the cells . A set of surface proteins called the major histocompatibility complexes ( MHCs ) play a major role in this process , but there are also many other surface proteins that play more minor roles . In 2002 , researchers developed a method that can trigger the symptoms of systemic scleroderma in mice . This method involves transplanting bone marrow from one mouse into another mouse . Both mice have identical MHC proteins on the surfaces of their cells , but have some differences in other cell surface proteins , and so the bone marrow from the donor mouse triggers an immune response in the recipient . To better understand how this mouse “model” of systemic scleroderma works , Ogawa , Morikawa et al . refined the method so that they could just transplant specific types of bone marrow cells into the recipient mice . The experiments reveal that bone marrow stromal stem cells , but not so-called “hematopoietic stem cells” , from a donor mouse are responsible for triggering the immune response and disease symptoms in the recipients . Ogawa , Morikawa et al . ’s findings show that mismatched minor cell surface proteins on bone marrow stromal stem cells can trigger symptoms of systemic scleroderma in mice . Further studies are required to find out how these cells encourage T cells to trigger an autoimmune response . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"immunology",
"and",
"inflammation"
] | 2016 | MHC-compatible bone marrow stromal/stem cells trigger fibrosis by activating host T cells in a scleroderma mouse model |
Multiple hypothalamic neuronal populations that regulate energy balance have been identified . Although hypothalamic glia exist in abundance and form intimate structural connections with neurons , their roles in energy homeostasis are less known . Here we show that selective Ca2+ activation of glia in the mouse arcuate nucleus ( ARC ) reversibly induces increased food intake while disruption of Ca2+ signaling pathway in ARC glia reduces food intake . The specific activation of ARC glia enhances the activity of agouti-related protein/neuropeptide Y ( AgRP/NPY ) -expressing neurons but induces no net response in pro-opiomelanocortin ( POMC ) -expressing neurons . ARC glial activation non-specifically depolarizes both AgRP/NPY and POMC neurons but a strong inhibitory input to POMC neurons balances the excitation . When AgRP/NPY neurons are inactivated , ARC glial activation fails to evoke any significant changes in food intake . Collectively , these results reveal an important role of ARC glia in the regulation of energy homeostasis through its interaction with distinct neuronal subtype-specific pathways .
The central nervous system comprises an elaborate network of neuronal populations that have been identified to control energy homeostasis ( Morton et al . , 2006; Pang and Han , 2012 ) . This extensive network includes neural circuits that operate within ( Aponte et al . , 2011; Atasoy et al . , 2012; Jennings et al . , 2013; Klöckener et al . , 2011; Krashes et al . , 2011; Lee et al . , 2013; Vong et al . , 2011; Yamanaka et al . , 2003 ) and beyond ( Georgescu et al . , 2005; Hommel et al . , 2006; Zhan et al . , 2013 ) the hypothalamus . Of these , the arcuate nucleus of the hypothalamus ( ARC ) is most actively studied . The key components of the feeding circuit in ARC have been identified to comprise agouti-related protein/neuropeptide Y ( AgRP/NPY ) -expressing neurons and pro-opiomelanocortin ( POMC ) -expressing neurons ( Aponte et al . , 2011; Atasoy et al . , 2012; Krashes et al . , 2011; Zhan et al . , 2013 ) , although an earlier study ( Kong et al . , 2012 ) suggests other cell types may function as additional regulators . In particular , little is known about the functional role of glial fibrillary acidic protein ( GFAP ) -expressing glia that co-exist in the same brain circuit as these neurons . Glia represent a major cell population in the brain , with their numbers at least equaling those of neurons . Traditionally viewed as passive support elements , recent research has unraveled their important roles in the modulation of multiple physiological functions ( Chen et al . , 2012; Gourine et al . , 2010; Halassa et al . , 2009; Nedergaard et al . , 2003; Schummers et al . , 2008 ) . GFAP-expressing hypothalamic glia , particularly astrocytes , have extensive structural connections with neurons ( Horvath et al . , 2010 ) and also express the receptor for the satiety hormone , leptin ( Cheunsuang and Morris , 2005; Hsuchou et al . , 2010 , 2009; Jayaram et al . , 2013; Kim et al . , 2014; Pan et al . , 2008 ) and for insulin ( García-Cáceres et al . , 2016 ) . Another type of GFAP-expressing hypothalamic glia , tanycytes , are interestingly glucosensitive and are activated by transmitters related to arousal and feeding ( Bolborea and Dale , 2013; Dale , 2011; Rodríguez et al . , 2005 ) . These findings suggest a potential role of glia in the regulation of energy homeostasis , thereby opening up several key questions: Do GFAP-expressing hypothalamic glia play a role in the regulation of feeding ? Do they functionally interact with AgRP/NPY and/or POMC neurons ? If so , how does manipulation of glia in ARC affect food intake and feeding behavior ? Here we demonstrate that direct and acute glial activation in the mouse ARC facilitates the activity of AgRP/NPY but not POMC neurons and is sufficient to reversibly evoke feeding . In the presence of AgRP/NPY neuronal inactivation , acute ARC glial activation did not evoke any change in feeding . In addition , disruption of Ca2+ signaling in ARC glia leads to reduced food intake . This reveals an important role of ARC glia-AgRP/NPY neuron circuit in the regulation of energy homeostasis .
To acutely activate glia in vivo , we selectively expressed designer receptors exclusively activated by designer drugs ( DREADDs ) ( Alexander et al . , 2009; Armbruster et al . , 2007 ) in GFAP-expressing glia in ARC of C57BL/6 adult mice . This was achieved by stereotaxic injection in ARC with an adeno-associated virus ( AAV ) containing a Gfap promoter-driven gene that encodes the evolved human M3-muscarinic receptor fused to mCherry ( hM3D ( Gq ) -mCherry ) ( Figure 1A ) . The hM3D ( Gq ) couples to the Gq protein-mediated signaling to activate glia upon binding of clozapine-N-oxide ( CNO ) . CNO is an exclusively specific ligand of DREADD ( Agulhon et al . , 2013 ) and is inert when used at low concentrations ( see CNO dose experiments in Figure 2—figure supplement 1B–D ) . Successful bilateral viral injection in ARC was confirmed by localized expression of mCherry ( Figure 1B ) . The hM3D ( Gq ) -mCherry was selectively expressed in glia and not in neurons . This was evidenced by the co-localization of mCherry with anti-GFAP ( Figure 1C ) and anti-S100B ( Figure 1—figure supplement 1A: 93 . 9% of mCherry-expressing cells ( n = 401/430 cells in 12 animals ) were immunopositive for S100B ) but not with anti-neuronal nuclei ( NeuN ) immunohistochemistry stains ( Figure 1D ) . There was also no expression of mCherry in specific neuronal sub-types: orexigenic AgRP/NPY neurons and anorexigenic POMC neurons in Npy-hrGFP and Pomc-EGFP mice respectively ( Figure 1—figure supplement 1A: 99 . 1% of GFP-expressing neurons in Npy-hrGFP mice ( n = 733/744 neurons in 11 animals ) and 99 . 3% of GFP-expressing neurons in Pomc-EGFP mice ( n = 439/448 neurons in 9 animals ) did not express mCherry ) . The hM3D ( Gq ) -mCherry glial population comprised both astrocytes and tanycytes that line the third ventricle ( Figure 1—figure supplement 1B: 14 . 9% of the mCherry-expressing cells co-localized with anti-vimentin , a marker for tanycytes ( n = 80/512 mCherry-expressing cells in 4 animals ) . In vivo injections of CNO ( 0 . 3 mg/kg ) ( Koch et al . , 2015; Krashes et al . , 2011 , 2013; Pei et al . , 2014 ) induced significantly greater Fos immunoreactivity ( Palkovits et al . , 2007; Ramírez et al . , 2015 ) as compared to saline injections in ARC astrocytes ( Figure 1E–F: n = 95 mCherry-expressing astrocytes in 5 animals each for saline and CNO groups , p=0 . 0114 , unpaired t-test , comparing averaged responses across astrocytes in each saline- and CNO-injected animal ) but not in tanycytes ( Figure 1—figure supplement 1C ) . CNO , but not saline application in ARC slices ( Figure 1G ) also evoked robust Ca2+ responses in hM3D ( Gq ) -mCherry-expressing astrocytes ( Figure 1H–I: n = 42 mCherry-expressing astrocytes in 3 animals each for saline and CNO groups , p=5 . 49E-16 , unpaired t-test , comparing averaged responses across astrocytes in the saline- and CNO-injected animals ) . A greater complexity of astrocytic processes was also observed in CNO-injected animals as compared to saline-injected animals ( Figure 1—figure supplement 1D–F ) . 10 . 7554/eLife . 18716 . 003Figure 1 . CNO-dependent activation of hM3D ( Gq ) -mCherry expressing ARC glia evokes increased Fos immunoreactivity and elevated intracellular Ca2+ . ( A ) Design of the AAV construct expressing hM3D ( Gq ) -mCherry under the Gfap promoter . ( B ) ( Left ) Schematic drawing showing the location of ARC ( red ) in a coronal brain slice ( Right ) An example DAPI-stained ( blue ) coronal brain slice containing hM3D ( Gq ) -mCherry ( red ) expressing glia in ARC . Scale bar , 1 mm . ( C–D ) DAPI , anti-GFAP and anti-NeuN immunohistochemistry showing expression of hM3D ( Gq ) -mCherry in glia but not in neurons . ( Inset ) Magnified image of a hM3D ( Gq ) -mCherry expressing glial cell and non hM3D ( Gq ) -mCherry expressing neurons . Scale bars , 10 µm . ( E ) In vivo injection of CNO but not saline induces Fos immunoreactivity in hM3D ( Gq ) -mCherry-expressing ( red ) , S100B-positive ( green ) ARC astrocytes . Animals were perfused for Fos quantification 2 hr post injection . Scale bar , 10 µm . ( F ) Population mean of normalized Fos expression in astrocytes following saline or CNO injection . ( G ) Configuration of calcium imaging of OGB-1-AM loaded , hM3D ( Gq ) -mCherry expressing astrocytes in ARC coronal slices during CNO application . ( H ) Local application of CNO but not saline ( red dot; 10 mM , 20 psi , 200 ms ) evoked a robust Ca2+ increase in ARC astrocytes . ( I ) The population average of mean% fluorescence change ( dFF ) of astrocytes when CNO or saline was applied . *p<0 . 05 , ****p<0 . 0001 . Error bars represent SEM . See also Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 00310 . 7554/eLife . 18716 . 004Figure 1—figure supplement 1 . DREADDs are specifically expressed in glial cells . Astrocytes , but not tanycytes , are activated by DREADDs and fasting . Morphological changes in astrocytes are also induced by DREADDs and fasting . ( A ) Expression of hM3D ( Gq ) -mCherry ( red ) ( Left ) co-localizes with anti-S100B ( green ) immunohistochemistry but did not co-localize with ( Middle ) GFP ( green ) in Npy-hrGFP and ( Right ) Pomc-EGFP slices . Scale bars , 20 µm . ( B ) Anti-vimentin immunohistochemistry showing expression of hM3D ( Gq ) -mCherry in some of the vimentin-expressing tanycytes lining the third ventricle ( 3V ) . Scale bars , 50 µm ( left ) , 10 µm ( right ) . ( C ) Population mean of normalized Fos expression in hM3D ( Gq ) -mCherry-expressing tanycytes 2 hr after saline or CNO ( 0 . 3 mg/kg ) injection . n = 51 mCherry-expressing tanycytes in 6 saline-injected animals , n = 25 mCherry-expressing tanycytes in 4 CNO-injected animals , p=0 . 364 , unpaired t-test , comparing averaged responses across tanycytes in each saline- and CNO-injected animal ( D ) anti-GFAP immunohistochemistry ( green ) revealing the morphology of hM3D ( Gq ) -mCherry ( red ) expressing ARC astrocytes 2 hr after in vivo injection of ( Left ) saline and ( Right ) CNO ( 0 . 3 mg/kg ) . Scale bars , 10 µm . ( E ) Sholl analysis of length and ( F ) intersections made by processes of hM3D ( Gq ) -mCherry ARC astrocytes revealed greater complexity in astrocytic morphology after in vivo injection of CNO compared to that of saline . n = 15 mCherry-expressing astrocytes in 5 animals in each group . ( G ) Fasting induces greater Fos immunoreactivity in S100B-positive ARC astrocytes and S100B-negative putative ARC neurons ( but not in tanycytes ) as compared to fed animals . Population mean of normalized Fos expression in ARC astrocytes , neurons and tanycytes of fasted and fed animals is shown . Fed: n = 440 astrocytes , 442 neurons and 244 tanycytes in 11 animals , Fasted: n = 400 astrocytes , 402 neurons and 226 tanycytes in 10 animals , p=0 . 0393/ p=0 . 0164/ p=0 . 279 , unpaired t-test , comparing averaged responses of astrocytes/neurons/tanycytes between the fasted and fed animals . ( H ) anti-GFAP immunohistochemistry revealing the morphology of ARC astrocytes in ( Left ) fed and ( Right ) fasted animals . Scale bars , 10 µm . ( I ) Sholl analysis of length and ( J ) intersections made by the processes of ARC astrocytes revealed greater complexity in astrocytic morphology in fasted as compared to fed animals . n = 18 astrocytes in 6 animals in each group . In Figure 1—figure supplement 1G–J , fasted and fed animals were perfused 16–18 hr post initiation of fasting in fasted animals . Fed animals were given ad libitum access to food . In Figure 1—figure supplement 1E–F , 1I–J , two-way ANOVA followed by Bonferroni post hoc tests was used . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ****p<0 . 0001 . N . S . , not significant . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 004 Notably , the physiological activation of ARC during fasting also induced Fos immunoreactivity in ARC astrocytes and neurons but not in tanycytes ( Figure 1—figure supplement 1G ) . A greater complexity of astrocytic processes was also observed in fasted animals as compared to fed animals ( Figure 1—figure supplement 1H–J ) . These data suggest that activation of hM3D ( Gq ) -mCherry in astrocytes can induce both the functional and morphological changes observed during physiological activation of astrocytes after fasting . Collectively , these results confirm the specific and selective expression and activation of hM3D ( Gq ) -mCherry in ARC glia ( mainly astrocytes ) , thereby allowing investigation of the functional effects of physiological activation of ARC glia . To assess the effect of ARC glial activation on feeding , we compared the food intake of viral-injected , ad libitum fed mice in custom-designed cages ( Figure 2—figure supplement 1A ) on days with CNO injection to that under control conditions . The control conditions included ( 1 ) cage acclimatization without injection ( baseline ) , ( 2 ) with saline injection before CNO injection ( pre-CNO saline ) and ( 3 ) after CNO injection ( post-CNO saline ) . The comparison of food intake following saline and CNO injections allowed the same animal to be used as its own control ( Figure 2A ) , since CNO , but not saline , can activate hM3D ( Gq ) -mCherry-expressing glia specifically ( Figure 1E–I ) ( Agulhon et al . , 2013 ) . The CNO dose was selected to be 0 . 3 mg/kg ( Koch et al . , 2015; Krashes et al . , 2011 , 2013; Pei et al . , 2014 ) for all experiments , as concentration experiments in control mice with low ( 0 . 3 mg/kg ) and high ( 5 mg/kg ) ( Yang et al . , 2015 ) CNO concentrations revealed non-specific inhibition of food intake at high but not low CNO concentration ( Figure 2—figure supplement 1B–D ) . Saline or CNO injections were performed at 09:00 , 2 hr after the start of light phase when mice were in a calorically replete state . Feeding measurements were made between 09:00–17:00 . Interestingly , the average total food intake during days of CNO administration was about three- to four-fold greater than that during baseline , pre-CNO saline and post-CNO saline ( Figure 2B: p<0 . 0001 , n = 11 animals , unpaired t-test comparing total food intake during three days of CNO with that during three days of baseline , three days of pre-CNO saline and four days of post-CNO saline ) . This feeding response was reversible and reproducible across repeated CNO administration in the same animal ( Figure 2B ) , suggesting that glial activation may serve as an intrinsic regulatory mechanism of food intake . 10 . 7554/eLife . 18716 . 005Figure 2 . CNO-dependent activation of hM3D ( Gq ) -mCherry expressing ARC glia evokes increased day food intake , time spent at food chamber and food seeking attempts in C57BL/6 mice . ( A ) Schematic of experimental paradigm . The mice were allowed to recover and for hM3D ( Gq ) -mCherry to be expressed 2–3 weeks post viral injection before acclimatization in custom cages for 3–7 days ( baseline ) and to saline injection for three days ( pre-CNO saline ) . CNO injections were repeated for three days , each separated by two days of saline injection ( post-CNO saline ) to allow the CNO effects to clear . All injections were performed at 09:00 while the food intake was measured at specific time points between 09:00–17:00 . All mice were housed in custom cages between 09:00–17:00 and returned to the standard cages after 17:00 daily . ( B ) Total food intake between 09:00–17:00 during baseline , pre-CNO saline , CNO and post-CNO saline averaged across animals . Dotted line 1 refers to the averaged total food intake across baseline , pre- and post- CNO saline while dotted line 2 refers to two folds of this average . ( C ) Food intake during hourly time points from 09:00 to 17:00 ( except from 09:00–09:30 and 09:30–10:00 where 30 min time points were used ) during pre-CNO saline and CNO administration . ( D ) The percentage of time mice spent at food chamber relative to other cage areas during specific time points following pre-CNO saline and CNO administration . ( E ) The frequency of attempts made to access the food chamber during specific time points following pre-CNO saline and CNO administration . In Figure 2C–E , values between 09:00–09:30 and 09:30–10:00 were normalized to hourly values . Pre-CNO saline and CNO values were averaged across three days of repeats before computing the average across animals . Two-way ANOVA followed by Bonferroni post hoc tests was used . *p<0 . 05 , **p<0 . 01 , ****p<0 . 0001 . Error bars represent SEM . See also Figure 2—figure supplements 1–6 . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 00510 . 7554/eLife . 18716 . 006Figure 2—figure supplement 1 . CNO administered at 5 mg/kg ( but not at 0 . 3 mg/kg ) induces non-specific effects on feeding in animals without DREADD expression . CNO ( 0 . 3 mg/kg ) evoked robust feeding behavior in animals with hM3D ( Gq ) -mCherry-expressing ARC glia . ( A ) Setup of custom-designed , open-top cages comprising a food chamber , a drinking spout/water bottle , mouse igloo and sterile gauzes . The mouse igloo was positioned opposite to the food chamber in each cage to serve as an alternative location for rest and play while the animal was not feeding at the food chamber . The bedding was removed and replaced with sterile gauzes to prevent contamination of food in the chamber and therefore ensure accurate measurement of food intake . Video monitoring of the food behavior was enabled with web-cameras positioned above the cages and connected to a laptop for data storage . ( B–D ) Administration of CNO at 5 mg/kg but not 0 . 3 mg/kg induces non-specific decrease in feeding in C57BL/6 animals without hM3D ( Gq ) -mCherry expression . ( B ) Schematic of experimental paradigm . Mice were acclimatized to custom cages before saline injections were performed for four days . CNO injections were repeated for two days , separated by one day of saline injection to allow the CNO effects to clear . All injections were performed at 19:00 while the food intake was measured at specific time points between 19:00–23:00 . All mice were housed in custom cages between 19:00–23:00 and returned to the standard cages after 23:00 daily . Cumulative food intake in C57BL/6 animals was assessed from 19:00–23:00 at the specific time points , averaged across five days of saline and two days of CNO before taking the average across animals . ( C ) Animals were injected with CNO at lower dose ( 0 . 3 mg/kg ) . n = 8 animals , ANOVA , Drug: p>0 . 5 , F ( 1 , 7 ) = 0 . 4824; Time: p<0 . 0001 , F ( 4 , 28 ) = 30 . 17; Interaction between drug and time: p<0 . 02 , F ( 4 , 28 ) = 3 . 481 , comparing the population and trial averaged cumulative food intake at specific time points between 19:00–23:00 during saline and CNO administration . ( D ) Animals were injected with CNO at higher dose ( 5 mg/kg ) . n = 7 animals , ANOVA , Drug: p<0 . 03 , F ( 1 , 6 ) = 8 . 59; Time: p<0 . 0001 , F ( 4 , 24 ) = 80 . 2; Interaction between drug and time: p=0 . 0005 , F ( 4 , 24 ) = 7 . 41 , comparing the population and trial averaged cumulative food intake at specific time points between 19:00–23:00 during saline and CNO administration . ( E–F ) CNO administration ( 0 . 3 mg/kg ) , when compared to saline administration , induces an increase in food intake , frequency of shorter-duration feeding attempts and maximum duration of feeding episodes in animals with hM3D ( Gq ) -mCherry-expressing ARC glia . ( E ) Cumulative food intake from 09:00 to17:00 at the specific time points , averaged across three days of pre-CNO saline and three days of CNO administration before taking the average across animals . n = 11 animals , ANOVA , Drug: p<0 . 0001 , F ( 2 , 24 ) = 19 . 5; Time: p<0 . 0001 , F ( 8 , 192 ) = 39 . 2; Interaction between drug and time: p<0 . 0001 , F ( 16 , 192 ) = 21 . 5 , comparing the population and trial averaged cumulative food intake at specific time points between 09:00–17:00 during pre-CNO saline and CNO administration . ( F ) Frequency histograms of the food-seeking attempts of all animals during pre-CNO saline and CNO administration binned by the duration spent at the food chamber . Following CNO administration , the mode duration of food-seeking events was shorter while the skew of the distribution of duration of attempts tended towards more positive duration values than that following saline administration ( p<0 . 0001 , Kolmogorov-Smirnov test , comparing the distribution of durations following saline and CNO administration ) . In Figure 2—figure supplement 1C–E , two-way ANOVA followed by Bonferroni post hoc tests was used . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , ****p<0 . 0001 . N . S . , not significant . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 00610 . 7554/eLife . 18716 . 007Figure 2—figure supplement 2 . Activation of ARC glia evokes increased food intake within 3 hr after CNO application ( 0 . 3 mg/kg ) but does not increase total food intake in the dark phase . ( A ) Schematic of experimental paradigm . Mice were acclimatized to custom cages before saline injections were performed for three days . CNO injections were repeated for three days , separated by two days of saline injection to allow the CNO effects to clear . All injections were performed at 22:00 while the food intake was measured at specific time points between 22:00–06:00 . All mice were housed in custom cages between 22:00–06:00 and returned to the standard cages after 06:00 daily . ( B ) Cumulative food intake from ( Left ) 22:00–01:00 , ( Middle ) 02:00–06:00 and ( Right ) 22:00–06:00 at the specific time points , averaged across seven days of saline and three days of CNO before taking the average across animals . n = 6 animals , ANOVA , 22:00–01:00/02:00–06:00/22:00–06:00 - Drug: p=0 . 0099/0 . 5852/0 . 339 , F ( 1 , 10 ) = 10 . 07/ F ( 1 , 10 ) = 0 . 318/ F ( 1 , 10 ) = 1 . 01; Time: p<0 . 0001/ p<0 . 0001/p<0 . 0001 , F ( 3 , 30 ) = 195 . 4/F ( 4 , 40 ) = 180/ F ( 8 , 80 ) = 239; Interaction between drug and time: p=0 . 0036/0 . 341/0 . 739 , F ( 3 , 30 ) = 5 . 60/ F ( 4 , 40 ) = 1 . 16/ F ( 8 , 80 ) = 0 . 643 , comparing the population and trial averaged cumulative food intake at specific time points during saline and CNO administration . ( C ) Food intake at specific time points from ( Left ) 22:00–01:00 , ( Middle ) 02:00–06:00 and ( Right ) 22:00–06:00 , averaged across seven days of saline and three days of CNO administration before taking the average across animals . n = 6 animals , ANOVA , 22:00–01:00/02:00–06:00/22:00–06:00 - Drug: p=0 . 0228/0 . 150/0 . 717 , F ( 1 , 10 ) = 7 . 22/ F ( 1 , 10 ) = 2 . 43/ F ( 1 , 10 ) = 0 . 139; Time: p<0 . 0001/ p<0 . 0001/ p<0 . 0001 , F ( 3 , 30 ) = 38 . 2/ F ( 4 , 40 ) = 8 . 21/ F ( 8 , 80 ) = 17 . 2; Interaction between drug and time: p=0 . 0266/0 . 862/0 . 0007 , F ( 3 , 30 ) = 3 . 53/ F ( 4 , 40 ) = 0 . 322/ F ( 8 , 80 ) = 3 . 85 , comparing the population and trial averaged food intake at specific time points during saline and CNO administration . Food intake during 22:00–22:30 and 22:30–23:00 were normalized to hourly values . ( D ) Total food intake between 22:00–06:00 during pre-CNO saline , CNO and post-CNO saline averaged across animals . p>0 . 3 , n = 6 animals , unpaired t-test comparing total food intake during three days of CNO administration with that during three days of pre-CNO saline and four days of post-CNO saline administration . In Figure 2—figure supplement 2B–C , two-way ANOVA followed by Bonferroni post hoc tests was used . *p<0 . 05 , **p<0 . 05 , N . S . , not significant . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 00710 . 7554/eLife . 18716 . 008Figure 2—figure supplement 3 . Analysis of viral expression showed that mCherry is specifically expressed within the ARC , although there were a few cases where the expression spread into adjacent areas in the hypothalamus . ( A–B ) A trend where CNO induced greater increase in food intake with increased size of viral expression within the ARC was observed . CNO-induced increase in food intake was computed by taking the averaged total food intake across days with CNO administration subtracted by that across days with saline administration . The size of viral expression is quantified by the spread of viral expression along the ( A ) caudal-rostral axis and ( B ) medial-lateral axis of ARC . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 00810 . 7554/eLife . 18716 . 009Figure 2—figure supplement 4 . Administration of CNO ( 0 . 3 mg/kg ) in mice lacking hM3D ( Gq ) -mCherry in GFAP-expressing ARC glia does not evoke increased food intake , frequency of feeding attempts and duration of feeding . Viral injection of AAV-Gfap-hM3D ( Gq ) -mCherry was attempted in these animals but failed . All experimental procedures were identical in mice with and without successful viral injection . ( A ) Example DAPI-stained ( blue ) coronal brain slice where ARC glia did not express hM3D ( Gq ) -mCherry . ( B ) Schematic of experimental paradigm . The mice were allowed to recover after surgery before being acclimatized to custom cages for 3–7 days ( baseline ) and to saline injection for three days ( pre-CNO saline ) . CNO injections were repeated for three days , each separated by two days of saline injection ( post-CNO saline ) to allow the CNO effects to clear . All injections were performed at 09:00 while the food intake was measured at specific time points between 09:00–17:00 . All mice were housed in custom cages between 09:00–17:00 and returned to the standard cages after 17:00 daily . ( C ) Total food intake between 09:00–17:00 during baseline , pre-CNO saline , CNO and post-CNO saline averaged across animals . p>0 . 1 , n = 8 animals , unpaired t-test comparing total food intake during three days of CNO administration with that during three days of baseline , three days of pre-CNO saline and 4 days of post-CNO saline administration . ( D ) Food intake during hourly time points ( except from 09:00–09:30 and 09:30–10:00 where 30 min time points were used ) averaged across three days of pre-CNO saline and three days of CNO administration before taking the average across animals . n = 8 animals , ANOVA , Drug: p>0 . 4 , F ( 2 , 21 ) = 0 . 725; Time: p>0 . 3 , F ( 8 , 168 ) = 1 . 16; Interaction between drug and time: p>0 . 1 , F ( 16 , 168 ) = 1 . 33 , comparing the population and trial averaged food intake at specific time points between 09:00–17:00 during pre-CNO saline and CNO administration . Food intake during 09:00–09:30 and 09:30–10:00 were normalized to hourly values . ( E ) Cumulative food intake from 09:00 to 17:00 at the specific time points , averaged across three days of pre-CNO saline and three days of CNO before taking the average across animals . n = 8 animals , ANOVA , Drug: p>0 . 3 , F ( 2 , 21 ) = 1 . 08; Time: p<0 . 0001 , F ( 8 , 168 ) = 17 . 8; Interaction between drug and time: p>0 . 2 , F ( 16 , 168 ) = 1 . 29 , comparing the population and trial averaged cumulative food intake at specific time points between 09:00–17:00 during pre-CNO saline and CNO administration . ( F ) The percentage of time mice spent at food chamber relative to other cage areas during specific time points averaged across three days of pre-CNO saline or three days of CNO before taking the average across animals . n = 8 animals , ANOVA , Drug: p>0 . 3 , F ( 1 , 14 ) = 1 . 03; Time: p<0 . 0001 , F ( 8 , 112 ) = 11 . 0; Interaction between drug and time: p>0 . 4 , F ( 8 , 112 ) = 0 . 971 , comparing the population and trial averaged percentage of time spent at specific time points between 09:00–17:00 during pre-CNO saline and CNO administration . ( G ) The frequency of attempts made to access the food chamber during specific time points averaged across three days of pre-CNO saline or three days of CNO before computing the average across animals . Values between 09:00–09:30 and 09:30–10:00 were normalized to hourly values . n = 8 animals , ANOVA , Drug: p>0 . 9 , F ( 1 , 14 ) = 0 . 00391; Time: p<0 . 0001 , F ( 8 , 112 ) = 20 . 6; Interaction between drug and time: p>0 . 9 , F ( 8 , 112 ) = 0 . 203 , comparing the population and trial averaged frequency of attempts at specific time points between 09:00–17:00 during pre-CNO saline and CNO administration . In Figure 2—figure supplement 4D–G , two-way ANOVA followed by Bonferroni post hoc tests was used . ( H ) Frequency histograms of food-seeking attempts of all animals during pre-CNO saline and CNO administration binned by the duration spent at the food chamber . p>0 . 1 , Kolmogorov-Smirnov test , comparing the distribution of durations during saline and CNO administration . N . S . , not significant . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 00910 . 7554/eLife . 18716 . 010Figure 2—figure supplement 5 . Administration of CNO ( 0 . 3 mg/kg ) in control mice with mCherry-expressing ARC glia does not evoke increased food intake . ( A ) Design of the AAV construct expressing mCherry under the Gfap promoter . ( B ) Schematic of experimental paradigm . The mice were allowed to recover after surgery before being acclimatized to custom cages for three days ( baseline ) and to saline injection for three days ( pre-CNO saline ) . CNO injections were repeated for three days , each separated by two days of saline injection ( post-CNO saline ) to allow the CNO effects to clear . All injections were performed at 09:00 while the food intake was measured at specific time points between 09:00–13:00 . All mice were housed in custom cages between 09:00–13:00 and returned to the standard cages after 13:00 daily . ( C ) Total food intake between 09:00–13:00 during pre-CNO saline , CNO and post-CNO saline averaged across animals . p>0 . 5 , n = 7 animals , unpaired t-test comparing total food intake during three days of CNO administration with that during seven days of saline administration . ( D ) Food intake during hourly time points ( except from 09:00–09:30 and 09:30–10:00 where 30 min time points were used ) averaged across three days of pre-CNO saline and three days of CNO before taking the average across animals . n = 7 animals , ANOVA , Drug: p=0 . 701 , F ( 1 , 12 ) = 0 . 154; Time: p<0 . 0001 , F ( 4 , 48 ) = 8 . 90; Interaction between drug and time: p=0 . 225 , F ( 4 , 48 ) = 1 . 47 , comparing the population and trial averaged food intake at specific time points between 09:00–13:00 during pre-CNO saline and CNO administration . Food intake during 09:00–09:30 and 09:30–10:00 were normalized to hourly values . Two-way ANOVA followed by Bonferroni post hoc tests was used . N . S . , not significant . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 01010 . 7554/eLife . 18716 . 011Figure 2—figure supplement 6 . DREADD activation of glia does not lead to changes in body weight . ( A ) Averaged body weight across animals with ( black , n = 5 animals ) or without ( grey , n = 8 animals ) expression of hM3D ( Gq ) -mCherry in ARC glia . Injections of CNO or saline were performed once a day at 09:00 ( ANOVA , Without and with expression: p>0 . 9 , F ( 1 , 11 ) = 0 . 0160; Time: p<0 . 0001 , F ( 12 , 132 ) = 18 . 2; Interaction: p>0 . 8 , F ( 12 , 132 ) = 0 . 578 , comparing the population averaged body weight without and with hM3D ( Gq ) -mCherry expression . Two-way ANOVA followed by Bonferroni post hoc tests was used . ( B ) Averaged body weight across animals with expression of hM3D ( Gq ) -mCherry in ARC glia . Injections of CNO or saline were performed twice a day at 09:00 and 17:00 . n = 8 animals , p>0 . 06 , paired t-test comparing body weight measured during days of pre-CNO saline and CNO injections . N . S . , not significant . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 011 The CNO effect was also time-dependent when the population and trial-averaged food intake at specific time points between 09:00 and 17:00 during CNO and saline administration were examined . The time course revealed increased food intake within 30 min post CNO injection with peak feeding within 1 hr post injection . This effect persisted for at least 3 more hours ( Figure 2C: n = 11 animals , two-way analysis of variance ( ANOVA ) , Drug: p<0 . 0001 , F ( 2 , 30 ) = 19 . 0; Time: p<0 . 0001 , F ( 8 , 240 ) = 9 . 71; Interaction between drug and time: p<0 . 0001 , F ( 16 , 240 ) = 11 . 6; see also Figure 2—figure supplement 1E ) . At peak feeding , the food intake was more than 13 times greater than that during saline administration ( Figure 2C ) . It is worth mentioning that CNO administration evoked increased food intake within 3 hr post CNO application in the dark phase when majority of food intake occurs ( Figure 2—figure supplement 2A–C ) . The total food intake in the 8 hr of dark phase was however not significantly different between CNO and saline application ( Figure 2—figure supplement 2D ) . This may be due to a ceiling effect in the drive to feed when glia may already be activated at night ( see section below ) . We next compared the percentage of time viral-injected mice spent at the food chamber relative to other areas of the cage after CNO and saline administration ( Figure 2D ) . The average percentage of time that mice spent at the food chamber following CNO administration was significantly greater than that following saline administration in the same mice ( Figure 2D: n = 11 animals , ANOVA , Drug: p<0 . 0001 , F ( 1 , 20 ) = 47 . 8; Time: p<0 . 0001 , F ( 8 , 160 ) = 26 . 6; Interaction between drug and time: p<0 . 0001 , F ( 8 , 160 ) = 22 . 7 , comparing the population and trial-averaged time spent at food chamber at specific time points between 09:00 and 17:00 during pre-CNO saline and CNO administration ) . The same was true for the frequency of food seeking attempts ( Figure 2E: n = 11 animals , ANOVA , Drug: p=0 . 003 , F ( 1 , 20 ) = 11 . 4; Time: p<0 . 0001 , F ( 8 , 160 ) = 17 . 2; Interaction between drug and time: p<0 . 0001 , F ( 8 , 160 ) = 10 . 1 , comparing the population and trial-averaged frequency of attempts at specific time points between 09:00 and 17:00 during pre-CNO saline and CNO administration ) . CNO administration also increased the frequency of shorter-duration feeding attempts and increased the maximum duration of the feeding episodes ( Figure 2—figure supplement 1F ) . Collectively , these observations suggest that glial activation can trigger mice to devote greater time for feeding . A trend of increased CNO-induced food intake was observed with increased size of hM3D ( Gq ) -mCherry expression in the ARC ( Figure 2—figure supplement 3 ) . In mice lacking hM3D ( Gq ) -mCherry expression in ARC due to failed viral injection attempts ( identified in post hoc analysis ) ( Figure 2—figure supplement 4 ) as well as mice injected with the control virus AAV-Gfap-mCherry in ARC ( Figure 2—figure supplement 5 ) , CNO administration did not increase food intake . Although glial activation resulted in the pronounced feeding response , DREADD activation of glia did not lead to changes in the body weight ( Figure 2—figure supplement 6 ) . Collectively , these data support an active role of ARC glia in the modulation of feeding behavior and suggest that direct glial activation is sufficient to evoke acute feeding even during the daytime when the mice are normally resting and in calorically replete state . The presence of activated ARC astrocytes during fasting ( Figure 1—figure supplement 1G ) suggests that glia may play a critical role in modulating feeding under physiological conditions . To address this question , we assessed the effect of reduced ARC glial activity on feeding by first injecting an AAV comprising a Gfap promoter-driven gene that encodes the pleckstrin homology ( PH ) domain of phospholipase C ( PLC ) -like protein p130 ( p130PH ) fused with mRFP ( AAV-Gfap-p130PH-mRFP ) in ARC . Glial expression of the p130PH construct was previously shown to disrupt the Ca2+ signaling in astrocytes in vivo by acting as a mobile cytosolic IP3 buffer to inhibit release of Ca2+ from internal stores ( Xie et al . , 2010 ) . Control animals were injected with AAV-Gfap-mRFP without the p130PH construct ( Figure 3A ) . We compared the Fos activity of mRFP-expressing , S100B-positive ARC glia in the AAV-Gfap-p130PH-mRFP and AAV-Gfap-mRFP ( control ) injected animals after fasting . Indeed , the Fos activity of astrocytes in the AAV-Gfap-p130PH-mRFP injected animals was significantly lower than that in the control animals ( Figure 3B , C: p130PH: n = 847 mRFP-expressing astrocytes , Control: n = 508 mRFP-expressing astrocytes , 8 animals each , p=0 . 0264 , unpaired t-test , comparing averaged responses across astrocytes in each Gfap-p130PH-mRFP and control animal ) . The Fos activity of tanycytes , however , was not significantly different between the experimental and control groups ( Figure 3—figure supplement 1A ) . The Fos activity of neurons was significantly lower in experimental than in control animals ( Figure 3—figure supplement 1B ) , suggesting possible glial-neuronal modulation ( see below ) . 10 . 7554/eLife . 18716 . 012Figure 3 . Disruption of Ca2+ signaling in ARC glia with selective expression of p130PH-mRFP leads to decreased Fos immunoreactivity . ( A ) Design of the AAV constructs expressing Gfap promoter-driven p130PH-mRFP and mRFP ( control ) respectively . ( B ) Fos immunoreactivity in mRFP-expressing ( red ) , S100B-positive ( green ) ARC astrocytes is lower in fasted animals injected with AAV-Gfap-p130PH-mRFP than in control animals injected with AAV-Gfap-mRFP . Animals were perfused for Fos quantification 16–18 hr post initiation of fasting . Scale bar , 20 µm . ( C ) Population mean of normalized Fos expression in ARC astrocytes of AAV-Gfap-p130PH-mRFP and AAV-Gfap-mRFP injected animals . *p<0 . 05 . Error bars represent SEM . See also Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 01210 . 7554/eLife . 18716 . 013Figure 3—figure supplement 1 . Disruption of Ca2+ signaling in ARC glia with selective expression of p130PH-mRFP leads to decreased Fos immunoreactivity in neurons ( but not in tanycytes ) with no change in the body weight of fasted animals . ( A ) Fos immunoreactivity in mRFP-expressing tanycytes is not significantly different in fasted animals injected with AAV-Gfap-p130PH-mRFP than in control animals injected with AAV-Gfap-mRFP . Animals were perfused for Fos quantification 16–18 hr post initiation of fasting . Population mean of normalized Fos expression in ARC tanycytes of AAV-Gfap-p130PH-mRFP and AAV-Gfap-mRFP injected animals . p130PH: n = 22 tanycytes in 8 animals , Control: n = 24 tanycytes in 8 animals , p=0 . 651 , unpaired t-test , comparing averaged responses across tanycytes in AAV-Gfap-p130PH-mRFP and AAV-Gfap-mRFP injected animals . ( B ) Fos immunoreactivity in non mRFP-expressing , S100B-negative ARC putative neurons is lower in fasted animals injected with AAV-Gfap-p130PH-mRFP than in control animals injected with AAV-Gfap-mRFP . Population mean of normalized Fos expression in ARC putative neurons of AAV-Gfap-p130PH-mRFP and AAV-Gfap-mRFP injected animals . p130PH: n = 718 putative neurons in 8 animals , Control: n = 600 putative neurons in 8 animals , p=0 . 0030 , unpaired t-test , comparing averaged responses across putative neurons in AAV-Gfap-p130PH-mRFP and AAV-Gfap-mRFP injected animals . ( C ) Averaged body weight across AAV-Gfap-p130PH-mRFP ( red , n = 8 animals ) and AAV-Gfap-mRFP ( blue , n = 8 animals ) injected animals , measured after viral injection surgery . ANOVA , Gfap-p130PH-mRFP and Gfap-mRFP: p=0 . 805 , F ( 1 , 14 ) = 0 . 0630; Days: p<0 . 0001 , F ( 17 , 238 ) = 32 . 6; Interaction: p=0 . 980 , F ( 17 , 238 ) = 0 . 420 , comparing the population averaged body weight of AAV-Gfap-p130PH-mRFP and AAV-Gfap-mRFP injected animals . Two-way ANOVA followed by Bonferroni post hoc tests was used . **p<0 . 01 , N . S . , not significant . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 013 We then compared the food intake of the Gfap-p130PH-mRFP-injected animals and control animals during both dark and light phases . Interestingly , the population and trial-averaged food intake was significantly lower in Gfap-p130PH-mRFP-injected animals than control animals in the dark phase but not light phase ( Figure 4A: n = 8 animals , ANOVA , Drug: p<0 . 0001 , F ( 1 , 14 ) = 30 . 3; Time: p<0 . 0001 , F ( 19 , 266 ) = 72 . 9; Interaction between drug and time: p=0 . 0131 , F ( 19 , 266 ) = 1 . 92 ) . This reduction of food intake during the dark phase was not compensated by an increase in food intake in the light phase as the cumulative food intake of Gfap-p130PH-mRFP-injected animals at the 24-hr time point ( 19:00 ) remained significantly lower than control animals ( Figure 4B: n = 8 animals , ANOVA , Drug: p<0 . 0001 , F ( 1 , 14 ) = 44 . 7; Time: p<0 . 0001 , F ( 19 , 266 ) = 1091; Interaction between drug and time: p<0 . 0001 , F ( 19 , 266 ) = 16 . 4 ) . Similar to the DREADD experiments , no significant difference in the body weight was observed between Gfap-p130PH-mRFP-injected animals and control animals ( Figure 3—figure supplement 1C ) . Collectively , these findings support an essential role for glia in the modulation of feeding during physiological conditions . 10 . 7554/eLife . 18716 . 014Figure 4 . Disruption of Ca2+ signaling in ARC glia with selective expression of p130PH-mRFP leads to decreased food intake in C57BL/6 mice during the dark phase . ( A ) Food intake during hourly time points ( except from 19:00–19:30 and 19:30–20:00 where 30 min time points were used ) in AAV-Gfap-p130PH-mRFP or AAV-Gfap-mRFP injected animals . Values between 19:00–19:30 and 19:30–20:00 were normalized to hourly values . ( B ) Cumulative food intake from 19:00–19:00 ( 24 hr , both dark and light phases ) at the specific time points in AAV-Gfap-p130PH-mRFP or AAV-Gfap-mRFP injected animals . Values were averaged across three days of repeats before computing average across animals . Two-way ANOVA followed by Bonferroni post hoc tests was used . *p<0 . 05 , ***p<0 . 001 , ****p<0 . 0001 . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 014 There is emerging evidence that astrocytes are integral components of brain circuits linked to specific brain functions and are capable of modulating neuronal responses ( Eroglu and Barres , 2010; Halassa and Haydon , 2010 ) . It is thus possible that the activation of DREADD-expressing glia can consequently modulate nearby ARC neurons to contribute to the observed evoked feeding behavior . To test if this mechanism underlies the glial activation-induced increase in feeding , we investigated the potential role of the glial modulation of AgRP/NPY and POMC neurons . We first injected the hM3D ( Gq ) -mCherry virus in the ARC of Npy-hrGFP mice and Pomc-EGFP mice . Next , we performed ex vivo whole-cell current clamp recordings from NPY neurons ( GFP-positive neurons in Npy-hrGFP mice ) and POMC neurons ( GFP-positive neurons in Pomc-EGFP mice ) in brain slices containing hM3D ( Gq ) -mCherry-expressing ARC glia ( Figure 5A ) . Interestingly , CNO activation of the hM3D ( Gq ) -mCherry-expressing ARC glia depolarized and/or increased the firing rate of AgRP/NPY neurons ( Figure 5B , C: n = 15 neurons in 15 slices from 9 animals , p<0 . 0001 , unpaired t-test , comparing CNO-induced responses with null responses; averaged resting membrane potential: −52 . 2 ± 2 . 65 mV [van den Top et al . , 2004] ) . In POMC neurons , CNO did not induce a significant response at a population level ( Figure 5B , C: n = 14 neurons in 14 slices in 6 animals , p=0 . 370; unpaired t-test , comparing CNO-induced responses with null responses; averaged resting membrane potential: - 44 . 1 ± 2 . 09 mV [Cowley et al . , 2001] ) . These findings were further validated by first injecting both Npy-hrGFP and Pomc-EGFP mice ( with hM3D ( Gq ) -mCherry-expressing ARC glia ) with either CNO or saline before performing Fos immunohistochemistry . Indeed , greater Fos immunoreactivity was observed after CNO injection as compared to saline injection in NPY but not POMC neurons ( Figure 5D–E , F: n = 5 Npy-hrGFP animals each for CNO and saline injections , p=0 . 0428 , unpaired t-test , comparing CNO and saline induced responses; n = 5 and n = 6 Pomc-EGFP animals for CNO and saline injections respectively , p=0 . 736 , unpaired t-test , comparing CNO and saline induced responses ) . Collectively , these observations show that direct and acute activation of ARC glia facilitates the activity of AgRP/NPY but not POMC neurons . 10 . 7554/eLife . 18716 . 015Figure 5 . CNO activation of hM3D ( Gq ) -expressing glia evokes facilitatory responses in NPY but not POMC neurons . ( A ) ( Top ) Configuration of whole-cell patch-clamp recordings of GFP-labeled NPY or POMC neurons in the ARC of coronal brain slices infected with AAV-Gfap-hM3D ( Gq ) -mCherry during CNO application . ( Bottom ) Merged green fluorescence ( GFP ) , red fluorescence ( mCherry ) and differential interference contrast images of ( Left ) a GFP positive NPY neuron and ( Right ) a GFP positive POMC neuron patched in viral injected Npy-hrGFP and Pomc-EGFP slices respectively . Relative positions of ( 1 ) patch pipette and ( 2 ) CNO drug pipette were as indicated . Scale bars , 10 µm . ( B ) Local CNO application ( red dot; 10 mM , 20 psi , 200 ms ) evoked depolarizing response in ( Top ) an NPY ( Bottom ) but not in POMC neuron . ( C ) A population average of mean membrane potential ( Vm ) of NPY or POMC neurons when CNO was applied . ( D–E ) In vivo injection of CNO induced greater Fos immunoreactivity in NPY neurons than saline injection . This was not observed in POMC neurons . Scale bars , 20 µm . ( F ) Population mean of normalized Fos expression in NPY and POMC neurons following saline or CNO injection . *p<0 . 05 , ****p<0 . 0001 , N . S . , not significant . Error bars represent SEM . See also Figure 5—figure supplement 1–2DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 01510 . 7554/eLife . 18716 . 016Figure 5—figure supplement 1 . CNO does not evoke any response in NPY neurons when calcium is chelated in hM3D ( Gq ) -mCherry-expressing ARC astrocytes . ( A ) Electrophysiological characteristics of an astrocyte with the I-V curve showing absence of active membrane currents . ( B ) ( Left ) Spread of A633 ( and BAPTA ) in the syncytium within 30–45 mins of patching an astrocyte identified by its small round soma with thin radiating processes . Scale bar , 200 µm . ( Right Top ) NPY neurons were patched immediately after BAPTA/A633 dialysis within the hM3D ( Gq ) -mCherry-expressing astrocyte syncytium . ( Right Bottom ) Individual hM3D ( Gq ) -mCherry-expressing astrocytes were loaded with A633 ( and BAPTA ) . Scale bar , 20 µm . ( C ) Local CNO application ( red dot; 10 mM , 20 psi , 200 ms ) did not evoke any response in NPY neurons after BAPTA/A633 dialysis of adjacent astrocytes through whole cell patch clamping of a passive astrocyte . ( D ) Population average of mean Vm of NPY neurons when CNO was applied in the absence and presence of BAPTA/A633 dialysis of adjacent astrocytes ( p=1 . 18 E-5 and p=0 . 957 , unpaired t-test , comparing CNO-induced responses with null responses in the absence and presence of BAPTA dialysis respectively ) . **p<0 . 01 , ****p<0 . 0001 , N . S . , not significant . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 01610 . 7554/eLife . 18716 . 017Figure 5—figure supplement 2 . Administration of CNO ( 0 . 3 mg/kg ) in Agrp-Ires-cre mice with hM4D ( Gi ) -mCherry-expressing AgRP neurons and hM3D ( Gq ) -mCherry-expressing ARC glia does not evoke any change in food intake . ( A ) Design of the 2 AAV constructs ( 1 ) AAV construct expressing hM3D ( Gq ) -mCherry under the Gfap promoter ( 2 ) AAV construct expressing hM4D ( Gi ) -mCherry using FLEX switch strategy to induce Cre-mediated transgene inversion and expression in AgRP neurons . ( B ) Expression of mCherry ( red ) co-localizes with anti-NeuN ( blue , presumably AgRP/NPY neurons ) and anti-S100B ( green ) immunohistochemistry . Scale bar , 10 µm . ( C ) Schematic of experimental paradigm . Mice were allowed to recover after surgery before being acclimatized to custom cages ( baseline ) and to saline injection for three days ( pre-CNO saline ) . CNO injections were repeated for three days , each separated by a day of saline injection ( post-CNO saline ) to allow the CNO effects to clear . All injections were performed at 09:00 while the food intake was measured at specific time points between 09:00–15:00 . ( D ) Total food intake between 09:00–15:00 during pre-CNO saline , CNO and post-CNO saline administration , averaged across animals . p>0 . 2 , n = 9 animals , unpaired t-test comparing total food intake during three days of CNO administration with that during five days of saline administration . ( E ) Food intake during hourly time points ( except from 09:00–09:30 and 09:30–10:00 where 30 min time points were used ) averaged across three days of pre-CNO saline and three days of CNO before taking the average across animals . n = 9 animals , ANOVA , Drug: p=0 . 684 , F ( 1 , 16 ) = 0 . 172; Time: p=0 . 0072 , F ( 6 , 96 ) = 3 . 15; Interaction between drug and time: p=0 . 520 , F ( 6 , 96 ) = 0 . 870 , comparing the population and trial averaged food intake at specific time points between 09:00–15:00 during saline and CNO administration . Food intake during 09:00–09:30 and 09:30–10:00 were normalized to hourly values . Two-way ANOVA followed by Bonferroni post hoc tests was used . N . S . , not significant . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 017 To confirm that the CNO-evoked facilitatory responses in NPY neurons indeed have an astrocytic origin , we performed chelation of astrocytic Ca2+ by patch-clamping electrophysiologically characterized astrocytes ( Figure 5—figure supplement 1A ) with the cell-impermeable Ca2+ chelator 1 , 2-Bis ( 2-aminophenoxy ) ethane-N , N , N′ , N′-tetraacetic acid ( BAPTA ) . The spread of BAPTA within the local syncytium of astrocytes through gap junctions ( Jourdain et al . , 2007 ) , visualized by including Alexa Fluor 633 ( A633 ) in the patch pipette , was confirmed to be about 200 μm from the patched astrocyte after 20–50 min of dialysis ( Figure 5—figure supplement 1B ) . Indeed , BAPTA chelation of Ca2+ in hM3D ( Gq ) -mCherry-expressing ARC astrocytes blocked the CNO-evoked depolarization in NPY neurons ( Figure 5—figure supplement 1C–D: Control: n = 15 neurons in 15 slices in 9 animals , BAPTA: n = 5 neurons ( patched after BAPTA dialysis of astrocyte syncytium ) in 5 slices in 3 animals , p=0 . 0050 , unpaired t-test , comparing responses in the absence and presence of BAPTA dialysis ) . Together these findings demonstrate that astrocytic Ca2+ activation contributes to NPY neuronal facilitation . Although we have observed preferential activation of AgRP/NPY neurons over POMC neurons during ARC astrocytic activation , it is unclear how this cell-type specific modulation arises . Is this due to the intimate organization between astrocytes and specific neuronal subtypes ? Or do astrocytes indiscriminately facilitate both AgRP/NPY and POMC neurons but their net responses shaped by the presence of a strong AgRP to POMC inhibition ( Atasoy et al . , 2012 ) ? To dissect these possibilities , we first recorded spontaneous inhibitory postsynaptic currents ( sIPSCs ) from POMC neurons during CNO activation of hM3D ( Gq ) -mCherry-expressing ARC glia . Indeed , CNO evoked an increase in the frequency ( but not amplitude ) of sIPSCs in POMC neurons ( Figure 6A top , 6B: n = 7 neurons in 7 slices in 3 animals , Amplitude: p=0 . 324; Frequency: p=0 . 0078 , paired t-test , comparing normalized responses before and after CNO application ) . This observation reveals strong inhibition received by POMC neurons during glial activation , possibly via NPY neuronal activation . To test if facilitatory responses in POMC neurons evoked by glial activation can be unmasked when the strong inhibition they receive is blocked , we recorded the CNO-evoked responses in POMC neurons in the presence of GABA receptor ( GABAR ) blockers ( 50 μM Picrotoxin , a GABAA receptor antagonist and 20 μM SCH-50911 , a GABAB receptor antagonist , see Materials and methods: Slice physiology ) . Interestingly , CNO evoked strong depolarization in POMC neurons when patched in the presence of GABAR blockers ( Figure 6A bottom , Figure 6C: Control: n = 14 neurons in 14 slices in 6 animals , GABAR blockers: n = 7 neurons in 7 slices in 5 animals , p=0 . 0046 , unpaired t-test , comparing responses between in the absence and presence of GABAR blockers ) . These data support our hypothesis that astrocytic activation facilitates both AgRP/NPY and POMC neurons indiscriminately . Unlike AgRP/NPY neurons , however , the presence of a strong inhibitory input , possibly due to the activation of AgRP/NPY-POMC inhibitory connection ( Atasoy et al . , 2012 ) during glial activation , can further shape the net responses of POMC neurons . Our data thus suggest that ARC glial activation indiscriminately facilitates both AgRP/NPY neurons and POMC neurons but the presence of an inhibitory input leads to no net response in POMC neurons , thereby providing a mechanistic understanding of the glial activation-induced increase in feeding . 10 . 7554/eLife . 18716 . 018Figure 6 . CNO activation of hM3D ( Gq ) -expressing glia evokes increased frequency of sIPSCs in POMC neurons . In the presence of GABAergic blockers , hM3D ( Gq ) -expressing glia activated depolarization in POMC neurons is revealed . ( A ) ( Top ) Local CNO application ( red dot; 10 mM , 20 psi , 200 ms ) evoked increased frequency of sIPSCs in POMC neurons patched in high chloride internal solution in the presence of bath application of NBQX . ( Bottom ) Local CNO application ( red dot; 10 mM , 20 psi , 200 ms ) evoked a facilitatory response in POMC neurons patched in the presence of picrotoxin and SCH-50911 ( see Materials and methods: Slice physiology ) . ( B ) Population average of mean normalized amplitude and frequency of sIPSCs in POMC neurons before and after CNO application . ( C ) Population average of mean Vm of POMC neurons when CNO was applied in the absence and presence of GABAergic blockers ( p=0 . 370 and p=0 . 0460; unpaired t-test , comparing CNO-induced responses with null responses ) . *p<0 . 05 , **p<0 . 01 , N . S . , not significant . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 01810 . 7554/eLife . 18716 . 019Figure 6—figure supplement 1 . Administration of CNO ( 0 . 3 mg/kg ) in mice with hM4D ( Gi ) -mCherry-expressing ARC glia does not evoke any change in food intake . ( A ) Design of the AAV construct expressing hM4D ( Gi ) -mCherry under the Gfap promoter . ( B ) Schematic of experimental paradigm . Mice were allowed to recover after surgery before being acclimatized to custom cages ( baseline ) and to saline injection for three days . CNO injections were repeated for three days , each separated by two days of saline injection to allow the CNO effects to clear . All injections were performed at 19:00 while the food intake was measured at specific time points between 19:00–23:00 . All mice were housed in custom cages between 19:00–23:00 and returned to the standard cages after 23:00 daily . ( C ) Total food intake between 19:00–23:00 during saline and CNO administration , averaged across animals . p>0 . 2 , n = 7 animals , unpaired t-test comparing total food intake during three days of CNO administration with that during seven days of saline administration . ( D ) Cumulative food intake from 19:00–23:00 at the specific time points , averaged across seven days of saline and three days of CNO administration before taking the average across animals . n = 7 animals , ANOVA , Drug: p=0 . 888 , F ( 1 , 12 ) = 0 . 0207; Time: p<0 . 0001 , F ( 4 , 48 ) = 27 . 8; Interaction between drug and time: p=0 . 998 , F ( 4 , 48 ) = 0 . 0333 , comparing the population and trial averaged cumulative food intake at specific time points between 19:00–23:00 during saline and CNO administration . ( E ) Food intake during hourly time points ( except from 19:00–19:30 and 19:30–20:00 where 30 min time points were used ) averaged across seven days of saline and three days of CNO before taking the average across animals . n = 7 animals , ANOVA , Drug: p=0 . 854 , F ( 1 , 12 ) = 0 . 0354; Time: p<0 . 0001 , F ( 4 , 48 ) = 12 . 4; Interaction between drug and time: p=0 . 983 , F ( 4 , 48 ) = 0 . 0980 , comparing the population and trial averaged food intake at specific time points between 19:00–23:00 during saline and CNO administration . Food intake during 19:00–19:30 and 19:30–20:00 were normalized to hourly values . Two-way ANOVA followed by Bonferroni post hoc tests was used . N . S . , not significant . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18716 . 019 To test if AgRP/NPY activation is necessary for glial activation-induced increase in feeding , we measured the food intake during simultaneous activation of ARC glia and inactivation of AgRP/NPY neurons . This was achieved by stereotaxic injection of both AAV-Gfap-hM3D ( Gq ) -mCherry and AAV-hSyn-DIO-hM4D ( Gi ) -mCherry in the ARC of Agrp-Ires-cre mice that allows selective expression of hM3D ( Gq ) -mCherry in ARC glia and hM4D ( Gi ) -mCherry in AgRP/NPY neurons ( Figure 5—figure supplement 2A–B ) . The food intake of these viral-injected , ad libitum fed mice was then compared on days with CNO and saline injections ( Figure 5—figure supplement 2C ) . The average total food intake ( Figure 5—figure supplement 2D ) as well as population and trial-averaged time course of food intake ( Figure 5—figure supplement 2E ) during days of CNO administration were not significantly different from those during saline administration . This finding that AgRP/NPY activation is required to mediate the glial activation-induced increase in feeding further establishes the crucial role of the ARC glia-AgRP/NPY circuit in the regulation of energy balance .
Our work shows that ARC glial activation enables increase in food intake via AgRP/NPY neurons . Interestingly , ARC glial activation facilitates both AgRP/NPY and POMC neurons . Unlike AgRP/NPY neurons , however , the responses in POMC neurons are further shaped by an inhibitory input , possibly due to activation of the AgRP/NPY neuron-POMC neuron inhibitory pathway ( Atasoy et al . , 2012 ) . The individual POMC neuronal responses are variable ( see Figure 5c ) , suggesting varying relative strengths of the astrocyte-POMC and astrocyte-AgRP/NPY-POMC pathways . These competing pathways , however balance , revealing no significant response at the population level . Our findings show that although glial cells interact with distinct neuronal subtypes non-specifically , it can confer neuronal subtype-specific modulation through direct modulation of distinct neuronal circuits . This may be a common mechanism to explain the glial-neuronal subtype specific modulation observed in other brain regions ( Perea et al . , 2014 ) . The lack of astrocyte-neuronal subtype wiring also suggests that glial activation may modulate other neuronal subtypes within the ARC . This thus opens the possibility of glial modulation of energy expenditure-promoting neurons in ARC e . g . GABAergic RIP-Cre neurons which do not overlap with the POMC and AgRP/NPY neuronal population ( Kong et al . , 2012 ) and can potentially prevent the expected body weight gains during AgRP/NPY neuronal activation ( Krashes et al . , 2011 ) . This may explain the absence of weight changes during ARC glial modulation observed in our study ( Figure 2—figure supplement 6 , Figure 3—figure supplement 1C ) . There are a few possible mechanisms that may underlie the non-specific , glial activation-stimulation of AgRP neurons and POMC neurons . These include: ( 1 ) astrocytic release of glutamate or D-serine gliotransmitters ( Halassa et al . , 2007; Haydon and Carmignoto , 2006; Parpura and Zorec , 2010; Scofield et al . , 2015; Volterra and Meldolesi , 2005 ) ; ( 2 ) regulation of extracellular transmitters ( Pannasch et al . , 2014 ) or extracellular potassium ( Wang et al . , 2012 ) possibly through changes in the activity of transporters in the astrocyte membrane ( Bazargani and Attwell , 2016 ) . Further investigation is required to dissect these possibilities . Under physiological conditions , glial activation may occur via neuronal-glial interactions ( Allen and Barres , 2009; Bazargani and Attwell , 2016; Fields and Burnstock , 2006; Lalo et al . , 2006 ) , possibly evoked by active AgRP/NPY neurons in the presence of a strong drive to feed ( Chen et al . , 2015; Mandelblat-Cerf et al . , 2015 ) . This , together with the strong glial-neuronal interactions observed in our work ( Figure 5A–C , Figure 1—figure supplement 1G , Figure 3B–C , Figure 3—figure supplement 1B ) suggest modulation of feeding by bidirectional communication ( Allen and Barres , 2009 ) between glia and neurons . Direct glial modulation by orexigenic molecules ( Murphy and Bloom , 2006 ) such as ghrelin via astrocytic ghrelin receptors ( García-Cáceres et al . , 2014 ) may also be a possible mechanism to glial activation . We manipulated glial Ca2+ activity bi-directionally with viral-mediated Gq-DREADD technology ( Agulhon et al . , 2013; Armbruster et al . , 2007 ) and viral-mediated p130PH disruption of glial Ca2+ signaling ( Xie et al . , 2015 , 2010 ) . Both methods allow local and specific manipulation of Ca2+ signaling pathway of glia in ARC . The Gq-DREADD technology mimics acute activation of astrocytic GPCRs that triggers Ca2+ activation ( Agulhon et al . , 2013 , 2008 ) while the p130PH technology enables disruption of the endogenous PLC/IP3 glial Ca2+ signaling ( Li et al . , 2013; Xie et al . , 2015 , 2010 ) during physiological feeding states . We have concerns using the Gi-DREADD technology to manipulate glial activity . Although the Gi-DREADD technology has been shown to inhibit neuronal firing via induction of hyperpolarization and inhibition of presynaptic neurotransmitter release ( Roth , 2016 ) , it is unclear whether Gi-DREADD can affect glial activity or glial Ca2+ signaling pathways as these cells are electrically non-excitable and do not fire action potentials . In addition , existing work that activated Gi-coupled GPCR pathways in glia do not reveal any blockade of both astrocytic activity and astrocyte-neuronal modulation ( Agulhon et al . , 2012 ) . Indeed , we did not observe any effect of glial Gi-DREADD activation on feeding ( Figure 6—figure supplement 1 ) . Previous work has provided evidence that support the role for astrocytes in feeding . This includes experiments showing chronic modification of feeding by synaptic remodeling ( Horvath et al . , 2010 ) during conditional deletion of astrocytic leptin receptors ( Kim et al . , 2014 ) ( see also Jayaram et al . , 2013 ) as well as observation of reactive gliosis ( Horvath et al . , 2010 ) and activation of astrocytic inflammatory signaling pathways after feeding of high fat diet ( Buckman et al . , 2015 ) . These studies however do not address the question of whether direct glial-activation can evoke acute changes in feeding . The behavioral changes in feeding in these studies also cannot be solely isolated to the hypothalamus due to body-wide deletion of leptin receptors in GFAP-expressing cells ( expression of GFAP ( Apte et al . , 1998; Buniatian et al . , 1998; Nolte et al . , 2001 ) and leptin receptors ( Bjørbaek and Kahn , 2004; Gautron and Elmquist , 2011 ) occur in both the central nervous system and periphery ) . Our work directly addresses these gaps in knowledge by demonstrating a causal relationship between arcuate glia and acute as well as chronic modulation of food intake . We also further provide mechanistic evidence to explain how glial activation can differentially modulate AgRP/NPY and POMC neurons through direct modulation of distinct neuronal pathways and demonstrate the necessity of active AgRP/NPY neurons in mediating the glial activation-induced feeding . A previous study ( Yang et al . , 2015 ) investigating glial regulation of feeding by using the CNO-DREADD system reported distinct conclusions from our study . We attribute the differences to the use of different CNO doses and show that a high CNO dose ( 5 mg/kg ) used in that study ( Yang et al . , 2015 ) can in fact induce non-specific inhibition in feeding ( see Figure 2—figure supplement 1B–D ) , possibly due to CNO-induced ataxia ( Sigma-Aldrich’s toxicological studies [Sigma-Aldrich , 2016] ) . Our work , together with previous findings , collectively demonstrates a critical role of glia in the regulation of feeding and further establishes ARC as an important site of glial-mediated regulation . These findings also imply a possible causal link between increased ARC glial Ca2+ during astrogliosis ( Kanemaru et al . , 2013 ) and hyperphagia during ARC inflammation ( Dorfman and Thaler , 2015 ) . Future work in this area may reveal glia as a possible target of therapeutic intervention .
All mice were housed at room temperature with a 12-hr light and 12-hr dark cycle ( lights on: 07:00; lights off: 19:00 ) . Mice were housed with cage mates except after surgery , during food behavior , Fos , glial morphological quantification and acute slice experiments where they were housed individually . For food behavior experiments , C57BL/6 ( Taconic , Germantown , NY ) and Agrp-Ires-cre ( Agrptm1 ( cre ) Lowl/J; JAX 012899; Bradford B . Lowell , Harvard ) adult mice of both gender , between 7–14 weeks old were used . For Fos experiments , acute slice whole cell patch-clamp , calcium imaging and immunohistochemistry experiments , C57BL/6 , Npy-hrGFP ( van den Pol et al . , 2009 ) ( B6 . FVB-Tg [Npy-hrGFP]1Lowl/J; JAX 006417; Bradford B . Lowell , Harvard ) , Pomc-EGFP ( Cowley et al . , 2001 ) ( C57BL/6J-Tg ( Pomc-EGFP ) 1Low/J; JAX 009593; Malcolm J . Low , University of Michigan Medical School ) mice of both gender , between 7 weeks–8 months old were used . All mice were housed in standard cages except during behavior experiments where custom-designed cages with no bedding ( Figure 2—figure supplement 1A ) were used . Standard mouse chow in pellet form ( Prolab RMH 3000 , 5P00*; Gross energy: 4 . 19 kcal/g , Metabolizable energy: 3 . 18 kcal/g , Physiological fuel value: 3 . 46 kcal/g; Protein: 25 . 999% , Fat: 14 . 276% , Carbohydrates: 59 . 725% ) and water were provided ad libitum . All experiments were performed under protocols ( 0513-044-16 ) approved by the Animal Care and Use Committee at MIT and conformed to NIH guidelines . The AAV-Gfap-hM3D ( Gq ) -mCherry virus ( Construct: Bryan Roth , UNC Chapel Hill , Addgene plasmid # 50478; AAV Serotype 2/8 , UNC Vector Core ) , AAV-Gfap104-mCherry virus ( Construct: Ed Boyden , MIT , AAV Serotype 2/8 , UNC Vector Core ) , AAV-Gfap-p130PH-mRFP and AAV-Gfap-mRFP viruses ( Shinghua Ding ( Xie et al . , 2010 ) , Serotype 2/5 ) , AAV-hSyn-DIO-hM4D ( Gi ) -mCherry virus ( Construct: Bryan Roth , UNC Chapel Hill , Addgene plasmid # 44362; AAV Serotype 2/8 , UNC Vector Core ) and AAV-Gfap-hM4D ( Gi ) -mCherry virus ( Construct: Bryan Roth , UNC Chapel Hill , Addgene plasmid # 50479; AAV Serotype 2/5 , UNC Vector Core ) were injected and specifically expressed in ARC glia . Mice were initially anesthetized with 4% isoflurane in oxygen and maintained on 1 . 5–2% isoflurane on a stereotaxic apparatus . The skull was exposed and a small hole was drilled above each side of ARC . Bilateral injection was performed with a glass micropipette ( 20–30 μm diameter ) filled with 200 nl of virus at the following coordinates: bregma: AP:−1 . 40 mm , DV:−5 . 80 mm , L: ±0 . 30 mm . The injection speed was controlled at 100 nl/min with a micromanipulator ( Quintessential Stereotaxic Injector , Stoelting ) . Mice were injected intraperitoneally with meloxicam ( 1 mg/kg ) for postoperative care . Experiments were performed 2–3 weeks post-injection to allow for recovery and viral expression . Ad libitum fed mice were transferred from their standard cages to the open-top , custom-designed cages ( Figure 2—figure supplement 1A ) daily during the duration of the study . Each cage contained a food chamber , a drinking spout ( water bottle external to the cage ) , mouse igloo and sterile gauze to substitute the removed bedding . The mouse igloo was positioned opposite to the food chamber in each cage and served to be an alternative location for rest and play . The weight of the food chamber was taken at specific time points using a weighing balance ( CAS MWP-300N , 300 g , 0 . 01 g precision ) . In some experiments , measurements were taken after intraperitoneal injections of saline or clozapine-N-oxide ( CNO , C0832 , Sigma-Aldrich , St . Louis , MO ) at 09:00 ( Figure 2B-E , Figure 2—figure supplement 1E–F , Figure 2—figure supplements 4–5 , Figure 5—figure supplement 2 ) , 19:00 ( Figure 2—figure supplement 1B–D , Figure 6—figure supplement 1 ) or 22:00 ( Figure 2—figure supplement 2 ) . The feeding behavior was continuously recorded during the experiments with web cameras ( Logitech , Newark , CA ) positioned above the cages and the open source iSpy webcam software ( iSpyConnect , Perth , Australia ) . Mice were transferred back to the standard cages after these experiments . CNO was administered at 0 . 30 mg/kg of body weight ( see Figure 2—figure supplement 1B–D ) while saline was given at the same volume as control . At the conclusion of the behavior study , all mice were perfused and the fixed brains were sectioned and imaged to confirm the stereotaxic accuracy of the viral injection site . Some of the sectioned slices were immunostained to verify the expression specificity of hM3D ( Gq ) -mCherry in glia . Perfusion: Mice were anesthetized with 4% isoflurane and perfused transcardially with 0 . 1 M PBS followed by chilled 4% paraformaldehyde in 0 . 1 M PBS . The brains were then postfixed in 4% paraformaldehyde in 0 . 1 M PBS ( <4°C ) overnight . Fixed brains were sectioned into 50–70 µm coronal slices ( containing ARC ) with a vibratome . Localization of viral expression: The fixed slices were mounted on a glass slide with the Vectashield Hardset mounting media ( Vector Labs , Burlingame , CA ) . The slides were imaged using a confocal microscope ( Zeiss LSM 5 Pascal Exciter , Carl Zeiss , Oberkochen , Germany ) and the images were analyzed for localization of mCherry ( Figures 1–2 , Figure 5—figure supplement 2 ) or mRFP ( Figures 3–4 ) expression in ARC . Anti-GFAP/Anti-S100B and anti-NeuN immunohistochemistry: The fixed slices were blocked in 10% normal goat serum with 0 . 5% triton in PBS ( 1 hr , room temperature ) before being stained with mouse anti-GFAP ( 1:400 , MAB360; Millipore , Billerica , MA ) or mouse anti-S100B ( 1:600 , S2532; Sigma ) and rabbit anti-NeuN ( 1:250 , ABN78; Millipore ) in blocking buffer overnight ( at 4°C ) . This was followed by a 3-hr incubation in Alexa Fluor 488 goat anti-mouse ( 1:200 , Invitrogen , Carlsbad , CA , A11029 ) and Alexa Fluor 488 goat anti-rabbit ( 1:250 , Invitrogen , A11034 ) in 1% normal goat serum and 0 . 3% triton , before being mounted on a glass slide with mounting media . Confocal imaging was performed and the images were analyzed for presence of co-localization of mCherry/mRFP and GFAP /S100B ( Figure 1C , E , Figure 3B , Figure 1—figure supplement 1A , Figure 5—figure supplement 2B ) , absence/presence of co-localization of mCherry and NeuN ( Figure 1D , Figure 5—figure supplement 2B ) . Mice were perfused 2 hr post CNO/saline injection or 16–18 hr post initiation of fasting in fasted animals . The brains were post-fixed in paraformaldehyde over 2 nights before being sectioned into 100 µm coronal slices . The fixed slices were first incubated in 1 . 2% Triton-X-100 for 15 min before being blocked in 5% normal goat serum with 2% BSA and 0 . 2% Triton-X-100 in PBS ( 1 hr , room temperature ) . For anti-Fos immunohistochemistry , the slices were then incubated in blocking buffer containing rabbit anti-Fos ( 1:1000 , PC38; Calbiochem , San Diego , CA ) and in some experiments with mouse anti-S100B ( 1:600 , S2532; Sigma ) overnight at 4°C . For morphological analysis of glia , slices were incubated in blocking buffer containing rabbit anti-GFAP ( 1:400 , G9269 , Sigma ) overnight at 4°C . This was followed by a 2-hr incubation in blocking buffer containing Alexa Fluor 647 goat anti-rabbit ( 1: 1000 , Invitrogen , A21245 ) and Alexa Fluor 488 goat anti-mouse ( in some anti-Fos immunohistochemistry experiments , 1:1000 , Invitrogen , A11001 ) as well as 0 . 3 µM DAPI in PBS before being mounted on a glass slide with mounting media . Confocal imaging was performed with similar optical parameters for all slices imaged to ensure fair comparisons . For anti-Fos immunohistochemistry experiment , MetaMorph Basic Offline ( v . 7 . 7 . 0 . 0 , Molecular Devices , Sunnyvale , California ) was used for image quantification . Individual astrocytic glial cells were identified by the S100B and in some cases with mCherry/mRFP labeling ( Figure 1E-F , Figure 3B–C , Figure 1—figure supplement 1G ) while NPY/AgRP and POMC neurons were identified by their GFP expression ( Figure 5D-F ) . In some cases , putative neurons were identified by DAPI and their lack of S100B and/or mRFP labeling ( Figure 1—figure supplement 1G and Figure 3—figure supplement 1B ) . Tanycytic glial cells were identified by their location along the third ventricle next to the ARC ( Figure 1—figure supplement 1C , Figure 1—figure supplement 1G , Figure 3—figure supplement 1A ) and mCherry/mRFP labeling ( Figure 1—figure supplement 1C , Figure 3—figure supplement 1A ) . These cells were then circled manually before the mean Fos signal for each cell was computed . Normalized Fos values were computed by normalizing raw Fos signal by the averaged Fos signal across control cells . For morphological analysis of glia ( Reeves et al . , 2011 ) , ARC glia with clear soma and processes were identified by GFAP labeling ( Figure 1—figure supplement 1D , Figure 1—figure supplement 1H ) and in some cases with mCherry expression ( Figure 1—figure supplement 1D ) . Image stacks in 1 μm planes were then collected under 60x oil objective with 2x digital zoom . The glial cells were 10–90 μm from the surface . Image stacks were traced and 3D-reconstructed using the Neurolucida software ( MBF Bioscience , Williston , VT ) . The coordinate files generated by the reconstruction were further analyzed by Sholl analysis using Neuroexplorer software ( MBF Bioscience , Williston , VT ) . Virtual circles at increasing radii of 3 . 9 μm increments were drawn from the soma of each glia . The total length of GFAP-labeled processes passing through each circle and the number of intersections of processes with each circle were quantified ( Figure 1—figure supplement 1E–1F , Figure 1—figure supplement 1I–J ) 100 µm coronal fixed slices , prepared as described above , were stained with mouse monoclonal anti-vimentin ( 2 µg/ml , DSHB Hybridoma Product 40E-C , deposited to DSHB ( Iowa City , Iowa ) by Alvarez-Buylla , Arturo ) followed by Alexa Fluor 488 goat anti-mouse ( 1:1000 , Invitrogen , A11001 ) . The amount of food intake during a specific time segment for each mouse was computed by subtracting the weight of the food chamber at the beginning of the time point by that at the end of the time point . The time segment for total food intake was from 09:00–17:00 ( Figure 2B , Figure 2—figure supplement 4C ) or 09:00–13:00 ( Figure 2—figure supplement 5C ) or 09:00–15:00 ( Figure 5—figure supplement 2D ) or 19:00–23:00 ( Figure 6—figure supplement 1C ) or 22:00–06:00 ( Figure 2—figure supplement 2D ) . The total food intake data was averaged across animals ( Figure 2B , Figure 2—figure supplement 2D , Figure 2—figure supplement 4C , Figure 2—figure supplement 5C , Figure 5—figure supplement 2D , Figure 6—figure supplement 1C ) as well as averaged across trials and animals ( Figure 2C , Figure 4 , Figure 2—figure supplement 1C–E , Figure 2—figure supplement 2B–C , Figure 2—figure supplement 4D–E , Figure 2—figure supplement 5D , Figure 5—figure supplement 2E , Figure 6—figure supplement 1D–E ) . Processing of the videos was done using in-house code written in MATLAB ( Natick , MA ) . The average percentage of time spent at the food chamber was computed as ( total time spent at the chamber in the time segment ) / ( total duration in the time segment ) *100% for each animal before taking the average across trials and animals ( Figure 2D , Figure 2—figure supplement 4F ) . The average frequency of food-seeking attempts ( /hr ) was computed by counting the number of times each mouse approach ( and stay ) at the food chamber in the time segment before normalizing the count by the duration of the time segment considered before taking the average across trials and animals ( Figure 2E , Figure 2—figure supplement 4G ) . The frequency-histogram was computed by binning the food-seeking attempts for all animals by the duration spent on the food chamber ( Figure 2—figure supplement 1F , Figure 2—figure supplement 4H ) . Coronal brain slices of ARC ( 300 µm ) were prepared from AAV-Gfap-hM3D ( Gq ) -mCherry injected Npy-GFP and Pomc-EGFP mice with a vibratome ( Leica VT 1200S , Leica , Wetzlar , Germany ) . Mice were anesthetized with Avertin solution ( 20 mg/ml , 0 . 5 mg/g body weight ) and transcardially perfused with 15–20 ml of ice-cold carbongenated ( comprising 95% O2/5% CO2 , pH 7 . 33–7 . 38 ) cutting solution containing ( in mM ) : 194 sucrose , 30 NaCl , 4 . 5 KCl , 1 . 2 NaH2PO4 , 0 . 2 CaCl2 , 8 MgCl2 , 26 NaHCO3 , and 10 D- ( + ) -glucose ( 360 mOsm ) . The slices were incubated in artificial cerebral spinal fluid ( aCSF ) at 32°C for 10 min followed by in fresh aCSF at room temperature for at least 1 hr . These slices were then transferred to a slice recording chamber for electrophysiology . The aCSF contained ( in mM ) : 119 NaCl , 2 . 3 KCl , 1 . 0 NaH2PO4 , 26 . 2 NaHCO3 , 11 Glucose , 1 . 3 MgSO4 , 2 . 5 CaCl2 ( pH 7 . 4 , 295–305 mOsm ) . Whole-cell patch clamp recordings were performed with IR-DIC . All recordings were conducted at room temperature . The electrophysiological current-clamp recordings were filtered at 10 kHz and sampled at 10 kHz at gain = 1 while that for voltage-clamp recordings were filtered at 2 kHz and sampled at 10 kHz at gain = 5 . The intracellular pipette solution for patching neurons in Figure 5A–C , Figure 5—figure supplement 1C–D , Figure 6A bottom , Figure 6C contained ( in mM ) : 131 KGluconate , 17 . 5 KCl , 9 NaCl , 1 MgCl2 . 6H2O , 10 HEPES , 1 . 1 EGTA , 2 MgATP , 0 . 2 Na2GTP ( pH 7 . 3 , 300 mOsm ) . The intracellular pipette solution for patching astrocytes in Figure 5—figure supplement 1A contained ( in mM ) : 50 KGluconate , 8 NaCl , 1 MgCl2 , 10 HEPES , 13 K2SO4 , 2 MgATP , 0 . 4 NaGTP , 40 BAPTA tetrapotassium salt ( sc-202076 , Santa Cruz Biotech , Texas , USA ) ( pH 7 . 3 , 300 mOsm ) . Alexa Fluor 633 Hydrazide ( A633 , A30634 , ThermoFisher Scientific , MA , USA ) was also included to allow the observation of the spread of BAPTA in the astrocyte syncytium . The A633 was imaged with an orange-red excitation filter/far red emission filter . The intracellular pipette solution for patching neurons in Figure 6A top and Figure 6B contained ( in mM ) : 103 CsCl , 12 Cs-methanesulfate , 5 TEA-Cl , 4 NaCl , 10 HEPES , 0 . 5 EGTA , 4 MgATP , 0 . 3 Na2GTP , 10 Phosphocreatine , 5 Lidocaine N-ethyl chloride ( pH 7 . 4 , 300 mOsm ) . Glass pipettes ( 3–5 MΩ , KG33 , King Precision ) were pulled with a Sutter P-97 puller ( Sutter instruments ) . CNO ( 10 mM ) was applied by pressure injection with a picospritzer ( 20 psi , 200 ms ) . In Figure 6A top , Figure 6B , 2 , 3-Dioxo-6-nitro-1 , 2 , 3 , 4-tetrahydrobenzo[f]quinoxaline-7-sulfonamide ( 10 µM NBQX , ab120045 , Abcam , MA , USA ) was bath-applied . In Figure 6A bottom , Figure 6C , picrotoxin ( 50 µM PTX , 1128 , Tocris , Bristol , UK ) was included in the intracellular pipette solution while SCH 50911 ( 20 µM , 9084 , Tocris , Bristol , UK ) was bath applied . In a select number of neurons , both PTX and SCH 50911 were included in the intracellular pipette solution and the data are also included in the population data in Figure 6C . GFP labeled NPY or POMC ARC neurons were visualized with an Olympus BX61WI microscope ( Olympus , Tokyo , Japan ) coupled with a 40x water immersion lens ( Olympus ) . Recordings were performed with a multiclamp 700B amplifier and digidata 1440A data acquisition system , with pClamp software in the current-clamp mode . Analysis was performed with the Clampfit 10 . 2 . 0 . 12 software ( Molecular Devices ) . The input resistances of the AgRP and POMC neurons patched in Figures 5–6 , Figure 5—figure supplement 1C–D were comparable and agreed with previous literature ( Fu and van den Pol , 2010 ) . The CNO-induced change in mean Vm in Figures 5C , 6C and Figure 5—figure supplement 1D was defined as ( Averaged membrane potential ) after drug - ( Averaged membrane potential ) before drug where ( Averaged membrane potential ) was computed as the averaged membrane potential over 10 s . In Figure 6B , the amplitude and frequency of each event in the 120 s segment before and after CNO application was quantified before the population average was taken over all events in the respective segments . The population averaged amplitude or frequency of each segment was then normalized by the population averaged amplitude or frequency computed across events within the 120 s segment before CNO application . Coronal brain slices of ARC ( 300 µm ) were prepared from AAV-Gfap-hM3D ( Gq ) -mCherry injected C57BL/6 mice as described above . The slices were placed on a porous membrane in a 3 . 5 mm dish bathed in carbogenated aCSF . Oregon Green 488 BAPTA-1 AM ( OGB ) dye solution was prepared by adding 1 µl Pluronic F-127 ( 20% Solution in DMSO , P3000MP , ThermoFisher Scientific , MA , USA ) and 49 µl DMSO to 50 µg of OGB ( O6807 , ThermoFisher Scientific ) . Bulk loading of the dye solution was performed by adding 3 µl of this dye solution over the ARC . After incubation at room temperature in the dark for 40 min , the slices were transferred to fresh aCSF for an additional 30 min before the experiment . hM3D ( Gq ) -mCherry glia in ARC were visualized with an Olympus BX61WI upright microscope under a 40x/0 . 80 W Olympus LUMPlanFL N water immersion lens and an Olympus CellSens dimension ( v 1 . 6 ) -driven Hamamatsu ORCA-R2 digital CCD camera ( C10600 ) . Fluorescence was continuously excited using a Lumen 200/220 ( Prior Scientific , MA , USA ) . OGB1-AM and mCherry were imaged with a blue excitation filter/green emission filter and a green excitation filter/red emission filter , respectively . Images were continuously acquired at 1 . 25 Hz , using 1x1 binning . Images were saved by the CellSens software as tagged image fileformat files and analyzed with both ImageJ ( NIH ) and custom-written MATLAB scripts . Individual cells were circled manually based on the mCherry expression . The mean fluorescence for each cell was computed from frame to frame , giving a time-varying intensity signal for each cell . After correction for lamp flicker noise and dye photobleaching , dFF was calculated by subtracting the local corresponding baseline ( F ) from the peak of the response to get dF before taking the ratio by F . Increases in dFF in ROIs indicated an increase in Ca2+ concentration . Analysis was performed using GraphPad Prism ( La Jolla , CA ) . Two-way ANOVA followed by Bonferroni post hoc tests and standard two-tailed paired or unpaired t-test were used as indicated in text . A p value of less than 0 . 05 was considered significant in these studies . Error bars indicate S . E . M . Blind experiments were performed for the Fos and slice physiology studies . Blind experiments were not performed for the behavior studies but the same criteria were applied to all allocated groups for comparisons . No randomization was performed for the study . No statistical methods were used to predetermine sample sizes , but our sample sizes are similar to those reported to previous publications ( Cowley et al . , 2001; Krashes et al . , 2011; Wu et al . , 2014 ) . | Neurons in an area of the brain called the hypothalamus control how much an animal eats . However , it is not clear what role other brain cells , such as glial cells , might play in influencing feeding . Glial cells do not send nerve impulses like neurons , but instead they mostly serve to support and protect the neurons . Now , Chen et al . changed the activity of a particular kind of glial cell , known as astrocytes , to explore what effect this has on how much mice eat . Astrocytes are unique amongst glial cells because they can respond to neuronal activity and release chemicals that change the activity of other cells , including neurons . The experiments revealed that switching astrocytes on in the hypothalamus made mice eat more , while turning them off had the opposite effect and reduced feeding . Chen et al . also found that glial cells partner with and change the activity of a particular group of neurons , known as the AgRP/NPY-expressing neurons . These neurons were already known to increase feeding activity when they become more active . In contrast , Chen et al . showed that glial cells do not affect the activity of another group of neurons , known as POMC-expressing neurons . Previous research had shown that mice eat less when their POMC-neurons are more active . Together the findings reveal that , within the hypothalamus , an interaction between glial cells and neurons influences how much an animal will eat . Further work is now required to understand the exact interaction between the glial cells and neurons , and to find out if other kinds of glial cells also have a role in controlling feeding . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Direct modulation of GFAP-expressing glia in the arcuate nucleus bi-directionally regulates feeding |
HAC1 encodes a transcription factor that is the central effector of the unfolded protein response ( UPR ) in budding yeast . When the UPR is inactive , HAC1 mRNA is stored as an unspliced isoform in the cytoplasm and no Hac1 protein is detectable . Intron removal is both necessary and sufficient to relieve the post-transcriptional silencing of HAC1 mRNA , yet the precise mechanism by which the intron prevents Hac1 protein accumulation has remained elusive . Here , we show that a combination of inhibited translation initiation and accelerated protein degradation—both dependent on the intron—prevents the accumulation of Hac1 protein when the UPR is inactive . Functionally , both components of this fail-safe silencing mechanism are required to prevent ectopic production of Hac1 protein and concomitant activation of the UPR . Our results provide a mechanistic understanding of HAC1 regulation and reveal a novel strategy for complete post-transcriptional silencing of a cytoplasmic mRNA .
The unfolded protein response ( UPR ) is a eukaryotic stress response pathway that is activated when unfolded proteins accumulate in the endoplasmic reticulum ( ER ) lumen ( Gardner et al . , 2013 ) . In the budding yeast Saccharomyces cerevisiae , the central effector of the UPR is the transcription factor Hac1p ( Cox and Walter , 1996; Mori et al . , 1996; Nikawa et al . , 1996 ) . HAC1 ( and its metazoan ortholog Xbp1 ) is unique among eukaryotic genes in that it contains an intron that is excised through an unconventional cytoplasmic splicing reaction mediated by two proteins , Ire1p and tRNA ligase , rather than the spliceosome ( Cox and Walter , 1996; Kawahara et al . , 1997; Sidrauski and Walter , 1997; Sidrauski et al . , 1996 ) . In the absence of ER protein-folding stress , the intron-containing mRNA ( denoted HAC1u; ‘u’ for UPR ‘uninduced’ ) is transcribed and exported to the cytoplasm but does not give rise to detectable protein due to the presence of the inhibitory intron ( Cox and Walter , 1996; Chapman and Walter , 1997 ) . The accumulation of unfolded proteins in the ER lumen activates the ER-resident transmembrane kinase–endonuclease Ire1p , which cleaves out the intron from HAC1 mRNA via its cytoplasmic nuclease domain ( Sidrauski and Walter , 1997 ) . After the exons are joined by tRNA ligase , the resulting spliced mRNA ( denoted HAC1i; ‘i’ for UPR ‘induced’ ) is now translated into Hac1ip . This active transcription factor is imported into the nucleus , where it activates the expression of UPR target genes involved in restoring protein-folding homeostasis in the ER ( Chapman et al . , 1998 ) . Intron removal is both necessary and sufficient to relieve the post-transcriptional silencing of HAC1 that otherwise prevents Hac1p accumulation ( Chapman and Walter , 1997 ) . The post-transcriptional silencing of HAC1 and its subsequent reversal by cytoplasmic splicing together enable a rapid UPR that does not depend on de novo transcription ( Rüegsegger et al . , 2001 ) . At the same time , a robust silencing mechanism is required to prevent ectopic accumulation of Hac1up from the abundant cytoplasmic pool of HAC1u mRNA that might otherwise turn on UPR target genes in the absence of ER stress . The current model for silencing is that elongating ribosomes are stalled on the mRNA during translation , thereby preventing synthesis of full-length Hac1p ( Rüegsegger et al . , 2001 ) . According to this model , the mediator of translational attenuation is a long-range base-pairing interaction between the 5′ untranslated region ( UTR ) and intron of HAC1u mRNA . The key data supporting the stalled elongation model is that the majority of HAC1u mRNA sediments in the polysome region of a sucrose gradient ( Arava et al . , 2003; Chapman and Walter , 1997; Cox and Walter , 1996; Kuhn et al . , 2001; Mori et al . , 2010; Park et al . , 2011; Payne et al . , 2008; Rüegsegger et al . , 2001; Sathe et al . , 2015 ) despite no detectable Hac1up . Furthermore , the heavy-sedimenting HAC1u mRNA is distributed in a discontinuous pattern with peaks and valleys that precisely match the peaks and valleys observed for polysomes ( Rüegsegger et al . , 2001 ) . These data provide convincing evidence that heavy-sedimenting HAC1u mRNA reflects ribosome association rather than another high-molecular-weight complex that co-sediments with polysomes , or so-called ‘pseudo-polysomes’ ( Thermann and Hentze , 2007 ) . Given this apparent ribosome association of HAC1u mRNA , an alternative explanation for the absence of Hac1up is that Hac1up is synthesized but immediately degraded ( Cox and Walter , 1996 ) . However , Hac1up and Hac1ip are thought to have similar half lives ( Chapman and Walter , 1997; Kawahara et al . , 1997 ) , arguing against differential protein degradation as the primary mechanism that prevents Hac1up accumulation yet allows Hac1ip accumulation . Despite widespread acceptance of the stalled elongation model ( Richter and Coller , 2015 ) , the mechanism by which base pairing between untranslated regions causes ribosomes to stall in the open reading frame ( ORF ) is unknown . The reduced efficiency of translational attenuation in vitro suggested that additional factors might be involved in the inhibitory mechanism ( Rüegsegger et al . , 2001 ) , perhaps acting to transduce the signal from the untranslated regions to translating ribosomes . However , the sequence of the base-pairing region can be changed without affecting silencing if base pairing is preserved ( Rüegsegger et al . , 2001 ) , making it unlikely that any sequence-specific RNA-binding proteins are involved . In addition , the more recent discovery of the No-Go Decay ( NGD ) pathway that recognizes stalled ribosomes and targets the associated mRNA for degradation ( Doma and Parker , 2006 ) raises the question of how the ribosome–HAC1u mRNP evades detection and subsequent turnover by the quality-control machinery . For these reasons , and others described below , we revisited the stalled elongation model , which led us to instead identify an entirely different mechanism of post-transcriptional silencing of HAC1u mRNA that reconciles these issues .
We recently reported improved ribosome-footprint profiles ( Ingolia et al . , 2009 ) and mRNA-abundance measurements from exponentially growing S . cerevisiae ( Weinberg et al . , 2016 ) . Under these growth conditions HAC1 mRNA is almost entirely unspliced ( Figure 1—figure supplement 1A ) and is distributed across a sucrose gradient with ~50% in the non-translating fractions and the remaining ~50% extending across all of the translating ( i . e . , 80S and larger ) fractions without substantial enrichment in any particular fraction ( Figure 1A ) , similar to previous observations ( Arava et al . , 2003; Chapman and Walter , 1997; Cox and Walter , 1996; Kuhn et al . , 2001; Mori et al . , 2010; Park et al . , 2011; Payne et al . , 2008; Rüegsegger et al . , 2001; Sathe et al . , 2015 ) . In contrast , the well-translated actin ( ACT1 ) mRNA is essentially absent from the non-translating fractions and is found mostly in large polysomes . Based on the sedimentation of HAC1 mRNPs , we predicted that the mRNA would generate a large number of ribosome-protected fragments , in a quantity only ~2-fold fewer than similarly abundant mRNAs ( based on the fraction of HAC1 mRNA in the untranslated fractions ) . Strikingly , however , after normalizing for mRNA abundance HAC1 generates the fewest ribosome-protected fragments among all expressed yeast genes ( Figure 1B ) —and ~50-fold fewer than expected from the polysome profile . Rather than providing evidence for stalled ribosomes on HAC1u mRNA , instead these observations suggest that either ribosomes are stalled on the mRNA in a closely packed configuration that prevents nuclease cleavage between ribosomes , which would eliminate the ~28-nucleotide fragments that are sequenced in the ribosome-profiling method ( Figure 1C , middle ) ; or that there are not stalled ribosomes on HAC1u mRNA ( Figure 1C , bottom ) . 10 . 7554/eLife . 20069 . 003Figure 1 . Ribosome density on unspliced HAC1 mRNA . ( A ) Polysome analysis of HAC1 and ACT1 mRNAs . Extracts prepared from exponentially growing yeast cells were fractionated on 10–50% sucrose gradients , with absorbance at 260 nm monitored ( top ) . The relative distributions of HAC1 and ACT1 mRNAs across fractions were determined by qRT-PCR ( bottom ) . Shown are the mean ± SEM with n = 2 ( i . e . , the range ) , expressed as a fraction of the total mRNA detected . ( B ) Histogram of ribosome densities measured by ribosome profiling and RNA-seq . The ratio of the number of ribosome-protected fragments ( RPFs ) to the number of RNA-seq reads ( mRNA counts ) was calculated for each of 4838 expressed yeast genes ( data from Weinberg et al . , 2016 ) . Shown is the distribution of log-transformed ratios in bins of 0 . 05 , with the position of HAC1 indicated . ( C ) Possible scenarios to explain a lack of RPFs . While an mRNA with average ribosome density will generate many ~28 nucleotide ( nt ) RPFs ( top ) , the close packing of stacked ribosomes could inhibit the RNase digestion between ribosomes required to generate ~28 nt RPFs ( middle ) . Alternatively , an mRNA that does not contain translating ribosomes would not generate RPFs ( bottom ) . ( D ) Polysome analysis of HAC1 mRNA variants with shortened ORFs . G-to-T point mutations were introduced into the first exon of HAC1 to generate premature stop codons , with the resulting ORFs shown as thick colored boxes ( constitutive 5′- and 3′-UTRs located within exons 1 and 2 , respectively , are shown as thin black lines; other untranslated regions are depicted as thin colored boxes; and the coding regions of exons 1 ( teal ) and 2 ( purple ) are labeled as 'HAC1 exon1' and ‘exon2’ , respectively ) . The maximum number of ribosomes that could be accommodated was calculated based on each ribosome occupying 28 nt . Polysome analysis was performed as in ( A ) , with data for wild-type HAC1 from ( A ) duplicated for comparison . ( E ) Effects of heparin on polysome analysis . Purified uncapped luciferase ( luc ) RNA was added to either lysate or lysis buffer in the absence ( – ) or presence ( + ) of 0 . 2 mg/ml heparin . Polysome analysis was performed as in ( A ) with absorbance at 260 nm monitored ( top ) , and the relative distributions of exogenous luc RNA ( middle ) and endogenous HAC1 and ACT1 mRNAs ( bottom; in lysate only ) were determined . ( F ) Refined polysome analysis of HAC1 mRNAs . Extracts were prepared in heparin-containing lysis buffer from strains shown in ( D ) . Polysome analysis was performed as in ( A ) . ( G ) Polysome analysis of HAC1 mRNAs during the UPR . Strains shown in ( D ) were grown to mid-log phase and treated with 8 mM DTT for 20 min before harvesting . Extracts were prepared in heparin-containing lysis buffer , and polysome analysis was performed as in ( A ) . Dotted lines indicate the fractions after which the corresponding color-coded mutant mRNAs would not be expected to sediment based on ORF length . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 00310 . 7554/eLife . 20069 . 004Figure 1—figure supplement 1 . Splicing of HAC1 mRNA . ( A ) Design of RT-PCR assay for HAC1 splicing . To differentiate between the HAC1u and HAC1i mRNAs , reverse-transcription products are PCR amplified using a pair of primers that flank the intron . The resulting PCR products are separated by agarose gel electrophoresis , with a size difference of 252 base pairs ( bp ) corresponding to the size of the intron . ( B ) Splicing status of HAC1 mRNA . Total RNA was extracted from exponentially growing yeast cells of the indicated genotypes . RT-PCR across the intron was used to differentiate between unspliced ( black arrow ) and spliced ( red arrow ) HAC1 mRNA . The small percentage of spliced HAC1 mRNA observed in wild-type cells is IRE1 dependent , as expected . ( C ) Splicing of HAC1 mRNA variants upon DTT treatment . Strains expressing the indicated HAC1 variants were grown to mid-log phase and either left untreated ( –DTT ) or treated with 8 mM DTT for 20 min ( +DTT ) before harvesting . Total RNA was extracted and used for RT-PCR analysis as in ( A ) . The same cultures ( two of each strain ) were used for polysome analysis in Figure 1F and G . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 004 Although the polysome-like sedimentation of HAC1u mRNA indicates ribosome association , it does not reveal if the ribosomes associated with the mRNA were ever engaged in its translation . Alternatively , the associated ribosomes might be bound in a conformation that is unrelated to translation of HAC1u mRNA . We therefore devised an experiment to definitively determine whether the ribosomes bound to HAC1u mRNA reflect ribosomes that were engaged in its translation . To do so , we took advantage of the observation that HAC1 mRNA is distributed across all of the translating fractions of a sucrose gradient ( Figure 1A ) . If the deep sedimentation is due to multiple translating ribosomes being stalled on a single mRNA , then reducing the number of translating ribosomes that can fit on the mRNA should shift the sedimentation pattern toward lighter fractions . We designed a series of constructs containing point mutations in the first exon of HAC1 that created premature termination codons , which reduce the size of the ORF and thereby limit the number of translating ribosomes ( Figure 1D , top ) . To ensure that the mutant alleles were expressed at near wild-type levels , we replaced the endogenous HAC1 allele without disrupting flanking regulatory regions . Remarkably , each of the mutant mRNAs had a sedimentation pattern that was indistinguishable from the wild-type mRNA ( Figure 1D , bottom ) . In the most extreme case , the mRNA containing a 21-nucleotide ORF that can only accommodate a single translating ribosome still co-sedimented with polysomes containing upwards of 10 ribosomes ( fraction 14 of the gradient ) . These data provide direct evidence that the polysome-like sedimentation of HAC1u mRNA is not due to stalled ribosomes on the mRNA . We hypothesized that the ribosome association of HAC1u mRNA was instead due to non-specific interactions between the mRNA and bona fide polysomes formed on other mRNAs . To evaluate the extent of such non-specific interactions , we introduced an exogenous control RNA that should not be translated: an uncapped luciferase-encoding RNA purified from an in vitro transcription reaction . We analyzed the sedimentation behavior of this control RNA when it was added to lysis buffer compared to when it was added to the yeast lysate prior to centrifugation . Surprisingly , some of the exogenous RNA was found in the translating fractions of the lysate—a behavior not observed in lysis buffer alone ( Figure 1E , middle ) . Through extensive optimization , we found that the addition of heparin to the lysis buffer ( at a concentration of 0 . 2 mg/ml ) was sufficient to largely prevent the deep sedimentation of exogenous RNA . The addition of heparin also had a major effect on the sedimentation of HAC1 mRNA , as most ( 83% ) now sedimented in the non-translating fractions of the gradient ( Figure 1E , bottom ) . In contrast , the sedimentation of ACT1 mRNA ( Figure 1E , bottom ) and the overall polysome profile ( Figure 1E , top ) were largely unchanged , suggesting that heparin competed away non-specific interactions without disrupting bona fide polysomes . Based on these results , we re-analyzed the HAC1 mutants with shortened ORFs using heparin-containing lysis buffer . Under these conditions , the polysome co-sedimentation of each of the mRNAs was greatly reduced ( from ~50% to < 20% ) but there was still no difference among the constructs ( Figure 1F ) , providing further evidence against stalled ribosomes on HAC1u mRNA . Importantly , when we treated cells with the reducing agent dithiothreitol ( DTT ) to induce the UPR and concomitant splicing of HAC1 mRNA ( Figure 1—figure supplement 1C ) and repeated the experiment , the variant mRNAs now displayed the expected differential sedimentation based on ORF length ( Figure 1G ) , validating our experimental design . Together , these results demonstrate that at steady state HAC1u mRNA is not associated with either actively elongating or stalled ribosomes . This indicates that the primary block to production of Hac1up is at the stage of translation initiation , not translation elongation as previously proposed ( Chapman and Walter , 1997; Richter and Coller , 2015; Rüegsegger et al . , 2001 ) . The absence of ribosomes on HAC1u mRNA suggests that translation initiation is inhibited . To further understand the mechanism of this inhibition , we used a green fluorescent protein ( GFP ) reporter system that has been previously shown to recapitulate post-transcriptional silencing by the HAC1 intron ( Chapman and Walter , 1997; Rüegsegger et al . , 2001 ) . We replaced the first exon of HAC1 with the GFP ORF lacking its own stop codon ( Figure 2B ) , which allowed us to quantitatively analyze post-transcriptional silencing of GFP using a combination of flow cytometry ( protein abundance ) , quantitative RT-PCR ( mRNA abundance ) , and sucrose gradient fractionation ( ribosome density ) . When GFP was embedded in an otherwise wild-type HAC1 context the mRNA sedimented almost entirely in the non-translating fractions ( Figure 2A ) , and there was no detectable fluorescence above background ( Figure 2C ) . In contrast , cells expressing a reporter construct missing the entire intron displayed a strong fluorescence signal ( Figure 2C ) , and most of the mRNA was found in the translating fractions ( Figure 2A ) . Thus , the reporter GFP mRNA behaves similarly to endogenous HAC1 mRNA . 10 . 7554/eLife . 20069 . 005Figure 2 . Contribution of long-range base pairing to intron-dependent silencing . ( A ) Polysome analysis of reporter mRNAs . Extracts were prepared in heparin-containing lysis buffer from strains expressing the GFP reporter mRNAs depicted in ( B ) . Polysome analysis was performed as in Figure 1A , with data for wild-type HAC1 from Figure 1F duplicated for comparison . ( B ) Design of reporter mRNAs . Constructs are depicted as in Figure 1D , with the dotted line indicating a deleted region . Colored stars indicate mutations to the base-pairing region , with specific nucleotide changes shown below in red . ( C ) Flow cytometry analysis of reporter strains . Strains expressing the GFP reporter mRNAs depicted in ( B ) were grown to mid-log phase and analyzed by flow cytometry . Plotted is the median GFP intensity ( normalized to cell size ) of the cell population relative to background fluorescence in the wild-type ( no GFP ) strain with error bars indicating quartiles of the cell population , all averaged across replicates ( n = 2–7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 00510 . 7554/eLife . 20069 . 006Figure 2—figure supplement 1 . Characterization of GFP reporter strains . ( A ) RNA abundance measurements for GFP reporter mRNAs . Total RNA was extracted from strains expressing the indicated mRNAs , using the same cultures as for flow cytometry analyses in Figure 2C . qRT-PCR was used to measure the abundances of GFP reporter mRNAs relative to ACT1 mRNA , with all data normalized to the abundance of the ‘wild-type’ GFP reporter mRNA ( construct 2 ) . Shown are the mean ± SD ( n = 2–7 ) . ( B ) Immunoblotting analysis of reporter strains . Proteins were extracted from mid-log-phase cultures of strains expressing the indicated reporter mRNAs depicted in ( A ) . Immunoblots for the GFP reporter and actin loading control are shown . No signal for GFP is observed in the base-pairing mutants ( constructs 5 and 6 ) even after overexposing the blot , consistent with the corresponding flow cytometry results ( Figure 2C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 006 To determine whether 5′-UTR–intron base pairing is required to prevent ribosome loading on the reporter mRNA , we designed constructs in which the sequence implicated in base-pairing interactions in either the 5′-UTR or the intron was mutated to its complement to disrupt the interaction ( Figure 2B , bottom ) , as was done previously ( Rüegsegger et al . , 2001 ) . Both mutant mRNAs sedimented mostly in the translating fractions in a manner that was similar to the intronless construct ( Figure 2A ) . When we combined the 5′-UTR and intron mutations and thereby restored base pairing , the mRNA was now found almost exclusively in the non-translating fractions and resembled the original reporter mRNA . These results demonstrate that base pairing between the 5′-UTR and intron is required to prevent ribosome loading by directly impeding the binding or progress of the scanning ribosome . Given that the base-pairing mutant mRNAs were loaded with ribosomes similarly to the intronless mRNA ( Figure 2A ) , we expected to observe GFP expression from the mutant mRNAs that was similar to that of the intronless construct . Remarkably , however , neither of the strains expressing a base-pairing mutant mRNA had any GFP detectable by either flow cytometry ( Figure 2C ) or immunoblotting ( Figure 2—figure supplement 1 ) . The lack of GFP signal despite polysome sedimentation was not due to low mRNA abundance , as the mutant mRNAs were present at similar levels compared to the intronless mRNA ( Figure 2—figure supplement 1A , compare construct 3 with constructs 5–6 ) . These results suggest that an additional silencing mechanism acting downstream of translation initiation prevents GFP accumulation when base pairing is disrupted . In addition to removing the intron portion of the base-pairing interaction , splicing of HAC1 mRNA also alters the C-terminal tail of the encoded protein: The ORF in HAC1u mRNA has a 10-amino-acid tail encoded by the intron , which in HAC1i mRNA is replaced by an 18-amino-acid tail encoded by the second exon ( Figure 3A ) . The fact that the polypeptide tails encoded by the HAC1u and HAC1i mRNAs are different suggests that the 10-amino-acid tail unique to Hac1up may be functionally important . We therefore hypothesized that the intron-encoded C-terminal tail may be involved in the additional silencing mechanism revealed by our base-pairing mutant reporters ( Figure 2 ) . 10 . 7554/eLife . 20069 . 007Figure 3 . Post-translational silencing mediated by the intron-encoded C-terminal tail . ( A ) Schematic of HAC1 mRNA splicing . The proteins encoded by HAC1u and HAC1i mRNAs differ in their C-terminal tails , with the amino acid sequences indicated . ( B ) Design of reporter mRNAs . Black shading indicates recoding , with the original and recoded sequences depicted below ( mutations in red ) . Untranslated regions colored red correspond to those of the TMA7 mRNA , with the reporter gene integrated at the TMA7 rather than HAC1 locus . Otherwise constructs are depicted as in Figure 2B . ( C ) Flow cytometry analysis of reporter strains . Strains expressing the GFP reporter mRNAs depicted in ( B ) were analyzed as in Figure 2C , with data for the first three strains duplicated from Figure 2B for comparison . ( D ) Polysome analysis of reporter mRNAs . Extracts were prepared in heparin-containing lysis buffer from strains expressing the GFP reporter mRNAs indicated in ( B ) . Polysome analysis was performed as in Figure 1A , with data for the wild-type and intronless GFP reporters from Figure 2A duplicated for comparison . ( E ) Differentiating between co-translational and post-translational silencing mechanisms . Left: Schematic of reporter construct that generates two separate polypeptides from each round of translation . Right: Extracts were prepared from strains expressing reporter mRNAs that either encoded the 10-amino-acid C-terminal tail of Hac1up ( +tail ) or contained a stop codon just before the tail ( –tail ) . Immunoblotting was used to detect HA-tagged mRuby ( top ) and GFP ( bottom ) , with actin as a loading control . Two biological replicates are shown for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 00710 . 7554/eLife . 20069 . 008Figure 3—figure supplement 1 . Additional analyses of GFP reporter constructs . ( A ) RNA abundance measurements for GFP reporter mRNAs , analyzed as in Figure 2—figure supplement 1A using the same cultures as for flow cytometry analyses in Figure 3C . ( B ) Flow cytometry and RNA analyses of additional reporter strains . Strains expressing the indicated GFP reporter mRNAs ( left ) were analyzed by flow cytometry as in Figure 2C ( middle ) or qRT-PCR as in Figure 2—figure supplement 1A ( right ) , with data for the first four strains duplicated from Figure 3C and Figure 3—figure supplement 1A for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 00810 . 7554/eLife . 20069 . 009Figure 3—figure supplement 2 . Identifying a 2A peptide sequence that is active in S . cerevisiae . ( A ) Design of the 2A reporter construct . In the absence of 2A activity , a 57–58 kDa polypeptide ( * ) is produced that contains both an HA tag and GFP . When the 2A sequence is active , translation of the reporter generates a 30–31 kDa HA-tagged protein and a 27 kDa GFP . Shown are the 30-amino-acid sequences of the 2A peptides that were tested , with each preceded by a GlySerGly linker ( gray ) and ‘cleaved’ at the site marked with an arrow . The region corresponding to a minimal 20-amino-acid T2A peptide ( ‘minT2A’ ) that has been used previously in yeast ( Beekwilder et al . , 2014 ) is indicated . Minimal versions of the F2A peptide have also been used in yeast ( Doronina et al . , 2008; de Felipe et al . , 2003; Sharma et al . , 2012; Voordeckers et al . , 2012 ) but with ‘cleavage’ efficiencies that were low or not directly measured . ( B–C ) Analysis of 2A activity by immunoblotting . BY4741 was transformed with a pRS413-derived centromeric plasmid containing the indicated reporter construct under control of the GPD promoter . Proteins were extracted from mid-log-phase cultures and analyzed for HA- ( B ) and GFP- ( C ) tagged proteins by immunoblotting . Only the P2A peptide causes efficient separation of the upstream and downstream proteins ( indicated by arrows ) . The nucleotide sequences encoding the 30-amino-acid peptides are:F2A: CACAAACAAAAGATTGTTGCGCCTGTGAAACAGCTTTTGAACTTTGACCTGCTCAAGTTGGCAGGAGACGTCGAGTCCAACCCTGGGCCTE2A: AGACATAAATTTCCCACTAACATCAACAAACAGTGTACTAATTACTCTCTCCTCAAATTGGCTGGAGATGTTGAGAGCAACCCTGGC CCCP2A: GCTATGACTGTGATGACATTCCAGGGACCAGGTGCAACAAACTTCTCCCTCTTGAAACAAGCAGGAGATGTTGAGGAAAATCCCGGGCCTT2A: CGGGGGCCTCGCCCCCAAAACCTTGGGGTAAGGGCCGAGGGCAGGGGAAGTCTTCTAACATGCGGGGACGTGGAGGAAAATCCCGGCCCCDOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 009 To investigate this possibility , we designed a reporter construct in which we removed the entire intron except for the 5′ end that codes for the 10-amino-acid tail of Hac1up ( Figure 3B , fourth construct ) . Remarkably , cells expressing this construct had no detectable GFP ( Figure 3C ) , despite the corresponding mRNA being abundant ( Figure 3—figure supplement 1A ) and detected on polysomes due to the absence of the base-pairing interaction ( Figure 3D ) . Thus , the coding region at the 5′ end of the intron is sufficient for complete silencing of the GFP reporter . Introducing a termination codon between GFP and the intron restored robust fluorescence comparable to that observed for the intronless construct , which implicated translation of the 5′ end of the intron as required for silencing . Furthermore , making 10 nucleotide changes that maintained the coding potential of the intron-encoded tail ( Figure 3B , bottom ) had no effect on silencing ( Figure 3C ) , suggesting that the amino-acid sequence encoded by the 5′ end of the intron is more important than the nucleotide sequence itself . To determine whether the 10-amino-acid element alone was sufficient for silencing in the absence of any other HAC1 sequences , we expressed GFP from a different locus in the yeast genome ( TMA7 ) either with or without the 10-amino-acid tail . When the 10-amino-acid sequence was either absent or not translated due to a premature stop codon , we observed robust GFP signal ( Figure 3C ) . In contrast , there was no fluorescence detected in cells expressing GFP containing the 10-amino-acid tail . Thus , the C-terminal tail of Hac1up is sufficient for silencing independently of the rest of the HAC1 intron and any other HAC1 sequences . Having established the effects of the 10-amino-acid tail in isolation , we next examined the effects of the tail when translation initiation was inhibited by the base-pairing interaction . In an otherwise wild-type HAC1 context , preventing translation of the 10-amino-acid tail either by deleting the entire sequence or by introducing a stop codon caused GFP to accumulate to low but detectable levels ( Figure 3C ) without greatly affecting the sedimentation of the corresponding mRNAs ( Figure 3D ) . Thus , even when base pairing is intact there is some low-level accumulation of GFP that is normally suppressed by the 10-amino-acid silencing element . In the context of reporter constructs lacking the intron-encoded C-terminal tail , mutations in either the 5′-UTR or intron that disrupted base pairing greatly increased the amount of GFP , while the compensatory double-mutant construct with restored base pairing had only low levels of GFP ( Figure 3C and Figure 3—figure supplement 1B ) . These results are in agreement with previous GFP reporter experiments , which used constructs that contained a stop codon between the GFP and intron sequences and therefore had inadvertently eliminated the effects of the 10-amino-acid tail ( Chapman and Walter , 1997; Rüegsegger et al . , 2001 ) . Together , our GFP reporter experiments indicate that the HAC1 intron mediates post-transcriptional silencing through a pair of independent but partially redundant mechanisms: base-pairing interactions with the 5′-UTR that inhibit translation initiation , and a novel silencing mechanism mediated by the intron-encoded C-terminal tail of Hac1up . Robust expression requires that both silencing mechanisms be inactivated , as would happen simultaneously when the intron is removed by Ire1p-dependent splicing . What is the mechanism by which the intron-encoded tail of Hac1up silences gene expression ? Because disrupting translation of the tail elevated protein levels ( Figure 3C ) without affecting polysome formation ( Figure 3D ) , we inferred that the tail was exerting its effect downstream of translation initiation . We therefore reasoned that the 10-amino-acid sequence was acting either by halting translation across the entire mRNA ( after polysome formation ) ; or by promoting protein degradation . Our inability to detect GFP containing the 10-amino-acid tail ( Figure 3C ) prevented us from directly comparing the half lives of GFP with and without the tail . Instead , to distinguish between stalled elongation and protein degradation , we designed a reporter construct that generates two separate polypeptides from a single round of translation through co-translational ‘cleavage’ mediated by a viral 2A peptide ( Sharma et al . , 2012 ) . After screening for a 2A peptide sequence that functions efficiently in S . cerevisiae ( Figure 3—figure supplement 2 ) , we designed a construct containing HA-mRuby and GFP sequences separated by the P2A peptide ( derived from porcine teschovirus-1 ) . The GFP sequence downstream of P2A was appended with the 10-amino-acid tail ( or was not , as a control ) , which should lead to the absence of detectable GFP regardless of the mechanism of action . In contrast , the upstream HA-tagged mRuby should only be detected if translation itself is not affected by the 10-amino-acid sequence , since it reports on the number of rounds of translation but is not covalently linked to the inhibitory tail . Assaying for GFP by immunoblotting revealed that accumulation of the protein was suppressed by the 10-amino-acid tail , as expected ( Figure 3E ) . In contrast , HA-mRuby was detected at similar levels whether or not the tail was included in the construct . These results indicate that the tail functions downstream of translation , likely by acting as a ‘degron’ ( Varshavsky , 1991 ) that targets the protein for immediate degradation after synthesis ( Cox and Walter , 1996 ) . If the C-terminal tail of Hac1up functions as a degron , additional proteins may be involved in recognizing the degron and targeting the covalently linked protein for degradation . To identify such trans-acting factors , we used a genetic approach that took advantage of the strong silencing phenotype imparted by the 10-amino-acid tail alone ( Figure 3C ) . We constructed strains in which the first exon of HAC1 was replaced by HIS3 , which we reasoned might behave like the analogous GFP reporter genes and be completely silenced by the 10-amino-acid sequence . Similarly to GFP , replacing the first exon of HAC1 with HIS3 but keeping the HAC1 locus otherwise intact prevented expression of His3p as evidenced by histidine auxotrophy ( Figure 4A ) . Removing the entire intron restored His3p expression and growth of the corresponding strain on medium lacking histidine , suggesting that silencing was mediated by the HAC1 intron . Strains containing a stop codon between HIS3 and the intron ( to prevent translation of the 10-amino-acid degron ) had a low but detectable level of growth on histidine-lacking medium ( Figure 4A ) , consistent with the weak fluorescence signal observed from the corresponding GFP reporter construct ( Figure 3C ) . On the other hand , strains containing His3p appended with the C-terminal tail of Hac1up but no other elements of the HAC1 intron could not grow on medium lacking histidine . These results indicate that the degron is sufficient for functional silencing of HIS3 expression , providing a useful genetic tool to identify additional genes involved in the degron-dependent silencing mechanism . 10 . 7554/eLife . 20069 . 010Figure 4 . Identification of DUH1 through a genetic selection . ( A ) Evaluating a genetic reporter for intron-dependent silencing . Strains expressing the indicated HAC1-based HIS3 reporter mRNAs ( depicted as in Figure 2B ) either without ( – ) or with ( + ) an N-terminal HA tag were grown to saturation , and 10-fold dilution series were plated on either SC or SC–His media . ( B ) Flowchart of genetic selection for dds mutants . After selecting for spontaneous mutants that could grow on medium lacking histidine , restreaked clones were filtered out for cis mutants , verified to be expressing HA-His3p , and a subset analyzed by whole-genome sequencing . ( C ) Chromosome map of mutations identified by whole-genome sequencing . Each color corresponds to a different dds mutant strain . Shown on the right is the DUH1 locus ( YJL149W ) , with locations of nucleotide changes ( with respect to DUH1 start codon ) and corresponding amino changes listed below . ( D ) Effect of DUH1 disruption on expression of the genetic reporter . Strains expressing the indicated reporter mRNAs depicted in ( A ) in either a DUH1 or duh1∆ ( two independent clones ) background were grown to mid-log phase . Extracts were prepared and immunoblotted for HA-His3p and actin loading control . ( E ) Flow cytometry analysis of reporter strains . Strains expressing the indicated GFP reporter mRNAs in either a DUH1 or duh1∆ background were analyzed as in Figure 2C . Data is plotted as in Figure 2C ( n = 2 for all strains ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 01010 . 7554/eLife . 20069 . 011Figure 4—figure supplement 1 . Results and validation of the genetic selection . ( A ) Table of cis-acting mutations identified by Sanger sequencing . In the parental strain , the Hisp3-coding sequence is shown in gray while the 10-amino-acid degron is shown in black . Amino-acid changes caused by nucleotide mutations are shown in red . ( B ) Analysis of HA-His3p abundance in dds mutant strains . Protein extraction and immunoblotting were performed as in Figure 4D , using untagged BY4741 as a negative specificity control for the HA antibody . Strains indicated in red contain mutations in the HA-His3p +tail reporter gene . Strains that were subjected to whole-genome sequencing are indicated in blue and gray , with blue indicating those that contain a mutation in DUH1 and gray indicating those that do not . ( C ) Table of trans-acting mutations identified by whole-genome sequencing . Strains containing DUH1 mutations are listed in order of increasing total number of mutations . ( D ) Representative example of Sanger sequencing confirmation of DUH1 mutations . Shown are sequencing traces across a region of DUH1 for the parental selection strain and two dds mutants . Each dds mutant contains a nucleotide substitution ( highlighted in yellow ) , compared to both the reference sequence and parental selection strain , corresponding to the mutation identified by whole-genome sequencing . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 01110 . 7554/eLife . 20069 . 012Figure 4—figure supplement 2 . Additional analysis of GFP reporter constructs . RNA abundance measurements for GFP reporter mRNAs in the indicated strain backgrounds , analyzed as in Figure 2—figure supplement 1A using the same cultures as for flow cytometry analyses in Figure 4E . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 012 Although we had initially intended to use chemical mutagenesis to generate silencing-defective mutants , upon streaking out the selection strain on histidine-lacking medium we noticed that a small number of slow-growing colonies appeared even without mutagen treatment . We reasoned that such spontaneous suppressor strains would have very few mutations , making it possible to identify suppressor mutations by whole-genome sequencing without requiring backcrossing or forming complementation groups . Thus , to isolate mutants with defective degron-dependent silencing ( 'dds mutants' ) we simply plated the selection strain expressing HA-tagged His3p with the 10-amino-acid tail on histidine-lacking medium and isolated the rare single colonies that grew for further analysis ( Figure 4B ) . From five independent platings we isolated a total of 123 mutant strains that , when re-streaked , could grow on medium lacking histidine . Sanger sequencing of the C-terminal region of the reporter gene in each mutant identified 15 strains harboring a cis mutation that either altered the sequence of the degron , introduced a premature stop codon before or within the degron , or removed the stop codon of the degron resulting in a six-amino-acid C-terminal extension ( Figure 4—figure supplement 1A ) —all of which provided confirmation that the genetic selection worked as desired . From the strains that displayed histidine prototrophy , we picked 35 ( including four cis mutants ) to analyze by immunoblotting and found that all had detectable HA-His3p ( Figure 4—figure supplement 1B ) . We selected 20 of these strains with unknown mutations for whole-genome sequencing ( as well as the parental selection strain as a reference , and three strains with known mutations in the degron as positive controls for our variant-calling procedure ) , taking care to select strains that varied widely in HA-His3p abundance or growth rate to minimize our chances of sequencing the same mutation in multiple strains . We sequenced the 24 genomes together in a single lane of a HiSeq sequencer using 50-nucleotide single-end reads , which provided 38–74X coverage ( 9 . 3–18 . 0 million reads ) of each yeast genome . We then used standard mapping and variant-calling tools ( BWA and FreeBayes , respectively ) to identify variants that were absent from the parental genome , which successfully recovered the positive-control cis mutants . Remarkably , 17 out of the 20 strains that we sequenced contained a mutation in the same gene YJL149W ( Figure 4C and Figure 4—figure supplement 1C ) , and in each case we confirmed the mutation by Sanger sequencing ( Figure 4—figure supplement 1D ) . Because no other ORF was found mutated in more than one of the 20 mutant strains ( Figure 4—figure supplement 1C ) , we focused our follow-up efforts on YJL149W . YJL149W had previously been named DAS1 for 'Dst1-delta 6-Azauracil Sensitivity 1' when it was identified in a genetic screen unrelated to the UPR ( Gómez-Herreros et al . , 2012 ) . Based on the protein’s domain structure ( containing an F-box domain and leucine-rich repeats ) and physical interactions with the SCF core components Cdc53p and Skp1p ( Willems et al . , 1999 ) , YJL149W was annotated as a 'putative SCF ubiquitin ligase F-box protein' ( Cherry et al . , 2012 ) . F-box proteins act as adapters to target substrates for ubiquitination and subsequent degradation by the proteasome ( Skaar et al . , 2013 ) . We therefore propose to rename this gene DUH1 for 'Degrader of Unspliced HAC1 gene product 1' to reflect its role in the degradation of Hac1up , as we demonstrate later . The DUH1 variants that we identified in our 17 strains comprised 16 different mutations , of which 8 were nonsense , 7 were missense , and 1 was a frameshift ( Figure 4C , right ) . The nonsense mutations tended to cluster in the first half of the ORF , suggesting that they likely functioned as null alleles . In addition , three of the 17 strains containing mutations in DUH1 did not contain any other mutations , implicating the DUH1 mutations as causative for the phenotype . We therefore tested whether knocking-out DUH1 ( which is non-essential ) in the original selection strain recapitulated the de-silencing phenotype . Disrupting DUH1 led to a dramatic increase in the steady-state abundance of HA-His3p containing the 10-amino-acid tail , while having no effect on HA-His3p lacking the tail or constructs repressed by long-range base pairing ( Figure 4D ) . Analogously , deleting DUH1 in the GFP reporter strains completely eliminated the silencing effect of the Hac1up tail but had no effect on base-pairing-mediated silencing ( Figure 4E and Figure 4—figure supplement 2 ) . Notably , in the absence of DUH1 the GFP reporter in a wild-type HAC1 context was now expressed at the same leaky level as previously seen for the reporter in which a stop codon was positioned between GFP and the intron , indicating that degron-dependent silencing is required to suppress leaky GFP expression . Together , these results provide genetic evidence that DUH1 is the adapter protein that recognizes the Hac1up tail and targets the covalently attached protein for degradation . Having established that DUH1 is required for degron-dependent silencing of two different reporter genes , we returned to HAC1 itself . To detect Hac1p we introduced a 3xHA tag at the extreme N terminus of the endogenous HAC1 allele ( Figure 5A ) , which did not interfere with its post-transcriptional silencing in the absence of the UPR ( Figure 5B and C ) . To determine which Hac1p isoforms are regulated by DUH1 , we constructed a set of HA-tagged strains that constitutively produced either Hac1ip containing the 18-amino-acid exon 2–encoded tail ( construct 3 ) , Hac1up containing the 10-amino-acid intron-encoded tail ( construct 5 ) , or Hac1Δtailp containing no tail at all ( constructs 4 and 6 ) . All of the mRNAs encoding HA-tagged Hac1p variants were expressed at similar levels in the presence versus the absence of DUH1 ( Figure 5B ) . Strikingly , at the protein level only the abundance of Hac1up was affected by disruption of DUH1 , increasing by ~five fold in the knock-out strain ( Figure 5C ) . The increased abundance of Hac1up in the absence of DUH1 was not due to increased translation , as evidenced by deletion of DUH1 having no impact on ribosome density on any of the HA-tagged reporter mRNAs ( Figure 5D ) . Instead , our results suggest that Duh1p specifically affects the turnover of Hac1up due to its 10-amino-acid tail , as was suggested by the results of our reporter experiments . 10 . 7554/eLife . 20069 . 013Figure 5 . Effects of DUH1 on expression and stability of Hac1p . ( A ) Design of 3xHA-tagged HAC1 mRNA variants . Constructs are depicted as in Figure 2B , with the location of the N-terminal 3xHA tag indicated . ( B ) RNA abundance measurements for HAC1 mRNA variants . Total RNA was extracted from strains expressing the indicated mRNAs in either a DUH1 or duh1∆ background . qRT-PCR was used to measure the abundances of HAC1 variants relative to ACT1 mRNA , with all data normalized to the abundance of construct 1 in strain BY4741 . Shown are the mean ± SD ( n = 2 ) . ( C ) Effect of DUH1 disruption on protein abundances . Strains expressing the indicated mRNAs depicted in ( A ) in either a DUH1 or duh1∆ background were grown to mid-log phase . Extracts were prepared and immunoblotted for 3xHA-Hac1p and actin loading control . ( D ) Polysome analysis of 3xHA-tagged HAC1 mRNA variants . Extracts were prepared in heparin-containing lysis buffer from strains expressing the mRNAs indicated in ( A ) in either a DUH1 or duh1∆ background . Polysome analysis was performed as in Figure 1A . ( E ) Analysis of protein degradation kinetics . Strains expressing construct 5 ( depicted in A ) in either a DUH1 or duh1∆ background were grown to mid-log phase before being treated with cycloheximide ( CHX ) to halt translation . At the indicated time points , aliquots of cells were quenched in dry-ice-cold methanol and harvested by centrifugation . Protein extraction and immunoblotting were performed as in ( C ) , except that a high-sensitivity antibody was used to detect 3xHA-Hac1up . Shown are the mean ± SD ( n = 3 ) , expressed as a fraction of protein detected at t = 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 013 Because we were able to detect Hac1up by immunoblotting ( using a high-sensitivity antibody ) even when DUH1 was intact ( unlike the corresponding GFP reporter ) , we could use cycloheximide ( CHX ) shut-off experiments to directly assay the impact of DUH1 on the turnover of Hac1up . In both the presence and absence of DUH1 , Hac1up was degraded so rapidly that we could not accurately measure its half life even using a rapid harvesting procedure , due to the ~2 minutes required for CHX to accumulate in cells and halt translation ( Gerashchenko and Gladyshev , 2014 ) . However , the protein-degradation kinetics allowed us to calculate an upper bound for the half life of Hac1up , which was 50 seconds when DUH1 was present ( Figure 5E ) . Deletion of DUH1 stabilized Hac1up and increased its half-life upper bound to 2 minutes . These results demonstrate that DUH1 is required for the extremely short half life of Hac1up that normally limits its accumulation . The true half-life difference upon DUH1 deletion is likely to be greater than the ~2-fold difference in upper bounds that we measured , based on the ~5-fold difference in steady-state protein levels that could not be accounted for by differences in either mRNA abundance or ribosome density ( Figure 5B and D ) . It was previously observed that disrupting the base-pairing interaction between the 5′-UTR and intron of HAC1 mRNA was sufficient to allow accumulation of Hac1up , which led to a model in which base-pairing alone was responsible for the post-transcriptional silencing phenomenon ( Rüegsegger et al . , 2001 ) . Our results using constructs in which the base-pairing region was deleted ( Figure 5 ) suggested that the previously observed accumulation of Hac1up was unknowingly being buffered by Duh1p-dependent degradation . To directly address this possibility , we generated HA-tagged constructs in which the base-pairing region was disrupted by mutations in either the 5′-UTR or intron or was reconstituted by the compensatory mutations ( Figure 6A ) and determined the effect of DUH1 deletion on steady-state protein levels . Because the 5′ and 3′ splice sites remained intact in these constructs , we introduced them into an ire1Δ background to eliminate any potential confounding effects of background splicing ( Figure 1—figure supplement 1B ) . As previously observed , mutating the base-pairing region in the presence of DUH1 resulted in detectable levels of Hac1up ( Figure 6C ) . However , the accumulation of Hac1up was greatly stimulated by deletion of DUH1 , which was not explained by a corresponding increase in mRNA abundance ( Figure 6B and C ) . Thus , although disrupting the base pairing produces detectable amounts of Hac1up as previously reported ( Rüegsegger et al . , 2001 ) , Duh1p-dependent degradation restricts the steady-state level of the protein . 10 . 7554/eLife . 20069 . 014Figure 6 . Relationship between base pairing– and degron-dependent repression . ( A ) Design of 3xHA-tagged HAC1 mRNA variants . Constructs are depicted as in Figure 2B . Colored stars indicate mutations to the base-pairing region , with specific nucleotide changes shown in Figure 2B ( red and blue stars ) or below ( green and black stars ) in red . ( B ) RNA abundance measurements for HAC1 mRNA variants in the indicated strain backgrounds , analyzed as in Figure 5B . ( C–D ) Effect of DUH1 disruption on protein abundances . ire1∆ strains expressing the indicated mRNAs depicted in ( A ) in either a DUH1 or duh1∆ background were analyzed as in Figure 5C , except that a high-sensitivity antibody was used to detect 3xHA-Hac1up . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 014 Remarkably , even a single nucleotide change in the center of the base-pairing region ( Sathe et al . , 2015 ) was sufficient for some accumulation of Hac1up , which again was enhanced by deletion of DUH1 ( Figure 6D ) . This result suggests that the 5′-UTR–intron base-pairing interaction is only marginally stable , which may be required for efficient dissociation of the intron after splicing ( see Discussion ) . Collectively , we have shown that a pair of silencing mechanisms , one translational and the other post-translational , prevents spurious production of Hac1up in the absence of the UPR . The existence of such a fail-safe silencing mechanism implies that ectopic production of Hac1up has physiological consequences that negatively impact cellular fitness . However , a previous study suggested that Hac1up is itself not an active transcription factor because it lacks the activating 18-amino-acid tail found in Hac1ip ( Mori et al . , 2000 ) , raising the question as to why Hac1up accumulation would need to be tightly regulated . We hypothesized that Hac1up was in fact an active transcription factor but that in the previous study its accumulation had been prevented due to the experiments being performed in a DUH1 background . To test this hypothesis , we evaluated the ability of strains that constitutively produced either Hac1ip , Hac1up , or Hac1Δtailp to grow under conditions of chronic ER stress induced by the drug tunicamycin . Strains expressing Hac1ip or Hac1Δtailp grew on tunicamycin-containing medium regardless of whether DUH1 ( or IRE1 ) was present ( Figure 7A ) , consistent with both proteins being active transcription factors that are not targeted by Duh1p . In contrast , strains expressing Hac1up only grew robustly on tunicamycin-containing medium when DUH1 was knocked out ( but independently of IRE1 ) . These results confirm our hypothesis that Duh1p-dependent degradation normally masks the activity of Hac1up . Our findings also explain how HAC1 was able to be initially identified as a high-copy activator of the UPR in an Δire1 strain , since the UPR activity detected in this strain had to have resulted from Hac1up produced from unspliced HAC1 mRNA ( Chapman and Walter , 1997; Cox and Walter , 1996 ) . 10 . 7554/eLife . 20069 . 015Figure 7 . Requirement for DUH1 to suppress Ire1p-independent activation of the UPR . ( A ) Analysis of Hac1p activity in the UPR . Strains expressing the indicated HAC1 mRNA variants , with ( + ) or without ( ∆ ) IRE1 and/or DUH1 present , were grown to saturation . 10-fold dilution series were plated on YPD without ( –Tm ) or with ( +Tm ) 400 ng/ml tunicamycin to induce ER stress . ( B ) Impact of DUH1 on detection of Hac1up . Strains expressing the indicated mRNAs , with ( + ) or without ( ∆ ) IRE1 and/or DUH1 present , were analyzed as in Figure 6C . Black arrow indicates the position of Hac1up , which migrates more slowly than Hac1ip . Construct 1 , which lacks a 3xHA tag , was used as a negative control for anti-HA immunoblotting . ( C ) Effect of Duh1p-dependent degradation on the UPR . Strains expressing wild-type HAC1 with an N-terminal 3xHA tag , with ( + ) or without ( ∆ ) IRE1 and/or DUH1 present , were grown to saturation . 10-fold dilution series were plated on YPD without tunimacyin ( –Tm ) or containing the indicated concentration of tunicamycin ( +Tm ) . Plates were imaged at days 2 ( top ) , 3 ( middle ) , and 6 ( bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 01510 . 7554/eLife . 20069 . 016Figure 7—figure supplement 1 . Ire1p-independent accumulation of Hac1p . Strains expressing the indicated mRNAs , with ( + ) or without ( ∆ ) IRE1 and/or DUH1 present , were analyzed and presented as in Figure 7B . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 016 Because Hac1up has UPR-inducing activity ( Figure 7A ) , we reasoned that the fail-safe mechanism we discovered is required to prevent leaky production of Hac1up that would otherwise cause Ire1p-independent activation of the UPR . However , we initially failed to detect Hac1up produced from unspliced HAC1 mRNA even when DUH1 was disrupted ( Figure 5C ) . On the other hand , our results from the analogous GFP and HIS3 reporter gene studies demonstrated that translational repression mediated by 5′-UTR–intron base pairing was incomplete and allowed a low level of protein synthesis that was normally 'cleaned up' by Duh1p-dependent degradation ( Figure 4D and E ) . This led us to hypothesize that Hac1up itself was also being occasionally produced from HAC1u mRNA and rapidly degraded in a Duh1p-dependent manner , but that the short half life of Hac1up ( Figure 5E ) —relative to both GFP ( Natarajan et al . , 1998 ) and His3p ( Belle et al . , 2006 ) —further reduced the steady-state abundance of Hac1up to an extremely low level that was initially undetectable . To enhance detection of HA-tagged Hac1up , we made two modifications: We used a more sensitive anti-HA antibody for immunoblotting; and we knocked out IRE1 in our strains , which eliminates the background Ire1p-dependent splicing of HAC1 mRNA ( Figure 1—figure supplement 1B ) and concomitant production of Hac1ip that can otherwise dominate the signal on immunoblots . With these modifications we could now detect Hac1up being produced from unspliced HAC1 mRNA , but only when the protein was stabilized by deletion of DUH1 ( Figure 7B ) and at levels far below even those of Hac1up being constitutively produced in the presence of DUH1 ( Figure 7—figure supplement 1 ) . Despite the relatively low level of leaky translation product we detected , these results provide molecular evidence that Hac1up is being continuously produced from HAC1u mRNA but rapidly degraded due to its C-terminal degron . Does the leaky production of Hac1up unmasked by deletion of DUH1 have any functional consequences ? The ability of constitutively produced Hac1up to promote survival under UPR-inducing conditions ( Figure 7A ) suggested that small amounts of Hac1up might also induce the UPR to some extent . To test this possibility , we examined how strains expressing HA-tagged but otherwise unmodified HAC1 mRNA grew on media containing different concentrations of tunicamycin . At the highest tunicamycin concentration only strains expressing IRE1 could grow regardless of whether DUH1 was also present ( Figure 7C ) , suggesting that growth was dependent on the abundant Hac1ip produced from HAC1i mRNA . In contrast , at lower concentrations of tunicamycin the IRE1 deletion strain showed some growth , but this was reproducibly enhanced by simultaneous deletion of DUH1 . These results indicate that the low level of Hac1up detectable in strains lacking DUH1 ( Figure 7B ) is sufficient to activate the UPR enough to facilitate cell survival under stress , even in the absence of Ire1p . Thus , degron-dependent degradation of Hac1up mediated by Duh1p is normally required to prevent ectopic Ire1p-independent activation of the UPR .
Altogether , we have shown that the ability of unspliced HAC1 mRNA to be stored in the cytoplasm without giving rise to detectable protein depends on two layers of post-transcriptional silencing , which together comprise a fail-safe mechanism ( Figure 8 ) . The initial silencing mechanism that acts on HAC1u mRNA is a block to translation initiation , which is caused by secondary structure in the 5′-UTR that inhibits binding or progression of the scanning ribosome . However , this silencing mechanism does not entirely prevent translation , allowing the production of small amounts of Hac1up . Because Hac1up is an active transcription factor , even small amounts of this protein produced by leaky translation could activate the UPR and affect cellular physiology . To prevent this , a second silencing mechanism exists to rapidly degrade any Hac1up that is produced by leaky translation . By relying on recognition of the unique C-terminal tail of Hac1up by the F-box protein Duh1p , the protein-degradation mechanism selectively targets Hac1up for ubiquitination and subsequent degradation by the proteasome . Because both translational repression and protein degradation rely on sequences found in the intron of HAC1 , removal of the intron by Ire1p-dependent splicing simultaneously results in the inactivation of both silencing mechanisms . In this way , splicing allows the rapid accumulation of Hac1ip from the existing pool of HAC1 mRNA . 10 . 7554/eLife . 20069 . 017Figure 8 . Fail-safe post-transcriptional silencing of unspliced HAC1 mRNA . See the main text for a description . DOI: http://dx . doi . org/10 . 7554/eLife . 20069 . 017 One surprising finding of our studies is that the polysome-like sedimentation of HAC1u mRNA is not due to translation of the mRNA . Instead , our results using mRNAs containing premature stop codons and using a modified lysis buffer ( Figure 1 ) indicate that translation initiation on HAC1u mRNA almost never occurs due to the 5′-UTR–intron base-pairing interaction . The lack of appreciable translation of HAC1u mRNA suggests that the substrate for Ire1p-dependent splicing is untranslated mRNA , rather than polysomal mRNA containing stalled ribosomes as originally proposed ( Rüegsegger et al . , 2001 ) . A corollary of this is that the synthesis of Hac1ip requires de novo translation initiation on HAC1i mRNA , rather than just the resumption of translation by stalled ribosomes that initiated on HAC1u mRNA . The initial hint that HAC1u mRNA is translationally repressed at the initiation rather than elongation stage came from ribosome-profiling data , which revealed a lack of ribosome-protected fragments derived from HAC1u mRNA ( Weinberg et al . , 2016 ) . At least in this case , our findings suggest that ribosome profiling can provide a more accurate measurement of translation than traditional polysome profiling . What is the molecular basis for the polysome-like sedimentation of HAC1u mRNA ? The disruption of the pseudo-polysomes by addition of heparin , which is more negatively charged than RNA itself , suggests that electrostatic interactions are responsible . Moreover , the discontinuous distribution of HAC1u mRNA that matches the peaks and valleys observed for polysomes ( Rüegsegger et al . , 2001 ) indicates that HAC1u mRNA is associated with integral numbers of actual 80S ribosomes . On the basis of these data , we propose that electrostatic interactions between HAC1u mRNA and either non-translating 80S ribosomes or , more likely , bona fide polysomes ( containing both ribosomes and mRNA ) are responsible for its polysome-like sedimentation . HAC1u mRNA may be especially prone to such non-specific interactions because of the strong initiation block , though other untranslated mRNA molecules ( including the small fraction of ribosome-free mRNA molecules observed even for well-translated genes ) may also behave similarly . In this regard , the analysis method ( i . e . , use of constructs containing premature stop codons ) and experimental tools ( i . e . , exogenous RNA controls combined with heparin-containing lysis buffer ) that we utilized here to definitively establish the translation state of HAC1u mRNA will be generally useful in polysome-profiling studies in the future . After the unexpected discovery that eliminating the 5′-UTR–intron base pairing alone did not result in detectable protein expression , we uncovered the effect of protein degradation caused by an intron-encoded C-terminal degron . The genetic selection that we performed to identify components of the protein-degradation pathway yielded just a single gene with recurrent mutations: DUH1 . Our failure to isolate mutations in other components of the ubiquitin–proteasome machinery ( i . e . , the E2 ubiquitin-conjugating enzyme that presents the ubiquitin to the Duh1p-containing SCF complex for transfer onto Hac1up; and subunits of the proteasome itself ) may reflect the pleiotropic nature of such mutations , which would hinder growth despite the HIS3 reporter gene being de-silenced; an extreme case is the essential gene encoding the sole E1 ubiquitin-activating enzyme in yeast , UBA1 . In addition , our reliance on spontaneous mutagenesis made it unlikely that we would have isolated multiple mutations in redundant sets of genes , which may also explain our failure to isolate E2 mutants given their sometimes overlapping functions ( Chen et al . , 1993 ) . Nonetheless , the fact that we isolated 16 different mutations in DUH1 in 17 of our 20 sequenced strains illustrates the power of our genetic selection system . A noteworthy feature of our selection was that by combining spontaneous mutagenesis with whole-genome sequencing , we were able to rapidly go directly from phenotypic mutants to genotypic mutations without any crosses and relying on only simple computational tools for analysis . Compared to an alternative silencing mechanism involving only protein degradation with no translational repression ( Cox and Walter , 1996 ) , the fail-safe mechanism we discovered has the advantage of not wasting resources on the production of Hac1up that will anyway be rapidly degraded . But why does budding yeast rely on a pair of silencing mechanisms to prevent accumulation of Hac1up , rather than just completely blocking translation in the first place ? An intriguing possibility is that the marginal stability of the 5′-UTR–intron base-pairing interaction ( Figure 6D ) is the cause of leaky translation but is also required to make the translational repression reversible ( Rüegsegger et al . , 2001 ) . If the base-pairing interaction were more stable and prevented translation altogether , splicing might not be sufficient for the intron to dissociate from the 5′-UTR despite their being covalently unlinked . Another possibility is that the act of translating the mRNA has a function . For example , a pioneer round of translation on HAC1u mRNA might be required to form the 5′-UTR–intron interaction that inhibits subsequent rounds of translation . Such a requirement might be due to the ribosome physically bringing the distant sequences together through a tethering mechanism , or due to the ribosome unwinding alternative competing structures that initially form with the ORF . A third possibility is that the small amount of Hac1up being constantly produced and degraded has a function . In particular , the ability to regulate accumulation of Hac1up through Duh1p without relying on Ire1p activation would provide a means of rapidly producing a small pool of Hac1up in the absence of ER protein-folding stress . A constitutively produced yet transient pool of Hac1up might also contribute to the switch-like behavior of the UPR , based on the predicted ability of Hac1up to heterodimerize with Hac1ip and thereby facilitate proteasomal degradation of any Hac1ip produced from background splicing ( Figure 1—figure supplement 1 ) . Alternatively , the leaky translation of HAC1u mRNA that we discovered may reflect a more fundamental property of translational repression: that it is inherently incomplete . The two predominant mechanisms known to inhibit cap-dependent translation are upstream ORFs ( uORFs ) and 5′-UTR secondary structure ( Gebauer and Hentze , 2004 ) . Inhibition by uORFs has been shown to be only partial , due to both leaky scanning enabling bypass of the uORF and reinitiation facilitating translation of the downstream ORF ( Meijer and Thomas , 2002 ) . Our findings suggest that inhibition by 5′-UTR secondary structure may also be incomplete , potentially due to the action of RNA helicases that are constantly unwinding RNA structures in vivo ( Rouskin et al . , 2014 ) . Future work may shed light on both the causes and consequences of the incomplete translational repression of unspliced HAC1 mRNA .
Saccharomyces cerevisiae strains used in this study were derived from BY4741 ( MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0 ) and are listed in Supplementary file 1 . Strains were cultured at 30°C with shaking in rich yeast-peptone-dextrose ( YPD ) medium , synthetic complete ( SC ) medium , or the appropriate drop out medium ( e . g . , SC–His ) as indicated . Cultures were grown to mid-log phase ( OD600 ~0 . 5 ) before harvesting by either vacuum filtration ( polysome analysis ) , methanol quenching ( protein half-life ) , or brief centrifugation at 4°C ( all other experiments ) and then snap frozen in liquid nitrogen . For growth tests under UPR stress ( Figure 7 ) , 10-fold serial dilutions of cells were spotted onto YPD plates containing either the indicated concentration of tunicamycin ( Sigma , St . Louis , MO ) dissolved in DMSO ( +Tm ) or DMSO alone ( –Tm ) . To facilitate generation of HAC1 variant and reporter strains , the entire transcribed region of the HAC1 gene – as well as additional flanking sequence – was amplified from genomic DNA ( extracted from BY4741 ) and cloned into the pCR4-TOPO vector backbone ( Invitrogen , Carlsbad , CA ) using Gibson Assembly Master Mix ( New England Biolabs , Ipswich , MA ) according to the Gibson assembly method ( Gibson et al . , 2009 ) . GFP and HIS3 reporter constructs were generated from this plasmid by replacing the HAC1 ORF with the reporter gene using the Gibson assembly method . Variants of these constructs were generated using inverse PCR with 5′-phosphorylated primers ( designed to introduce point mutations or deletions ) and Phusion High-Fidelity DNA Polymerase ( New England Biolabs ) followed by ligation with Quick Ligase ( New England Biolabs ) ; or using the Gibson assembly method with either PCR products or synthetic gBlocks ( Integrated DNA Technologies , Coralville , IA ) . All plasmid inserts were verified by Sanger sequencing before PCR amplification with Phusion DNA polymerase for integration into the yeast genome . Yeast transformations were performed using the PEG–lithium acetate method ( Gietz and Woods , 2002 ) . HAC1 was deleted from the genome using the URA3 cassette of pRS416 to replace the entire transcribed region of HAC1 ( from 366 nucleotides upstream of the start codon , to 750 nucleotides downstream of the stop codon in exon 2 ) . IRE1 and/or DUH1 were deleted using the HIS3MX or URA3 cassettes of pFA6a-3HA-HIS3MX6 or pRS416 , respectively , as indicated in Supplementary file 1 . HAC1 variant and reporter strains were generated from hac1Δ::URA3 strains using 5-FOA counterselection . Transformants were screened by PCR to identify integrants , which were subsequently verified by PCR and Sanger sequencing of the entire integrated cassette . Polysome analyses by sucrose gradient fractionation are described in more detail at Bio-protocol ( Aboulhouda et al . , 2017 ) . Yeast strains were grown to mid-log phase ( OD600 ~ 0 . 5 ) in YPD and harvested by vacuum filtration as previously described ( Weinberg et al . , 2016 ) . The frozen cell pellet was transferred into a pre-chilled mortar that was surrounded by and filled with liquid nitrogen . The pellet was ground to a fine powder by hand using a pre-chilled pestle , transferred into a 50 ml conical tube filled with liquid nitrogen , and after boiling off the liquid stored at –80°C . Crude lysate was prepared by briefly thawing the cell powder on ice for 1 min and then resuspending in 1 ml polysome lysis buffer ( 20 mM HEPES-KOH [pH 7 . 4] , 5 mM MgCl2 , 100 mM KCl , 1% Triton X-100 , 2 mM DTT , 100 μg/mL cycloheximide , 20 U/mL SuperUPERase-In [Thermo Fisher Scientific , Waltham , MA] , 1X cOmplete EDTA-free Protease Inhibitor Cocktail [Roche , Switzerland] ) . Where indicated , the lysis buffer also contained 200 μg/mL heparin ( Sigma ) . The lysate was centrifuged at 1300 g for 10 min , and the supernatant was transferred to a new tube taking care to avoid the thick top lipid layer . The resulting clarified lysate was diluted to a concentration of 25 OD260 units/ml , flash frozen in liquid nitrogen in single-use aliquots , and stored at –80°C . A 10–50% continuous sucrose gradient was prepared in polysome gradient buffer ( 20 mM HEPES-KOH [pH 7 . 4] , 5 mM MgCl2 , 100 mM KCl , 2 mM DTT , 100 μg/mL cycloheximide , 20 U/mL SUPERase-I ) and allowed to cool to 4°C while lysate was thawed gently on ice . Where indicated , 100 ng luciferase RNA ( Promega , Madison , WI ) was then added to 100 μl of thawed cell lysate . 100 μl of cell lysate ( at 25 OD260 units/ml ) was loaded on top of the gradient , and gradients were spun in an SW41-Ti rotor at 36000 rpm for 2 . 5 hr at 4°C . Gradients were fractionated into 15 fractions using a Biocomp Piston Gradient Fractionator , and fractions were flash frozen and stored at –80°C . RT-qPCR was performed directly on gradient fractions using the TaqMan Gene Expression Cells-to-CT Kit ( Life TechnologiesThermo Fisher Scientific ) . Briefly , 10 μl Lysis Buffer containing DNase was combined with 0 . 1 μl synthetic XenoRNA control ( provided with the TaqMan Cells-to-CT Control Kit ) and 1 μl of the fraction . Reactions were incubated at room temperature for 5 min , and then quenched on ice by addition of 1 μl STOP Solution followed by incubation at room temperature for an additional 2 min . 4 . 5 μl of this reaction was added to 5 . 5 μl RT Master Mix , and reverse transcription was carried out at 37°C for 1 hr followed by a 5 min incubation at 95°C before cooling to 4°C . RT reactions were diluted 6-fold in nuclease-free water prior to qPCR . qPCR was carried out in duplicate for each fraction using the TaqMan Gene Expression Master Mix ( Thermo Fisher Scientific ) according to the manufacturer’s instructions , using 10 μl reactions containing 4 . 5 μl of six-fold diluted RT reaction per qPCR reaction . XenoRNA was analyzed using the primer-probe assay provided with the TaqMan Cells-to-CT Control Kit . ACT1 and luciferase RNAs were analyzed using predesigned TaqMan Gene Expression Assays from Life TechnologiesThermo Fisher Scientific ( Sc04120488_s1 and Mr03987587_mr , respectively ) . HAC1 and GFP RNAs were analyzed using the following primer-probe qPCR assays from Integrated DNA Technologies ( containing 6-FAM/ZEN/IBFQ quenchers with a primer-to-probe ratio of 1-to-2 ) : GenePrimer 1Primer 2ProbeHAC1TCAAGAGCTATGTTCAGTGTCGGGTTTCTACTGTTCTGTCTCCG56-FAM/CGCGCCCTC/ZEN/CTACAATTATTTGTGG/3IABkFQGFPGTTTGCCATAAGTTGCGTCCTGGTAGAATTGGATGGCGAC56-FAM/CTGTGAGTG/ZEN/GTGAGGGTGAAGGG/3IABkFQ Relative mRNA abundances in each fraction ( i . e . , Cq values ) were first normalized to XenoRNA to account for differences in qRT-PCR efficiency among fractions , and then calculated as a percentage of total mRNA detected across the gradient . All strains were analyzed in duplicate , beginning with separate cultures of the same strain , to confirm the reproducibility of each result . ACT1 mRNA was analyzed in every gradient and confirmed to co-sediment with polysomes in all cases . Yeast strains were grown to mid-log phase ( OD600 ~ 0 . 5 ) in SC medium , then diluted 10 fold in fresh medium before quantifying GFP fluorescence using a LSR II flow cytometer ( Becton Dickinson , Franklin Lakes , NJ ) and 530/30 filter . Raw data from 10 , 000 cells per sample was analyzed using FlowJo and gated to remove debris before being exported to Microsoft Excel . Fluorescence in each cell was normalized to cell size using the side scatter measurement before calculating the median and quartiles of the population . All strains were analyzed at least twice , beginning with separate cultures of the same strain , to confirm reproducibility . The equivalent of 5 OD600 units was harvested for both protein and RNA analyses from the same cultures used for flow cytometry . Protein samples were prepared using the NaOH/TCA method ( Riezman et al . , 1983 ) , separated by SDS-PAGE using 12% or 4–12% Bolt Bis-Tris gels ( Thermo Fisher Scientific ) , and transferred in 1X CAPS Buffer onto 0 . 22 micron PVDF membrane ( Bio-Rad , Hercules , CA ) . Blots were probed with the following antibodies diluted in 1X TBS-T containing 5% nonfat dry milk: mouse anti-GFP ( RRID:AB_390913 , Roche #11814460001 , 1:1000 ) , mouse anti-HA ( RRID:AB_627809 , Santa Cruz Biotechnology [Dallas , TX] sc-7392 , 1:3000 ) , rat anti-HA high sensitivity ( RRID:AB_390918 , Roche #11867423001 , 1:5000 ) , mouse anti–beta actin ( RRID:AB_449644 , AbCam [] ab8224 , 1:10000 ) , HRP-conjugated goat anti-mouse IgG ( RRID:AB_631736 , Santa Cruz Biotechnology sc-2005 , 1:10000 ) , and HRP-conjugated goat anti-rat IgG ( RRID:AB_631755 , Santa Cruz Biotechnology sc-2032 , 1:10000 ) . Blots were developed using Clarity ECL Western Blotting Substrate ( Bio-Rad ) , and chemiluminescence was detected on a ChemiDoc Imaging System ( Bio-Rad ) . Total RNA was isolated using the hot acid phenol method ( Ares , 2012 ) and resuspended in nuclease-free water . For each sample , 2 . 5 μg of RNA was treated with RQ1 DNase ( Promega ) according to the manufacturer’s instructions . cDNA synthesis and qPCR were performed using the Cells-to-CT Kit ( Thermo Fisher Scientific ) as described above . For RT-PCR analysis of HAC1 splicing , cDNA generated from input samples ( as described in 'Sucrose gradient analysis' ) was used as a template for PCR amplification using the following intron-flanking primers: forward , ACGACGCTTTTGTTGCTTCT; reverse , TCTTCGGTTGAAGTAGCACAC . PCR products were analyzed by agarose gel electrophoresis . Yeast strains YRDS221 and YRDS241 ( Supplementary file 1 ) were grown to mid-log phase ( OD600 ~ 0 . 5 ) in SC medium . Cycloheximide ( Sigma ) was added to a final concentration of 100 μg/ml and samples collected at the indicated time points . To rapidly harvest samples , 5 ml of culture was directly added to a prechilled 50 ml tube ( in a dry ice–ethanol bath ) containing 25 ml Quench Solution ( 60% methanol , 10 mM HEPES-KOH pH 7 . 4 ) and mixed well . Cells were collected by centrifugation at 4°C , snap frozen , and processed for immunoblotting as above . Quantification of immunoblot signal was performed by densitometry with ImageJ . An upper bound on protein half-life was calculated from 3 independent experiments as the time point at which ~50% of the protein remained . Yeast strain YRDS57 ( Supplementary file 1 ) was grown overnight in liquid YPD , plated on SC–His solid medium , and allowed to grow for 5–7 days at 30°C . Spontaneous mutant colonies were isolated from five independent platings and verified for growth on SC–His by restreaking and for expression of the HA-His3p by immunoblotting . Clones were then examined for cis mutations by Sanger sequencing the 3′ end of the HIS3 reporter gene . Of the remaining clones that had a ‘trans’ mutation , 20 were chosen for analysis by genomic DNA sequencing along with 4 control strains ( the parent strain YRDS57 and 3 strains with cis mutations ) . Genomic DNA was isolated from individual saturated overnight cultures using the MasterPure Yeast DNA Purification Kit ( Epicentre ) according to the manufacturer’s instructions . Purified genomic DNA was treated with RNase A/T1 Cocktail ( Thermo Fisher Scientific ) , phenol extracted , precipitated with ethanol , and quantified with the Qubit dsDNA HS Assay Kit ( Thermo Fisher Scientific ) . For each strain 1 ng of genomic DNA was used as input into the Nextera XT DNA Library Preparation Kit ( Illumina , San Diego , CA ) , and genomic DNA libraries were prepared according to the manufacturer’s instructions using the Nextera XT Index Kit for barcoding during PCR . Genomic DNA libraries from each strain were quantified with the Qubit dsDNA HS Assay Kit ( Invitrogen ) , and pooled into a single tube in equal proportions . Pooled DNA was purified using the DNA Clean and Concentrator-5 Kit ( Zymo Research , Irvine , CA ) , and 200–350 bp fragments were isolated using a BluePippin Gel Cassette ( 2% agarose dye-free with internal standards; ) . The size-selected library was sequenced on a single lane of a HiSeq 4000 sequencer using 50-nucleotide single-end reads . Data analysis was performed using the Galaxy web server ( Afgan et al . , 2016 ) . Reads in FASTQ format were aligned to the sacCer3 reference genome using BWA ( version 0 . 1 ) with default 'Commonly Used' settings . Sequence differences between the reference genome and aligned reads were identified using FreeBayes ( version 0 . 4 ) to generate a variant call format ( VCF ) file for each strain . The VCF-VCFintersect tool ( with options '-i --invert' ) was then used to compare VCF files for mutants against the VCF file for the parental selection strain , in order to identify variants that were unique to the mutants . Finally , variants were manually annotated using the UCSC Genome Browser . For proof-of-principle analysis of cis mutants , the same mapping and variant-calling procedures were used with a custom reference genome containing only the reporter gene . Genome-sequencing data from the genetic selection have been deposited in the NCBI Sequence Read Archive ( SRA ) under accession SRP081128 . | Molecular machines called ribosomes read the genetic instructions in an mRNA molecule and then translate them to make proteins . However , cells do not translate all of the template mRNAs that they have available into proteins; instead they have a number of ways to block the process to control when certain proteins are made . In budding yeast , the mRNA that codes for a protein called Hac1 is always present in the cell but the protein is normally not detected . The Hac1 protein is responsible for helping the cell deal with certain types of stress , so it only accumulates when the cell is experiencing such stresses . The mRNA that encodes Hac1 ( referred to as HAC1 mRNA ) contains a sequence called an intron . These sequences are normally cut out of mRNAs before they are read by the ribosome . However , the intron in the HAC1 mRNA is unusual , because it is only removed when cells are subjected to stress . The rest of the time , this intron serves to block the production of Hac1 through a poorly understood mechanism . Now , Di Santo et al . show the HAC1 mRNA uses two strategies to keep itself fully repressed—both of which involve its intron . One strategy relies on a structure formed in the HAC1 mRNA that prevents ribosomes from starting translation in the first place . However , this block is occasionally bypassed , causing some Hac1 protein to be produced when it should not be . To deal with this , the Hac1 protein that is produced contains a short protein sequence , encoded by the intron , that targets this unneeded protein for degradation . These two strategies together comprise a “fail-safe” mechanism to completely repress the HAC1 mRNA . Following on from these findings , it will be important to determine whether other mRNAs – both in budding yeast and in other species including humans – use a similar fail-safe strategy to block proteins from being made when they should not be . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2016 | The fail-safe mechanism of post-transcriptional silencing of unspliced HAC1 mRNA |
Pluripotency transcription programs by core transcription factors ( CTFs ) might be reset during M/G1 transition to maintain the pluripotency of embryonic stem cells ( ESCs ) . However , little is known about how CTFs are governed during cell cycle progression . Here , we demonstrate that the regulation of Oct4 by Aurora kinase b ( Aurkb ) /protein phosphatase 1 ( PP1 ) during the cell cycle is important for resetting Oct4 to pluripotency and cell cycle genes in determining the identity of ESCs . Aurkb phosphorylates Oct4 ( S229 ) during G2/M phase , leading to the dissociation of Oct4 from chromatin , whereas PP1 binds Oct4 and dephosphorylates Oct4 ( S229 ) during M/G1 transition , which resets Oct4-driven transcription for pluripotency and the cell cycle . Aurkb phosphor-mimetic and PP1 binding-deficient mutations in Oct4 alter the cell cycle , effect the loss of pluripotency in ESCs , and decrease the efficiency of somatic cell reprogramming . Our findings provide evidence that the cell cycle is linked directly to pluripotency programs in ESCs .
Embryonic stem cells ( ESCs ) have unique transcriptional programs for self-renewal and pluripotency which differentiates into all types of cells . Core transcription factors—Oct4 , Sox2 , Nanog ( OSN ) —govern such pluripotency transcriptional programs ( Jaenisch and Young , 2008; Young , 2011 ) . ESCs grow rapidly and undergo an unusual cell cycle , characterized by a very short G1 phase and a long S phase in mouse and human ( Kapinas et al . , 2013; Savatier et al . , 1994; White and Dalton , 2005 ) . The duration of G1 in mouse ESCs and human ESCs determines their fate with regard to differentiation and pluripotency ( Coronado et al . , 2013; Mummery et al . , 1987; Pauklin and Vallier , 2013 ) . A recent study revealed that the S and G2 phases tend to maintain the pluripotent state at early time of differentiation ( Gonzales et al . , 2015 ) . Thus , cell cycle regulation in ESCs should be linked to pluripotency in maintaining ESC identity . Our understanding of the molecular associations between the cell cycle and pluripotency in ESCs is limited . Robust Cdk2 activity shortens G1 phase by inducing rapid G1-S transition and promotes pluripotency and self-renewal in ESCs ( Neganova et al . , 2009; Van Hoof et al . , 2009 ) . Nanog controls S phase entry by targeting Cdc25c and Cdk6 ( Zhang et al . , 2009 ) . Despite the growing evidence on the direct connection between the cell cycle and pluripotency , it remains unknown how ESCs preserve their pluripotency through cell cycle progression and reset pluripotency transcriptional programs during the transition from mitosis to G1 phase . Oct4 is considered a master regulator of ESC pluripotency through its cooperation with other core transcription factors ( Jerabek et al . , 2014 ) . Post-translational modifications to Oct4 affect its transcriptional activity and lead to ESC pluripotency . For example , we previously reported that O-GlcNAcylatoin of murine Oct4 ( T228 ) is important for ESC pluripotency and somatic cell reprogramming ( Jang et al . , 2012 ) . Also , Oct4 is controlled by phosphorylation ( Brumbaugh et al . , 2012; Saxe et al . , 2009; Spelat et al . , 2012 ) , but there is no evidence that phosphorylation-mediated regulation of Oct4 during the cell cycle affects Oct4-mediated pluripotency programs in ESCs . The Aurora kinase b ( Aurkb ) -protein phosphatase 1 ( PP1 ) axis is critical for kinetochore assembly/disassembly during the cell cycle , regulating the balance between phosphorylation and dephosphorylation of kinetochore substrates ( Emanuele et al . , 2008; Kim et al . , 2010 ) . Specifically , PP1 mediates the M/G1 transition and ensures proper resetting of the subsequent G1 phase by dephosphorylating cell cycle machinery ( Ceulemans and Bollen , 2004 ) . When the cell cycle resets , transcriptional programs for ESC pluripotency should be reset , because transcriptional programs are generally switched off at the onset of mitosis and subsequently reestablished during entry into the next G1 phase ( Delcuve et al . , 2008; Egli et al . , 2008; Martinez-Balbas et al . , 1995 ) . Thus , during the cell cycle , the Aurkb-PP1 axis might be linked directly to the post-translational modification of pluripotency factors with regard to the resetting of pluripotency in ESCs . In this study , we demonstrate that the Aurkb-PP1 axis regulates Oct4 during the cell cycle over time and by location . We found that Oct4 contains a well-conserved Aurkb phosphorylation residue ( S229 ) and PP1 binding motif ( RVXF ) in its homeodomain . Aurkb phosphorylates Oct4 ( S229 ) during G2/M phase and dissociates p-Oct4 ( S229 ) from chromatin , and PP1 dephosphorylates p-Oct4 ( S229 ) during the M/G1 transition , which prompts Oct4 to reset pluripotency transcription on re-entry into the following G1 phase . We found that mutating the Aurkb-phosphorylation residue S229 and the PP1-binding residue F271 of Oct4 in ESCs led to a significant loss of pluripotency and altered the cell cycle . Transduction of these mutants into MEFs significantly decreased the reprogramming efficiency . Based on these findings , we propose that the spatiotemporal regulation of Oct4 by the Aurkb-PP1 axis during the cell cycle is critical for resetting pluripotency and cell cycle genes in determining the identity of ESCs .
To understand the function of the phosphorylation of Oct4 , we examined its phosphorylation sites by transient transfection of Flag-Oct4 into E14 ESCs , analyzed the phosphorylation state of purified Oct4 by mass spectrometry , and identified 4 phosphorylation sites ( Figure 1—figure supplement 1A and B ) . We then generated phosphor-mimetic mutants and measured their transcriptional activities by transfecting them into NIH-3T3 cells that stably harbored Oct4-driven luciferase reporter genes ( Figure 1—figure supplement 1C ) . Only the S229D mutant significantly reduced Oct4 transcriptional activity . Notably , serine 229 lies in the N-terminal region of the homeodomain of Oct4 and is well conserved throughout many species ( Figure 1—figure supplement 1D and E ) . Next , we generated a rabbit polyclonal antibody against phosphorylated Oct4 ( S229 ) [thereafter p-Oct4 ( S229 ) ] and confirmed its specificity by dot blot and western blot ( Figure 1—figure supplement 2A and B ) . We then examined p-Oct4 ( S229 ) expression by confocal microscopy in undifferentiated E14 ESCs ( Figure 1A upper panel ) . p-Oct4 ( S229 ) in E14 ESCs was detected locally around mitotic cells . From this result above , we wondered whether Oct4 phosphorylation at serine 229 occurs in a cell cycle-dependent manner . 10 . 7554/eLife . 10877 . 003Figure 1 . Phosphorylated Oct4 at serine 229 is enriched in G2/M phase and dissociated from chromatin . ( A ) Immunostaining of E14 ESCs treated with or without nocodazole ( NOC , 200 ng/ml ) for 10 hr . Oct4 was stained with anti-Oct4 ( green ) , p-Oct4 ( S229 ) was stained with anti-p-Oct4 ( S229 ) ( red ) , and DNA was stained with DAPI ( blue ) . White boxes represent cells at various stages . Shown are interphase ( 1 ) , metaphase ( 2 ) , and anaphase ( 3 ) cells . Scale bars were shown . ( B ) E14 ESCs were treated with nocodazole ( 200 ng/ml ) for the indicated times and immunoblotted with the indicated antibodies . Phosphorylation levels of Oct4 at serine 229 were gradually induced during nocodazole treatment . ( C ) Histograms of the proportions of nocodazole-treated ( 200 ng/ml ) E14 ESCs at various stages in the cell cycle . Cells were stained with PI and DNA contents were analyzed by FACS ( 1x104 cells/sample ) . ( D and E ) Fluorescence images of E14 ESCs expressing mKO2-Cdt1 and mAG-Geminin ( FUCCI reporter ) . Shown are green ( mAG-geminin ) and red ( mKO2-Cdt1 ) fluorescence . E14 ESCs expressing FUCCI reporter were left untreated or treated with nocodazole ( NOC , 200 ng/ml ) for 10 hr . p-Oct4 ( S229 ) was stained with anti-p-Oct4 ( S229 ) ( red , Figure 1E; green , Figure 1F ) , and DNA was stained with DAPI ( blue ) . Scale bars , 30 μm ( F ) ChIP-qPCR assay was performed with anti-IgG , anti-Oct4 , and anti-p-Oct4 ( S229 ) in E14 ESCs with or without nocodazole ( NOC , 200 ng/ml ) for 10 hr . Values represent mean ± standard deviation ( n≥3 ) . ( **p<0 . 01 , ***p<0 . 001 ) DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 00310 . 7554/eLife . 10877 . 004Figure 1—figure supplement 1 . Identification of Oct4 phosphorylation at serine 229 residue . ( A ) Sequence alignment of detected phosphorylation regions of Oct4 between mouse and human . ( B ) Oct4 is phosphorylated at S229 in ESCs . Oct4 was purified using stably expressed Flag-Oct4 in ZHBTc4 ESCs by immunoprecipitation with anti-Flag . Phosphorylation sites on Oct4 were analyzed by nano-LC-ESI-MS/MS . ( C ) Oct4-driven transcriptional activity was measured with Oct4 mutants . Ten copies of Oct4-responsive element ( 10X Oct4 RE ) -driven luciferase reporter gene was incorporated into the genome of NIH 3T3 cells . These stable cells were infected with retroviruses expressing Oct4 wild-type ( WT ) and phosphor mimic mutants . Luciferase activity was measured 4 days after infection . Values represent mean ± standard deviation ( n≥3 ) . ( D ) A schematic shows identified Oct4 phosphorylation sites . The S229 residue is located at POUh domain on Oct4 . ( E ) Sequence alignment of Oct4 phosphorylated region ( S229 ) between species . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 00410 . 7554/eLife . 10877 . 005Figure 1—figure supplement 2 . Characterization of Oct4 phosphorylation site at serine 229 and anti-p-Oct4 ( S229 ) antibody . ( A and B ) Characterization of the antibody to p-Oct4 ( S229 ) . By dot blot analysis , the antibody ( p-Oct4 ( S229 ) ) specifically recognized phosphorylated peptide ( A ) . Lysates of HEK293T cells transfected with Flag-Oct4 wild type ( WT ) , S229A , and S229D mutant were immunoblotted with indicated antibodies . The p-Oct4 ( S229 ) antibody only recognized wild type Oct4 ( B ) . ( C ) By treatment of E14 ESCs with aphidicholin , which induced G1 phase arrest , phosphorylation of Oct4 at S229 was decreased . p-Oct4 ( S229 ) was detected by Western blot and Actin was used as an internal control . ( D ) The level of p-Oct4 ( S229 ) was only increased after treatment of nocodazole ( Noc ) in P19 ECC cells . P19 ECC cells were treated with aphidicholin ( Aphi ) or nocodazole ( Noc ) for 10 hr . Lysates from these cells were immunoblotted with indicated antibodies and Actin was used as an endogenous control . ( E ) DNA damage in E14 ESCs was induced by Adriamycin treatment . Gradual decrease of p-Oct4 ( S229 ) during adriamycin treatment was detected by Western blot . p53 was used as an positive control for inducing DNA damage in E14 ESCs . ( F ) p-Oct4 ( S229 ) was pulled down with anti-p-Oct4 ( S229 ) antibody under the same condition we performed ChIP assay followed by western blot . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 005 To this end , we treated cells with various agents that are related to the cell cycle and DNA damage . Notably , treatment of ESCs with nocodazole significantly enhanced Oct4 phosphorylation at S229 , but aphidicolin and adriamycin decreased p-Oct4 ( S229 ) levels ( Figure 1—figure supplement 2C–E and Figure 1B ) . In previously published data , 18 hr incubation with 200ng/ml of nocodazole was enough to synchronize hESCs at G2/M phase without inducing differentiation ( Zhang et al . , 2009 ) . In the case of E14 ESCs , nocodazole treatment during 10 hrs completely arrested cells at G2/M phase . On treatment with nocodazole , p-Oct4 ( S229 ) began to rise in late S phase or early G2/M phase , peaking at G2/M phase ( Figure 1A–C ) . To confirm the localization of p-Oct4 ( S229 ) , we adapted the fluorescence ubiquitination cell cycle indicator ( FUCCI ) reporter system to ESCs ( Sakaue-Sawano et al . , 2008 ) . We generated E14 ESCs that stably expressed GFP-mAG-geminin during S-G2-M phase and examined p-Oct4 ( S229 ) by confocal microscopy on nocodazole treatment ( Figure 1D ) . As expected , p-Oct4 ( S229 ) overlapped with GFP-geminin in G2/M phase . In contrast , p-Oct4 ( S229 ) did not merge with Red-mKO2-Cdt1 , which is highly expressed in G1 phase ( Figure 1E ) . We then confirmed that Oct4 dissociates from the binding region of Oct4 and Nanog at G2/M phase by ChIP-qPCR ( Figure 1F ) . Under the same conditions , p-Oct4 ( S229 ) rarely bound to the same locus , despite p-Oct4 ( S229 ) was successfully pulled down with the antibody ( Figure 1—figure supplement 2F ) . This result is consistent with a previous report that a human phosphor-mimetic form of Oct4 ( S325E ) [homolog of mouse Oct4 ( S229 ) ] binds to DNA more weakly than Oct4 ( WT ) by in vitro EMSA ( Brumbaugh et al . , 2012 ) . The loss of DNA binding affinity of Oct4 by phosphorylation might be induced by steric and electrostatic clashes ( Saxe et al . , 2009 ) . Based on these findings , Oct4 is specifically phosphorylated at serine 229 , and p-Oct4 ( S229 ) dissociates from chromatin in G2/M phase . To identify the kinases that phosphorylate Oct4 ( S229 ) , we selected 19 candidates using a group-based prediction system ( Xue et al . , 2008 ) and among Oct4-interacting kinases ( Ding et al . , 2012 ) ( Figure 2—figure supplement 1A ) . We examined the phosphorylation of S229 by these 19 recombinant kinases by in vitro kinase assay and western blot with anti-pOct4 ( S229 ) —6 kinases could phosphorylate S229 ( Figure 2—figure supplement 1B ) . 10 . 7554/eLife . 10877 . 006Figure 2 . Aurkb binds and phosphorylates Oct4 at serine 229 during G2/M phase . ( A ) Radioactive in vitro kinase assay using recombinant Aurkb to phosphorylate GST-Oct4 WT and S229A mutant . Coomassie staining of purified proteins and autoradiogram showing incorporation of γ-32P ATP . ( B ) Cold in vitro kinase assay reactions using recombinant Aurkb with purified GST , GST-Oct4 WT , and S229A mutant as substrate followed by western blot . ( C and D ) Nocodazole-arrested E14 ESCs ( 200 ng/ml for 10 hr ) were treated with the Aurora kinase inhibitors AT9283 ( inhibits Aurka and Aurkb ) , hesperadin ( inhibits Aurkb ) , and MLN8237 ( inhibits AurkA ) . Gradual decreases in p-Oct4 ( S229 ) levels with increasing concentrations of Aurkb inhibitors in E14 ESCs were seen by western blot ( C ) . FACS analysis was performed under the same condition ( D ) ( 1x104 cells/sample ) . ( E ) Coimmunoprecipitation of Oct4 with Aurka and Aurkb from E14 ESCs stably expressing Flag-tagged Aurora kinases . ( F ) Changes in Oct4 interaction with Aurkb during cell cycle progression . Whole-cell lysates from Flag-Oct4-expressing ZHBTc4 ESCs were pulled down with anti-Flag beads . Bound proteins were immunoblotted with the indicated antibodies . ( G ) DNA content analysis of Flag-Oct4 expressing ZHBTc4 ESCs by FACS . Flag-Oct4-expressing ZHBTc4 ESCs , treated with nocodazole ( 200 ng/ml ) for 6 hr , were released for 2 and 4 hr and DNA contents were counted ( 1x104 cells/sample ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 00610 . 7554/eLife . 10877 . 007Figure 2—figure supplement 1 . Screening of kinases responsible for phosphorylation on Oct4 at S229 . ( A ) A schematic diagram represents strategy for kinase screening . ( B ) Cold in vitro kinase assay was performed using predicted 19 kinases with purified GST-Oct4 WT and S229A mutant . 50nM of kinases were used in all reactions . Kinase responsible for Oct4 phosphorylation at S229 was detected by the p-Oct4 ( S229 ) antibody and PonceauS staining was used as a loading control . ( C and D ) Identification of kinase responsible for phosphorylation at S229 on Oct4 in vivo Each kinase was stably knocked down in E14 ESCs by infection of indicated lentiviral shRNA and only Aurkb-knockdown ESCs harbored decreased p-Oct4 ( S229 ) level . p-Oct4 ( S229 ) level was detected by Western blot after treatment of nocodazole for 10 hr in each E14 ESCs and Oct4 was used as an internal control ( C ) . Stable knockdown of indicated kinases in E14 ESCs were analyzed by qRT-PCR ( D ) . ( E ) Transient knockdown of Aurkb in E14 ESCs reduced p-Oct4 ( S229 ) level . By infection of lentiviral shRNAs targeting Aurka and Aurkb in E14 ESCs , knockdown levels were detected with indicated antibodies 2 days after infection . p-Oct4 ( S229 ) level in each E14 ESCs was detected by Western blot after treatment with nocodazole for 10 hr and Actin was used as an internal control . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 007 To identify the kinases that mediate Oct4 ( S229 ) phosphorylation during G2/M phase , we screened the kinases by administering nocodazole to ESCs that were knocked down with cognate lentivirally expressed shRNAs of kinases ( Figure 2—figure supplement 1C and D ) . p-Oct4 ( S229 ) levels declined significantly on knockdown of Aurkb . Further , we confirmed that recombinant Aurkb phosphorylated GST-Oct4 ( S229 ) by in vitro 32P-ATP-labeled kinase assay and western blot with anti-p-Oct4 ( 229 ) ( Figure 2A and B ) . To verify the Aurkb-mediated phosphorylation of Oct4 ( S229 ) , we treated nocodazole-pretreated E14 ESCs ( 10 hr ) with various aurora kinase inhibitors for 15 min . An Aurkb-specific inhibitor , hesperadin , completely blocked the phosphorylation , but an Aurka-specific inhibitor , MLN8237 , did not . AT9283 , an inhibitor of both Aurka and Aurkb , prevented phosphorylation ( Figure 2C ) . Under this condition , Aurkb inhibition did not alter cell cycle profile ( Figure 2D ) . Aurkb preferentially phosphorylates serine when arginine lies 2 residue upstream of a phosphoserine ( -2 position ) ( Sugiyama et al . , 2002 ) . In Oct4 , we found arginine-227 , residing 2 residues upstream of S229 ( Figure 1—figure supplement 1E ) . We then observed that Flag-Aurkb interacts with endogenous Oct4 in E14 ESCs by immunoprecipitation ( Figure 2E ) . To determine the cell cycle phases during which Oct4 preferentially interacts with Aurkb , Flag-Oct4-expressing ZHBTc4 ESCs were pretreated with nocodazole for 6 hr , maintaining them in G2/M phase , and released on removal of nocodazole for the cell cycle progression . Notably , Flag-Oct4 interacted strongly with endogenous Aurkb in G2/M phase in Flag-Oct4-expressing ZHBTc4 ESCs ( Figure 2F and G ) , consistent with our result that Oct4 ( S229 ) is heavily phosphorylated in G2/M phase ( Figure 1 ) . These findings demonstrate that Aurkb is the kinase that phosphorylates Oct4 ( S229 ) in G2/M phase . When nocodazole treated ZHBTc4 ESCs were released into normal serum , the Aurkb-Oct4 interaction weakened and p-Oct4 ( S229 ) levels declined ( Figure 2F ) , indicating that certain phosphatases catalyze the dephosphorylation of p-Oct4 ( S229 ) during the M/G1 transition . In examining the amino acid sequence of Oct4 , we found that it contains a protein phosphatase 1 ( PP1 ) -binding sequence ( 268-RVWF-271 ) in its homeodomain , near the S229 Aurkb phosphorylation site in the 3-dimensional structure ( Figure 3A and B ) . This motif is well conserved among many species ( Figure 3—figure supplement 1A ) . Thus , we studied the interaction of Oct4 with 3 isoforms of PP1: PP1α , PP1β , and PP1γ . We found that Oct4 interacted more strongly with endogenous PP1β and PP1γ than with PP1α in ZHBTc4 ESCs ( Figure 3C ) . 10 . 7554/eLife . 10877 . 008Figure 3 . PP1 binds and dephosphorylates Oct4 at serine 229 during G1 phase . ( A ) Sequence alignment of Oct4 . Oct4 contains a conserved PP1 docking motif ( RVXF ) . ( B ) Three-dimensional structure of Oct4 and DNA complex ( MMDB ID: 87311 ) was adapted from the Molecular Modeling Database ( MMDB ) of NCBI . Each yellow region indicates S229 and an RVWF PP1-binding domain . ( C ) Coimmunoprecipitation assay revealing the endogenous interaction between Oct4 and PP1 catalytic subunits . Proteins were immunoprecipitated from Flag-Oct4-expressing ZHBTc4 ESCs with Flag antibody , followed by western blot . ( D ) Changes in Oct4 interaction with PP1 catalytic subunits during cell cycle progression . Whole-cell lysates from Flag-Oct4-expressing ZHBTc4 ESCs were pulled down with anti-Flag beads . Immunoprecipitated proteins were immunoblotted with the indicated antibodies . ( E ) Purified GST-Oct4 ( WT ) or GST-Oct4 ( F271A ) mutant was incubated with purified ( His ) 6-PP1β and PP1γ and then pulled down with GST beads . Immunoblot shows that PP1β and γ directly bind GST-Oct4 ( WT ) . PP1β and PP1γ show weaker interaction with GST-Oct4 ( F271A ) than wild-type Oct4 . ( F ) In vitro phosphatase assay using PP1β or PP1γ with phosphorylated Oct4 as substrate . Okadaic acid ( OKA ) treatment decreased PP1-mediated dephosphorylation of Oct4 . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 00810 . 7554/eLife . 10877 . 009Figure 3—figure supplement 1 . PP1 dephosphorylates Oct4 at S229 in vitro and in vivo . ( A ) Sequence alignment of PP1 binding motif in Oct4 between species . ( B ) In vitro phosphatase assay using purified PP1 isoforms with GST Oct4 ( WT ) . rAurkb-mediated phosphorylation of Oct4 at S229 was decreased by incubation with PP1β and PP1γ , but not by PP1α . Reduced p-Oct4 ( S229 ) levels were detected by Western blot with indicated antibodies . ( C and D ) Treatment of okadaic acid ( OKA ) in E14 ESCs increased p-Oct4 ( S229 ) . E14 ESCs were incubated with 50nM of OKA for indicated time and increased p-Oct4 ( S229 ) level was detected by Western-blot with indicated antibodies ( Left panel ) . DNA contents were analyzed by FACS ( right panel ) ( C ) . E14 ESCs were treated with indicated concentrations of OKA for 2 hr and p-Oct4 ( S229 ) level was increased by dose-dependent of OKA . p-Oct4 ( S229 ) was detected by Western blot ( left panel ) and DNA contents were analyzed by FACS ( right panel ) ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 009 Next , we examined the interaction between Oct4 and PP1 isoforms during the M/G1 transition after nocodazole treatment and release into normal serum ( Figure 3D ) . p-Oct4 ( S229 ) disappeared quickly during the M/G1 transition . In parallel , PP1β and PP1γ interacted strongly with Oct4 in G1 phase ( 2 hr after release ) . However , PP1α bound weakly to Oct4 regardless of cell cycle stage , suggesting that PP1α might be not a true p-Oct4 ( S229 ) phosphatase during the M/G1 transition . Thus , we focused on PP1β and PP1γ with regard to the dephosphorylation of Oct4 in subsequent experiments . We altered phenylalanine-271 to alanine in Oct4 [Oct4 ( F271A ) ] by site-directed mutagenesis and measured the in vitro interaction between PP1 and Oct4 ( WT ) or Oct4 ( F271A ) using bacterially purified recombinant ( His ) 6-PP1β , ( His ) 6-PP1γ , GST-Oct4 ( WT ) , and GST-Oct4 ( F271A ) . ( His ) 6-PP1β and γ bound more robustly to GST-Oct4 ( WT ) but weakly to GST-Oct4 ( F271A ) ( Figure 3E ) , indicating that Oct4 interacts directly with PP1 through a PP1-binding motif ( RVWF ) . We then determined whether PP1 dephosphorylates the Aurkb-catalyzed p-Oct4 ( S229 ) by preincubating recombinant GST-Oct4 with recombinant Aurkb , adding purified PP1 , and measuring the phosphorylation state of p-Oct4 ( S229 ) by western blot . Recombinant PP1β and PP1γ , but not PP1α , dephosphorylated Aurkb-mediated phospho-Oct4 ( S229 ) ( Figure 3—figure supplement 1B ) . Further , pretreatment with okadaic aid ( OKA ) , a PP1 inhibitor , blocked the dephosphorylation of p-Oct4 ( S229 ) ( Figure 3F ) , indicating that the interaction between Oct4 and PP1 is important for dephosphorylation of phospho-S229 in Oct4 . Treatment of E14 ESCs with 50 nM OKA increased p-Oct4 ( S229 ) levels after 4 hr ( Figure 3—figure supplement 1C ) . Rising concentrations of OKA ( from 0–200 nM ) gradually increased p-Oct4 ( S229 ) levels after 2 hr of treatment ( Figure 3—figure supplement 1D ) , suggesting that PP1 activity regulates the phosphorylation state of S229 in Oct4 . These findings demonstrate that PP1 isoforms have an opposite activity with Aurkb by binding to Oct4 and by dephosphorylating S229 of Oct4 during the M/G1 transition in ESCs . Based on the findings that p-Oct4 ( S229 ) dissociates from condensed chromatin ( Figure 1 ) and PP1 dephosphorylates p-Oct4 ( S229 ) during the M/G1 transition ( Figure 3 ) , We hypothesized that PP1-mediated dephosphorylation of Oct4 ( S229 ) is required for the resetting of Oct4 for pluripotency gene expression during the M/G1 transition . To examine this possibility , we arrested E14 ESCs in G2/M phase with nocodazole and released them into normal serum ( Figure 4A and B ) . The high amounts of p-Oct4 ( S229 ) that accumulated in G2/M phase vanished quickly after G1 phase ( after 1 hr release ) , reappeared at late S phase ( after 6 hr release ) and enriched at the next M phase ( 10 hr after release ) . In addition , to confirm weather Oct4 binding to target genes are regulated by phosphorylation dependent manner throughout the S-G2/M phase , we chased p-Oct4 level and binding of Oct4 to target genes ( Figure 4—figure supplement 1A and B ) . Binding of Oct4 to target chromatin declined through S-G2/M phase ( 7–9 hr after release ) , and increased at the M/G1 transition ( 10 hr after release ) in parallel with the level of p-Oct4 ( S229 ) accumulated . This result is consistent with recent report that Aurkb is active during S phase in ESCs ( Mallm and Rippe , 2015 ) . 10 . 7554/eLife . 10877 . 010Figure 4 . PP1-mediated dephosphorylation of Oct4 ( S229 ) correlates with the resetting of pluripotency genes in the next G1 phase . ( A ) p-Oct4 ( S229 ) levels after release of E14 ESCs from M-phase arrest . Shown are immunoblots for the indicated proteins . ( B ) Nocodazole-treated E14 ESCs were released and analyzed for DNA content by FACS ( 1x104 cells/sample ) . ( C ) OKA treatment retards dephosphorylation of p-Oct4 ( S229 ) during the M/G1 phase transition . The experimental strategy is shown ( upper panel ) . The same strategy was applied to ( D–F ) . Whole-cell lysates from E14 ESCs were collected and assessed by western blot . ( D ) Histogram shows cell cycle state of E14 ESCs without ( upper panel ) or with ( lower panel ) OKA treatment . ( E ) ChIP-qPCR analysis of E14 ESCs with anti-Oct4 in regions of pluripotency-associated Oct4 target genes during the M/G1 phase transition with or without OKA treatment . IgG was used as a control . Values represent mean ± standard deviation ( n≥3 ) . t-test was used to calculate the statistical significance of differences in enrichment levels of Oct4 at pluripotency-associated Oct4 target genes in ESCs during the M/G1 transition with or without OKA . ( *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 ) ( F ) Nascent RNA of pluripotency-associated Oct4 target genes from E14 ESCs were collected and analyzed by real-time qPCR during the M/G1 phase transition with or without OKA . Levels of each nascent RNA were normalized by those in asynchronous E14 ESCs . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 01010 . 7554/eLife . 10877 . 011Figure 4—figure supplement 1 . Dissociation of p-Oct4 ( S229 ) from chromatin occurs independent to chromatin status . ( A ) p-Oct4 ( S229 ) levels after release of E14 ESCs from M-phase arrest . Shown are immunoblots for the indicated proteins . ( B ) ChIP-qPCR analysis of E14 ESCs with anti-Oct4 in regions of pluripotency-associated Oct4 target genes after release of E14 ESCs from M-phase arrest . IgG was used as a control . Values represent mean ± standard deviation ( n≥3 ) . ( C ) Anti-Oct4 ChIP-qPCR of ZHBTc4 ESCs that express exogenous Oct4 in regions of pluripotency-associated Oct4 target genes during the M/G1 phase transition with or without OKA treatment . Values represent mean ± standard deviation ( n≥3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 011 Interestingly , the overall levels of Oct4 protein did not change significantly throughout the cell cycle . However , unlike Oct4 , the levels of Nanog , which has a short half-life ( Ramakrishna et al . , 2011 ) , rose in G1 phase , declined through S-G2/M phase , and reappeared at the start of the next G1 phase . This finding indicates that the synchronization of the phosphorylation state of Oct4 ( S229 ) with the cell cycle is linked to the resetting of Oct4 to its target genes . To test that PP1 is required for resetting the transcription of Oct4 target genes , we administered 50 nM OKA to nocodazole-pretreated E14 ESCs for 4 hr and released them into normal serum . As expected , OKA retarded the dephosphorylation of p-Oct4 ( S229 ) and re-entry into the next G1 phase ( Figure 4C and D ) . To examine the binding of Oct4 to its targeting pluripotency genes on chromatin during the M/G1 transition , we performed the ChIP-qPCR assay . As a result , Oct4 bound weakly to target genes in G2/M phase , strengthening its association during entry into G1 phase . On treatment with OKA , Oct4 binding to target genes declined significantly during entry into G1 phase ( Figure 4E ) . In addition , to elucidate the cell cycle effect induced by OKA treatment to Oct4 binding , we performed ChIP-qPCR assay in Zhbtc4 ESCs stably expressing wild-type Oct4 ( WT ) and phosphor-defect mutant ( S229A ) . When cells were released into G1 phase , OKA treatment significantly prevented the binding of Oct4 ( WT ) to target genes . On the other hand , binding of Oct4 ( S229A ) was relatively less affected even in treatment of OKA ( Figure 4—figure supplement 1C ) . However , Oct4 ( S229A ) mutant affects the O-GlcNAcylation of Oct4 , which is critical for Oct4 activity , thereby Oct4 ( S229A ) mutant fail to self-renew ( Jang et al . , 2012 ) . To analyze Oct4-depenent transcriptional resetting during the M/G1 transition , we measured nascent RNA levels ( Figure 4F ) . When E14 ESCs were arrested in G2/M phase , nascent RNA levels of a subset of Oct4-targeting pluripotency genes declined significantly versus asynchronous ESCs . Nascent RNA levels of certain Oct4 target genes were upregulated when cells entered G1 phase ( until 2 hr after release ) . Complementing the ChIP data , the nascent RNA levels of target genes were retarded after OKA treatment . Thus , we conclude that dephosphorylation by PP1 is critical for the transcriptional resetting of Oct4 to pluripotency genes during the M/G1 transition . We next wondered whether Oct4 resets a subset of cell cycle related genes during the M/G1 transition . To address this , we first narrowed down the putative 1258 Oct4 target genes by crossover between 5824 genes co-occupied by OSN and 4617 genes decreased by Oct4 depletion in ZHBTc4 ESCs ( Figure 5A and Figure 5—source data 1 ) using publically-available ChIP-seq and RNA-seq data ( Boo et al . , 2015; Whyte et al . , 2013 ) . By gene ontology analysis of the putative Oct4 target genes using DAVID ( http://david . abcc . ncifcrf . gov ) , we identified some Oct4 target genes associated with various cell cycle related functional categories ( Figure 5B and Figure 5—source data 2 ) . 10 . 7554/eLife . 10877 . 012Figure 5 . Oct4 regulates cell cycle related genes by direct targeting and resetting during the M/G1 transition . ( A ) A Venn diagram shows overlapped genes between proximal genes of Oct4 binding sites ( green , n=5824; ( Whyte et al . , 2013 ) ) and downregulated genes ( fold changes≤0 . 75 ) in ZHBTc4 ESCs after Oct4 depletion by doxycycline treatment for 2 days ( red , n=3617; [Boo et al . , 2015] ) . ( B ) Gene ontology ( GO ) functional categories for putative Oct4 target genes . Cell cycle related GO functional categories are enriched . ( C ) RNA-seq reads of Bub1 and Rif1 of E14 ESCs during ESC differentiation upon LIF withdrawal ( upper panel; [Xiao et al . , 2012] ) and ChIP-seq binding profiles of Oct4 at the Bub1 and Rif1 locus in undifferentiated E14 ESCs ( lower panel; [Whyte et al . , 2013] ) . ( D ) ChIP-qPCR analysis of E14 ESCs with anti-Oct4 in regions of Bub1 and Rif1 during the M/G1 phase transition with or without OKA treatment . IgG was used as a control . Values represent mean ± standard deviation ( n≥3 ) . t-test was used to calculate the statistical significance of differences in enrichment levels of Oct4 in ESCs during the M/G1 transition with or without OKA treatment . ( *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 ) ( E ) Nascent RNA levels of Bub1 and Rif1 in E14 ESCs were nalyzed by real-time qPCR during the M/G1 phase transition with or without OKA treatment . Levels of nascent RNA were divided by those in asynchronous state of E14 ESCs . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 01210 . 7554/eLife . 10877 . 013Figure 5—source data 1 . Identification of putative Oct4 target genes . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 01310 . 7554/eLife . 10877 . 014Figure 5—source data 2 . Cell-cycle related genes in putative Oct4 target genes . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 014 Intriguingly , we found that Oct4 governs the cell cycle genes related to S/G2/M phase . Thus , among these genes related to S/G2/M phase , we focused on Bub1 and Rif1 because loss of Bub1 and Rif1 are known to induce differentiation in the ESC-based knockdown experiment ( Dan et al . , 2014; Lee et al . , 2012 ) and expression levels of both genes decrease upon ESC differentiation ( Figure 5C upper ) in previously published RNA-seq study ( Xiao et al . , 2012 ) . Furthermore , both Oct4 enrichment score and expression level of Bub1 and Rif1 were one of the top10 genes among putative Oct4 targeting cell cycle related genes ( lower , Figure 5C and Figure 5—source data 2 ) . To investigate whether Bub1 and Rif1 are reset by PP1-mediated dephosphorylation of Oct4 during the M/G1 transition , we first investigated the resetting of Oct4 to target genes during the M/G1 transition by ChIP-qPCR ( Figure 5D ) . Oct4 binding to both genes on chromatin was weakly sustained in G2/M phase and rapidly increased during the M/G1 transition ( 0 . 5 hr after release into normal serum ) , and thereafter Oct4 binding is either saturated or declined , while induced Oct4 binding to both genes during the M/G1 transition were relatively retarded when cells were released into normal serum with treatment of OKA ( Figure 5D ) . Likewise , nascent RNA levels of Bub1 and Rif1 were downregulated when cells were arrested in G2/M phase and upregulated during the M/G1 transition , but rapid increase of nascent RNA levels of both genes during entry into G1 phase was retarded by OKA treatment ( Figure 5E ) . Considered together , we concluded that Oct4 directly controls cell cycle related genes by its resetting to cell cycle related genes during the M/G1 transition . To identify which regions are reset by Oct4 , we mapped the genomewide occupancy of Oct4 at G2/M and G1 phase in E14 ESCs by ChIP-seq . We found high-confidence peaks ( with p value<10−5 ) at G2/M ( 9204 ) and G1 ( 24548 ) phase ( Figure 6A and Figure 6—source data 1 ) . Expectedly , Oct4 bound much more regions of genome at G1 phase rather than G2/M phase ( Figure 6A ) . Furthermore , mean peak density of Oct4 bound regions was higher in G1 phase than G2/M phase ( Figure 6B ) implicating that Oct4 genome-widely resets the target genes at G1 phase . Next to identify reset region by Oct4 , we discovered 9092 of genomic regions enriched by Oct4 more than two-fold increase in G1 phase compared to G2/M phase ( Figure 6—source data 2 ) . We categorized these as the Oct4 resetting peaks , which are significantly enriched in not only pluripotency but also cell cycle categories ( Figure 6C ) . For example , we showed that Oct4 more strongly binds to the regions of both Nanog and Sox2 at G1 phase rather than G2/M phase , supporting that Oct4 resets its target genes at G1 phase ( Figure 6D ) . 10 . 7554/eLife . 10877 . 015Figure 6 . Oct4 ChIP-seq at G2/M and G1 phase of cell cycle . ( A ) Venn diagram of overlap between G2/M and G1 ChIP-seq peaks . ( B ) Mean ChIP-seq density of Oct4 around previously published Oct4-occupied regions ( Ang et al . , 2011 ) between G2/M and G1 phase . The level of Oct4 increased after release into M/G1 transition compared to G2/M phase . ( C ) Gene ontology ( GO ) functional categories for genes which are reset by Oct4 . Pluripotency and cell cycle related GO functional categories are significantly enriched . ( D ) Integrated genomics viewer ( IGV ) screenshots for ChIP-seq data of Nanog and Sox2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 01510 . 7554/eLife . 10877 . 016Figure 6—source data 1 . Identification of G2/M or G1 specific Oct4 binding peaks . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 01610 . 7554/eLife . 10877 . 017Figure 6—source data 2 . Candidate genes reset by Oct4 . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 017 To determine the significance of recycling Oct4 through Aurkb/PP1 in ESC pluripotency , we generated ZHBTc4 ESCs that stably expressing wild-type Oct4 ( WT ) , a phosphor-mimic Oct4 ( S229D ) mutant , and a PP1-binding-defective Oct4 ( F271A ) mutant . We confirmed that ectopic wild-type Oct4 and Oct4 mutants were expressed when endogenous Oct4 was removed by doxycycline ( dox ) treatment ( Figure 7A ) . 10 . 7554/eLife . 10877 . 018Figure 7 . Oct4 mutants—Oct4 ( S229D ) and Oct4 ( F271A ) —effect the loss of pluripotency and alter the cell cycle by impeding gene expression . ( A ) Flag Oct4 WT and mutants were stably incorporated into the genome of ZHBTc4 ESCs . These cells were treated with doxycycline for the indicated days . Stable expression of exogenous Oct4 was confirmed by western blot . ( B ) After 2 days of doxycycline treatment in ZHBTc4 ESCs ( Mock , wild-type Oct4 ( WT ) - , Oct4 ( F271A ) - , and Oct4 ( S229D ) - backup cells ) , DNA content ( 1x104 cells/sample ) were analyzed in each ESC by FACS . Values represent mean ± standard deviation ( n≥3 ) . t-test was used to calculate the statistical significance of differences in G1 phase of ZHBTc4-Oct4 ( WT ) versus -Oct4 ( S229D ) and -Oct4 ( F271A ) . ( *p<0 . 05 , ***p<0 . 001 ) ( C ) Anti-Oct4 ChIP-qPCR of ZHBTc4 ESCs that express exogenous Oct4 at 2 days after doxycycline treatment . F271A and S229D mutants showed decreased binding to target genes compared with Oct4 WT . Oct4-depleted ZHBTc4 cells ( Mock ) were used as a control . Values represent mean ± standard deviation ( n≥3 ) . t-test was used to calculate the statistical significance of differences in enrichment levels of Oct4 in ZHBTc4-Oct4 ( WT ) versus Oct4 ( S229D ) and Oct4 ( F271A ) ESCs . ( *p<0 . 05 , **p<0 . 01 ) ( D ) Levels of nascent RNA were measured by qRT-PCR and each nascent RNA levels were normalized by the levels in Oct4 ( WT ) backup ZHBTc4 ESCs in the asynchronous state after doxycycline treatment for 2 days . The experimental scheme is shown ( upper panel ) . ( E ) Relative expression of genes targeted by Oct4 related to pluripotency and cell cycle in ZHBTc4 . mRNA levels of the indicated genes decreased significantly in Mock , Oct4 ( F271A ) , and Oct4 ( S229D ) backup cells after doxycycline treatment but not in Oct4 ( WT ) backup cells . ( F ) Indicated ZHBTc4 ESCs were stained for alkaline phosphatase ( AP ) activity after 7 days of doxycycline treatment . ( G ) Reprogramming of MEFs into iPS cells driven by Oct4 , Sox2 , and Klf4 . Oct4 wild-type was replaced by F271A , S229A , and S229D mutants . Reprogrammed cells were identified by AP staining and counted . Results from 3 independent experiments are presented . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 01810 . 7554/eLife . 10877 . 019Figure 7—figure supplement 1 . Both S229D and F271A mutants of Oct4 decrease Oct4 activity . ( A ) Oct4-transcriptional activity was measured using NIH3T3 cells harboring ten copies of Oct4-responsive element ( 10X Oct4 RE ) -driven luciferase reporter gene stably . These stable cells ( NIH 3T3 ) were infected with retroviruses expressing Oct4 wild-type ( WT ) , F271A , S229A and S229D mutants and luciferase activity was measured 4 days after infection . Both S229D and F271A mutants of Oct4 barely induced Oct4-driven luciferase activities . Values represent mean ± standard deviation ( n≥3 ) . ( B and C ) Endogenous Oct4 in ZHBTc4 ESCs was replaced by infection with indicated retroviral Oct4 mutants . AP staining was performed after selection of infected cells ( 2 days ) in the presence of doxycycline 7 days later . AP-positive colony numbers were assessed as relative percent mean ± standard deviation ( n≥3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 019 Under the same conditions ( dox for 2 days ) , we analyzed the cell cycle patterns in these ESCs ( Figure 7B ) . Oct4-depleted ZHBTc4 ESCs ( Mock ) harbored significantly more cells in G1 phase and fewer S-phase populations , whereas Oct4 ( WT ) -backup cells had a typical ESC cell cycle profile . Notably , the cell cycle profile of Oct4 ( F271A ) -backup cells resembled that of ZHBTc4 ESCs ( Mock ) . Oct4 ( S229D ) -backup cells contained larger G1-phase populations than Oct4 ( WT ) -backup cells but fewer than Oct4 ( F271A ) -backup cells , indicating that the PP1-binding motif in Oct4 is essential for maintaining the pluripotency and cell cycle progression of ESCs . The binding of Oct4 mutants to target genes—including pluripotency-related and cell cycle genes—decreased significantly versus Oct4 ( WT ) after dox treatment for 2 days ( Figure 7C ) . To determine the reset patterns of Oct4 mutants during the M/G1 transition , we analyzed nascent RNA levels of a subset of Oct4 target genes in ZHBTc4 ESCs harboring Oct4 ( WT ) and Oct4 mutants that were pretreated with dox for 2 days , given nocodazole ( 6 hr ) , and released into normal serum ( Figure 7D ) . Nascent RNA transcripts of certain Oct4 target genes in Oct4 ( WT ) -backup cells were upregulated during the M/G1 transition , whereas those in Oct4-mutated cells did not increase versus ZHBTc4 Oct4 ( WT ) , indicating that Oct4 mutations ( S229D , F271A ) impede the prompt resetting of Oct4 during the entry into the next G1 phase . To determine the long-term effects of Oct4 mutations on ESC pluripotency and cell cycle progression , we cultured ZHBTc4 ESCs containing Mock , Oct4 ( WT ) , Oct4 ( S229D ) , and Oct4 ( F271A ) for 7 days with dox . Ectopic Oct4 ( WT ) and Oct4 mutants ( S229D , F271A ) were continuously expressed until dox treatment for 7 days ( Figure 7A ) . Total mRNA expression of a subset of Oct4 target genes in Oct4 ( WT ) cells was continuous , whereas that in Oct4-depleted and Oct4 mutant-backup cells was significantly downregulated ( Figure 7E ) . Consistent with this finding , alkaline phosphatase ( AP ) positive colonies in Oct4-depleted and Oct4 mutant-backup cells was much lower than in Oct4 ( WT ) -backup cells ( Figure 7F ) . To determine the effects of mutation on Oct4 transcriptional activity , we infected retroviral wild-type and mutant Oct4 into NIH-3T3 cells in which a 10X Oct4 response element ( RE ) -driven luciferase reporter was stably incorporated and measured luciferase activity . Oct4 ( F271A ) and Oct4 ( S229D ) showed little luciferase activity ( Figure 7—figure supplement 1A ) . Oct4 ( S229A ) also had weaker activity than Oct4 ( WT ) . Consistent with these data , during Oct4 depletion in ZHBTc4 ESCs , retroviral infection with Oct4 ( F271A ) and Oct4 ( S229D ) failed to rescue the maintenance of ZHBTc4 ESC self-renewal compared with Oct4 ( WT ) infected cells ( Figure 7—figure supplement 1B and C ) . In addition , these Oct4 mutants impeded somatic cell reprogramming ( Figure 7G ) . 10 . 7554/eLife . 10877 . 020Figure 8 . Schematic . A model describing the dissociation and resetting of Oct4 on chromatin by Aurkb/PP1 during the cell cycle . Aurkb phosphorylates Oct4 ( S229 ) , leading to dissociation of Oct4 from chromatin during G2/M phase . On mitotic exit , PP1 binds to Oct4 and dephosphorylates Oct4 ( S229 ) , which resets Oct4-driven transcription to maintain pluripotency and cell cycle progression . DOI: http://dx . doi . org/10 . 7554/eLife . 10877 . 020 Based on our findings , we propose that the recycling of Oct4 by Aurkb/PP1 over time and by location is pivotal for the transcriptional resetting of Oct4 during entry in to the subsequent G1 phase , ultimately maintaining ESC pluripotency and cell cycle progression ( Figure 8 ) .
In this study , we have demonstrated that Oct4 , a master pluripotency transcription factor , is spatiotemporally regulated by the Aurkb-PP1 axis during the cell cycle . In G2/M phase , Aurkb phosphorylates Oct4 extensively at serine 229 , leading to its dissociation from chromatin ( Figure 1 and Figure 2 ) . The detachment of Oct4 from chromatin is consistent with findings from previous reports . Most transcriptional machinery proteins , such as RNA pol II ( Gottesfeld and Forbes , 1997; Parsons and Spencer , 1997 ) , and many transcription factors dissociate from condensed chromatin at the onset of mitosis , and many studies have implied that mitotic dissociation of transcription factors occurs through phosphorylation ( Delcuve et al . , 2008; Dephoure et al . , 2008 ) . Aurka regulates ESC pluripotency through phosphorylation-mediated inhibition of p53 ( Lee et al . , 2012 ) but is not involved in the phosphorylation-mediated recycling of Oct4 ( Figure 2 and Figure 2—figure supplement 1 ) . Thus , both Aurk isoforms—Aurka and Aurkb—appear to be related to ESC pluripotency , but in a different molecular mechanism . We also found that PP1 mediates the resetting of Oct4 during the M/G1 transition . PP1 governs the resetting of cell cycle machinery ( Ceulemans and Bollen , 2004 ) , binding specific sequences—the RVxF motif—and dephosphorylating interactors ( Hendrickx et al . , 2009 ) . We identified an RVWF motif in the C-terminal POU-h domain ( Figure 3A ) and an Aurkb phosphorylation site ( S229 ) in the N-terminal POU-h domain of Oct4 , which come into close proximity in the 3-dimensional structure ( Figure 3B ) . This arrangement convinces us that PP1 binds Oct4 through the RVxF motif and dephosphorylates Oct4 in vitro ( Figure 3E and F ) . In addition , we revealed that PP1 inhibition by OKA treatment delayed the Oct4 dephosphorylation at S229 residue and consequently resetting to the target genes in E14 ESCs during M/G1 transition ( Figure 4C and E ) . We further found that OKA treatment relatively less affected the binding of phosphor-defect Oct4 ( S229A ) to target genes , significantly impeding the resetting of Oct4 ( WT ) at M/G1 transition ( Figure 4—figure supplement 1C ) , implying that dephosphorylation of S229 residue by PP1 is paramount for the resetting of Oct4 to target genes . Nonetheless , we cannot perfectly rule out the possibility that OKA treatment may affect the function of broad spectra of cell cycle regulators during M/G1 transition . We confirmed that Aurkb-phosphor-mimetic and PP1-binding-defective mutations lead to a loss of pluripotency , even in the presence of LIF , and that introduction of these mutants into somatic cells lowers the reprogramming efficiency ( Figure 7G ) , supporting that Aurkb/PP1 is a critical pair of regulators in resetting Oct4 on chromatin during the cell cycle . Mitotic phosphorylation of a transcription factor affects its transcriptional activity and other properties . For example , Oct1 , a member of the POU domain transcription factors , is phosphorylated during mitosis and localizes to the spindle matrix , forming a complex with lamin B1 at the midbody ( Kang et al . , 2011 ) . Mitotic phosphorylation of Sp1 protects itself from ubiquitin-dependent degradation ( Chuang et al . , 2008 ) . Notably , we found that Oct4 protein levels were unchanged , even though nascent Oct4 RNA levels fluctuated during the cell cycle ( Figure 4A and F ) , implying that phosphorylated Oct4 is probably protected from degradation during the cell cycle . We are interested in determining the mechanism of how phosphorylated Oct4 escapes protein degradation . The molecular association between pluripotency and the cell cycle in ESCs has garnered attention with regard to identifying the mechanism by which the cell fate of ESCs is determined . In particular , the protraction of G1 in naive ESCs by knockdown of cyclin E1 causes spontaneous differentiation; in contrast , promoting G1/S transition through overexpression of cyclin E1 enhances self-renewal ( Coronado et al . , 2013 ) . G1 phase regulators , cyclin D proteins control the differentiation of hESCs into various lineages via the TGF-β-Smad2/3 pathway ( Pauklin and Vallier , 2013 ) . Sox2 was identified as a cell cycle regulator that suppresses p21 and p27 and induces cyclin D3 expression ( Herreros-Villanueva et al . , 2013 ) . Oct4 downregulation lengthens G1 phase and upregulates p21 in ESCs ( Lee et al . , 2012 ) . A nontranscriptional function of Oct4 in mitotic entry has recently been reported ( Zhao et al . , 2014 ) . A recent study performed functional screening of human embryonic stem cells under various differentiation conditions and identified that genes involved in S and G2 phases are gatekeepers of differentiation ( Gonzales et al . , 2015 ) . Furthermore , S and G2 phases of cell cycle possess an intrinsic propensity for maintaining pluripotency . In this study , in addition to observing that Aurkb/PP1 control the dynamics of Oct4 during the cell cycle , we found that Oct4 governs the cell cycle of ESCs by directly targeting genes that are related to cell cycle regulation and pluripotency . In particular , 2 cell cycle genes—Bub1 and Rif1—are reset by Oct4 during the cell cycle ( Figure 5E and Figure 7D ) . Considering that Bub1 and Rif1 are important cell cycle regulators that control the checkpoints for mitosis and replication ( Bolanos-Garcia and Blundell , 2011; Yamazaki et al . , 2013 ) and are crucial for maintaining pluripotency ( Dan et al . , 2014; Lee et al . , 2012 ) , the regulation of S-G2-M phase by Oct4 might also be critical for ESC pluripotency . Our study might provide insights into why ESCs reset Oct4 during the cell cycle . Like , this mechanism might bifurcate the fate of ESCs: tight resetting of Oct4 on chromatin during the cell cycle strengthens the pluripotency of ESCs at the ground state , but on differentiation , loose resetting of Oct4 at the next M/G1 transition leads to lineage differentiation . During reprogramming by nuclear transfer , mitotic chromosomal condensation is required to reset the origins of replication of differentiated donor cells in embryonic DNA replication ( Lemaitre et al . , 2005 ) , transfer of a mitotic genome into a zygote in mitosis-enhanced reprogramming ( Egli et al . , 2007 ) , and mitotic chromatin induces core pluripotency factors more rapidly than interphase nuclei ( Halley-Stott et al . , 2014 ) , suggesting that a genome can be exchanged during mitosis which is an open window that allows transcription factors to occupy target genes on mitotic exit and thus enabling postmitotic cell fate changes to be induced . We have provided evidence that cell cycle machinery cooperates with pluripotency transcriptional programs . The resetting of Oct4 occurs rapidly during the exit from mitosis and that delayed resetting alters the cell cycle and effects the loss of pluripotency—ie , prompt resetting of Oct4 prevents postmitotic cell fate changes in ESCs . Based on our results , we suggest that the potential of ESCs to differentiate might be derived from the small window of the M/G1 transition by which the resetting of Oct4 is the central mechanism to determine the maintenance of ESC pluripotency or lineage commitments .
E14 and ZHBTc4 ESCs were cultured as described ( Jang et al . , 2012 ) in 0 . 1% gelatin ( Sigma-Aldrich , St . Louis , Missouri ) coated plates . ZHBTc4 ESCs were kindly provided by Hitochi Niwa ( RIKEN , Japan ) . The mouse ESC medium was composed of DMEM ( Hyclone , Logan , Utah ) and 15% ( v/v ) fetal bovine serum ( FBS; Gibco , Grand Island , New York ) , supplemented with 2 mM L-glutamine , 55 μM β-mercaptoethanol , 1% ( v/v ) nonessential amino acids , 100 U/ml penicillin , 100 μg ml-1 streptomycin ( all from Gibco ) , and 1000 U/ml ESGRO ( Millipore , Germany ) . Aurora kinase inhibitors AT9283 , Hesperadin and MLN8237 were purchased from Selleckchem ( Houston , Texas ) and okadaic acid from Sigma . For G2/M phase synchronization , E14 cells were treated with 200 ng/ml nocodazole ( Calbiochem , Germany ) for 10 hr and ZHBTc4 cells were treated with same concentrations of nocodazole for 6 hr . In order to release , synchronized cells were washed three times with PBS , and incubated for the indicated time in fresh culture media . For cell cycle analysis , the collected cells at the indicated time were immediately fixed in 70% ethanol and stained with propidium iodide ( PI; Sigma , P4170 ) for 1 hr at room temperature in the dark . Cell cycles were analyzed using FACSCalibur flow cytometer and LSRII ( SORP ) ( Becton Dickinson , Franklin Lakes , New Jersey ) . Analysis of cell cycle data was performed with FlowJo ( Tree Star Inc . , Ashland , Oregon ) . Cells grown on coverslips were fixed in 4% ( w/v ) paraformaldehyde and permeabilized in 0 . 5% ( w/v ) Triton X-100 in PBS for 30 min at room temperature ( RT ) . After permeabilization , the cells were blocked with 3% ( w/v ) BSA for 30 min . Subsequently , they were incubated in primary antibody for 1 hr at RT . Antibody dilutions were 1:500 for anti-Oct4 ( Santa Cruz , Dallas , Texas , sc-5279 ) , 1:200 for anti-p-Oct4 ( S229 ) , 1:200 for pH3S10 ( Millipore , 05–598 ) . Secondary antibodies used in immunostaining were Alexa Fluor 488 , 568 ( Invitrogen , Carlsbad , California ) . Confocal micro-images were obtained by a confocal laser scanning microscope ( Carl Zeiss , Germany , LSM 510 META ) . The plasmid pGAE-mKO2:Cdt1 and pGAE-mAG:Geminin were generously provided by Savatier , P . ( INSERM U846 , France ) . For stable expression in ESCs , fragments containing mKO2:Cdt1 and hmAG1:Geminin coding sequences , respectively , were generated by PCR amplification and were sub-cloned between the SalI and AgeI sites into pCAG-IP vector to generate pCAG-mKO2:Cdt1-IP and pCAG-mAG:Geminin-IP . PP1α , β and γ were amplified by PCR from cDNA of E14 ESCs . The PCR products were digested with XhoI and AgeI then subcloned into pCAG-Flag-IP vector . All Oct4 mutants were generated by site-directed mutagenesis ( Intron , Korea ) . For long-term transgene expression in ZHBTc4 ESCs , Flag-tagged Oct4 was cloned into pCAG-Flag-IP , which was generated by inserting a Flag tag into pCAG-IP , kindly provided by Hitoshi Niwa ( RIKEN , Japan ) . To generate ESCs stably expressing Flag-Oct4 , pCAG-Flag-Oct4-IP was transfected into ESCs using Lipofectamine ( Invitrogen ) . After 48 hr of transfection , selection with 2μg/ml of puromycin was performed to determine stable integration . Puromycin resistant cells were expanded and analyzed for Oct4 by Western blot . The reporter gene assay was done as described ( Jang et al . , 2012 ) . Briefly , Ten copies of Oct4-responsive element ( 10X Oct4 RE ) -driven luciferase reporter gene was incorporated into the genome of NIH 3T3 cells by retroviral infection . To stably incorporate reporter gene into genomic DNA , cells were selected with puromycin for at least 2 weeks . These stable cells were infected with retroviruses expressing Oct4 . Luciferase activity was measured 4 days after infection of Oct4 . 0 . 2 μg of purified GST-Oct4 proteins were used in cold in vitro kinase assays with purified recombinant kinases . All kinases were purchased from Proqinase ( Germany ) . Kinase reactions were performed in kinase buffer ( 60mM HEPES-NaOH pH 7 . 5 , 3 mM MgCl2 , 3 mM MnCl2 , 3 μM Na-orthovanadate , 1 . 2 mM DTT , 0 . 5mM ATP ) for 30 min at 30°C . Then reactions were stopped by the addition of 5X SDS-PAGE loading buffer and assessed by Westernblot . For radioactive in vitro kinase assay GST-Oct4 was incubated with Aurkb in kinase buffer ( 60 mM HEPES-NaOH pH 7 . 5 , 3 mM MgCl2 , 3mM MnCl2 , 3 μM Na-orthovanadate , 1 . 2 mM DTT , 0 . 25 mM ATP ) with 0 . 1 µM γ-32P-ATP ( NEG002A250UC , purchased from PerkinElmer , Waltham , Massachusetts ) for 30 min at 30°C . Reactions were then stopped by the addition of 5X SDS-PAGE loading buffer and loaded for separation on 8% SDS-PAGE gel . After staining with Coomassie Blue , the gels were dried and exposed to films . In vitro phosphatase assay was performed as described ( Kassardjian et al . , 2012 ) . Briefly , phosphorylated GST-Oct4 was pulled down with Glutathione Sepharose 4B ( Sigma ) for 4 hr at 4°C and suspended in phosphatase buffer . GST-Oct4 beads phosphorylated by Aurkb were incubated with 1 μg of purified His-tagged PP1 for 1 hr at 30°C , with mild shaking . Immunoprecipitation and Western blot were performed as described ( Jang et al . , 2012 ) . Anti-p-Oct4 ( S229 ) antibody was made by GenScript ( Piscataway , New Jersey ) . Anti-Oct4 ( sc-5279 ) , anti-PP1α ( sc-6104 ) and anti-PP1γ ( sc-6108 ) were acquired from Santa Cruz Biotechnology; anti-Nanog ( ab14959 ) and anti-PP1β ( ab53315 ) were purchased from Abcam ( UK ) ; anti-Aurka ( 610938 ) and anti-Aurkb ( 611082 ) were acquired from BD Transduction Laboratories ( San Jose , California ) ; anti-Esrrb ( H6707 ) was obtained from R&D System ( Minneapolis , Minnesota ) ; and anti-phospho-Histone H3 ( Ser10 ) ( 05–598 ) was purchased from Millipore . To prepare nascent RNA , Click-iT Nascent RNA Capture Kit ( Life Technologies , Carlsbad , California ) was used . First , ESCs were incubated with 100 μM of 5-ethynyl uridine ( EU ) for 15 min . After incubation , EU-labeled RNA was isolated and converted into biotinylated RNA by Click reaction . The biotinylated RNA was pulled down with streptavidin magnetic beads . cDNA was synthesized with RNA bound to the beads as a template and analyzed by qRT-PCR . The primers that span an intron–exon boundary were used for specifically detecting of nascent RNA . The primer sequences used in the analysis are given in Supplementary file 1 . Lentiviruses were produced using PLKO-puro constructs that express shRNAs; Aurka#2 ( TRCN0000025140 ) , Aurka#5 ( TRCN00000251430 ) , Aurkb#2 ( TRCN0000374361 ) , Aurkb#3 ( TRCN0000321718 ) , Pak1#4 ( TRCN0000025258 ) , Pak2#1 ( TRCN0000025209 ) , Pak2#5 ( TRCN0000413412 ) , Pkca#2 ( TRCN0000022875 ) and Pkca#5 ( TRCN00002187830 ) were purchased from Sigma . 293FT cells were cotransfected with 0 . 5 μg each of pMD2 . G , pMDLg/pRRE , pRSV-rev , and 1 μg of pLKO-shRNA using Lipofectamine ( Invitrogen ) in a 6-well plate . 48 hr after transfection , virus-containing medium was collected and passed through 0 . 45-μm filters . Polybrene ( 10 μg/ml ) was added to target cells immediately prior to infection , and infection was performed for 5 hr . Target cells were selected with puromycin ( 2 μg/ml ) 48 hr after infection . Preparation of RNAs , reverse transcription PCR , real-time qPCR , and chromatin immunoprecipitation ( ChIP ) assay were done as described ( Jang et al . , 2012 ) . The primer sequences used are given in Supplementary file 1 . We ChIPed with Oct4 antibody at G2/M ( nocodazole-treated cells ) and G1 phase ( release of nocodazole-treated cells ) in E14 ESCs . ChIPed DNAs were sequenced by LAS ( Korea , http://www . lascience . co . kr ) . For the ChIP-seq analysis , reads were mapped to the mouse genome ( NCBI build 37/ mm9 ) . The detailed analysis was done as described ( Kim et al . , 2015 ) . Self-renewal assay ( colony-forming assay ) was performed as described ( Chambers et al . , 2007 ) . ESCs were trypsinized to a single cell and re-plated 500 cells in a well of 6-well plates . After incubation for 7 days with/without doxycycline , the plates were stained for alkaline phosphatase and counted . Reprogramming was performed as described ( Jang et al . , 2012 ) . Briefly , equal amounts of retrovirus encoding Oct4 , Sox2 , and Klf4 were applied to MEFs in 10% FBS DMEM media containing 8 ng/ml polybrene . After 24 hr , fresh mESC culture media were added , and the culture was then maintained for up to 21 days . Reprogrammed cells were identified by alkaline phosphatase ( AP ) staining and scored . Nano-LC-ESI-MS/MS Analysis of Phosphorylation Sites in Oct4 . Nano-LC-ESI-MS/MS analysis was previously performed as described ( Jang et al . , 2012 ) . | Embryonic stem cells can give rise to any type of cell in the body – an ability known as pluripotency . These cells rapidly divide and self-renew until they are exposed to signals that cause them to mature into a particular specialized cell type . As cells prepare to divide , they transition through a series of phases known as the cell cycle . In embryonic stem cells , these phases are often shorter than in other cell types . This altered timing is thought to be important for maintaining the pluripotency of the stem cells . Proteins called core transcription factors also help stem cells to remain pluripotent . Evidence suggests that the activity of some of these proteins affects the timing of the different cell cycle phases . However , it is not clear exactly how they do so or how the activity of the transcription factors is controlled . A core transcription factor called Oct4 is thought to be a “master regulator” of pluripotency that controls the activity of many of the other core transcription factors . Shin , Kim , Kim , Kim et al . have now studied the activity of Oct4 around the point of cell division . This revealed that a protein called aurora kinase B modifies Oct4 by adding a phosphate group to it just before a cell divides . This modification causes Oct4 to detach from chromatin , the protein structure in which DNA is packaged inside cells . Following cell division , another protein called PP1 removes the phosphate group from Oct4 . This “resets” the pluripotency of the stem cell , allowing it to continue to self-renew . Cells that contain only mutant forms of Oct4 that cannot bind to aurora kinase B or PP1 lose their pluripotency . The mutant Oct4 proteins also alter the cell cycle of the stem cells . Overall , Shin et al . ’s findings suggest that Oct4 regulates the cell cycle of embryonic stem cells as well as their pluripotency . How Oct4 activity affects the specialization of the stem cells into mature cell types remains to be investigated in future studies . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2016 | Aurkb/PP1-mediated resetting of Oct4 during the cell cycle determines the identity of embryonic stem cells |
Although changes in brain activity during learning have been extensively examined at the single neuron level , the coding strategies employed by cell populations remain mysterious . We examined cell populations in macaque area V4 during a rapid form of perceptual learning that emerges within tens of minutes . Multiple single units and LFP responses were recorded as monkeys improved their performance in an image discrimination task . We show that the increase in behavioral performance during learning is predicted by a tight coordination of spike timing with local population activity . More spike-LFP theta synchronization is correlated with higher learning performance , while high-frequency synchronization is unrelated with changes in performance , but these changes were absent once learning had stabilized and stimuli became familiar , or in the absence of learning . These findings reveal a novel mechanism of plasticity in visual cortex by which elevated low-frequency synchronization between individual neurons and local population activity accompanies the improvement in performance during learning .
Although perceptual learning has been a phenomenon studied for many decades , the neuronal mechanisms underlying the improvement in sensory discrimination after practice remain mysterious . Classical theories of perceptual learning have proposed that the improvement in performance during learning relies on a fine retuning and an overrepresentation of the neurons involved in the processing of the trained stimuli ( Sagi and Tanne , 1994; Karni and Bertini , 1997 ) . However , experiments in sensory cortex failed to find an expansion in the cortical representation of the trained stimulus ( Schoups et al . , 2001; Ghose et al . , 2002 ) and reported relatively modest changes in neuronal responses and sensitivity after learning ( Schoups et al . , 2001; Ghose et al . , 2002; Sigala and Logothetis , 2002 ) . An alternative possibility is that learning causes pronounced changes in the way in which information is encoded in population activity , but weaker changes at the single-cell level . Examining the changes in population activity related to learning is motivated by the fact that sensory information is actually encoded in a distributed manner across populations of neurons . Thus , behavioral performance in visual , auditory , or motor tasks ( Sparks et al . , 1976; Georgopoulos et al . , 1986; Lee et al . , 1988 ) is known to be much more accurate than would be predicted from the responses of single neurons ( Paradiso , 1988 ) . Furthermore , theoretical studies have demonstrated that coding strategies based on the responses of a population of neurons encode more information than coding strategies based on single-cell responses ( Zohary et al . , 1994; Abbott and Dayan , 1999; Pouget et al . , 2000; Sompolinsky et al . , 2001 ) . Unfortunately , despite the clear importance of analyzing population activity , whether and how networks of cells exhibit changes in population activity to influence the accuracy of behavioral responses during learning has rarely been investigated experimentally . The major limitation that prevented our understanding of how the information encoded by populations of neurons changes during the time course of learning has been the inability to record from the same population of neurons for extended periods of time , while animals improve behavioral performance during learning . To overcome this limitation , we examine here the relationship between learning and population activity by devising a task in which monkeys rapidly learn ( within one session ) to discriminate between consecutive , briefly flashed images slightly rotated with respect to each other . This task offers the advantage that the same population of neurons can be examined during the entire time course of learning , thus greatly reducing the sampling bias characterizing the day-by-day acute recordings of neuronal activity . To examine the neural network correlates of rapid learning , we focused on the population response in mid-level visual cortex ( area V4 , Desimone and Schein , 1987; Hegde and Van Essen , 2005 ) . Among all sensory cortical areas , the visual cortex is the best understood in terms of receptive field properties and circuitry ( Hubel and Wiesel , 1969 ) , thus , it provides a unique opportunity for investigating the impact of perceptual learning on neuronal responses . Importantly , area V4 sends inputs to downstream areas involved in perceptual decisions , and individual neuron responses in extrastriate cortex are more strongly correlated to behavior than those in V1 ( Nienborg and Cumming , 2006; Gu et al . , 2014 ) . In addition , lesion studies have suggested that area V4 plays a key role in perceptual learning ( Schiller and Lee , 1991 ) .
We simultaneously recorded single units and local field potentials ( LFP ) using multiple electrodes in mid-level visual cortex ( area V4 ) of two macaque monkeys . Monkeys were trained to perform an image discrimination task ( Figure 1A ) in which a target image presented for 300 ms was compared , after a 800–1200-ms delay , with a 300-ms test image ( either identical to the target or rotated by 3o , 5o , 10o , or 20o; the monkeys were required to complete at least 800 trials in one session; all test orientations were randomly interleaved ) . After monkeys were able to accurately ( >85% ) perform the task with a set of 10 prototype stimuli ( to which they were exposed for many weeks of practice ) , we introduced novel stimuli at the beginning of each session . Behavioral and electrophysiological data were analyzed in parallel after dividing trials into 96-trial blocks in order to detect changes in performance due to learning ( a complete session lasted about 90 min ) . 10 . 7554/eLife . 08417 . 003Figure 1 . Individual neuron and LFP responses in area V4 during rapid learning . ( A ) Schematic description of the rapid learning protocol . After 500 ms of fixation , a target image was presented for 300 ms , followed , after a 1000-ms blank , by a 300-ms test image ( monkeys maintained fixation within a 1 × 1 deg window ) . Monkeys were required to hold a lever for 1500 ms if the test image was rotated relative to the target , and release it within 500 ms if the target and test were identical . ( B ) Changes in mean behavioral discrimination threshold binned in 96-trial blocks for monkeys 1 and 2 . Error bars represent s . e . m . ( C , D ) Two example inter-spike interval ( ISI ) distributions for each block in two single units . Insets show average spike waveforms in each block . ( E ) Raster plots represent the spike timing of two V4 neurons recorded throughout the time course of learning . ( F ) Examples of local field potential ( LFP ) responses at two recording sites . Each curve represents the average response across all trials and test orientations within the same block . The horizontal bars mark the 300-ms time windows during which the target and test stimuli were presented . ( G ) Mean single unit firing rates and discrimination performance ( d' ) across blocks of learning . The error bars represent SEM . ( H ) Mean relative LFP power in the theta , alpha , beta , and gamma bands in different blocks . The LFP power of individual channel in each block across all test orientation was normalized by the mean in the first block; the mean population LFP power was calculated by averaging across all channels . LFP power in different frequency bands was scaled differently since recordings from different sites depended on electrode impedance . DOI: http://dx . doi . org/10 . 7554/eLife . 08417 . 00310 . 7554/eLife . 08417 . 004Figure 1—figure supplement 1 . Control experiment—Monkey 2 was passively exposed for 10 sessions to novel natural scenes ( similar to those in Figure 1A ) while the animal performed a color detection task ( red–green task ) in the contralateral hemifield . After 150 trials of passive ( unattended ) exposure to the image , the monkey was engaged in the rapid learning experiment described in the manuscript ( Figure 1A ) . However , even though images themselves were familiar to the animal , the fact that the monkey did not practice the image orientation discrimination task led to behavioral effects similar to those reported in the manuscript . That is , we found a gradual improvement in discrimination threshold at the end of each session—the gradual learning curve looked similar to that obtained in the original experiments ( asterisks indicate p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08417 . 004 Figure 1B shows that , as expected , behavioral discrimination threshold gradually decreases during the time course of learning ( monkey M1: n = 31 sessions , p = 1 . 42 10−5; monkey M2: n = 17 sessions , p = 0 . 028 , ANOVA test; each of blocks 2–4 is characterized by a lower threshold than block 1 , p < 0 . 05 , post hoc multiple comparisons ) . Interestingly , the largest improvement in behavioral performance occurred between blocks 1 and 2 , followed by performance saturation after block 2 ( on average , the image orientation discrimination threshold decreased from 18 . 9o to 4 . 5o , that is , a 76 . 2% decrease at the end of the session ) . We first examined whether learning induces changes in neuronal responses at individual V4 sites . Figure 1C , D shows two examples of spike waveforms and inter-spike interval distribution that demonstrate that our recordings were stable in time . Figure 1E , F shows single unit and average LFP traces of example recording channels . We measured the block-by-block changes in mean firing rate , neuronal discrimination performance ( d' , defined as the capacity to discriminate between nearby test image orientations ) , and mean LFP power across blocks of trials . Remarkably , learning did not influence individual neuron responses and LFPs ( during the test presentation averaged across all stimulus orientations ) in a significant manner . Indeed , Figure 1G shows that the population mean firing rate and d' throughout the test stimulus presentation ( n = 105 sites ) did not differ during the time course of learning ( balanced one-way parametric ANOVA test , p > 0 . 9 for mean firing rate; p > 0 . 1 for d' ) . For individual cells , we found that 15 . 4% cells increased their firing rates , whereas 17 . 3% showed a decrease ( p < 0 . 05 , ANOVA test , post hoc comparing blocks 1 and 4 ) . Additionally , we compared LFP power changes relative to block 1 but did not find significant differences across blocks of trials in any frequency band ( Figure 1H , ANOVA and Kruskal–Wallis tests , p > 0 . 15; we also examined the event-related potentials , or ERPs , for the target and test stimuli , but they did not change across blocks; target: p = 0 . 66 , and test: p = 0 . 59 , ANOVA tests ) . Next , we directly tested our hypothesis that the improvement in behavioral performance during learning is accompanied by synchronous firing of neurons with their neighbors . We thus examined the timing relationship between the spikes of single neurons and the ongoing LFP oscillation by quantifying the spike-field coherence ( SFC ) . First , we computed the spike-triggered average ( STA , triggered from same number of subsampled spikes in order to avoid bias ) by averaging the LFP signal within a window centered ±150 ms on each elicited spike in each block . Second , we computed SFC by dividing the power spectrum of the STA to the average of all power spectra of the LFP segments used to obtain the STA . SFC varies as a function of frequency and yields values between 0 and 1 . The larger the SFC , the more accurately the spikes follow a particular phase of this frequency . We calculated the SFC separately for each block ( for each single-unit-LFP pair , n = 625 ) . Figure 2A , B shows two examples of cross-channel SFC early in the session ( block 1 ) and during learning ( blocks 2–4 ) . Clearly , rapid learning is associated with an increase in low-frequency SFC ( particularly in the theta band , 4–8 Hz ) during the intervals when the two stimuli are presented ( 0–300 ms and 1300–1600 ms . In contrast , SFC at higher frequency bands ( alpha , beta , and gamma bands ) was either unchanged or slightly decreased during the time course of learning . These results were confirmed for the population of spike-LFP pairs ( Figure 2C ) —learning was associated with an increase in theta SFC during the intervals when the two stimuli are presented ( p < 0 . 05 , Wilcoxon signed-rank test , by comparing theta SFC in blocks 2–4 vs block 1 for time intervals 150–270 ms and 1430–1550 ms; SFC was calculated within a 300-ms window sliding every 10 ms ) , whereas coherence in the high-frequency bands did not change across blocks of learning ( p > 0 . 05 ) . 10 . 7554/eLife . 08417 . 005Figure 2 . Rapid learning increases spike-LFP theta synchronization . ( A , B ) Spike-field coherence ( SFC ) from two example pairs of recording sites during blocks of learning . Each row shows the mean SFC in the low-frequency bands for the two example pairs in a particular block . ( C ) SPC—population average . The two panels show the population average ( median change ) of normalized SFC change in blocks 2–4 relative to block 1 throughout the trial . For each block , SFC was calculated within a 300-ms window sliding every 10 ms , and then the results were normalized for each session . The left panel shows SFC changes for the low frequencies , and the right panel represents frequencies within the gamma band . The x-axis represents time relative to the onset of the target stimulus . The two white vertical bars mark the onset of the target and test stimuli . The horizontal bars represent the time interval when the target and test stimuli are presented . DOI: http://dx . doi . org/10 . 7554/eLife . 08417 . 005 Previous studies have reported synchrony modulation mainly during the sustained neuronal response ( Gieselmann and Thiele , 2008 ) . Therefore , we focused our SFC analysis between 150 and 350 ms after stimulus onset since the transient spike responses , which are highly synchronized to stimulus onset , might possibly induce artifacts related to event-related synchrony . In addition , since neurons have different strengths of their stimulus onset transients , and the SFC measure is sensitive to spike counts , focusing the analysis on the period immediately following stimulus presentation might contaminate our measure of spike-LFP synchronization . We thus computed the mean SFC values ( across all pairs and sessions ) and found that whereas SFC in the higher frequency bands showed no significant change across blocks ( Kruskal–Wallis , p > 0 . 2 ) , theta-band SFC was significantly increased during the time course of learning ( Figure 3A; n = 625 pairs , comparing blocks 2–4 with block 1 , Kruskal–Wallis , p < 10−7 ) . Figure 3B shows the distribution of SFC values for all the pairs in our population and reveals that learning is accompanied by a pronounced increase in theta SFC in blocks 2–4 relative to block 1 ( p < 0 . 05 , t-test ) . Furthermore , post hoc analysis shows a significant SFC increase in each of blocks 2 , 3 , and 4 relative to block 1 ( p < 0 . 05; the median theta SFC was increased by 34% ) . The increase in theta SFC during learning was statistically significant in both animals ( Figure 3C , p < 0 . 001 , Kruskal–Wallis test , post hoc analysis ) . 10 . 7554/eLife . 08417 . 006Figure 3 . Changes in spike-LFP coherence during learning . ( A ) Relative change of mean SFC in each block and frequency band . The error bars represent s . e . m . across spike-LFP pairs . Asterisks for each point show the statistical significance of the difference between SFC in each of blocks 2 , 3 , and 4 relative to block 1 in each frequency band . ( B ) Distribution of mean theta SFC values in blocks 2–4 vs block 1 . Color indicates the number of spike-LFP pairs measured in blocks 2–4 vs block 1 in 0 . 02 bins . ( C ) Block-by-block change in theta SFC relative to block 1 for each monkey . Error bars represent s . e . m . ( * represents p < 0 . 05; *** represents p < 0 . 001 ) . ( D ) Relative change in theta SFC when the pairs originating from the same electrode are included ( blue , all pairs ) or excluded ( red , different electrodes ) . The error bars represent s . e . m . ( E ) Near pairs ( electrode distance <2 mm ) : Kruskal–Wallis test showed no significant changes induced by learning in alpha , beta , and gamma bands ( p = 0 . 09 , 0 . 19 , and 0 . 12 , respectively ) , but a significant increase in theta band ( p = 2 . 29 10−5 ) . ( F ) Far pairs ( electrode distance >2 mm ) : Kruskal–Wallis test showed no significant changes in alpha , beta , and gamma bands ( p = 0 . 58 , 0 . 10 , and 0 . 13 respectively ) , but a significant increase in theta band ( p = 1 . 73 10−6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08417 . 006 One potential concern is that the sensitivity to SFC may differ for different frequency ranges irrespective of the changes observed during learning . For instance , while the theta power is constant across blocks , the power in other bands declines somewhat ( Figure 1H ) . In addition , the sensitivity of the coherence results may be related to the number of action potentials falling within a period at each frequency . However , this is not an issue in our analysis . The STAs used in the calculation of SFC is measured by summing all LFP segments and then dividing by the number of spikes . Even though the power spectrum of the STA depends on the power spectrum of the LFP signal ( decreasing LFP amplitude decreases the STA power despite the absence of spike-LFP synchronization ) , SFC is obtained by normalizing the power spectrum of the STA by the average of all power spectra of all LFP segments that were averaged to obtain the STA . This normalization ensures that SFC is independent of the spiking rates and the power spectrum of the LFP . Thus , the small block-by-block changes in LFP power in the alpha , beta , and gamma bands ( Figure 1H ) are unlikely to alter the sensitivity of our SFC measure during learning in the corresponding frequency band . Another concern is that the changes in theta SFC during learning may simply reflect an ERP with a predominant theta component . Thus , we shuffled the trials and recomputed SFC ( in each frequency band ) —if the change in theta SFC across blocks simply reflects the stimulus effect , we would expect to find a prominent increase in spike-LFP synchronization even when trials were shuffled . However , as shown in Figure 3A , the increase in theta SFC is unrelated to stimulus presentation ( although only theta SFC is shown in Figure 3A , we did not find statistically significant block-by-block changes in SFC in any frequency band , p > 0 . 05 , ANOVA test ) . We also separated our spike-LFP coherence analysis for the pairs originating from different electrodes and pairs originating from the same or different electrodes . However , although the increase in theta SFC during learning was slightly larger when the spike-LFP pairs were taken from different electrodes ( n = 523 pairs ) , the effects were highly significant in both cases ( Figure 3D , Kruskal–Wallis test ) . Since LFPs are composed of extracellular voltage fluctuations including local excitatory and inhibitory intracortical inputs ( Leopold and Logothetis , 2003 ) originating from recording sites within 2 mm or less ( Katzner et al . , 2009 ) , the modulation in SFC during learning was expected to be more pronounced when the recording sites are close to each other . Although we divided our SFC pairs into near ( <2 mm ) and far ( >2 mm ) groups ( 345 ‘near’ and 280 ‘far’ pairs ) , we found that the changes in theta SFC did not depend on the distance between electrode pairs ( Wilcoxon signed-rank test , p < 10−11 for both ‘near’ and ‘far’ pairs , see Figure 3E , F ) . Our analysis in Figure 2 suggests a sharp increase in theta spike-LFP coherence from block 1 to block 2 to match the changes in behavioral discrimination performance during learning ( Figure 1B ) . However , since this analysis was performed on individual blocks of trials , this might have occluded a gradual trial-by-trial transition in theta SFC . To address this issue , we computed the changes in theta SFC and the behavioral discrimination threshold relative to the first 64 trials in the session using a sliding window of 64 trials ( in 10-trial increments ) . As shown in Figure 4 , there was a gradual increase in theta spike-LFP coherence across trials that matched the time course of the behavioral improvement during learning . This indicates that both the changes in theta spike-LFP coherence and the improvement in discrimination occur gradually during learning . 10 . 7554/eLife . 08417 . 007Figure 4 . Gradual changes in theta spike-LFP coherence and behavioral performance during learning . ( A ) Mean behavioral discrimination threshold calculated throughout the session using a sliding window of 64 trials in steps of 10 trials . The solid line represents the exponential fit . The error bars represent s . e . m . ( B ) Median change in theta spike-LFP coherence ( blocks 2–4 vs block 1 ) calculated throughout the session using a sliding window of 64 trials in steps of 10 trials . The solid line represents the exponential fit . The error bars represent the distance between the first and third quartiles divided by the square root of n ( number of samples ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08417 . 007 Previous work in area V4 has shown that working memory influences theta power and the phase synchronization between spikes and LFPs ( Lee et al . , 2005 ) and theta coupling between areas V4 and prefrontal cortex . To test the possibility that learning might be accompanied by a change in spike-LFP coherence , particularly in the theta band , we calculated SFC in the delay period between the target and test by dividing the delay into three time windows ( early , middle , late ) , each of identical length to the stimulus period . However , we found that the median SFC in block 1 was not significantly different from that in blocks 2–4 in any of the windows of the delay period ( Figure 5 , p > 0 . 05 , Wilcoxon signed-rank test for each delay interval ) . This indicates that the improvement in behavioral performance during learning is unlikely to be explained by a change in the working memory load when novel images are presented during training . 10 . 7554/eLife . 08417 . 008Figure 5 . Changes in spike-LFP coherence ( SFC , spike-field coherence ) during the delay period . Median change in SFC ( blocks 2–4 vs block 1 ) during the delay period for each frequency band . The 1000-ms delay period was divided into three 333-ms windows referred as early , middle , and late . *** denotes p < 0 . 001; ** denotes p < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 08417 . 008 It is , in principle , possible that the low-firing neurons in our population may lead to measured SFC values biased toward high synchronization levels . To rule out this concern , for each neuronal pair , we examined the relationship between small changes in firing rate and theta SFC . However , we could not find a significant correlation between these two measures ( r = 0 . 0178 , p = 0 . 5711 , Pearson correlation ) . We subsequently computed the z-scores of SFC values in each block ( Jarvis and Mitra , 2001 ) by computing the distribution of coherence under null hypothesis given the number of degrees of freedom ( i . e . the number of spikes * number of tapers ) , then calculating how many standard deviations the observed coherence differs from zero . The z-score eliminated the bias due to low-firing rates and the small sample size . We found that even when we compared the z-scored coherence values , theta SFC was still significantly elevated after block 1 ( p < 0 . 05 , Wilcoxon signed-rank test ) , whereas the z-score coherence in the other frequency bands either showed no change ( in alpha and beta bands , p > 0 . 1 ) or decreased only slightly in the gamma band ( p < 0 . 01 ) . Altogether , these results further confirm that changes in theta SFC induced by learning are unrelated to changes in neuronal firing rates . The analysis in Figure 3 indicates that the increase in theta SFC is more prominent in block 2 when learning rate is the highest , followed by a decrease in blocks 3 and 4 when behavioral performance stabilizes ( post hoc test , p < 0 . 05 for the decrease in theta SFC in block 3 vs 2; p < 0 . 001 for block 4 vs 3 ) . This raises the possibility that the increase in spike-LFP theta coherence may only be required during the development of learning , but not after the behavioral threshold has reached the asymptote ( following block 2 ) . To test this possibility , we performed a new set of experiments in which the same image was presented in two consecutive sessions . Since after the first session the image became familiar ( behavioral performance was improved at the end of the session ) , this offered us the opportunity to examine the changes in theta spike-LFP coherence for novel and familiar images . Indeed , for familiar image sessions ( n = 23 ) , we found that , as expected , the orientation discrimination threshold was <5o in block 1 and performance did not improve further during the session ( p > 0 . 5 for each block comparison; ANOVA test , Figure 6A inset ) . Importantly , in contrast to novel images , the sessions in which we presented familiar images were not associated with significant changes in spike-LFP coherence in any frequency band ( Figure 6A; p > 0 . 05 , Kruskal–Wallis test , comparing SFC in block 1 vs blocks 2–4; n = 625 pairs ) . 10 . 7554/eLife . 08417 . 009Figure 6 . Relationship between spike-field theta synchronization and behavioral performance . ( A ) Relative change of SFC in different frequency bands during exposure to novel images ( left ) , during exposure to familiar images ( middle ) , and in the ‘no learning’ condition ( right ) . Error bars represents s . e . m . across spike-LFP pairs . The inset on top shows block-by-block behavioral performance in each type of session . ( B ) Correlation between the session-by-session change in monkey's orientation discrimination threshold and the mean change in theta synchronization in blocks 2–4 vs block 1 ( by averaging across the spike-LFP pairs in a given session ) . The dark line represents the linear regression fit . ( C ) Correlation coefficient between the session-by-session change in behavioral discrimination threshold and SFC in different frequency bands . Double asterisks show statistical significance ( p < 0 . 01 ) in theta band . DOI: http://dx . doi . org/10 . 7554/eLife . 08417 . 00910 . 7554/eLife . 08417 . 010Figure 6—figure supplement 1 . Control experiment— ( A ) Monkey 2 performed control experiments ( n = 12 sessions , 741 cell pairs ) in which novel natural scenes were flashed in the neurons' receptive fields , while the animal was engaged in a red/green color detection task in the contralateral hemifield . Briefly , after 500 ms of fixation , a 5-deg natural scene and a red square ( 3 deg in diameter ) were presented simultaneously for the same random duration ( 1000–1800 ms ) at two symmetric locations on the screen . The animal was required to signal the color change for the attended square ( from red to green ) within the next 3000 ms . Each session consisted of 400 trials . ( B ) We examined the block-by-block ( each block consisted of 100 trials ) changes in spike-LFP coherence ( SFC ) when the image was presented in the neurons' receptive fields ( during the first 1000-ms of stimulus presentation ) . The results ( shown as SFC changes with respect to SFC in block 1 ) confirm the results of our first control experiment ( Figure 6A ) , that is , we were unable to detect significant block-by-block changes in SFC in any frequency band during inattentive ( passive ) fixation ( theta: p = 0 . 99 , alpha: p = 0 . 21 , beta: p = 0 . 25 , gamma: p = 0 . 49 , Kruskal–Wallis test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08417 . 010 One important control is to ensure that the changes in theta SFC observed during rapid learning across blocks of trials are not due to the stimulus presentation itself . We thus collected data from one animal in which five control sessions were recorded before the monkey was able to accurately perform the behavioral task with the set of 10 prototype stimuli . However , despite the fact that animals performed a discrimination task identical to that described in Figure 1A and novel stimuli were presented in the same conditions , learning did not take place . Indeed , as shown in Figure 6A ( top ) , for the ‘no learning’ control sessions , behavioral discrimination threshold was high at the beginning of the session ( in block 1 ) and did not undergo statistically significant changes in subsequent blocks ( p > 0 . 1 , ANOVA test ) . Importantly , examining the changes in spike-LFP coherence for the population of 96 pairs , we found that SFC did not change across blocks of trials in any frequency band ( p > 0 . 1 , Kruskal–Wallis test , comparing SFC in block 1 vs blocks 2–4 ) . Similar results ( no statistically significant block-by-block changes in SFC in any frequency band , Figure 6—figure supplement 1 ) were found when natural images were flashed in the neurons' receptive fields during passive ( inattentive ) fixation experiments ( n = 12 sessions , 741 pairs ) . Altogether , these results indicate that the coordination of spike timing with the local theta LFP activity only occurs when animals improve their performance during learning , and that the increase in theta SFC does not continue with subsequent exposure to familiar stimuli after learning has stabilized , or in conditions in which learning does not take place . Finally , we examined whether the session-by-session increase in spike-LFP synchronization is related to the improvement in behavioral performance during learning . Thus , we measured the correlation between the average change in theta SFC across all the pairs recorded in a given session ( blocks 2–4 vs block 1 ) and the corresponding change in behavioral discrimination threshold ( Figure 6B ) . We found that the decrease in behavioral discrimination threshold is significantly correlated with the increase in theta SFC ( r = −0 . 49 , Pearson correlation , p < 0 . 01 ) . In contrast , despite slightly negative trends , there was no significant correlation between the change in SFC and behavioral performance in the higher frequency bands ( alpha: p = 0 . 42 , beta: 0 . 94 , and gamma: 0 . 17; Figure 6C ) .
We have demonstrated that the increase in behavioral discrimination performance during learning is associated with a tight coordination of spike timing with the local population activity . The spike-LFP synchronization associated with learning occurred in the theta band , while higher frequency synchronization was uncorrelated with changes in behavioral performance . Importantly , the changes in spike-LFP theta synchronization only characterized the learning of novel stimuli but were absent once stimuli became familiar or in situations when learning did not occur . These findings indicate that plasticity in visual cortex during the time course of learning is accompanied by elevated low-frequency synchronization between individual neuron responses and local population activity . Additionally , there was a significant increase in spike-LFP phase locking strength in the theta band during learning despite the fact that theta LFP power remained constant . Although our study reports neuronal changes associated with a rapid form of learning , it is possible that the improvement in behavioral performance occurring over larger time scales ( e . g . , weeks ) may be accompanied by similar changes in spike-LFP theta synchronization . Our results are consistent with the fact that learning induces rapid changes in the strength of synapses . Indeed , synaptic plasticity has been associated with coordinated action-potential timing across neuronal networks and oscillations of specific frequencies . In particular , neuronal oscillations in the theta frequency range ( 3–8 Hz ) have been associated with the induction of synaptic plasticity ( Buzsaki , 2002 ) . For instance , the timing between incoming stimuli and ongoing theta oscillations controls synaptic changes . In addition , oscillations in the theta range can be modulated by behavior and brain state ( Winson , 1978; Hasselmo et al . , 2002 ) and have been assigned a key role in the maintenance of information in short-term memory ( Lee et al . , 2005; Liebe et al . , 2012 ) . Although our study did not reveal an ongoing theta oscillation for the population activity , we demonstrate that the timing of neuronal spikes relative to the theta-filtered LFP activity during a rapid form of visual learning is correlated with the increase in behavioral performance . Our results are consistent with previous evidence in rabbit hippocampus ( Berry and Thompson , 1978 ) that animals that exhibit more theta activity show elevated classical conditioning of the nictitating membrane response . The increase in spike-LFP theta synchronization during learning is unlikely to be explained by changes in behavioral context , such as gradual changes in attention or alertness as the session progresses . Indeed , although attention and alertness have been previously associated with elevated neuronal firing in area V4 ( Fries et al . , 1997; McAdams and Maunsell , 1999; Fries et al . , 2001 ) , increased neuronal sensitivity ( McAdams and Maunsell , 1999 ) , increased gamma LFP power ( Fries et al . , 1997 , 2001 ) , and increased gamma spike-LFP synchrony ( Fries et al . , 1997; Jarvis and Mitra , 2001 ) , none of these measures changed significantly across blocks of trials ( p > 0 . 1 , for both animals ) . We also failed to find changes in eye movements during learning—there was no significant relationship between the horizontal/vertical saccade amplitude and frequency and block number ( p > 0 . 4 , Pearson correlation for each comparison ) . This indicates that the increase in theta coherence during learning is unlikely to have been contaminated by gradual changes in behavioral context or fixational eye movements during the time course of the session . In principle , it may be possible that the behavioral improvement in blocks 2 , 3 , and 4 could reflect a change in monkey's strategy to respond , for instance , by frequently holding the lever after block 1 until the end of the session , rather than learning to perform the image discrimination task . To rule out this possibility , we examined the block-by-block changes in behavioral performance in the match trials ( these trials require a bar release response ) . If monkey's strategy was to keep holding the lever as the session progressed , performance in the match trials ( randomly interleaved with the non-match trials ) would significantly deteriorate . However , we did not find statistically significant changes in match ( bar release ) responses across blocks of trials ( by analyzing all the sessions with an improvement in learning performance , p > 0 . 1; ANOVA test ) . Another possibility that could explain our behavioral results is that the poorer performance at the beginning of the session ( in block 1 ) may be due to animals being distracted by the novelty of stimulus presentation . To rule out this concern , we performed additional behavioral experiments ( Monkey 2 , n = 10 sessions ) in which the animal was passively exposed to a novel natural scene while he performed a color detection task ( red–green task ) in the contralateral hemifield ( Figure 1—figure supplement 1 ) . After 150 trials of passive ( inattentive ) exposure to the novel image , the monkey was switched to the image discrimination task ( Figure 1A ) . As expected , as the monkey did not actually practice the image orientation discrimination task for the stimuli he was exposed to , we found a gradual improvement in behavioral discrimination threshold during the session ( p < 0 . 01 , ANOVA test; each of blocks 2–4 was characterized by a lower threshold than block 1 , p < 0 . 05 , post hoc multiple comparisons ) . This control was repeated with images that were fixated ( without controlling attention ) , but the results were similar . Several possible mechanisms could modulate spike-timing accuracy in V4 . Indeed , it is possible that theta oscillations in areas outside V4 may modulate excitability in areas projecting to and receiving information from V4 . For instance , theta synchronization was associated with pattern recognition , working memory , sequence learning and navigation , most prominent in the temporal lobe ( von Stein et al . , 2000; Lengyel et al . , 2005; Rutishauser et al . , 2010; Hoerzer et al . , 2010; Benchenane et al . , 2010 ) . It also has been suggested that low-frequency synchronization is suitable for long-range or polysynaptic communication across distant brain areas ( von Stein et al . , 2000 ) . One possibility is that visual cortical neurons are coordinated with the theta rhythm in higher cortical areas , for example , activation of dopaminergic neurons of the ventral tegmental area or those in prefrontal cortex influences theta phase-locking of neurons ( Benchenane et al . , 2010 ) . Both of these types of neurons can be activated by novel stimuli , similar to those used in our experiments . In addition , theta synchronization within and between V4 and prefrontal cortex has been reported during the maintenance of visual short-term memory ( Raghavachari et al . , 2001; Liebe et al . , 2012 ) . The fact that we found a significant correlation between the accuracy of theta phase-locking during learning and behavioral performance suggests a relationship between synaptic plasticity in downstream areas and efficient information flow in visual cortex to facilitate learning .
All surgical procedures were overseen by UTHSC-H Animal Welfare Committee . A titanium head post was implanted in medial frontal region with the help of multiple anchor screws . Following a recovery period of about 10 days , monkeys were trained for 3–4 months on visual fixation and discrimination tasks . After the monkey learned the tasks , a recording chamber ( inner diameter of 19 mm ) for single unit multiple electrode recording was cemented over area V4 ( according to MRI map ) . A few stainless steel screws were inserted into the skull around the recording chamber and a thin stainless steel wire was wrapped around the screws for additional support . Two male rhesus monkeys ( Macaca mulatta ) were trained to perform a natural image orientation discrimination task . Stimuli were presented on a CRT color monitor ( Dell , Texas , United States , 60-Hz refresh rate , running MATLAB and using Psychophysics Toolbox ) , positioned 57 cm in the front of the animal . After monkeys triggered the trial by holding a bar , a small spot ( 0 . 1 deg ) was presented in the center of monitor . Monkeys were required to hold fixation within a 1-deg diameter window throughout stimulus presentation; the trial was automatically aborted if fixation instability ( microsaccade amplitude ) exceeded 0 . 25 deg at any time during stimulus presentation . All natural scenes were converted to equal-contrast gray scale circular images of 5 deg in diameter . After 500 ms of fixation , a target image was flashed for 300 ms . After a 1000-ms blank , a 300-ms test stimulus of random orientation ( rotated with respect to the target image by 0 , 3 , 5 , 10 , or 20° ) was flashed at the same visual location . The monkey was required to hold the bar for 1 . 5 s if the target and test stimuli were different and release the bar within 1 . 5 s if they are identical . The monkey was rewarded with 5 drops of juice for the correct choice . A full session consisted of 4 consecutive blocks of 96 trials each . Test image orientation was randomized across trials in each block ( each block was randomly composed by 48 match 12 non-match trials for each image orientation ) . Psychometric curves were obtained in each session . We averaged the percentage correct responses for each orientation difference between target and test . The probability of false alarms ( which is the value obtained for the 0o orientation difference ) represents the proportion of match trials in which the monkey's response was incorrect . The psychometric curves represent the best fit of the data using Weibull functions: P ( Δθ ) =1− ( 1−FA ) exp − ( Δθ /a ) b , where FA is the false alarm rate , and a and b are the offset and slope terms of the best Weibull fit . The threshold was computed as the orientation difference at which accuracy is 75% ( the threshold takes into account the false alarm rate ) . To rule out that the behavioral improvement in blocks 2 , 3 , and 4 could be due to a change in monkey's strategy to respond ( i . e . by repeatedly holding the bar ) , we examined the block-by-block changes in performance in the match trials ( these trials required a bar release response ) . If monkey's strategy was to keep holding the lever as the session progressed , performance in the match trials ( randomly interleaved with the non-match trials ) would significantly deteriorate . However , we did not find statistically significant changes in match ( bar release ) responses across blocks , by analyzing all the sessions in which found an improvement in learning performance ( p > 0 . 05; ANOVA test ) . Stimulus presentation and eye position monitoring was manipulated and synchronized with neuronal data using the ECM ( Experiment Control Module ) programmable device ( FHC Inc ) . Eye position was continuously monitored using an eye tracker system ( EyeLink II , SR Research Ltd . , Osgoode , ON , Canada ) that offers a binocular 1-kHz sampling rate . Eye position was calibrated before each experiment using a 5-point calibration procedure in which the animal was required to fixate on each one of 5 points ( 1 in the center , 2 in the vertical , and 2 in the horizontal axes or the diagonals ) in steps of 4 , 8 , and 12 deg from the central fixation spot . We analyzed the eye position on the x- and y-axis , as well as the number and speed of microsaccades . The eye-tracker gains were adjusted such as to be linear for the horizontal and vertical eye deflections . The fixation pattern was carefully analyzed offline . Microsaccades were analyzed every 10 ms by using a vector velocity threshold of 10 deg/s ( this corresponds to a 0 . 1 deg eye movement between consecutive 10-ms intervals ) . If a detected microsaccade exceeds 0 . 25 deg ( fixation instability ) , the trial is automatically aborted . We used two types of electrode systems in each monkey: ( i ) arrays of parylene-C-coated tungsten microelectrodes ( MPI , 1–2 MΩ at 1 KHz ) grouped in pairs and attached to several micro-drives ( Crist ) fixed on a grid , and penetrated transduraly through stainless steel guide tubes into the cortex; ( ii ) 16-channel U-probes ( Plexon ) with contacts spacing at 100 μm advanced using the NAN drive system ( Plexon ) attached to the recording chamber . In each session , we advanced up to 8 tungsten microelectrodes and/or 2 U-probes into area V4 . Real-time neuronal signals from multiple channels ( up to 32 , simultaneous 40 kHz A/D conversion on each channel ) were recorded and processed through Multichannel Acquisition Processor system ( MAP , Plexon Inc ) . The signals were first filtered by a preamplifier box into spike channels ( 150 Hz–8 kHz , 1 pole low-cut , 3 pole high-cut , with programmable referencing , 50× gain ) and field potential channels ( 0 . 07 , 0 . 7 , 3–170 , 300 , 500 Hz user selectable , 1 pole low-cut , 1 pole high-cut , 50× ) . Single-unit signals were further amplified , filtered , and viewed on an oscilloscope and heard through a speaker . The spike waveforms above threshold were saved and fine sorted after data acquisition was terminated using Plexon's offline sorter program . After a unit was isolated , its receptive field was mapped with dynamic gratings or using reverse correlation while the animal maintained fixation . As a measure of neuronal discrimination performance , we calculated the neurons' capacity to discriminate between each combination pair of test orientations ( d' , Green and Swets , 1966 ) , calculated by dividing the absolute difference between the neurons' mean firing rates by the rms standard deviation ) . Low-frequency field potential signals were amplified and digitized at 1 kHz . To correct for the time delays induced in the LFP signals by the filters in headstage and pre-amplification board , we used the software correction FPAlign provided by Plexon ( http://www . plexon . com/downloads . html ) . LFPs were further filtered between 0 . 5 Hz and 100 Hz using a fourth order Butterworth filter . In order to remove line artifacts , we applied a digital notch at 60 Hz ( fourth order elliptic filter , 0 . 1 db peak-to-peak ripples , 40 db stopband attenuation ) . All filtering was applied by using forward and backwards filtering to obtain zero phase shifts . We discarded all LFPs that had more than 3 points outside the mean ±4 standard deviations to avoid influence of irregular artifact noise from muscle activity or other sources . The amplitude of LFPs was measured in each trial by standard deviation , peak-valley amplitude , and average voltage for specific periods of interest . We estimated the LFP power density during the presentation of the stimuli using sliding windows of ±150-ms length in steps of 10 ms . In order to obtain optimal spectral concentration , we used the multitaper method by multiplying each data epoch with the tapers before Fourier transforming ( Mitra and Pesaran , 1999; Jarvis and Mitra , 2001; Pesaran et al . , 2002; Womelsdorf et al . , 2006 ) . The number of tapers was calculated according to the formula: K = 2 TW – 1 , where K is the highest number of tapers that can be used while preserving optimal time-frequency concentration of the data windowing available from the Slepian taper sequences , T is the length of the data in seconds , and W is the half-bandwidth of the multitaper filter . For our analysis , we applied spectral smoothing of TW = 4 , K = 7 tapers for frequencies greater than 30 Hz and a single Hanning taper for lower frequencies . The power spectral density was first normalized to the average power spectrum during the 300-ms fixation period before the presentation of the target stimulus ( averaged across all trials in a session ) . This helped balance the power spectrum between low and high frequencies within same amplitude range . The average power of theta , alpha , beta , and gamma frequency bands was calculated as the mean power at frequencies between 5 and ∼7 Hz , 8–13 Hz , 15–30 Hz , 35–80 Hz . To compare LFP power at individual channels across session , we further normalized the power in each block and frequency band to its mean value in the first block . To examine whether learning increases the coupling between the spike trains and LFPs , we calculated the STA and SFC . The coherence between two signals is a complex quantity whose magnitude is a measure of the phase synchrony for a given frequency . We computed STA by averaging the LFP signal within a window centered ±150 ms on each elicited spike . The 300-ms window enables us to measure the spectrum of the STA over entire theta frequency range . To quantify STA , we calculate its power spectrum ( i . e . the magnitude of all frequency components of the STA as a function of frequency ) . SFC was computed by dividing the power spectrum of the STA by the average of all power spectra of the LFP segments that were used to obtain the STA ( Fries et al . , 1997; Womelsdorf et al . , 2006 ) . Thus , SFC is independent of the firing rate of the single units and the power spectrum of the LFPs . SFC ranges from 0 , lack of synchronization , to 1 , perfect phase synchronization . We assessed the temporal SFC spectrum by using windows of ±150 ms that were moved over the data in 10-ms steps from 100 ms before the onset of the target stimulus to 100 ms after the disappearance of the test stimulus . To eliminate the onset transient response that could cause artifacts in the computation of SFC , we focused our analysis in the 150–350-ms window from the onset of each stimulus . To eliminate the bias caused by finite data set and eliminate the block-by-block bias in the calculation of SFC ( caused by the different number of spikes in each block ) , we randomly sampled the stimulus interval to extract the same number of spikes in each block , and then used these spikes to calculate the STA ( for each pair of electrodes ) . We also computed the z transformed SFC . It was defined by formula z = β* ( − ( V−2 ) *log ( 1−|C|2 ) −β ) , where β = 1 . 15 , C is the SFC , and V is number of degrees of freedom . Under null hypothesis , the z-transformed SFC value is distributed as a normal variate with variance equal to 1 . The z-scored SFC indicates how many standard deviations the observed SFC differs from zero . To validate our calculation of SFC , we used the Chronux function coherencycpt , which computes the multitaper SFC for a continuous signal ( LFP ) and point process data ( spike-train ) according to an optimal family of orthogonal tapers derived from Slepian functions . However , our results were identical when SFC was calculated using the STA method or the Chronux function coherencycpt . To calculate the alignment between spikes and LFPs , we computed the LFP phase in theta band for each spike . The LFP signal was filtered in the theta band by an equiripple FIR filter ( band edge 3 . 5–4 Hz , 8–8 . 5 Hz , attenuation 40 dB , 1 ripple ) . Similar filtering was applied for other frequency bands ( alpha 7–8 Hz , 14–15 Hz; beta 13–14 Hz , 30–31 . 5 Hz; gamma 28–30 Hz , 80–82 Hz ) . The filtered data were subsequently Hilbert transformed . The phase at each time was defined as the angle of the Hilbert complex . We analyzed the distribution of phase angles at each spike time within 150–350 ms after stimulus onset . Circular statistical analysis was performed using the Matlab CircStat toolbox . The significance of phase angle distribution non-uniformity was assessed by Rayleigh's and Omnibus tests at both individual pair and population level . Rayleigh's Z score is a measurement of the strength of non-uniform circular distribution . It was defined as the square of vector sum of all sample angles divided by the sample size . The difference in phase distribution was assessed for statistical significance by using the non-parametric circ_cmtest through all blocks and then Kuiper test between block pairs . To verify that the phase locking between spikes and LFPs in the theta band was not due to the stimulus itself , we calculated the theta phase of shuffled control . It was performed by shuffling the trials of LFP data to match original spike trains . Each trial was sampled once . We calculated the Rayleigh's Z value of shuffled spike-LFP phase . The median of 100 different reshuffle was considered as the shuffled Z value . | Throughout life , we learn and become better at many skills through repeated practice . However , how the brain cells enable us to adapt to changes in the environment and improve cognitive performance is poorly understood . The activity of a neuron can be recorded as a ‘spike’ of electrical activity . In the nervous system , neurons work together in networks . If a group of neurons fire in a synchronized manner , waves of activity may be recorded from that brain region . One important issue in neuroscience is whether the spikes of individual neurons are synchronized with the local network activity . Indeed , it is generally believed that it is functionally important for individual cells to synchronize their responses to the waves of population activity . The vast majority of studies aimed at understanding the behavior of neurons during learning have only recorded the activity of single neurons . This activity does not change much during learning , which suggests that learning may instead be encoded by the combined activity of a group of neurons . However , it is difficult to examine the same population of neurons as an animal practices and improves a skill . This is because the learning process typically takes longer than the length of time for which a single cell can be held in a stable condition and recorded from . To overcome these limitations , Wang and Dragoi briefly flashed images at monkeys and trained them to report when the images have been rotated . Monkeys learn to do this within a single-training session , which allows the responses of the same group of neurons—found in a part of the brain called the mid-level visual cortex—to be recorded throughout the learning process . Wang and Dragoi found that the improvement in behavioral performance during learning was accompanied by a tight synchronization between the spikes produced by individual neurons and the activity of groups of cells within a specific low-frequency band . This low-frequency activity had previously been linked to changes in the strength of functional connections between neurons in the hippocampus , which may be important for learning . The more synchronized this neural activity was , the better the monkeys were at the task . However , changes to the synchronization of spiking responses to local population activity in the higher frequency bands were unrelated to changes in performance . The changes to the level of synchronization were abolished once learning had stabilized and stimuli had become familiar . Although Wang and Dragoi have found that the mid-level visual cortex neurons fire in a more synchronized way throughout learning , it remains to be confirmed whether these changes in synchronization are causally related to learning . Future studies could test whether this is the case by electrically or optically stimulating neurons so that their activity synchronizes with the local population activity , and investigating whether this manipulation improves learning ability . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Rapid learning in visual cortical networks |
Many mediators and regulators of extravasation by bona fide human memory-phenotype T cells remain undefined . Mucosal-associated invariant T ( MAIT ) cells are innate-like , antibacterial cells that we found excelled at crossing inflamed endothelium . They displayed abundant selectin ligands , with high expression of FUT7 and ST3GAL4 , and expressed CCR6 , CCR5 , and CCR2 , which played non-redundant roles in trafficking on activated endothelial cells . MAIT cells selectively expressed CCAAT/enhancer-binding protein delta ( C/EBPδ ) . Knockdown of C/EBPδ diminished expression of FUT7 , ST3GAL4 and CCR6 , decreasing MAIT cell rolling and arrest , and consequently the cells’ ability to cross an endothelial monolayer in vitro and extravasate in mice . Nonetheless , knockdown of C/EBPδ did not affect CCR2 , which was important for the step of transendothelial migration . Thus , MAIT cells demonstrate a program for extravasastion that includes , in part , C/EBPδ and C/EBPδ-regulated genes , and that could be used to enhance , or targeted to inhibit T cell recruitment into inflamed tissue .
The phenotypes and functions of peripheral T cells are intimately linked to the cells’ localization and patterns of migration ( Masopust and Schenkel , 2013 ) . For memory-phenotype T cells , these relationships were the basis of the dichotomous schema describing central ( TCM ) and effector ( TEM ) memory cells ( Sallusto et al . , 1999 ) . The relationships among position , migration and function have also been emphasized in recent studies characterizing tissue resident memory T cells and recirculating memory T cells ( Bromley et al . , 2013; Fan and Rudensky , 2016; Gerlach et al . , 2016; Masopust and Picker , 2012; Masopust and Schenkel , 2013 ) . Elegant experiments in mice have shown that resident memory T cells can have critical roles in protection at barrier sites ( Jiang et al . , 2012 ) . Less well defined have been the cells within the memory-phenotype population that can be recruited from the blood to an inflamed site in the very early stages of an immune response ( Masopust and Picker , 2012 ) . Nonetheless , these ‘first responders’ can play essential roles in tissue defense ( Kohlmeier et al . , 2008; Maloy et al . , 2000; Wakim et al . , 2008 ) . The recruitment of leukocytes to tissue from blood has been described within the longstanding paradigm , derived primarily from studies of neutrophils , of a multistep process of the cells’ rolling , followed by arrest , crawling , and diapedesis ( Ley et al . , 2007 ) . An abundance of data support roles for selectins and their ligands in rolling , and for stimulated chemoattractant receptors and consequent integrin activation in firm arrest and extravasation ( Springer , 1994 ) . For trafficking of TEM cells to sites of inflammation , P- and E-selectins on endothelial cells and their glycosylated counter-ligands on T cells mediate rolling , which initiates the process ( Ley and Kansas , 2004; Sperandio et al . , 2009 ) . The best characterized ligands for P- and E-selectin are proteins , such as PSGL-1 ( mouse and human ) , CD43 ( mouse and human ) , ESL-1 ( mouse ) , and CD44 ( mouse ) that bear sugars containing sialyl Lewisx ( sLex ) , a tetrasaccharide consisting of N-acetylglucosamine , fucose , galactose , and sialic acid ( Mondal et al . , 2013; Phillips et al . , 1990 ) . The ability of leukocytes to display selectin ligands is determined primarily by the expression of glycosyltransferases responsible for synthesizing the appropriate glycoforms . These enzymes include core 2 β1 , 6-N-acetylglucosaminyltransferases , α1 , 3-fucosyltransferases , and α2 , 3-sialyltransferases ( Ley and Kansas , 2004; Sperandio et al . , 2009 ) . Although the necessary enzymes and a full complement of selectin ligands are expressed constitutively on myeloid cells , expression on lymphocyte populations is heterogeneous and can be affected by cellular activation and differentiation ( Austrup et al . , 1997; Ley and Kansas , 2004; Wagers et al . , 1998 ) . Studies of mouse CD4+ T cells and/or cell lines have shown high preferential expression of selectin ligands on Th1 versus Th2 cells ( Austrup et al . , 1997; Blander et al . , 1999 ) , and counter-regulation of the α1 , 3-fucosyltransferase gene Fut7 by T-bet and GATA-3 ( Chen et al . , 2006 ) ; and Fut7 can be induced in mouse CD4+ T cells in response to a number of cytokines , including IL-12 and TGF-β1 ( Ebel et al . , 2015; Ebel and Kansas , 2016 ) . Little else is known about the molecular mechanisms regulating the expression of these glycosyltransferases in T cells . Selectin-mediated rolling allows leukocytes to sample the endothelium for seven-transmembrane domain receptor agonists , principally chemokines , and for ligands , such as VCAM-1 , MAdCAM-1 , and ICAM-1 , for the α4β1 , α4β7 , and β2 integrins , respectively ( Springer , 1994 ) . Although signals induced by selectin ligands on neutrophils can yield an integrin conformation sufficient to support integrin-mediated rolling ( but not firm arrest ) , this is not observed for lymphocytes ( Alon and Ley , 2008 ) . Moreover , except for integrins on recently activated/effector T cells ( Shulman et al . , 2011 ) , integrin activation that is sufficient to induce firm arrest under flow requires chemoattractant receptor-transduced signals ( Alon and Ley , 2008 ) . Chemokine receptors not only induce integrin activation and leukocyte arrest , but also directly mediate transendothelial migration ( TEM ) ( Cinamon et al . , 2001; Shulman et al . , 2011 ) . Among the 19 G-protein-coupled chemokine receptors , only two , CXCR4 and CCR7 , are expressed on all naive T cells , whereas T cells with the effector/memory phenotype can express these and most of the remaining chemokine receptors , resulting in a high degree of combinatorial diversity ( Bachelerie et al . , 2014 ) . The expanded repertoire of chemokine receptors on these cells confers the potential to traffic to and within the wide range of inflammatory sites generated during host defense and injury . There is , however , relatively little understanding of how multiple chemokine receptors can cooperate to provide the functions required by specific T cell subsets , and how the expression of selectin ligands , chemokine receptors and integrins are co-regulated on memory-phenotype T cells in order to confer the ability to extravasate efficiently . Within the migratory T cell population , the initial cells to enter inflamed tissue should share a TEM phenotype , including not only MHC class I/II restricted cells , but also innate-like T cell such as blood-borne subsets of γ/δ T cells ( Hayday , 2000 ) , and mucosal-associated invariant T ( MAIT ) cells ( Gapin , 2014 ) . In our previous studies , we characterized the subset of human CD4+ T cells co-expressing the chemokine receptors CCR5 and CCR2 ( Zhang et al . , 2010 ) . These cells also express multiple inflammation-associated chemokine receptors and have features of a stable population of highly differentiated cells well equipped to serve as early responding TEM . In extending these observations to CD8+ T cells , as described below , we found that most human CD8α+CCR2+ T cells were MAIT cells . MAIT cells are innate-like T cells that express Vα7 . 2-Jα33 ( TRAV1-2-TRAJ33 according to the IMGT/GENE-DB nomenclature ( Giudicelli et al . , 2005 ) as part of a semi-invariant TCR ( Franciszkiewicz et al . , 2016; Porcelli et al . , 1993 ) , and recognize bacterial metabolites of riboflavin in the context of the non-polymorphic MR1 ( Kjer-Nielsen et al . , 2012 ) . Under homeostatic conditions , MAIT cells are found in the intestinal lamina propria and liver , and represent a significant percentage of CD8+ , memory-phenotype T cells in human blood . MAIT cells exhibit potent antibacterial activity and accumulate at sites of bacterial infections ( Gold et al . , 2010; Le Bourhis et al . , 2010; Meierovics et al . , 2013 ) . MAIT cells may also have roles in immune-mediated diseases ( Hinks , 2016 ) . Although based on their expression of chemokine receptors and other surface markers , MAIT cells have been characterized as tissue-homing ( Dusseaux et al . , 2011 ) , there are no detailed studies of their trafficking behavior . Human MAIT cells have generally been identified by their co-expression of TCRVα7 . 2 and the NK cell marker CD161 ( Franciszkiewicz et al . , 2016 ) . The discovery of the MR1-bound ligands for the MAIT cell TCRs has allowed for the development of MR1 tetramers ( Reantragoon et al . , 2013 ) . Although these tetramers represent a significant new tool for studying MAIT cells , recent data show that the tetramers of MR1 bound to riboflavin- and folate-derived ligands can also identify a heterogeneous collection of Vα7 . 2- non-MAIT cells , thereby defining a broader population including both MAIT cells and ‘atypical’ Vα7 . 2- MR1-restricted T cells ( Gherardin et al . , 2016; Meermeier et al . , 2016 ) . The vast majority of MAIT cells in human blood are CD8α+ ( many of which are also CD8βlow ) , and other than expression of CD8 , no differences have been noted between CD8+ and CD8- MAIT cells ( Franciszkiewicz et al . , 2016; Walker et al . , 2012 ) . For the sake of brevity , because the work described below deals only with the MAIT cells that are CD8α+ , we will use ‘MAIT’ in place of ‘CD8α+ MAIT’ . In our current work , we found that MAIT cells were highly efficient at extravasation across inflamed endothelium . We also discovered that MAIT cells selectively and highly express the bZIP transcription factor C/EBPδ . siRNA-mediated knockdown of C/EBPδ showed that C/EBPδ contributed to the expression of glycosyltransferases/selectin ligands as well as CCR6 on MAIT cells , and consequently was required for optimal rolling and arrest of these cells on activated endothelial cells . Although these effects led to a decrease in the number of MAIT cells crossing the endothelium , knocking down C/EBPδ had no separate effect on the final and critical step of transendothelial migration - which depended on CCR2 . Taken together , our data show that MAIT cells are efficient at trafficking across inflamed endothelium due to a coordinated program , regulated in part by C/EBPδ , that controls genes encoding proteins of disparate activities , each contributing to the migratory phenotype .
In expanding our previous studies of CCR2- and CCR5-expressing subsets of CD4+ T cells ( Zhang et al . , 2010 ) , we characterized CD8α+CCR2+ T cells from human peripheral blood . We found that approximately 80% of the CD8α+CCR2+ T cells expressed TCRVα7 . 2 and CD161 ( Figure 1A ) , which identify them as MAIT cells ( Martin et al . , 2009 ) , and approximately 75% of MAIT cells were CCR2+ . For additional analysis of chemokine receptor expression on MAIT cells , we divided the CD8α+ T cell population into naive and memory-phenotype cells , and further divided the memory-phenotype cells into CCR6- conventional ( non-MAIT ) cells , CCR6+ conventional ( non-MAIT ) cells , CCR2-/low MAIT cells , and CCR2+ MAIT cells ( see Figure 2—figure supplement 1A , below ) . In addition to CCR2 , we found that MAIT cells also prominently expressed CXCR4 , CXCR6 , CCR5 and CCR6 , but lacked CCR7 ( Figure 1B ) , consistent with published data ( Dusseaux et al . , 2011 ) . CCR6 and CD161 are co-expressed on Th17 cells ( Cosmi et al . , 2008 ) , and we found that CCR6 and CD161 marked virtually identical cells within the CD8α+TCRVα7 . 2+ subset , so that CCR6 and CD161 could be used interchangeably for identifying MAIT cells ( Figure 1C ) . Given the pattern of chemokine receptor expression on MAIT cells , our earlier data on CD4+CCR5+CCR2+ T cells as potential ‘first responders’ ( Zhang et al . , 2010 ) , and data from others identifying CCR5 and CCR2 as important receptors for TEM on effector/activated T cells ( Shulman et al . , 2011 ) , we considered whether MAIT cells - and in particular their CCR2+ subset - might exhibit efficient extravasation in the context of inflammation . We began the analysis using flow chamber assays with human umbilical vein endothelial cell ( HUVEC ) monolayers pre-treated overnight with TNFα . For studying T-cell-endothelial cell interactions , we purified subsets of CD8α+ T cells from peripheral blood using cell sorting , divided into MAIT cell and conventional cell subsets as described above and shown in Figure 2—figure supplement 1A . The staining with the anti-CCR2 antibody characteristically showed a continuum between CCR2- and CCR2+ MAIT cells ( Figure 1A and Figure 2—figure supplement 1A ) , and the limited number of CCR2- MAIT cells often required using a relaxed gate for obtaining sufficient numbers of these cells for trafficking experiments . Consequently , the ‘CCR2-’ MAIT cells typically contained CCR2low cells , and we have designated this subset as CCR2-/low MAIT cells accordingly . Even when using a severe gate for obtaining CCR2- and CCR2+ MAIT cells for other studies , such as analyzing gene expression , expression of CCR2 mRNA was found in the CCR2- MAIT cells ( see Figure 7I below ) , supporting the use of the CCR2-/low designation . We separated the CCR6+ from the CCR6- conventional memory-phenotype cells for purposes of comparison with MAIT cells , since all MAIT cells are CCR6+ ( Figure 1C ) . It is notable that the CCR2+ MAIT cells expressed the highest levels of CCR6 among the memory-phenotype subsets divided in this way ( Figure 2—figure supplement 1B ) . T cell subsets were introduced into the flow chambers at 0 . 75 dyn/cm2 for 4 min after which shear stress was increased to 5 dyn/cm2 and data on cell numbers were collected for 16 min . Naive CD8α+ T cells did not adhere to the HUVECs , and among the memory-phenotype cells there was , generally , a pattern of progressive increase in numbers of cells rolling , arrested , and transmigrating going from CCR6- conventional to CCR6+ conventional to CCR2-/low MAIT to CCR2+ MAIT cells ( Figure 2A ) . Although in our assays we detected TEM using differential interference contrast ( DIC ) microscopy , we were able to confirm that migration under the endothelial cell monolayer was occurring by using confocal microscopy with CFSE-stained CCR2+ MAIT cells and cell tracker Red CMTPX-stained HUVECs ( Figure 2—figure supplement 2A ) . Because rolling , arrest , and TEM are sequential , where arrest is predicated on rolling and TEM predicated on arrest , we calculated numbers of cells showing arrest and TEM as percentages of cells rolling or arresting , respectively . The data showed that the MAIT cells were most efficient at the step of arresting , and that the CCR2+ MAIT cells were particularly effective in the final step of TEM ( Figure 2B ) . δIn addition , among those cells that transmigrated , the cells within the CCR2+ MAIT subset took the shortest time between arrest and completion of transmigration . As an example , as was seen for cells from one donor , two out of three CCR2+ MAIT cells transmigrated in 23 s , and 1 min 55 s after arrest , whereas the CCR6+ conventional and CCR2-/low MAIT cells took 3 min 15 s and 2 min 42 s , respectively ( Figure 2—figure supplement 2B ) . The rapid transmigration of the CCR2+ MAIT cells can be seen in Video 1 , in which the white arrowheads mark cells at the initiation of TEM , and pooled data from five donors demonstrate significant differences among the times between arrest and TEM for the transmigrating cells from the CCR6+ conventional cells , CCR2-/low MAIT cells and CCR2+ MAIT cells ( Figure 2—figure supplement 2C ) . Overall , these patterns suggested coordinated and co-regulated capabilities in rolling , firm arrest , and TEM among the memory-phenotype subsets , with the CCR2+ MAIT cells best equipped for extravasation . In order to assess the ability of these subsets to traffic in vivo , we injected purified , CFSE-labeled cells into the left hearts of mice and , using flow cytometry , analyzed labeled cells remaining in inflamed ( and non-inflamed ) ears within a few minutes after injection . In these experiments , and some additional experiments that follow , the limitations in cell numbers discussed above prevented us from evaluating the CCR2-/low MAIT subset . We detected retention of CCR2+ MAIT cells , but not cells from the other subsets , in ears that had had prior intradermal injection of TNFα and IL-1β ( Figure 2C ) . We failed to detect any of these cells in non-inflamed ears . Confocal microscopy of ears after immuno-staining showed that labeled cells had extravasated into tissue ( Figure 2D and Video 2 ) . In order to understand the basis for the differences in rolling frequencies among the T-cell subsets , we analyzed adhesion molecules on the TNFα-treated HUVECs and the T cells . On the HUVECs , we detected up-regulation of E- , but not P-selectin , consistent with published data ( Yao et al . , 1996 ) , and up-regulation of both ICAM-1 and VCAM-1 ( Figure 3A ) . Staining the T-cell subsets using E-selectin-Fc and P-selectin-Fc chimeric proteins showed a progressive increase in staining from naive to CCR6− conventional to CCR6+ conventional to CCR2-/low MAIT to CCR2+ MAIT cells ( Figure 3B ) . A similar pattern was seen in staining the cells for sLex ( Figure 3C ) . Consistent with the critical role for sLex in mediating rolling , we showed that treating cells with an exo-sialidase , which eliminated staining for sLex ( Figure 3—figure supplement 1A ) , abolished rolling on TNFα-treated HUVECs ( Figure 3D ) . The differences in levels of selectin ligands among the T cell subsets could be due to differences in expression of the proteins that bear the relevant sugars and/or the degree of the appropriate glycosylation . Levels of surface PSGL-1 and CD43 could not explain the pattern of selectin ligand expression ( Figure 3—figure supplement 1B ) . We also checked expression of CD44 , even though CD44 has only been shown to be a selectin ligand on mouse neutrophils ( Hidalgo et al . , 2007; Mondal et al . , 2013 ) . CD44 could also not account for the pattern of selectin ligand expression , although CD44 was found at somewhat higher levels on the CCR6+ subsets as compared with the naive and CCR6- conventional cells . In general , expression of selectin ligands reflects regulation of the genes encoding the limiting glycosyltransferases ( Ley and Kansas , 2004 ) . We found that patterns of expression of GCNT1 , encoding core 2 β1 , 6-N-acetylglucosaminyltransferase-I , FUT4 , FUT6 , FUT9 , ST3GAL1 , 2 , 3 , and six were unremarkable ( Figure 3E and Figure 3—figure supplement 1C ) . However , expression of FUT7 and ST3GAL4 were highest in the MAIT cells , and particularly in the CCR2+ MAIT cell subset ( Figure 3E ) , consistent with the MAIT cells’ high expression of selectin ligands and the important roles for FUT7 and ST3GAL4 in the synthesis of selectin ligands in leukocytes ( Ellies et al . , 2002; Malý et al . , 1996; Mondal et al . , 2015 ) . Because leukocyte arrest on endothelium typically requires integrin activation in response to signals from chemoattractant receptors , we investigated the contributions of these two classes of proteins to the differences in the behaviors of the T cell subsets . As can be inferred from the expression patterns of integrin subunits ( Figure 4—figure supplement 1 ) , there were no differences in expression of LFA-1 , VLA-4 , and α4β7 that could explain the differences in efficiencies of arrest among the subsets . To determine the overall role of chemokine receptors , which couple to Gi/o G proteins , we analyzed T cells with and without pretreatment with pertussis toxin . Pertussis toxin had no effect on rolling , reduced the numbers of arrested memory-phenotype cells by approximately 50% , and eliminated TEM ( Figure 4A ) . In order to focus on the chemokine receptors important for arrest and TEM of MAIT cells in the flow chambers , we analyzed the TNFα-treated endothelial cells for expression of the mRNAs for the chemokine ligands of the receptors highly expressed on the MAIT cells . TNFα induced high expression of the genes for the CCR2 ligand CCL2 , the CCR5 ligand CCL5 , and the CCR6 ligand CCL20 , whereas CXCL16 was expressed at only a low level that was not augmented by TNFα ( Figure 4B ) . We tested a role for CCR6 by using antibody to CCL20 . Anti-CCL20 antibody significantly reduced numbers of cells arresting in the three CCR6+ subsets . Similarly , reflecting the effects specifically on the step of arresting , anti-CCL20 significantly reduced the percentage of rolling cells that arrested in two of the three CCR6+ subsets , but not in the CCR6- cells ( Figure 4C ) . Although convincing , the effects of anti-CCL20 were not large . It is notable in this regard that given the results using pertussis toxin , a 50% reduction in arresting cells is the maximum decrease that could have been expected by blocking a Gαi/o-coupled receptor such as CCR6 . Our data did show statistically significant differences between control- and anti-CCL20-treated CCR2+ MAIT cells in numbers of cells rolling and percentages of arrested cells that underwent TEM . Because similar results were not found for the other two CCR6-expressing subsets ( CCR6+ conventional and CCR2-/low MAIT cells ) , we concluded that CCR6 lacked a convincing activity in rolling or TEM of these cells . Taken together , the data suggest a role for CCR6 in arrest of the CCR6-expressing CD8α+ memory-phenotype T cells , and little or no effects on the steps of rolling and TEM . We addressed the role of CCR2 in TEM by using the CCR2 antagonist , BMS CCR2 22 . BMS CCR2 22 did not diminish the numbers of cells rolling or arrested - the lack of effect on firm arrest ruling out a general inhibition of chemokine receptor signaling . However , blocking CCR2 had a profound effect on TEM of the CCR2+ MAIT cells ( Figure 5A ) . The CCR2 inhibitor also significantly diminished TEM of the CCR6+ conventional cells , consistent with the expression of CCR2 on a subset of these cells ( Figure 2—figure supplement 1A ) . The studies using the CCR2 inhibitor were among those experiments in which we did not include CCR2-/low MAIT cells due to the low numbers of these cells that we could obtain from an individual donor . It is notable , however , that significant numbers of cells within the CCR2-/low MAIT cell samples were able to undergo TEM ( Figure 2 ) . We presume that CCR2 expressed on the CCR2low cells within these samples was responsible for this TEM . Given the high expression of CCR5 by MAIT cells ( Figure 1B ) and expression of CCL5 by the TNFα-activated endothelial cells , we also investigated a role for CCR5 using the CCR5 antagonist , maraviroc . Blocking CCR5 had no effect on the numbers of cells rolling , arrested , or transmigrated ( Figure 5B ) . However , among the transmigrating CCR2+ MAIT cells , blocking CCR5 significantly prolonged the time between when the cells initiated crawling and initiated TEM ( Figure 5C ) . There was no effect on the time interval between arrest and initiating crawling . These data on the activity of CCR5 may partly explain the differences in time between arrest and TEM that we measured for the CCR6+ conventional cells , CCR2-/low MAIT cells and the CCR2+ MAIT cells ( Figure 2—figure supplement 2C ) , because we found differences in the levels of CCR5 expression in the order CCR2+ MAIT cells > CCR2-/low MAIT cells > the CCR2+ subset of CCR6+ conventional cells ( Figure 5—figure supplement 1 ) . For measuring levels of CCR5 on the surface of the CCR6+ conventional cells , we limited the analysis to cells that were CCR2+ because the data in Figure 5A indicated that the CCR2+ cells were those undergoing TEM , and therefore were the cells scored in Figure 2—figure supplement 2C . For the CCR2-/low MAIT cells , in addition to having levels of CCR5 that were lower than on the CCR2+ MAIT cells , the reduced expression of CCR2 presumably also contributed to their delay in initiating TEM . Having determined some of the factors contributing to the trafficking behavior of MAIT cells , we next investigated how expression of these factors might be regulated . In previous , unpublished experiments we had characterized gene expression in the CCR5+CCR2+ subset of CD4+ memory-phenotype T cells that we had studied earlier and discovered that these cells expressed high levels of CEBPD , which encodes C/EBPδ . We found that MAIT cells also expressed high levels of CEBPD mRNA and C/EBPδ , and that these could be knocked down using CEBPD siRNAs ( Figure 6A and B ) . We found no selective expression in MAIT cells of the related genes CEBPA , CEBPB , CEBPE , CEBPG , and CEBPZ ( Figure 6—figure supplement 1 ) . siRNA-mediated knockdown of CEBPD diminished the overall trafficking of MAIT , but not non-MAIT cell subsets in the flow chamber assays ( Figure 6C ) . Knockdown of C/EBPδ had a significant effect on the rolling step ( Figure 6C ) and a modest , separate effect on firm arrest of the MAIT cells ( percentage of rolling cells undergoing arrest ) , although the effect on the CCR2+ MAIT cell subset did not reach statistical significance ( Figure 6D ) . Importantly , although knocking down C/EBPδ diminished the numbers of MAIT cells crossing the activated endothelial cells through effects on the initial trafficking steps , knocking down C/EBPδ had no consistent , independent effect on TEM ( percentage of arrested cells undergoing TEM ) in the MAIT cell subsets ( Figure 6D ) . To evaluate a role for C/EBPδ in trafficking in vivo , we co-injected differentially labeled CCR2+ MAIT cells that had been transfected with control or CEBPD siRNA into mice whose ears had been injected with TNFα and IL-1β . As compared to controls , significantly fewer cells with knockdown of C/EBPδ could be recovered from the inflamed ears ( Figure 6E ) . In order to understand the basis of the effects of knocking down C/EBPδ on rolling and arrest , we investigated if knockdown of C/EBPδ affected expression of sLeX/glycosyltransferases and chemokine receptors . Knockdown of C/EBPδ decreased surface levels of sLex and decreased expression of FUT7 and ST3GAL4 in MAIT , but not in conventional cells ( Figure 7A– C ) . Knockdown of C/EBPδ did not decrease expression of GCNT1 ( Figure 7—figure supplement 1 ) . Analysis of the 5’ flanking regions of FUT7 and ST3GAL4 ( Figure 7—figure supplement 2 ) , identified potential binding sites for C/EBPδ , and chromatin immunoprecipitation ( ChIP ) detected binding of C/EBPδ within the 5’ flanking regions of these genes , specifically in the MAIT cells ( Figure 7D ) . In analyzing the effects of knocking down C/EBPδ on chemokine receptors , we found that C/EBPδ supported the expression of surface CCR6 and CCR6 mRNA ( Figure 7E–G ) . The 5’ flanking region of CCR6 has multiple predicted binding sites for C/EBPδ ( Figure 7—figure supplement 2 ) , and just as for FUT7 and ST3GAL4 , ChIP showed binding of C/EBPδ to CCR6 ( Figure 7H ) . Again , the effect of knocking down C/EBPδ on the expression of CCR6 was limited to the MAIT cells , concordant with the pattern of C/EBPδ expression . Based on the absence of any effect of knocking down C/EBPδ on MAIT cell TEM , we anticipated that knocking down C/EBPδ would not affect expression of CCR2 . As shown in Figure 7I , knocking down C/EBPδ did not have a notable effect on the expression of either CCR2 or CCR5 in the MAIT cells .
Our data demonstrate heterogeneous , graded expression of the components mediating extravasation into peripheral tissue within populations of human T cells . The abilities to roll , arrest and migrate across an inflamed endothelial cell layer are not all or none and reflect both qualitative and quantitative differences in expression of the essential components , with naïve and MAIT cells occupying extreme positions along a spectrum . We found that the efficiency of the initial steps of rolling in the flow chamber assays were a function of the levels of surface selectin ligands and sLex , which in turn correlated with expression of FUT7 and ST3GAL4 . FucT-VII is the fucosyltransferase that is critical for synthesis of selectin ligands ( Knibbs et al . , 1996; Malý et al . , 1996; Smithson et al . , 2001 ) . In particular , the level of FucT-VII is the determining factor in synthesis of E-selectin ligands ( Knibbs et al . , 1996; Ley and Kansas , 2004; Wagers et al . , 1996 ) . The mouse and/or human genes for FucT-VII are induced in proliferating T cells ( Blander et al . , 1999; Knibbs et al . , 1996 ) , up-regulated by IL-12 ( Ebel et al . , 2015; Wagers et al . , 1998; White et al . , 2001 ) , and suppressed by IL-4 ( Wagers et al . , 1998 ) , related to the activities of T-bet and GATA-3 ( Chen et al . , 2006 ) . Among sialyltransferases , ST3Gal-IV ( Ellies et al . , 2002; Sperandio et al . , 2006 ) , together with ST3Gal-VI ( Yang et al . , 2012 ) , are the critical enzymes for synthesizing selectin ligands in mice , whereas ST3Gal-IV is the dominant sialyltransferase in synthesizing selectin ligands on myeloid cells in humans ( Mondal et al . , 2015 ) . Like the mouse Fut7 and/or human FUT7 , St3gal4 is induced by T cell activation ( Blander et al . , 1999 ) . However , little is known about how the mouse or human gene for ST3Gal-IV is regulated . In concert with the progressive increase that we found among the memory-phenotype T cell subsets in levels of selectin ligands and rolling , we found enhanced abilities for firm arrest that could not be explained by differences in integrin expression . Antibody neutralization of the CCR6 ligand , CCL20 , resulted in decreased numbers of cells arresting in flow chambers . Our findings are consistent with reports describing the ability of CCR6 to mediate arrest of lymphocytes , including , very recently , MAIT cells on adhesion-molecule coated plates and/or activated endothelial cells ( Alcaide et al . , 2012; Campbell et al . , 1998; Fitzhugh et al . , 2000; Ghannam et al . , 2011; Kim et al . , 2017 ) . Of particular interest , the other chemokine receptors on the CD8α+ T cells could not substitute for CCR6 in maintaining optimal firm arrest , nor was CCR6/CCL20 necessary for TEM of those cells showing CCR6-independent arrest . On the CD8α+ T cells , therefore , CCR6 has a particular role in triggering firm arrest . CCR6 is notable as the chemokine receptor that is expressed on all T cells that can make IL-17 ( Acosta-Rodriguez et al . , 2007; Singh et al . , 2008 ) , and CCR6 has been shown to be important for Th17 cell arrest on ICAM-1 ( Alcaide et al . , 2012 ) . In addition to the Th17 cell regulator RORγt , only STAT5A ( Tsuruyama et al . , 2016 ) and , from our own work , PLZF ( Singh et al . , 2015 ) have been described as controlling CCR6 . Based on our experiments using siRNA knockdown and ChIP , we have now identified C/EBPδ as an additional , direct activator of CCR6 . The pertussis-toxin resistant arrest that we observed in the memory-phenotype T cell subsets could be viewed as a ‘baseline’ activity on which inducible integrin activation can be superimposed . Although we are not aware of pertussis-toxin resistant arrest being reported previously for resting memory-phenotype cells , this phenomenon has been described for human effector T cells produced by activation ex vivo ( Shulman et al . , 2011 ) . For those cells , unlike for the CCR6+ CD8α+ memory-phenotype T cells , there was no pertussis toxin sensitive component to the firm arrest , and their behavior was ascribed to high levels of integrin expression together with constitutive activity of PLC-γ1 ( Shulman et al . , 2011 ) . Non-integrin mediated adhesion is another possibility ( Schneider-Hohendorf et al . , 2014 ) . We have not investigated the mechanism responsible for the pertussis toxin resistant component of arrest by these resting cells , but our data suggest that in this regard the memory-phenotype CD8α+ T cells occupy a position intermediate between naïve and activated effector cells . We found that blocking CCR5 did not affect the T cells’ arrest , nor did it prevent cells from undergoing TEM over the 20 min of observation . However , the data indicated that CCR5 functioned to shorten the time between the initiation of crawling and TEM , which , in vivo , would be presumed to speed extravasation . Consistent with our observations , CCR5 has been found not to contribute to leukocyte adhesion , either in flow chambers or in blood vessels ( Diacovo et al . , 2005; Shulman et al . , 2011; Weber et al . , 2001 ) , but nonetheless to have roles in completing the process of extravasation ( Diacovo et al . , 2005 ) . For CCR2 , our data showed a critical and specific role in TEM . We often observed significant TEM not only in the CCR2+ MAIT cells , but also in the CCR2-/low MAIT cells . We presume that the CCR2low MAIT cells within the CCR2-/low MAIT cell samples accounted for the ability of cells in these samples to perform TEM . Because the number of CCR2-/low MAIT cells that we obtained from an individual donor was often inadequate for the flow chamber studies , thereby limiting the number of studies that we could do with these cells , we have not used the CCR2 antagonist to test the possibility that CCR2 was mediating TEM in the CCR2-/low cells . Although CCR2 has been studied extensively in monocyte biology , relatively little is known about the role of CCR2 on T cells . It is of interest that on monocytes , CCR2 and CCL2 were also described as important for TEM ( Weber et al . , 1999 ) , but not for firm adhesion ( Huo et al . , 2001; Weber et al . , 1999 ) . Of particular relevance for our studies , CCR2 was reported to be essential for TEM across HUVECs and human dermal microvascular endothelial cells in experiments using a mixed population of activated T cells ( Shulman et al . , 2011 ) . We have previously shown that CCR2 is expressed on a subset of human CCR5+CD4+ T cells that have features of a stable population of highly differentiated long-term memory cells which , based on chemokine receptor expression , TCR activation threshold , and effector cytokine production , are ideally equipped for mediating rapid recall responses in tissue ( Zhang et al . , 2010 ) . In addition , CD4+CCR5+CCR2+ T cells are found in cerebrospinal fluid associated with episodes of relapse in multiple sclerosis , and these cells demonstrate an enhanced ability to migrate across a model of the blood-brain barrier ( Sato et al . , 2012 ) . Recent data in mice have suggested that CCR2 is important for the trafficking into the CNS of a pathological subset of Th17 cells that contributes to chronic and relapsing EAE ( Kara et al . , 2015 ) . Taken together , the data suggest that - unlike for some other chemokine receptors that are associated with individual Th cell lineages - up-regulation of CCR2 ( and CCR5 ) is part of a program directed specifically at conferring the capacity for migration of T cells into tissue . Given the high degree of ligand/receptor promiscuity , the co-expression of many chemokines during inflammation , the co-expression of multiple chemokine receptors on individual cells , and the shared pathways for chemokine receptor signaling , studies of the chemokine system have often confronted questions of functional redundancy . Our data suggest that CCR6 , CCR5 , and CCR2 serve sequential and distinct functions in MAIT cell extravasation . From our experiments using anti-CCL20 antibodies , we can conclude that CCL20 is displayed on the surface of the TNFα-treated endothelial cells , and from the work of Shulman et al . ( 2011 ) , we know that CCL2 is sequestered in endothelial cell vesicles and only available to CCR2 within T-cell-endothelial cell synapses . Taken together , these observations suggest that the separate roles for the receptors could be the result of anatomic segregation of their chemokine ligands , which might be a general feature of the chemokine system that limits functional redundancy . Our interest in understanding the relationships between the expression of chemokine receptors such as CCR6 and CCR2 and the biology of T cell subsets led us to these studies of MAIT cells and their trafficking behavior . Our and/or others’ data ( Dusseaux et al . , 2011; Kim et al . , 2017 ) showed that the MAIT cells were at the high end of a continuum among CD8α+ T cells as regards expression of selectin ligands and/or multiple tissue-homing chemokine receptors , and at the low end of a continuum as regards expression of CCR7 and CD62L , which are required for entering non-inflamed lymph nodes . The MAIT cells’ behavior on activated endothelial cells in the flow chamber assays and in vivo were fully consistent with an enhanced ability to enter inflamed tissue rapidly , without additional phenotypic changes and/or activation within lymphoid organs . MAIT cells are able to make effector cytokines , and they contain cytotoxic molecules such as perforin , granulysin and granzymes ( Dusseaux et al . , 2011; Franciszkiewicz et al . , 2016; Le Bourhis et al . , 2013 ) . MAIT cells are able to produce effector cytokines both in response to TCR activation and directly in response to cytokines such as IL-18 and IL-12 ( Franciszkiewicz et al . , 2016; Jo et al . , 2014; Slichter et al . , 2016; Ussher et al . , 2014 ) . These properties would allow MAIT cells to function locally within the earliest stages of antibacterial , and perhaps antiviral ( Loh et al . , 2016; van Wilgenburg et al . , 2016 ) defense . In support of such a role , MAIT cells are decreased in blood of patients with bacterial pneumonias , and can be identified in M . tuberculosis-infected lungs ( Le Bourhis et al . , 2010 ) . We found that in MAIT cells , both FUT7 and ST3GAL4 , as well as CCR6 , are regulated in part by C/EBPδ , and our ChIP data suggest that FUT7 , ST3GAL4 , and CCR6 are direct targets of C/EBPδ . Knockdown of C/EBPδ resulted in decreased surface expression of both sLex and CCR6 , although the resulting effects on numbers of rolling cells mediated by selectin ligands was more pronounced than on the specific step of firm arrest mediated in part by CCR6 . Since the multiple steps of extravasation occur sequentially and interdependently , C/EBPδ’s regulation of rolling and firm arrest significantly affected numbers of cells migrating across the activated endothelial cells in the flow chamber assays . We presume that the effects that we documented in the flow chamber assays were the basis for the decreased entry of MAIT cells transfected with CEBPD siRNA into inflamed ears . We have summarized our findings for CCR2-expressing MAIT cells in the cartoon shown in Video 3 . Knockdown of C/EBPδ had no independent effect on the final step of MAIT cell TEM in the flow chamber assays , nor , consistent with this finding , on expression of CCR2 ( or CCR5 ) . Nonetheless , it is of interest that Yamamoto et al . described C/EBP binding sites 3’ to the CCR2 transcriptional start site that were important for CCR2 promoter activity and that could bind C/EBP proteins , including C/EBPδ , found in nuclear extracts of a human monocytic cell line ( Yamamoto et al . , 1999 ) . These data suggest that our negative results notwithstanding , C/EBPδ might , in fact , have a role in regulating CCR2 expression in MAIT cells . Regardless , consideration of these data raises the important point that experiments using siRNA knockdown will typically underestimate roles of target genes given that mRNA and protein knockdown are incomplete both at the level of individual cells and across the heterogeneous population of siRNA-transfected cells . Consequently , our findings using siRNA knockdown of C/EBPδ reflect the lower limit of the activities of C/EBPδ in supporting expression of the glycosyltransferases and CCR6 , and thereby MAIT cell trafficking . We attributed the effect of knocking down C/EBPδ on the step of firm arrest to the decrease in expression of CCR6 . Nonetheless , it is of interest that studies in mice have shown that ST3Gal-IV sialylates CXCR2 ( Frommhold et al . , 2008 ) , and probably CCR1 , and CCR5 ( although not CCR2 ) ( Döring et al . , 2014 ) . For these receptors , sialylation is important for chemokine-dependent activation . There is no evidence for sialylation of CCR6 , but it remains possible that in addition to a decrease in expression of CCR6 , knockdown of C/EBPδ could have led to diminished CCR6 function through loss of ST3Gal-IV-dependent sialylation . A role for glycosylation in CCR6 function warrants further study . C/EBPδ is one of six members of the C/EBP family of bZIP transcription factors , each of which contains a basic DNA-binding domain and a ‘leucine zipper’ dimerization domain ( Ramji and Foka , 2002 ) . In mice , Cebpd is induced in many tissues in response to endotoxin or other inflammatory stimuli , including cytokines such as IL-6 and IL-1 , although expression in these contexts is typically transient ( Alam et al . , 1992; Balamurugan and Sterneck , 2013; Ko et al . , 2015 ) . Studies in Cebpd knockout mice have demonstrated pleiotropic roles for C/EBPδ , including in multiple models of inflammation and infection ( Yan et al . , 2013; Chang et al . , 2012; Duitman et al . , 2012; Litvak et al . , 2009 ) . C/EBPδ has been implicated in the acute phase response ( Alam et al . , 1992; Juan et al . , 1993; Ray and Ray , 1994 ) and in the phenotype and function of macrophages ( Balamurugan et al . , 2013; Chang et al . , 2012; Duitman et al . , 2012; Litvak et al . , 2009; Maitra et al . , 2011; Yan et al . , 2013; Litvak et al . , 2009; Maitra et al . , 2011; Balamurugan et al . , 2013 ) . Of possible relevance to our own work , C/EBPδ has been suggested to enhance the migration of macrophages into infected lung ( Duitman et al . , 2012 ) . A broader role for C/EBPδ in regulating leukocyte migration in inflammation is suggested by data on the C/EBPδ-mediated induction of chemokines , including the ligands for CCR6 ( Chang et al . , 2012 ) and CCR2 ( Ko et al . , 2014 ) , and in the transcriptional response of endothelial cells to inflammatory stimuli ( Hogan et al . , 2017 ) . In T cells , expression of CEBPD was detected in analyzing the transcriptome of human CD8+CD161+ T cells ( Billerbeck et al . , 2010 ) , which would have included MAIT cells , and a bioinformatic analysis of the transcriptome and epigenetic modifications of human CD4+ T cells suggested that C/EBPδ might function as one of the ‘master regulators’ of TEM differentiation ( Durek et al . , 2016 ) . However , as far as we are aware , functional studies of C/EBPδ in T cells have not been previously reported . Another transcription factor that has been described with an effect that is analogous but functionally inverse to C/EBPδ is Kruppel-like factor 2 ( KLF2 ) , which supports T cell trafficking to lymph nodes versus peripheral tissue by enhancing expression of CD62L and CCR7 , and suppressing CXCR3 ( Carlson et al . , 2006; Preston et al . , 2013 ) . Similarly , SOCS1-mediated inhibition of STATs favors T cell trafficking to lymphoid organs versus peripheral tissue by enhancing expression of CCR7 and suppressing expression of CCR6 and CXCR3 ( Yu et al . , 2008 ) . Based on our data , in contrast to these other factors , C/EBPδ is a positive regulator supporting the tissue-migrating phenotype . Our study has some clear limitations that affect the generalizability of our conclusions . We studied only human MAIT cells from blood . Our ex vivo experiments used only a single type of endothelial cell ( HUVEC ) treated with a single pro-inflammatory activator ( TNFα ) , and our in vivo experiments were limited to a simple and artificial model of inflammation at a single tissue site . It is possible that other endothelial cells and/or other activators would differentially affect the abilities of MAIT and non-MAIT cells to arrest and undergo TEM , and thereby reveal activities of other molecules , such as additional transcription factors and chemokine receptors , in these processes . Studies extended to more complex animal models would provide information on the performance of MAIT cells and the molecular mechanisms underlying their trafficking behavior in more biologically relevant models of inflammation . It will be of considerable interest to identify additional factors that control the program for extravasation and the interactions among overlapping networks specifying this activity in coordination with the fates and functions of effector-capable T cells . An understanding of the molecular species regulating the integration of migratory activity and other effector functions of such cells would suggest ways of enhancing or inhibiting T-cell mediated processes in peripheral tissues .
HUVECs were cultured according to the supplier’s instructions ( Promocell , Germany ) . Human CD8+ T cells were isolated from elutriated lymphocytes from healthy donors obtained by the Department of Transfusion Medicine , Clinical Center , National Institutes of Health , Bethesda , MD , under a protocol approved by the Institutional Review Board . Informed consent was obtained after explanation of the risks . For use in flow chamber experiments , isolated human CD8+ T cells were washed and kept overnight in RPMI 1640 ( Life Technologies , Waltham , MA ) containing 10% FBS ( Gemini Bio-Products , West Sacramento , CA ) , 2 mM L-glutamine , and penicillin-streptomycin ( Life Technologies ) at 37°C in 5% CO2 . C57BL/6J WT mice were purchased from The Jackson Laboratory ( Bar Harbor , ME ) . All mice were used at 8–12 weeks of age . Mice were housed under specific pathogen-free conditions at the National Institutes of Health in an American Association for the Accreditation of Laboratory Animal Care-approved facility . Animal study protocols were approved by the Animal Care and Use Committee , NIAID , NIH . All antibodies were against human antigens . Anti–CCR2-biotin ( clone 48607 ) , anti–PSGL-1-allophycocyanin ( 688101 ) , human E-selctin Fc chimera and human P-selectin Fc chimera were purchased from R and D Systems , Minneapolis , MN . Anti–CCR5-FITC ( 2D7 ) , anti–CCR5-PE-Cy5 , anti–CCR4-Alexa Fluor 647 ( 1G1 ) , anti–CCR7-FITC ( 3D12 ) , anti–CCR9-Alexa Fluor 488 ( 112509 ) , anti–CCR10-PE ( 1B5 ) , anti–CXCR1-FITC ( 5A12 ) , anti–CXCR2-allophycocyanin ( 6C6 ) , anti–CXCR3-allophycocyanin ( 1C6 ) , anti-CXCR4-allophycocyanin ( 12G5 ) , anti–CXCR5-Alexa Fluor 647 ( RF8B2 ) , anti–CD8-APC-Cy7 ( SK1 ) , anti–CD8-Alexa Fluor 700 ( RPA-T8 ) , anti–CD62L-PE-Cy5 ( DREG-56 ) , anti–CD62L-FITC , anti–CD45RO-PE-Cy5 ( UCHL1 ) , anti–CD45RO-Brilliant Violet 605 , anti–CD45RO-PE-Cy7 , anti–CCR6-PE-Cy7 ( 11A9 ) , anti–CCR6-allophycocyanin , non-conjugated anti-sLeX ( CSLEX1 ) , anti-CD161-allophycocyanin ( DX12 ) , and PE-conjugated streptavidin were purchased from BD Biosciences , Franklin Lakes , NJ . Anti-integrin α4-FITC ( 58XB4 ) , anti-integrin β1- Alexa Fluor 647 ( TS2/16 ) , anti-integrin β2-allophycocyanin ( TS1/18 ) , anti-integrin β7-allophycocyanin ( FIB504 ) , anti-TCRVα7 . 2-FITC ( 3C10 ) , anti–CXCR6-PE ( K041E5 ) , anti-CX3CR1-FITC ( 2A9-1 ) , anti–CD43-allophycocyanin ( 10G7 ) , anti–CD44-allophycocyanin ( BJ18 ) and anti-human IgG Fc-biotin were purchased from BioLegend , San Diego , CA . For phenotypic analysis of leukocyte subsets , cells were stained in whole blood , in preparations of PBMCs isolated from blood using Ficoll/Hypaque ( Amersham Biosciences , United Kingdom ) , or in preparations of CD8+ T cells purified from elutriated lymphocytes by negative selection using RosetteSep ( StemCell Technologies , Canada ) . For whole blood samples , red cells were removed using Pharm Lyse ( BD Biosciences ) according to the manufacturer’s protocol . For staining other samples , 1 × 105 cells were suspended in 100 μl of Hanks Balanced Salt Solution ( HBSS , Mediatech , Corning , NY ) containing 2% FBS . For each sample , cells were incubated with 1 μg of each fluorescent-conjugated primary antibody for 15 min at room temperature ( RT ) , and washed with HBSS/FBS . For cells stained with anti–CCR2-biotin , the cells were incubated with PE-conjugated streptavidin for an additional 15 min at RT . For staining sLeX , the cells were incubated with allophycocyanin-conjugated anti mouse IgM ( II/41 ) ( BD Biosciences ) for an additional 15 min at RT . For staining with E-selectin Fc chimera or P-selectin Fc chimera , the chimeric proteins were first incubated with anti-human IgG Fc-biotin in 100 μl binding buffer ( HBSS +5 mM calcium chloride +2 mg/ml BSA ) on ice for 10 min before cells were added and incubated on ice for 30 min . After washing with chilled binding buffer , the cells were incubated with streptavidin-PE on ice for 10 min before being washed and analyzed . Staining data were collected on an LSR II cytometer ( BD Biosciences ) . To set gates for defining positive and negative cells in multicolor staining , samples were stained with a mixture of all antibodies save one . Flow cytometry data were analyzed using FlowJo ( Ashland , OR ) . Approximately 1 . 5 × 108 CD8+ T cells were isolated from elutriated lymphocytes to approximately 90% purity by negative selection using RosetteSep human CD8+ T cell enrichment cocktail ( StemCell Technologies ) and incubated with anti–CCR2-biotin and anti–CCR6-PE-Cy7 in HBSS plus 4% FBS for 15 min at RT . Following washing , the cells were stained with streptavidin-PE , anti–CD8-Alexa 700 , anti-TCRVα7 . 2-FITC , anti-CD62L-PE-Cy5 , and anti–CD45RO-Brilliant Violet 605 for an additional 15 min at RT . The cells were washed and re-suspended in HBSS plus 4% FBS , and cell subsets were isolated to nearly 100% purity using an Aria cytometer ( BD Biosciences ) . HUVECs were plated at confluence on μ-Slide I 0 . 4 Luer parallel plate flow chambers ( Ibidi , LLC , Germany ) which were coated with 50 μg/ml fibronectin ( R and D Systems ) in PBS , and were stimulated for 18–20 hr with human recombinant TNFα ( 40 ng/ml ( R and D Systems ) . HUVEC-coated parallel plate flow chambers were assembled with a two-pump system ( Harvard Apparatus , Holliston , MA ) . Sorted T cells were re-suspended at 4 × 105 cells/ml in perfusion medium ( RPMI 1640 medium containing 2% FBS and 10 mM HEPES ) . Perfusion of T cells into the flow chambers was performed at 37°C under a force of 0 . 75 dyn/cm2 for 4 min to allow accumulation of T cells , followed by a constant shear stress of 5 dyn/cm2 for 16 min . Images were acquired at a rate of four frames per second with an integrated fluorescence microscope , Leica AF 6000LX ( Leica Microsystems Inc . ) with a 20 x DIC objective . For analysis of cell migration , we used Imaris software ( Bitplane , South Windsor , CT ) to track and categorized cells . We categorized arresting cells as cells that remained stopped on the HUVEC monolayer for more than 10 s under a sheer stress of 5 dyn/cm2; rolling cells as cells that rolled before arresting ( whether that arrest had initiated at 0 . 75 dyn/cm2 or at 5 dyn/cm2 ) , and cells that rolled under a sheer stress of 5 dyn/cm2 but then detached; and transmigrating cells as cells that underwent stepwise darkening under a sheer stress of 5 dyn/cm2 . Inflammation was induced in the ears of C57BL/6 mice by intradermal injection of murine TNFα ( 10 μg ( R and D Systems ) and IL-1β ( 1 μg ( R and D Systems ) in 20 μl PBS . Eighteen hours after injection , 1 × 106 human T cells were injected into the left ventricle , and mice were euthanized 8 min later . Ears were removed , ear sheets were split and cartilage and fat were scraped off . The sheets were then immediately fixed in cold acetone for 20 min for tissue staining or treated in DMEM ( Invitrogen , Carlsbad , CA ) containing 1 mg/ml DNase I ( Sigma-Aldrich St . Louis , MO ) and 250 μg/ml Liberase TM ( Roche Custombiotech , Indianapolis , IN ) for 50 min at 37°C to obtain cell suspensions . Cells were then filtered through a 70 μm nylon mesh and washed prior to counting using the flow cytometer . In some cases , the T cells were labeled with either CMTPX ( Life Technologies ) or CFSE ( Life Technologies ) prior to injection . Staining for human CD3 , and murine CD31 was done using acetone-fixed ear skin sheets . Skin sheets were blocked for 2 hr at RT with Fc-blocker ( BD Biosciences ) in PBS containing 4% bovine serum albumin ( Sigma-Aldrich ) . After washing in PBS , skin sheets were incubated with anti-mouse CD31-Alexa Fluor 647 and anti-human CD3-Brilliant Violet 605 antibodies ( BioLegend ) overnight at 4°C . After washing in PBS , the sections were incubated with DAPI nuclear stain ( Invitrogen ) . Images were acquired using the Carl Zeiss LSM510/Axio Observer confocal microscope and LSM510 version 4 . 2 software . For removing sialic acid residues , cell-sorted subsets of CD8+ T cells were treated with 0 . 1 units of sialidase ( from Vibrio cholera , Sigma-Aldrich ) in 1 ml RPMI 1640 ( Life Technologies ) containing 10% FBS ( Gemini Bio-Products ) and 2 mM L-glutamine for 2 . 5 hr at 37°C . For inhibiting Gi/o proteins , CD8+ T cells were pre-incubated with pertussis toxin ( 1 μg/ml ( R and D Systems ) in RPMI 1640 medium containing 10% FBS and 10 mM HEPES for 3 hr at 37°C . For blocking CCR2 and CCR5 , pre-incubation was with BMS CCR2 22 ( 2 μM ( Tocris , Minneapolis , MN ) or Maraviroc ( 10 μM ( Tocris ) , respectively , for 30 min at 37°C and inhibitors were left in the medium throughout the assay . For neutralizing CCL20 , HUVEC monolayers in flow chambers were pre-treated for 2 hr at 37°C with 20 μg/ml anti-human CCL20/MIP-3α antibody ( c67310; R and D Systems ) , and antibody was maintained at 10 ng/ml throughout the assay . Subsets of CD8+ T cells were purified by cell sorting as described above . Total cellular RNA was isolated using the TRIzol reagent ( Invitrogen ) . Real-time RT-PCR was performed with 20 ng of RNA as a template , using the qScript cDNA SuperMix for reverse transcription and PerfeCTa qPCR FastMix , UNG , and ROX for PCR ( Quanta Biosciences , Beverly , MA ) . Primer and probe sets ( FAM/VIC-labeled ) were purchased from Applied Biosystems , Foster City , CA . Results were normalized based on the values for GAPDH , detected using TaqMan GAPDH control reagents ( Applied Biosystems ) . Real-time qPCR analysis was performed on samples in duplicate using an ABI 7700 Sequence Detection System ( Applied Biosystems ) . For some assays , for cells from each donor , relative levels of expression , based on values for 2−ΔCT , are shown after normalization to the single highest value , which was set to 100 . For other assays , values of 2−ΔCT are shown without additional normalization . Transfections with Amaxa Human T Cell Nucleofector Kit ( Lonza , Walkersville , MD ) and siRNAs ( Dharmacon , Lafayette , CO ) were performed following the manufacturers’ protocols ( Lonza ) . Five nmol of SMARTpool , SMARTpool control , or individual siRNAs was reconstituted in 250 μl of siRNA buffer , and a total of 10–20 μl was added to a cuvette containing 2–4 × 106 purified human CD8+ T cells in 100 μl transfection reaction buffer ( prepared from Nucleofector Solution and Supplement ) . The cuvette with cell/siRNA suspension was inserted into the Nucleofector Cuvette Holder and subjected to Nucleofector Program V-024 . Five hundred μl of pre-equilibrated culture medium ( RPMI 1640 containing 10% FBS , 2 mM L-glutamine , and penicillin-streptomycin ) was added to the cuvette , the sample was transferred to the well of 12-well plate , and culture medium was added to a total volume of 2 ml/well . Cells were incubated for 3–4 days at 37°C in 5% CO2 . For the in vitro experiments , three experiments used the CEBPD SMARTpool containing four siRNAs ( catalogue number L-010453 ) and a SMARTpool siRNA control ( catalogue number D-001810-01-05 ) , two experiments used a single CEBPD siRNA ( catalogue number D-010453–01 ) that was not part of the CEBPD SMARTpool and a control siRNA ( catalogue number D-001210–01 ) , and one experiment used this same single CEBPD siRNA in a pool with two additional CEBPD siRNAs ( catalogue numbers D-010453–02 and D-010453–03 ) not found in the CEBPD SMARTpool and a control siRNA ( catalogue number D-001210–01 ) . For the experiments using human cells injected into mice , cells were transfected with CEBPD SMARTpool containing four siRNAs and a SMARTpool siRNA control . Subsets of CD8+T cells were purified by cell sorting as described above and lysed on ice in buffer ( 30 mM Tris HCl , pH 8 . 0; 75 mM NaCl; 10% glycerol; and 1% Triton X-100 ) containing 1:100 proteinase inhibitor cocktail ( Cell Signaling , Danvers , MA ) . Cellular lysates were centrifuged at 12 , 000 x g for 10 min at 4°C , and supernatants were collected after centrifugation . Protein content was quantified using the Micro BCA protein assay ( Pierce , Rockford , lL ) according to the manufacturer’s guidelines with BSA as a standard . Samples were prepared for SDS-PAGE by boiling at 100°C with 2 x Laemmli sample buffer ( Bio-Rad , Hercules , CA ) plus 5% β-Mercaptoenthol . A total of 40 μg of cellular proteins and an aliquot of the PageRuler Plus Prestained Protein Ladder ( Thermo Scientific , Waltham , MA ) were separted by SDS-PAGE in Any kD Mini-PROTEAN TGX Gel ( Bio-Rad ) at 100 V . After electrophoresis , protein was transferred to an Immun-Blot polyvinylidene difluoride membrane ( Bio-Rad ) over 1 hr at RT , using a Mini Trans-Blot Cell ( Bio-Rad ) . Following transfer , the membrane was washed in Tris-buffered saline ( 20 mM Tris , 136 mM NaCl , PH 7 . 4 ) with 0 . 1% Tween 20 ( TBST ) , blocked for 1 hr in TBST/5% nonfat dried milk , incubated overnight at 4°C in the same solution containing 1:10 , 000 dilution of mouse anti-human C/EBPδ , washed with TBST and incubated at RT with 1:10 , 000 dilution of HRP-conjugated sheep anti-mouse antibody ( GE Healthcare , United Kingdom ) in TBST/5% nonfat dried milk for 1 hr , and then washed with TBST at RT . Protein bands were visualized using SuperSignal West Pico Chemiluminescent Substrate ( Pierce ) . Anti-human C/EBPδ was a kind gift from Esta Sterneck , NCI , NIH . Quantification of bands was done using Adobe Photoshop ( San Jose , CA ) and the ImageJ ( NIH ) program as described on the following web site: https://www . lukemiller . org/journal/2007/08/quantifying-Western-blots-without . html ChIP experiments were performed using Magna ChIP A/G kit from Millipore , Burlington , MA , according to the manufacturer’s instruction . Sorted cells were subjected to protein-DNA crosslinking with 1% formaldehyde for 10 min at RT and the reaction was terminated by addition of glycine solution to a final concentration of 125 mM . Cells were re-suspended in cell lysis buffer containing protease inhibitor cocktail . Samples were centrifuged at 2000 rpm for 5 min at 4°C , and the cell pellet was re-suspended in nuclear lysis buffer containing protease inhibitor cocktail . Chromatin was sheared by Bioruptor sonicator ( Diagenode , Denville , NJ ) to generate DNA fragments between 200 and 1000 base pairs , centrifuged at 13 , 000 rpm for 10 min at 4°C to pellet debris and diluted 10-fold in ChIP dilution buffer containing protease inhibitor cocktail . After removing 1% of the sample for analyzing input DNA , the sheared chromatin was incubated at 4°C overnight under rotation with protein A/G magnetic beads and rabbit polyclonal anti-C/EBPδ ( Santa Cruz Biotechnology , Dallas , TX ) or normal rabbit IgG as a negative control ( Millipore ) , or mouse monoclonal anti-C/EBPδ ( C6 , Santa Cruz Biotechnology ) or normal mouse IgG as a negative control ( Millipore ) . The immunoprecipitates were washed sequentially for 5 min with low-salt immune complex wash buffer , high salt immune complex wash buffer , LiCl immune complex wash buffer and TE buffer . The protein A/G magnetic beads/antibody/chromatin complex was re-suspended in 100 μl of ChIP elution buffer containing 1 μl proteinase K . The crosslinking between DNA and proteins was reversed by incubating the sample at 62°C for 2 hr followed by 95°C for 10 min . DNA was purified by spin column . To analyze promoter regions of CCR6 , FUT7 and ST3GAL4 , we used primers at 1 kb intervals as noted in the legend for Figure 7 . Quantitative real-time PCR was performed on an Applied Biosystems 7900HT system to determine the relative abundance of target DNA using RT2 SYBR Green/ROX qPCR master mix according to the manufacturer’s instructions ( SA Biosciences , Frederick , MD ) . A ChIP PCR primer from IGX1A targeting open-reading-frame-free intergenic DNA ( SA Biosciences ) was used as a negative control . The percent input enrichment was calculated using ChIP PCR array data analysis software ( SA Biosciences ) . The positions of possible transcription factor binding sites ( TFBS ) in relevant genes were identified in a 20 , 000 bp region immediately upstream of the transcription start sites ( TSS ) using the Genomatix Genome Analyzer’s Common TF software ( Ann Arbor , MI ) with default settings . The input sequences were manually obtained from the February 2009 human reference sequence ( GRCh37/hg19 ) in FASTA format . An optimized CEBPD-binding weight matrix was constructed based on known binding sites . Sample sizes were chosen based on pilot experiments , and took into consideration the nature of the measurements being made , the magnitudes of differences among groups being compared , and the degrees of donor-to-donor variability . No explicit power analyses were used . In some experiments a minimum of six experiments were performed in order to satisfy the requirements of the Wilcoxon signed rank test . Sample identities were not routinely masked during data acquisition or analysis . Most experimental results were replicated by two investigators . In some cases , only data from one investigator are shown . Otherwise , experiments were excluded only in cases of technical problems that made the results uninformative . We did not define or exclude outliers . Numbers of experimental replicates are noted in the figure legends , and refer to independent experiments and not technical replicates . Given the low abundance of some of the T cells subsets that were studied , most experiments contained a single measurement made for a single sample from a single donor . All statistical tests were performed on directed pairwise comparisons . The tests included two-tailed t tests that were either paired , ratio paired , or unpaired , and the Wilcoxon signed rank test . SEMs are shown for the raw data that were analyzed using the t tests . For some experiments that included control and treated samples , the paired t test was used in place of the ratio paired t test due to the presence of values equal to 0 . Tests used for each data set are noted in the figure legends and the source files . No corrections were made for multiple comparisons . Significance is displayed as *p<0 . 05; **p<0 . 01; ***p<0 . 001 . Statistical analyses were done using Prism ( GraphPad , La Jolla , CA ) . | Lymphocytes are a type of cell found in the blood that can detect and fight infections: in particular , some of them can leave the bloodstream to enter infected or inflamed tissues . To do so , these lymphocytes use proteins on their surface to roll along the inside wall of the blood vessel; then they stick to this wall and finally they pass through it . For some types of lymphocytes the details of this mechanism – such as precisely which surface proteins are necessary – remain unclear . Here , Lee et al . collect human lymphocytes from the blood of healthy donors , and they identify a subgroup of lymphocytes , called MAIT cells , that are particularly good at moving from blood to infected or inflamed tissues , and further experiments reveal the types of surface proteins that help them do so . Some of these proteins , for example selectin ligands , are important so the MAIT cell can roll on the wall of the blood vessel . Others , like CCR6 , are essential for the cell to stop rolling and stick to the wall . Lee et al . also identify C/EBPδ , a regulatory protein inside the MAIT cell that controls how these other two types of proteins are produced . Finally , Lee et al . show that additional proteins , such as CCR2 , are necessary for the lymphocyte to cross the vessel wall . The proteins that help lymphocytes move from blood to tissues represent important targets to fight diseases . For example , blocking these proteins could prevent lymphocytes from invading and damaging healthy tissues , which happens in autoimmune diseases like multiple sclerosis . Alternatively , manipulating these proteins could help to engineer lymphocytes that can invade and kill tumor tissues in cancers . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"immunology",
"and",
"inflammation"
] | 2018 | C/EBPδ drives interactions between human MAIT cells and endothelial cells that are important for extravasation |
The high levels of serine ( S ) and threonine ( T ) residues within the Presenilin 1 ( PS1 ) N-terminus and in the large hydrophilic loop region suggest that the enzymatic function of PS1/γ-secretase can be modulated by its ‘phosphorylated’ and ‘dephosphorylated’ states . However , the functional outcome of PS1 phosphorylation and its significance for Alzheimer’s disease ( AD ) pathogenesis is poorly understood . Here , comprehensive analysis using FRET-based imaging reveals that activity-driven and Protein Kinase A-mediated PS1 phosphorylation at three domains ( domain 1: T74 , domain 2: S310 and S313 , domain 3: S365 , S366 , and S367 ) , with S367 being critical , is responsible for the PS1 pathogenic ‘closed’ conformation , and resulting increase in the Aβ42/40 ratio . Moreover , we have established novel imaging assays for monitoring PS1 conformation in vivo , and report that PS1 phosphorylation induces the pathogenic conformational shift in the living mouse brain . These phosphorylation sites represent potential new targets for AD treatment .
Senile plaques , comprised primarily of β-amyloid ( Aβ ) , are the major pathological hallmark of Alzheimer’s disease ( AD ) . Presenilin 1 ( PS1 ) is the catalytic component of γ-secretase ( Wolfe et al . , 1999; Esler et al . , 2000; Li et al . , 2000 ) , which is responsible for the final enzymatic cleavage of amyloid precursor protein ( APP ) to generate Aβ ( De Strooper et al . , 1998 ) . Over 180 familial AD mutations have been identified in the PS1 gene , and the majority of them lead to an increase in the Aβ42/40 ratio ( De Strooper , 2007 ) . PS1 knock-in mice carrying familial AD mutations exhibit decreased Aβ40 and Aβ42 levels but increased Aβ42/40 ratio and accelerated Aβ deposition ( Xia et al . , 2015 ) . This supports the idea that up-regulation of the Aβ42/40 ratio rather than the total amount of Aβ has a strong impact on Aβ deposition and AD pathogenesis . Using Förster resonance energy transfer ( FRET ) -based imaging techniques we have previously shown that familial AD mutations in PS1 increase the proximity of PS1 N-terminus ( NT ) to C-terminus ( CT ) or to the large cytoplasmic loop domain , causing the so called ‘closed’ pathogenic PS1 conformation ( Berezovska et al . , 2005; Uemura et al . , 2009 ) . On the other hand , several nonsteroidal anti-inflammatory drugs and PS1/γ-secretase modulators , which are known to decrease the Aβ42/40 ratio ( Weggen et al . , 2001; Page et al . , 2008 ) , drive PS1 into the ‘open’ conformation ( Lleó et al . , 2004; Uemura et al . , 2009; Ohki et al . , 2011 ) . These reports indicate that PS1 conformational changes are tightly linked to changes of the Aβ42/40 ratio , suggesting that modulation of the pathogenic ‘closed’ PS1 conformation is a potential target for AD treatment . A number of studies highlight the reciprocal relationship between elevation in intracellular Ca2+ levels and Aβ pathology . Notably , Aβ deposition is observed in brain regions exhibiting a high basal rate of neuronal activity in humans ( Sperling et al . , 2009 ) . The idea that Aβ causes Ca2+ elevation is supported by several studies ( Kuchibhotla et al . , 2008; Lopez et al . , 2008; Busche et al . , 2012; Higuchi et al . , 2012 ) . On the other hand , aberrant Ca2+ homeostasis affects Aβ production . For example , acute stimulation of the entorhinal cortex using an optogenetic approach increases the Aβ42 level specifically in the projection area of the perforant pathway ( Yamamoto et al . , 2015 ) . KCl-induced depolarization causes sustained release of Aβ42 from AD mouse model synaptoneurosomes ( Kim et al . , 2010 ) . KCl-triggered Ca2+ influx elevates intracellular Aβ42/40 ratio in primary neurons ( Pierrot et al . , 2004 ) . We recently reported that increases in Ca2+ levels induce the PS1 pathogenic ‘closed’ conformation in primary neurons , followed by an increase in the Aβ42/40 ratio ( Kuzuya et al . , 2016 ) . Furthermore , PS1 is found in the ‘closed’ conformation in sporadic AD brain , and this pathogenic PS1 conformational shift correlates with the proximity to deposited Aβ ( Wahlster et al . , 2013 ) . In this report , we address the detailed molecular mechanisms of the Ca2+-driven pathogenic ‘closed’ conformation of the PS1/γ-secretase . Phosphorylation , one of the crucial post-translational modifications , is a rapid and dynamic event in cells , and represents a common mechanism of regulating protein conformation and/or activity . In PS1 , fifteen phosphorylation sites have been identified ( Kirschenbaum et al . , 2001; Lau et al . , 2002; Fluhrer et al . , 2004; Kuo et al . , 2008; Ryu et al . , 2010; Matz et al . , 2015 ) . Although the phosphorylation of some of these sites affects the stability of PS1/γ-secretase ( Lau et al . , 2002; Kuo et al . , 2008; Ryu et al . , 2010 ) , the impact of phosphorylation on PS1 conformation and function has not yet been tested . Here , using FRET-based imaging we show that phosphorylation-inhibited mutations at the three domains ( domain 1: T74 , domain 2: S310 , S313 , and domain 3: S365 , S366 and S367 ) prevents the Ca2+-mediated PS1 ‘closed’ conformational change . Conversely , phosphorylation-mimicking mutations reveal that domain 1 and 2 phosphorylation is necessary , but domain 3 , and S367 in particular , is crucial for causing the PS1 pathogenic ‘closed’ conformation . Moreover , Protein Kinase A ( PKA ) inhibitors prevent the Ca2+-induced PS1 ‘closed’ conformational change whereas PKA activators trigger it . Therefore , we conclude that the Ca2+-increased PKA activity leads to phosphorylation of PS1 at the three domains , inducing the PS1 pathogenic ‘closed’ conformation . These results provide strong evidence that PS1 phosphorylation at certain residues can potentially be molecular targets for AD prevention .
Using the conformation sensitive FRET probe , GFP-PS1-RFP ( G-PS1-R ) ( Uemura et al . , 2009 ) , and the real-time FRET assay , we have previously shown that PS1 conformation is dynamically regulated by intracellular Ca2+ . Specifically , we found that KCl or glutamate-induced increases of intracellular Ca2+ levels led to PS1 adopting a pathogenic ‘closed’ conformation ( Kuzuya et al . , 2016 ) . Now , by employing an antibody based FRET assay , we detect changes in the conformation of endogenous PS1 in primary neurons . We show that treatment with KCl significantly increased the proximity between fluorescently labeled PS1 N-terminus ( NT ) and PS1 C-terminus ( CT ) ( increased red pixels mean higher FRET efficiency , Figure 1A ) , indicating that KCl treatment triggers the ‘closed’ conformation of endogenous PS1 . 10 . 7554/eLife . 19720 . 003Figure 1 . Ca2+ influx-triggered PS1 conformational change and increased phosphorylation . ( A ) FLIM analysis of PS1 conformation . Pseudo-colored FLIM images of primary neurons treated with vehicle control or 50 mM KCl for 5 min . The neurons were stained with the antibodies to PS1 NT ( Alexa488 ) and PS1 CT ( Cy3 ) . The colorimetric scale shows Alexa488 lifetime in picoseconds ( ps ) . A scale bar indicates 10 µm . The graph shows quantitative analysis of the FRET efficiency between fluorescently labeled PS1 NT and PS1 CT in neuronal processes ( total of 81–103 processes from 32–38 cells ) . Mean ± SEM , ***p<0 . 001 , Student’s t-test . ( B ) Immunoprecipitation/Western blot analysis of PS1 phosphorylation . Primary neurons treated with vehicle control or 50 mM KCl for 5 min were immunoprecipitated with mouse and rabbit anti-phosphoserine antibody mixture , followed by immunoblotting with the anti-PS1 loop antibody ( Upper panel , top row ) . Normal mouse and rabbit IgG mixture was used as negative control . Lower gel shows the total level of PS1 CTF in neuronal cell lysates . The graph presents a quantitative analysis of the band intensity for phosphorylated PS1 ( n = 6 ) . The relative PS1 phosphorylation level in KCl-treated neurons was normalized to that in vehicle-treated cells . Mean ± SEM , *p<0 . 05 , One sample t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 00310 . 7554/eLife . 19720 . 004Figure 1—figure supplement 1 . Glutamate treatment induces PS1 phosphorylation . Primary neurons treated with 5 mM glutamate or with vehicle control for 2 min , were immunoprecipitated with mouse and rabbit anti-phosphoserine antibodies mixture , followed by the immunoblotting with an anti-PS1 loop antibody . Glutamate does not affect total PS1 CTF levels . The graph shows the relative PS1 phosphorylation level in glutamate-treated neurons normalized to that in vehicle-treated neurons ( n = 6 ) . Mean ± SEM , *p<0 . 05 , One sample t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 004 Since phosphorylation is known to regulate protein conformation and/or activity , we tested whether Ca2+ influx may induce phosphorylation of PS1 , and whether PS1 phosphorylation is responsible for the PS1 pathogenic conformational change . For this , the amount of phosphorylated PS1 was measured by immunoprecipitation/Western blotting in primary neurons treated with KCl . We found that KCl treatment significantly increased the level of PS1 phosphorylation at its serine residues ( Figure 1B ) . We have confirmed this finding using another stimulant , glutamate ( Figure 1—figure supplement 1 ) . Collectively , these results indicate that Ca2+ influx leads to PS1 phosphorylation . To address whether PS1 phosphorylation is involved in the Ca2+-mediated PS1 conformational change , we used phosphorylation-inhibited mutants of PS1 . For this , 7 W cells were transiently transfected with wild type ( WT ) or mutant PS1 constructs , in which serine ( S ) or threonine ( T ) previously reported as phosphorylation sites were substituted by alanine ( A ) ( Table 1 ) . Then , transfected cells were treated with the Ca2+ ionophore A23187 to increase intracellular Ca2+ levels . Following the treatment , the cells were fixed and immunostained with anti-FLAG ( on PS1 NT ) and anti-PS1 CT antibodies to selectively detect the conformational change of ‘exogenous’ PS1 using the antibody-based FRET assay . Significant increases in Ca2+ levels were verified by Indo-1 Ca2+ imaging ( Grynkiewicz et al . , 1985 ) ( Figure 2—figure supplement 1A ) . There was no statistical difference between the FRET efficiencies of WT PS1 and phosphorylation-inhibited PS1 mutants in vehicle-treated cells ( Table 1 ) , indicating that the conformation of the phosphorylation-inhibited PS1 mutants is not different from that of WT PS1 in basal conditions . A23187 significantly increased the FRET efficiency between fluorescently labeled NT and CT of WT PS1 ( Table 1 ) . This indicates that WT PS1 adopts the ‘closed’ conformation following the treatment with the Ca2+ ionophore . On the other hand , A23187 did not change the FRET efficiencies in T74A , S310A , S313A , S310A/S313A , S365A , S366A , S367A , S365A/S367A and S366A/S367A PS1 transfected cells , suggesting that conformation of these PS1 mutants was resistant to Ca2+ ionophore treatment ( Table 1 ) . The A23187-resistant phosphorylation sites on PS1 could be divided into three sub-membrane domains: domain 1 ( T74 ) on NT of PS1 , domain 2 ( S310 , S313 ) and domain 3 ( S365 , S366 , S367 ) within the large hydrophilic loop ( Figure 2—figure supplement 2 ) . 10 . 7554/eLife . 19720 . 005Table 1 . FLIM analysis of the PS1 NT-CT proximity in phosphorylation-inhibited PS1 mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 005Construct*Relative FRET efficiency ( % ) p valueDMSOA23187‡vs WT ( in DMSO ) §DMSO vs A23187PS1 Wild type ( WT ) 100 ± 8 . 9 ( n = 18 ) 134 . 7 ± 7 . 5 ( n = 25 ) -†p<0 . 05 PS1 S28A106 . 9 ± 7 . 0 ( n = 19 ) 143 . 4 ± 8 . 6 ( n = 19 ) n . s . †p<0 . 05 PS1 T74A96 . 5 ± 13 . 9 ( n = 14 ) 98 . 5 ± 8 . 3 ( n = 22 ) n . s . n . s . PS1 S310A88 . 8 ± 10 . 2 ( n = 17 ) 95 . 1 ± 8 . 6 ( n = 24 ) n . s . n . s . PS1 S313A92 . 7 ± 8 . 4 ( n = 15 ) 108 . 0 ± 8 . 8 ( n = 21 ) n . s . n . s . PS1 S310A/S313A84 . 2 ± 6 . 7 ( n = 26 ) 87 . 8 ± 9 . 0 ( n = 17 ) n . s . n . s . PS1 S319A/T320A106 . 2 ± 8 . 1 ( n = 20 ) 131 . 8 ± 8 . 1 ( n = 20 ) n . s . †p<0 . 05 PS1 S324A93 . 8 ± 6 . 2 ( n = 19 ) 129 . 9 ± 8 . 4 ( n = 15 ) n . s . †p<0 . 05 PS1 S337A99 . 9 ± 8 . 8 ( n = 13 ) 132 . 4 ± 10 . 9 ( n = 16 ) n . s . †p<0 . 05 PS1 S346A100 . 1 ± 8 . 9 ( n = 18 ) 135 . 0 ± 5 . 2 ( n = 19 ) n . s . †p<0 . 05 PS1 S353A89 . 6 ± 7 . 4 ( n = 18 ) 129 . 8 ± 8 . 3 ( n = 23 ) n . s . †p<0 . 05 PS1 T354A74 . 4 ± 5 . 6 ( n = 14 ) 113 . 4 ± 7 . 7 ( n = 23 ) n . s . †p<0 . 05 PS1 S357A91 . 2 ± 7 . 9 ( n = 18 ) 119 . 8 ± 9 . 2 ( n = 23 ) n . s . †p<0 . 05 PS1 S353A/S357A98 . 9 ± 9 . 1 ( n = 14 ) 125 . 9 ± 7 . 1 ( n = 13 ) n . s . †p<0 . 05 PS1 S365A102 . 3 ± 6 . 4 ( n = 19 ) 87 . 3 ± 9 . 9 ( n = 21 ) n . s . n . s . PS1 S366A102 . 8 ± 6 . 7 ( n = 16 ) 101 . 6 ± 12 . 6 ( n = 16 ) n . s . n . s . PS1 S367A99 . 7 ± 6 . 1 ( n = 10 ) 99 . 1 ± 11 . 4 ( n = 15 ) n . s . n . s . PS1 S365A/S367A104 . 1 ± 6 . 7 ( n = 19 ) 102 . 8 ± 8 . 4 ( n = 19 ) n . s . n . s . PS1 S366A/S367A102 . 7 ± 11 . 2 ( n = 10 ) 108 . 2 ± 9 . 9 ( n = 19 ) n . s . n . s . *The FRET efficiency in DMSO-treated cells expressing WT PS1 is set as 100% , and relative FRET efficiency in phosphorylation-inhibited mutants of PS1 is shown . Mean ± SEM , Student’s t-test , n: cell number , †: p<0 . 05 , n . s . : not significant . ‡p-value is shown for the comparison between WT PS1 and phosphorylation-inhibited mutants of PS1 in DMSO-treated conditions , or §for the comparison between DMSO-treated and A23187 ( 5 µM for 15 min ) -treated cells expressing the same PS1 construct . To confirm this finding we used a complementary approach , for which we introduced phosphorylation-inhibited mutations within the three domains into GFP-PS1-RFP , a conformation sensitive FRET probe ( Uemura et al . , 2009 ) . Consistently , A23187 significantly increased the FRET efficiency of WT G-PS1-R ( Figure 2—figure supplement 1B ) . On the other hand , A23187 did not change the FRET efficiency between GFP and RFP of domain 1 , domain 2 , or domain 3 phosphorylation-inhibited G-PS1-R mutants ( Figure 2—figure supplement 1B ) . Collectively , these results demonstrate that phosphorylation of the three domains is necessary for the Ca2+ dependent PS1 pathogenic conformational shift . It still remains unknown whether simultaneous phosphorylation of all three domains is necessary for the pathogenic change in PS1 conformation , or if there is a sequence of phosphorylation events leading to the PS1 conformational change . To distinguish between these scenarios , we generated a set of phosphorylation-mimicking G-PS1-R mutants by replacing serine or threonine with aspartic acid ( D ) in single domains ( domain 1 , domain 2 or domain 3 ) , in double domains ( domains 1 + 2 , domains 1 + 3 or domains 2 + 3 ) , or in triple domains 1 + 2 + 3 . The conformation of G-PS1-R with domain 1 , or domain 2 only , or combined domain 1 and 2 mutants was not different from that of WT G-PS1-R ( Figure 2A ) . However , the FRET efficiency of the domain 3 phosphorylation-mimicking G-PS1-R isoform was significantly higher than that of WT G-PS1-R , indicating that this mutant forces PS1 into the ‘closed’ conformation ( Figure 2A ) . Domains 1 + 3 G-PS1-R , domains 2 + 3 G-PS1-R and domains 1 + 2 + 3 G-PS1-R also adopted the ‘closed’ conformation ( Figure 2A ) . Collectively , these results strongly suggest that phosphorylation of domain 3 is the critical mechanism leading to the PS1/γ-secretase pathogenic ‘closed’ conformation . 10 . 7554/eLife . 19720 . 006Figure 2 . Phosphorylation mimicking mutations at S365-S366-S367 result in the PS1 conformational shift . ( A ) FLIM analysis of the PS1 conformation in 7 W cells transfected with WT or phosphorylation-mimicking mutant G-PS1-R . The FRET efficiency between GFP and RFP in phospho-mutants is normalized to the average FRET efficiency of the WT G-PS1-R expressing cells ( n = 42–53 cells ) . Mean ± SEM , ***p<0 . 001 , one-way factorial ANOVA . ( B ) Spectral FRET analysis shows RFP/GFP ( R/G ) ratio in 7 W cells transfected with WT or single phosphorylation-mimicking mutant G-PS1-R ( n = 50–69 cells ) . Mean ± SEM , ***p<0 . 001 , one-way factorial ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 00610 . 7554/eLife . 19720 . 007Figure 2—figure supplement 1 . Calcium imaging and validation of the phosphorylation at the three domains-driven PS1 conformational change by the G-PS1-R FRET reporter probe . This figure is related to the data displayed in Table 1 . ( A ) Indo-1/Ca2+ imaging shows changes in intracellular Ca2+ levels of the 7 W cells transfected with FLAG-WT PS1 , following 15 min treatment with vehicle control or with 5 µM A23187 ( n = 250–297 cells ) . Mean ± SEM , ***p<0 . 001 , Student’s t-test . ( B ) FLIM analysis of the PS1 conformation in the 7 W cells transfected with WT or domain 1 , 2 or 3 phosphorylation-inhibited mutants PS1 , following 15 min treatment with vehicle control or with 5 µM A23187 . The FRET efficiency in each group is normalized to the average FRET efficiency of vehicle-treated cells expressing WT G-PS1-R ( n = 22–55 cells ) . Mean ± SEM , *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , one-way factorial ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 00710 . 7554/eLife . 19720 . 008Figure 2—figure supplement 2 . Schematic image of the three domains in PS1 involved in Ca2+-triggered pathogenic ‘closed’ conformation . This figure is related to the data displayed in Table 1 . Schematic image of PS1 molecule and the three domains involved in Ca2+-triggered PS1 pathogenic ‘closed’ conformation ( blue squares ) . Serine/threonine residues known to undergo phosphorylation are shown by red circles . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 008 Domain 3 of PS1 consists of three consecutive serine residues: S365 , S366 and S367 . To identify specific residues in domain 3 responsible for the pathogenic conformational change of PS1 , we next generated single phosphorylation-mimicking G-PS1-R mutants . The spectral FRET analysis revealed significantly higher FRET efficiency in S367D G-PS1-R but not in S365D G-PS1-R or S366D G-PS1-R , as compared to WT G-PS1-R ( Figure 2B ) . This indicates that phosphorylation of PS1 at serine 367 is responsible for the PS1 ‘closed’ conformational change . Several kinases involved in phosphorylation of the serine residues within the PS1 cytoplasmic loop region , such as PKA , JNK , PKC and GSK3β , have been identified ( Kirschenbaum et al . , 2001; Fluhrer et al . , 2004; Kuo et al . , 2008 ) . However , the kinase ( s ) responsible for the phosphorylation that is crucial for the PS1 conformational change remains unknown . To address this , WT G-PS1-R expressing 7 W cells were incubated with inhibitors of several kinases prior to A23187 treatment . The spectral FRET assay revealed that only the PKA inhibitor ( KT5720 ) specifically rescued the Ca2+-triggered ‘closed’ PS1 conformation , whereas JNK , PKC and GSK3β inhibitors did not ( Figure 3A ) . This was confirmed using a different PKA inhibitor ( H-89 ) and a complementary antibody-based FLIM assay ( Figure 3—figure supplement 1A ) . To verify this further in primary neurons , we monitored the change of endogenous PS1 conformation following pre-treatment with the two PKA inhibitors: H-89 or KT5720 , or PKA activators: forskolin or 8-Bromo-cAMP , prior to incubation with KCl . The cAMP response element binding protein ( CREB ) , a well known PKA-specific substrate ( Gonzalez and Montminy , 1989 ) , was used as a positive control for the Ca2+ influx-induced PKA activation ( Ferguson and Storm , 2004 ) ( Figure 3—figure supplement 1B ) . We found that the proximity between the fluorescently labelled PS1 NT and CT was comparable between vehicle and KCl-treated neurons in the presence of H-89 or KT5720 ( Figure 3B ) . On the other hand , the FRET efficiency of forskolin or 8-Bromo-cAMP treated neurons was significantly higher than that of vehicle-treated neurons ( Figure 3B ) , supporting the involvement of PKA in the PS1 conformational change . Of note , there was no additive effect with KCl and PKA activators ( Figure 3B ) . The inhibitory effect of KT5720 on the KCl/Ca2+-triggered PS1 conformational change was also confirmed by real-time spectral FRET imaging in living primary neurons ( Figure 3—figure supplement 1C ) . To further corroborate the results obtained by pharmacological inhibition of PKA , we employed a dominant negative form of the PKA regulatory α subunit ( G324D ) ( Olson et al . , 1993 ) . We found that co-expression of the G324D PKA mutant together with G-PS1-R prevents both Ca2+-triggered PS1 phosphorylation at S310 and the ‘closed’ PS1 conformation ( Figure 3—figure supplement 1D ) . 10 . 7554/eLife . 19720 . 009Figure 3 . PKA activity is involved in the Ca2+-driven PS1 conformational change . ( A ) Spectral FRET analysis of the PS1 conformation in 7 W cells transfected with WT G-PS1-R and pre-treated with 1 µM KT5720 ( KT , PKA inhibitor , 16 hr ) , 20 µM SP600125 ( SP , JNK inhibitor , 4 hr ) , 100 nM Ro 31–8220 ( Ro , PKC inhibitor , 16 hr ) , or 5 µM TDZD-8 ( TD , GSK3β inhibitor , 4 hr ) . Spectral FRET imaging was performed after the cells were treated with DMSO ( vehicle control ) or 5 µM A23187 for 15 min ( n = 31–50 cells ) . Mean ± SEM , *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , one-way factorial ANOVA . ( B ) Endogenous PS1 conformation was monitored by the antibody-based FLIM analysis . Primary neurons were pre-treated with PKA inhibitors: 30 µM H-89 or 1 µM KT5720 for 16 hr , or with PKA activators: 10 µM forskolin or 0 . 5 mM 8-Bromo-cAMP for 1 hr , followed by the treatment with vehicle control ( - ) or 50 mM KCl ( + ) for 5 min . The FRET efficiency in each group is normalized to the average FRET efficiency of vehicle control treated neurons ( n = 36–125 processes from n = 17–28 cells ) . Mean ± SEM , ***p<0 . 001 , one-way factorial ANOVA . ( C ) Ca2+ sequestration by EGTA affects PS1 conformation in vehicle but not PKA activator-treated neurons . Primary neurons were pre-treated with vehicle , 10 µM forskolin or 0 . 5 mM 8-Bromo-cAMP for 1 hr , followed by the treatment with 2 mM EGTA or vehicle control for 15 min . The FRET efficiency in each group is normalized to the average FRET efficiency of vehicle control neurons ( n = 19–34 cells , n = 40–103 processes ) . Mean ± SEM , **p<0 . 01 , ***p<0 . 001 , one-way factorial ANOVA . ( D ) FLIM analysis of the PS1 conformation in 7 W cells transfected with WT or phosphorylation-inhibited mutants G-PS1-R , and treated with 10 µM forskolin or vehicle control for 1 hr . The FRET efficiency is normalized to the average FRET efficiency of vehicle-treated cells expressing WT G-PS1-R ( n = 20–39 cells ) . Mean ± SEM , *p<0 . 05 , **p<0 . 01 , one-way factorial ANOVA . ( E ) FLIM analysis of the PS1 conformation in 7 W cells transfected with WT or domain 1 + 2 phosphorylation-mimicking G-PS1-R isoform , and treated with 5 µM A23187 , 10 µM forskolin or 0 . 5 mM 8-Bromo-cAMP . The FRET efficiency normalized to that in vehicle treated cells is shown ( n = 34–47 cells ) . Mean ± SEM , *p<0 . 05 , **p<0 . 01 , one-way factorial ANOVA . ( F ) KCl-induced increase of the Aβ42/40 ratio is prevented by PKA inhibitor . Primary neurons were pre-treated with vehicle control or 1 µM KT5720 for 16 hr , followed by the treatment with vehicle control or 50 mM KCl for 30 min . Aβ40 and Aβ42 levels were measured by ELISA . The Aβ42/40 ratio in each group is normalized to that of the vehicle control neurons ( n = 19–25 ) . Mean ± SEM , *p<0 . 05 , **p<0 . 01 , one-way factorial ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 00910 . 7554/eLife . 19720 . 010Figure 3—figure supplement 1 . PKA is involved in the Ca2+-triggered PS1 pathogenic ‘closed’ conformation . ( A ) FLIM analysis of the PS1 conformation in 7 W cells . The cells were pre-treated with 30 µM H-89 or 1 µM KT5720 for 16 hr , followed by the treatment with vehicle control ( − ) or 5 µM A23187 ( + ) for 15 min . The FRET efficiency in each group is normalized to the average FRET efficiency of vehicle control treated neurons ( n = 37–52 cells ) . Mean ± SEM , **p<0 . 01 , ***p<0 . 001 , one-way factorial ANOVA . ( B ) Western Blot analysis of PKA-dependent CREB S133 phosphorylation . Primary neurons were pre-treated with vehicle control or 1 µM KT5720 for 16 hr , followed by the treatment with vehicle control or 50 mM KCl for 5 min . Treatment with 10 µM forskolin was used as a positive control of PKA activation . ( C ) Spectral FRET assay of the PS1 conformation in primary neurons . AAV-G-PS1-R-transduced primary neurons were pre-treated with 1 µM KT5720 for 16 hr , followed by the treatment with 5 mM KCl for 5 min . Pre: before KCl application , Post: 5 min after KCl application . n = 14–17 cells , Mean ± SEM , *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 , one-way factorial ANOVA . ( D ) Dominant negative PKA regulatory subunit α ( G324D PKA ) was co-transfected with WT G-PS1-R into 7 W cells . The cells were treated with vehicle control or 5 µM A23187 for 15 min . Western Blot analysis of the PS1 S310 phosphorylation ( top panel ) . Spectral FRET assay of the PS1 conformation ( bottom panel ) . n = 83–88 cells , Mean ± SEM , *p<0 . 05 , ***p<0 . 001 , one-way factorial ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 010 To further verify that the Ca2+-induced PKA activation is an essential step in the PS1 conformational change , we pre-treated primary neurons with forskolin or 8-Bromo-cAMP , followed by treatment with EGTA to reduce intracellular Ca2+ . As expected , EGTA decreased the FRET efficiency between PS1 NT and CT , indicating that PS1 adopted an ‘open’ conformation ( Figure 3C ) . On the other hand , EGTA had no effect on PS1 conformation in the presence of PKA activators ( Figure 3C ) , indicating that PKA activation is an event down-stream of the Ca2+ influx . To ensure that PKA activation causes the PS1 pathogenic conformational change via PS1 phosphorylation , we treated 7 W cells expressing WT G-PS1-R or the phosphorylation-inhibited G-PS1-R mutants with forskolin . Whereas PKA activation by forskolin caused the PS1 ‘closed’ conformation in WT G-PS1-R expressing cells , it did not affect the conformation of domain 1 , domain 2 , or domain 3 phosphorylation-inhibited mutants ( Figure 3D ) . This indicates that blocking phosphorylation of either one of these domains precludes PS1 from adopting the ‘closed’ conformation via PKA activation . Phosphorylation-mimicking mutations in domain 1 or domain 2 only ( but not in domain 3 ) , or domain 1 and domain 2 combined had no effect on PS1 conformation ( Figure 2A ) . Thus , to verify that domain 3 needs to be phosphorylated for the PS1 conformational change , and to determine whether PKA is involved , we transfected 7 W cells with WT or the domains 1 + 2 phosphorylation-mimicking G-PS1-R mutant , followed by treatment with A23187 , forskolin , or 8-Bromo-cAMP . Similar to WT G-PS1-R expressing cells , the FRET efficiency of domains 1 + 2 phosphorylation-mimicking G-PS1-R was significantly increased upon A23187 , forskolin or 8-Bromo-cAMP treatments ( Figure 3E ) . This indicates that PKA is involved in the phosphorylation of S365-S366-S367 residues . Since allosteric changes in PS1 conformation underlie changes in the Aβ42/40 ratio , with the ‘closed’ conformation linked to the increased Aβ42/40 ratio ( Berezovska et al . , 2005; Isoo et al . , 2007; Uemura et al . , 2010; Wahlster et al . , 2013; Kuzuya et al . , 2016 ) , we next tested if the inhibition of PKA activity would prevent the KCl treatment-induced increase in the Aβ42/40 ratio . We found that treatment with KT5720 indeed abolished the KCl-induced elevation of the Aβ42/40 ratio ( Figure 3F ) . Collectively , these data demonstrate that phosphorylation of domain 1 or 2 is not sufficient for the PS1 conformational change , but is required for PKA-mediated phosphorylation of domain 3 , leading to the pathogenic conformational change in PS1/γ-secretase . We established that PS1 phosphorylation at domain 3 forces PS1 into the ‘closed’ state in vitro . To determine whether Ca2+ influx may trigger PS1 domain 3 phosphorylation and the pathogenic ‘closed’ conformation in vivo , we optimized our spectral FRET assay ( Uemura et al . , 2009 ) to monitor PS1 conformation in living mice using two-photon microscopy . First , we found that GFP emission was the strongest in the 875–900 nm wavelength range , whereas RFP emission significantly decreased at >800 nm excitation , with barely detectable RFP signal at 900 nm in 7 W cells expressing G-PS1-R ( Figure 4—figure supplement 1A ) . Therefore , 900 nm excitation was used in subsequent in vivo experiments . Using two different negative controls for FRET: G-PS1 ( donor only ) and PS1-R ( acceptor only ) , we ensured that the emission intensity of RFP at 900 nm excitation is caused neither by GFP fluorescence bleed-through , nor by direct RFP excitation ( Figure 4—figure supplement 1B ) . Next , to express the G-PS1-R probe in the brains of living mice , we sub-cloned the construct into an AAV8 vector under the neuron specific human Synapsin 1 ( hSyn1 ) promoter ( Figure 4A ) . pAAV8-hSyn1-G-PS1 ( donor only ) and pAAV8-hSyn1-PS1-R ( acceptor only ) as negative controls , and pAAV8-hSyn1-R-G fusion protein as a positive control were also generated for the FRET assay . G-PS1-R expression in neurons of the somatosensory cortex was verified by two-photon imaging using 775 nm wavelength by which both GFP and RFP were excited simultaneously ( Figure 4B ) . We selectively excited GFP at 900 nm , and found that the RFP/GFP ( R/G ) ratio was significantly higher in G-PS1-R expressing neurons than that in G-PS1 ( donor only ) ( Figure 4C ) , validating the FRET detection and establishment of the PS1 conformation assay in living mouse brain . 10 . 7554/eLife . 19720 . 011Figure 4 . Ca2+ triggers the PS1/γ-secretase pathogenic conformation via PS1 phosphorylation in vivo . ( A ) Schematic representation of the pAAV8-hSyn1-WT G-PS1-R construct . ( B ) Two-photon image of the WT G-PS1-R expression in the somatosensory cortex of WT mouse . Laser at 775 nm wavelength was used for the excitation . A scale bar indicates 10 µm . ( C ) Mice were injected with AAV8-hSyn1-G-PS1 ( as a negative control of FRET ) , AAV8-hSyn1-WT G-PS1-R or AAV8-hSyn1-R-G ( as a positive control of FRET ) . GFP was excited at 900 nm wavelength , and the R/G ratio was recorded ( n = 50–60 cells , n = 3–6 mice ) . Mean ± SEM , ***p<0 . 001 , one-way factorial ANOVA . ( D ) Spectral FRET analysis of the PS1 conformation in vivo . Mice were injected with AAV8-hSyn1-WT G-PS1-R ( n = 3 ) or AAV8-hSyn1-S365A/S366A/S367A G-PS1-R ( n = 4 ) , and 300 mM KCl was applied topically . The R/G ratio in vivo was monitored after two-photon excitation at 900 nm ( total n = 20–28 cells per condition ) for the duration of 2 min . Representative images of the pseudo-colored neurons are shown ( additional time traces/images are shown in Figure 4—figure supplement 3 ) . Mean ± SEM , **p<0 . 01 , two-way repeated-measures ANOVA . ( E ) Ex-vivo spectral FRET analysis of the endogenous PS1 conformation in mouse brain sections . Mice were injected with 100 mM 8-Bromo-cAMP ( right hemisphere ) or vehicle ( left hemisphere ) into the somatosensory cortex . The 565 nm/522 nm ratio was calculated in individual neurons as readout of the FRET efficiency that reflects the relative proximity of the PS1 NT ( A488 ) to PS1 CT ( Cy3 ) . The histogram shows cell numbers plotted against the 565 nm/522 nm ratios . n = 3 mice , total of 444 ( vehicle ) and 505 ( 8 Bromo-cAMP ) neurons . ***p<0 . 001 , Student’s t-test . ( F ) Ex-vivo spectral FRET analysis of the endogenous PS1 conformation in mouse brain sections . Mice were pre-injected with 100 µM KT5720 ( right hemisphere ) or vehicle ( left hemisphere ) into the somatosensory cortex . 75 min post-injection , 100 mM 8-Bromo-cAMP ( both hemispheres ) was delivered to the same area for 5 min . The histogram shows cell numbers plotted against the 565 nm/522 nm ratios . n = 3 mice , total 423 ( vehicle ) and 436 ( KT5720 ) neurons analysed . ***p<0 . 001 , Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 01110 . 7554/eLife . 19720 . 012Figure 4—figure supplement 1 . Establishment of the two-photon spectral FRET settings for monitoring PS1 conformation . ( A ) 7 W cells expressing G-PS1-R were excited by different wavelength laser from 750 nm to 975 nm in 25 nm steps . The top panel shows representative images of the emission intensities in green and red channel ( n = 41 ) . A scale bar indicates 20 µm . The graph shows quantitative analysis of the relative emission intensities , with the intensity at 900 nm ( green channel ) and 750 nm ( red ) as 100% . Mean ± SEM . ( B ) 7 W cells expressing G-PS1-R , G-PS1 or PS1-R were excited by 900 nm laser . Left panel shows the representative images of relative emission intensities in green or red channel . A scale bar indicates 20 µm . The intensity of G-PS1-R in green channel was set as 100% ( n = 36 each ) . Mean ± SEM , ***p<0 . 001 , one-way factorial ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 01210 . 7554/eLife . 19720 . 013Figure 4—figure supplement 2 . YC3 . 6-based Ca2+ imaging in vivo after KCl application . 300 mM KCl or vehicle ( PBS ) was topically applied under the cranial window to mice expressing YC3 . 6 . The YFP/CFP ratio was measured in neurons of the somatosensory cortex in vivo ( n = 18–22 cells , three different mice for each group ) . Representative pseudo-colored images were shown in the top panels . Mean ± SEM , ***p<0 . 001 , two-way repeated-measures ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 01310 . 7554/eLife . 19720 . 014Figure 4—figure supplement 3 . Spectral FRRT assay of PS1 conformation in vivo . Additional time traces and pseudo-coloured images of the neurons in vivo showing changes in PS1 conformation . A scale bar indicates 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 01410 . 7554/eLife . 19720 . 015Figure 4—figure supplement 4 . Spectral FRET assay for monitoring endogenous PS1 conformation in mouse brain sections . ( A ) Immunohistochemical detection of the phosphorylated CREB in mouse brain injected with 100 mM 8-Bromo-cAMP ( right hemisphere ) or vehicle ( left hemisphere ) . The brain sections were immunostained with a p-CREB S133 antibody . Strong fluorescence in cAMP injected hemisphere shows PKA activation . A scale bar indicates 50 µm . Mean ± SEM , ***p<0 . 001 , Student’s t-test . ( B ) Antibody-based spectral FRET validation . The brain sections were immunostained with PS1 NT-A488 ( donor only , negative control for FRET ) , PS1 NT-A488-Cy3 ( A488-Goat-anti-mouse IgG was detected with Cy3-Donkey-anti-goat IgG , positive control ) , or PS1 NT-A488/PS1 CT-Cy3 ( FRET , PS1 NT-PS1 CT proximity ) antibodies . The Y axis shows the ratio of fluorescence intensities within the 522 nm to 672 nm spectral range normalized by the emission intensity at 522 nm . The positive control ( in green ) shows large ‘hump’ at 565/522 nm , and the FRET signal in PS1 NT-PS1 CT stained cells ( in red ) shows smaller but significant increase in the 565 nm/522 nm ratio compared to the donor only ( n = 75–179 cells ) . Mean ± SEM , ***p<0 . 001 vs . Donor only , one-way factorial ANOVA . ( C ) Immunohistochemical detection of the phosphorylated CREB S133 in mouse brain injected with 100 µM KT5720 ( right hemisphere ) or vehicle ( left hemisphere ) , followed by the injection of 100 mM 8-Bromo-cAMP ( both hemispheres ) . A scale bar indicates 50 µm . Mean ± SEM , ***p<0 . 001 , Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 015 To ensure that the G-PS1-R probe was functional and able to reliably report PS1 conformational changes in vivo , we applied KCl topically , under the cranial window . First , we verified that topical KCl application significantly increased intracellular Ca2+ levels in the somatosensory cortex neurons by employing Yellow Cameleon 3 . 6 ( YC3 . 6 ) -based ( Nagai et al . , 2004 ) in vivo Ca2+ imaging ( Figure 4—figure supplement 2 ) . The actual concentration of Ca2+ in cell bodies , calculated as described previously ( Kuchibhotla et al . , 2008 ) , was increased from ~58 nM at the basal condition up to ~2793 nM immediately after the applied KCl reached the imaging site . It remained at ~166 nM during the 2 min of imaging ( mean = ~649 nM ) . Consistently , the R/G ratio was significantly enhanced by the KCl application ( Figure 4D and Figure 4—figure supplement 3 ) , indicating that elevated Ca2+ induces the pathogenic PS1 conformation in vivo . To confirm that Ca2+ triggers the ‘closed’ conformational shift in PS1/γ-secretase via PS1 domain 3 phosphorylation , we expressed the phosphorylation-inhibited mutant S365A/S366A/S367A G-PS1-R in mouse brain prior to KCl stimulation . Notably , no change in the R/G ratio was detected in neurons expressing this mutant G-PS1-R after KCl application ( Figure 4D and Figure 4—figure supplement 3 ) . These findings indicate that increased intracellular Ca2+ triggers the ‘closed’ conformation of PS1 via domain 3 phosphorylation in vivo . To ensure that PKA-mediated phosphorylation is involved in the conformational change of the endogenous PS1 , 8-Bromo-cAMP and vehicle control were injected into the somatosensory cortex of mice , in right and left hemispheres , respectively . The mice were sacrificed 5 min post-injection , and PKA activation was verified by immunostaining of the brain sections for p-CREB S133 phosphorylation ( Figure 4—figure supplement 4A ) . Endogenous PS1 was immunolabeled with PS1 NT ( A488 ) and PS1 CT ( Cy3 ) antibodies , and the 565 nm/522 nm fluorescence emission ratio was used as a readout of the FRET efficiency ( Figure 4—figure supplement 4B ) . The ex-vivo spectral FRET assay revealed that PKA activation by 8-Bromo-cAMP led to the ‘closed’ conformation of endogenous PS1 as indicated by the increased number of neurons with a higher 565 nm/522 nm ratio ( Figure 4E ) . Next , to determine if KT5720 pre-treatment would prevent the Bromo-cAMP-induced pathogenic collapse of PS1 , the PKA inhibitor KT5720 or vehicle control were injected into mouse somatosensory cortex 75 min prior to 8-Bromo-cAMP injection . The ex-vivo spectral FRET assay revealed that PKA inhibition could prevent the 8-Bromo-cAMP-triggered ‘closed’ conformation of PS1 in mouse brain ( Figure 4F ) . Immunostaining for CREB S133 phosphorylation confirmed that KT5720 significantly suppressed 8-Bromo-cAMP-induced PKA activation ( Figure 4—figure supplement 4C ) . Since PS1 adopts the pathogenic ‘closed’ conformation in sporadic AD brains ( Wahlster et al . , 2013 ) , we investigated whether PS1 phosphorylation is up-regulated in the sAD brain . To test this , we used the commercially available S310 ( domain 2 ) phosphorylation specific antibody to compare the amount of phosphorylated PS1 in AD brains and in age , gender and post mortem interval ( PMI ) -matched control brains ( Table 2 ) . 10 . 7554/eLife . 19720 . 016Table 2 . List of the human brain samples used in the study . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 016CaseAgePMISexBraakCerad Control 16015FControl 27320FControl 38820FIIControl 4918FIAControl 56316MControl 6858MIControl 78748MIControl 89119FIIAControl 98610MControl 109417MIControl 11546MControl 125818FControl 138816FIIControl 146014MControl 1592unknownMIIControl 166827MIControl 177648FIpossibly AControl 189212MIIControl 1985unknownMIIControl 209223MIIAControl average ( Mean ± SEM ) 79 . 15 ± 3 . 0 19 . 16 ± 2 . 6 ( F:M = 8:12 ) AD 15812FVICAD 27318FVCAD 38424FVICAD 4855FVICAD 56024MVICAD 67818FVICAD 78524MVICAD 88620MVICAD 9874FVICAD 10695MVICAD 119412FVICAD 128612MVIBAD 138918FVIBAD 147116FVICAD 157017MVICAD 169618MVICAD 17916MVICAD 188314FVICAD 198712FIVBAD 208713MVICAD 217314MVICAD 22788MVICAD 23798MVICAD average ( Mean ± SEM ) 80 . 39 ± 2 . 1 14 ± 1 . 3 ( F:M = 11:12 ) First , the specificity of the p-PS1 ( S310 ) antibody to detect Ca2+-induced PKA-mediated phosphorylation at the S310 residue was verified in 7 W cells . The cells were transfected with FLAG-tagged WT or S310A mutant PS1 constructs , treated with the calcium ionophore A23187 , and phospho-PS1 ( S310 ) immunoreactivity was detected by confocal microscopy ( Figure 5—figure supplement 1A ) . We show that A23187 increased WT PS1 phosphorylation at S310 was abolished by pre-treatment with the PKA inhibitor KT5720 or PS1 S310A expression ( Figure 5—figure supplement 1A ) . Increased PS1 S310 phosphorylation by the PKA activator forskolin was also detected by Western blotting in PS70 cells ( Figure 5—figure supplement 1B ) . In addition , we verified dose-dependent S310 phosphorylated PS1 ( p-PS1 ) in human brain lysates by loading variable amounts of the total protein ( Figure 5—figure supplement 1B ) , suggesting that the amount of PS1 phosphorylation is measurable in human brains . We found that even though a broad distribution of the p-PS1 ( S310 ) level is observed among the AD cases , the average PS1 phosphorylation is significantly increased in AD brains as compared to non-demented control brains ( Figure 5 ) . Correlation analysis showed that neither age nor PMI significantly affected the level of PS1 phosphorylation in both control and AD cases ( Figure 5—figure supplement 2A and B ) . 10 . 7554/eLife . 19720 . 017Figure 5 . Phosphorylated PS1 level is increased in AD brains . Western blot analysis of PS1 phosphorylated at S310 in 20 control brains and 23 AD brains . The level of PS1 S310 phosphorylation ( phospho-PS1/total PS1 ) in AD brains is normalized to that in the control brains . Mean ± SEM , *p<0 . 05 , Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 01710 . 7554/eLife . 19720 . 018Figure 5—figure supplement 1 . Validation of the PKA-mediated PS1 phosphorylation at S310 . ( A ) Confocal microscope images of the 7 W cells immunostained with anti-phosphorylated PS1 at S310 ( green ) or anti-FLAG antibodies ( red ) . 7 W cells transfected with FLAG-WT PS1 or FLAG-S310A PS1 were pre-treated with 1 µM KT5720 or vehicle control for 16 hr , followed by the treatment of 5 µM A23187 or vehicle control for 15 min . A scale bar indicates 20 µm . ( B ) The level of phosphorylated PS1 at S310 was compared between forskolin and vehicle-treated PS70 cells by Western blotting . PS1/2 dKO MEF cells were used as a negative control . The level of phosphorylated PS1 at S310 was measured in the AD brain sample loaded at different total protein concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 01810 . 7554/eLife . 19720 . 019Figure 5—figure supplement 2 . Correlation between the relative phosphorylation level of PS1 and age , or PMI . Correlation analysis of phosphorylation of PS1 at S310 with age in control and AD brains ( A ) , or with post mortem interval ( PMI ) ( B ) . Spearman's nonparametric correlation analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 019
Increasing evidence suggests interactions between Ca2+ overload and Aβ in the pathogenesis of AD . Although it has been suggested that aberrant Ca2+ homeostasis affects Aβ production , the effect of Ca2+ elevations on PS1/γ-secretase function is unknown . We have recently reported that increases in Ca2+ levels induce the pathogenic change in PS1 conformation and increase the Aβ42/40 ratio in cultured neurons ( Kuzuya et al . , 2016 ) . Considering the rapid and dynamic nature of the conformational changes , it is unlikely that this change is achieved through mechanisms involving transcription or reassembly of PS1 into de novo γ-secretase complexes . However , the detailed molecular mechanism remains unclear , and whether similar Ca2+-mediated PS1 conformational changes occur in vivo is unknown . Our current study addresses these questions and demonstrates that Ca2+/PKA-mediated PS1 phosphorylation at domain 1 ( T74 ) , domain 2 ( S310 , S313 ) and domain 3 ( S365 , S366 , S367 ) is responsible for the PS1/γ-secretase pathogenic conformational change in vitro and in vivo . The high level of serines and threonines in the PS1 N-terminus and in the large hydrophilic loop region suggests that the enzymatic function of PS1 can be modulated by its ‘phosphorylated’ and ‘dephosphorylated’ states . It has been reported that PS1 phosphorylation at several residues enhances stability of the PS1 CT fragment that is necessary for γ-secretase activity ( Lau et al . , 2002; Kuo et al . , 2008; Ryu et al . , 2010 ) . Recently , Matz and colleagues have identified eleven new phosphorylation sites in PS1 ( Matz et al . , 2015 ) . They showed that phosphorylation-inhibited mutations of these sites affect neither activity of the PS1/γ-secretase nor Aβ production . Our current study revealed that six amino acids: T74 , S310 , S313 , S365 , S366 and S367 , all , except for S310 , within the newly-discovered PS1 phosphorylation sites , are critical for the Ca2+-triggered PS1 pathogenic , ‘closed’ conformation linked to increased Aβ42/40 ratio ( Berezovska et al . , 2005; Isoo et al . , 2007; Uemura et al . , 2010; Wahlster et al . , 2013; Kuzuya et al . , 2016 ) . We found that the conformation of the phosphorylation-inhibited PS1 mutants at these residues is not different from that of the WT PS1 . This is consistent with the findings of unaltered Aβ42/40 ratio in cells expressing PS1 phosphorylation-inhibited mutants ( Matz et al . , 2015 ) . However , the conformation of the phosphorylation-inhibiting mutants and WT PS1 is very different in the Ca2+ influx-stimulated condition . This finding suggests that phosphorylation of PS1 at these residues as a result of the Ca2+ influx/PKA activation underlies its conformational change . Our data show that introducing phosphorylation-inhibited mutations within all six residues prevents the Ca2+ influx/PKA activation-triggered PS1 conformational change . This indicates that phosphorylation of domain 1 ( T74 ) , domain 2 ( S310-S313 ) and domain 3 ( S365-S366-S367 ) is important for PS1 adopting the ‘closed’ conformation . On the other hand , the results using phosphorylation-mimicking mutants revealed that phosphorylation of domain 1 and domain 2 , though necessary , is not sufficient for the structural change of PS1 . Whereas domain 3 , specifically S367 , is critical for PS1 achieving the ‘closed’ conformation . These results beg the questions of how domain 3 is phosphorylated and whether phosphorylation of domain 1 and domain 2 is involved in the Ca2+-triggered PS1 pathogenic conformational change . Although the domain 1 and domain two phosphorylation-mimicking PS1 isoform ( T74D/S310D/S313D G-PS1-R ) shows the same conformation as WT G-PS1-R , PKA activators were able to induce the ‘closed’ conformation of this mutant . Therefore , we conclude that PKA is involved in phosphorylation of domain 3 . We propose the following model: PKA first phosphorylates PS1 at domain 2 , which leads to spatial rearrangements around domain 1 , and allows its phosphorylation . This subsequently changes the local conformation around domain 3 , enabling its phosphorylation by PKA . This final phosphorylation of S365-S366-S367 , with S367 being most crucial , drives the pathogenic conformational alteration of the PS1/γ-secretase ( Figure 6—figure supplement 1A ) . We propose that local spatial rearrangement around domain three is a key step for PKA-induced domain three phosphorylation , and phosphorylation of domain 1 and domain 2 is necessary for this to occur . This model is consistent with the finding that PKA is able to phosphorylate only S310 within domain 2 in recombinant PS1 hydrophilic loop ( 263–407 ) peptide in vitro ( Fluhrer et al . , 2004 ) . We reason that lack of domain 1 in this recombinant PS1 fragment does not allow change in the local conformation around domain 3 , explaining why S365-S366-S367 are not phosphorylated by PKA in this recombinant PS1 . Moreover , a recent structural simulation study has demonstrated that 350–373 residues of PS1 containing domain 3 , are subject to major movements ( Somavarapu and Kepp , 2016 ) . It is highly likely that naturally occurring domain 1 and domain 2 phosphorylation events may trigger these local spatial rearrangements around the 350–373 residues of PS1 . We confirmed that domain 1 or domain 3 phosphorylation-inhibited mutant PS1 were still able to get phosphorylated by PKA at domain 2 ( Figure 6—figure supplement 1B ) , suggesting that domain 2 is the first step of the phosphorylation . Consistently , except for domain 2 , there are no known PKA substrate consensus sequences ( Ubersax and Ferrell , 2007 ) around domain 1 or domain 3 of PS1 ( Figure 6—figure supplement 1C ) . PS1 conformational shifts are reflected by consistent changes in the FRET efficiency , detected by both the antibody-based FLIM assay of endogenous PS1 and by spectral FRET using the G-PS1-R reporter . Of note , all FRET-based assays only report relative changes in the proximity between the fluorophores , which does not necessarily translate into comparable changes in the distance between the protein epitopes ( PS1 NT-CT and NT-loop , in this case ) . However , regardless of which assay we use , increased Ca2+/PKA activation reliably and reproducibly increased the proximity between donor and acceptor fluorophores tagging PS1 NT-CT or NT-loop domains . This strongly suggests that we observe conformational changes in the entire PS1 molecule rather than just a change in the orientation of the donor and acceptor fluorophores . We have also established and validated novel in vivo PS1 conformation assay . Topical KCl application significantly increased intracellular Ca2+ ( mean = ~649 nM , ranging from 166 nM to 2793 nM ) in WT mice brains , a concentration of Ca2+ consistent with that observed in AD mouse models displaying Ca2+ overload in vivo ( mean = ~499 nM , ranging from 162 nM to 1535 nM ) ( Kuchibhotla et al . , 2008 ) . We verified that Ca2+ influx induces the pathogenic ‘closed’ PS1 conformation in vivo via PS1 S365/S366/S367 phosphorylation . The cAMP-triggered PKA activation in mouse brain and consecutive conformational change of the wild type PS1 that we report confirms PKA-mediated PS1 phosphorylation at endogenous levels . More importantly , we show up-regulation of the domain 2 ( S310 ) phosphorylation in AD brains , demonstrating that Ca2+-driven PS1 phosphorylation at certain residues may be important factors contributing to the pathogenesis of AD . Of note , a wide distribution of the phosphorylated PS1 levels observed in AD cases did not correlate with either age or PMI . Since we have selected a relatively homogenous group of AD brains in terms of pathology , Braak and CERAD scores could not be used to explain the wide distribution . Further investigation is necessary to determine if other factors , such as medications taken , disease duration , ApoE genotype , etc . could have affected the p-PS1 levels . We propose a positive feed-forward mechanism for the pathogenesis of AD ( Figure 6 ) : once AD pathology , i . e . Aβ oligomers or deposited Aβ , triggers Ca2+ overload , domains 1 , 2 and 3 of PS1 can be hyper-phosphorylated by PKA . This induces the pathogenic ‘closed’ PS1 conformation , and leads to subsequent elevation of the Aβ42/40 ratio , followed by accelerated Aβ oligomerization or deposition . In line with these findings , we have previously reported that PS1 shows a ‘closed’ conformation around Aβ plaques in the sporadic AD brain ( Wahlster et al . , 2013 ) , and that higher Aβ42/40 ratio leads to the pathogenic conformation of PS1/γ-secretase ( Zoltowska et al . , 2016 ) . It is conceivable that selective inhibition of PS1 phosphorylation at the domain 1 , 2 and/or 3 may present a novel opportunity for therapeutics for AD . Since phosphorylation-inhibited PS1 mutations on these sites do not affect assembly , maturation , or activity of the PS1/γ-secretase ( Matz et al . , 2015 ) , inhibition of these PS1 phosphorylation residues would affect neither the cleavage of Notch nor other γ-secretase substrates . On the other hand , inhibition of PKA activity may not be a suitable approach to prevent PS1 phosphorylation , due to multiple PKA substrates and potential off target effects in the brain and in the periphery . In future studies , PS1-targeting agents that selectively block PS1 phosphorylation at either one of the three residues could be identified , and their efficacy could be examined in vivo using our novel PS1 conformation assays . 10 . 7554/eLife . 19720 . 020Figure 6 . Mechanism of the Ca2+-triggered PS1 pathogenic conformational change . The schematic image of the molecular events involved in the Ca2+-triggered pathogenic ‘closed’ conformational change and increase of the Aβ42/40 ratio . The elevated Ca2+ levels induce PKA activation , followed by the phosphorylation of PS1 at domain 1 , domain 2 and domain 3 . Domain 3 phosphorylation , particularly at S367 , induces the PS1 pathogenic conformation that leads to increase in the Aβ42/40 ratio . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 02010 . 7554/eLife . 19720 . 021Figure 6—figure supplement 1 . Model of the PKA-mediated PS1 phosphorylation . ( A ) Model . PKA first phosphorylates domain 2 , followed by changing local conformation around domain 1 . PKA then phosphorylates domain 1 , which subsequently leads to local rearrangement around domain 3 . Finally , PKA phosphorylates domain 3 , causing PS1 pathogenic conformation . ( B ) The level of phosphorylated PS1 at S310 was compared between forskolin and vehicle-treated 7 W cells expressing WT , domain 1 , 2 or 3 phospho-inhibited PS1 by Western blotting . ( C ) Amino acid sequences around domain 1 , 2 and 3 . Domain 2 includes the PKA substrate consensus sequence: R-R-V-S-K . DOI: http://dx . doi . org/10 . 7554/eLife . 19720 . 021
The N-terminally FLAG-tagged wild type ( WT ) PS1 and the N-terminally FLAG-tagged phosphorylation-inhibited PS1 mutants: S28A PS1 , T74A PS1 , S310A/S313A PS1 , S313A PS1 , S324A PS1 , S337A PS1 , S365A PS1 , S366A PS1 , S367A PS1 , S365A/S367A PS1 and S366A/S367A PS1 , were kind gifts from Dr . Patrick C Fraering ( Brain Mind Institute and School of Life Sciences , Ecole Polytechnique Fédérale de Lausanne , Lausanne , Switzerland ) ( Matz et al . , 2015 ) . The N-terminally FLAG-tagged phosphorylation-inhibited PS1 mutants; S310A PS1 , S319A/T320A PS1 , S346A PS1 , S353A PS1 , T354A PS1 , S357A PS1 and S353A/S357A PS1 , were generated from the N-terminally FLAG-tagged WT PS1 by site-directed mutagenesis using the QuikChange Site-directed mutagenesis kit ( Agilent Technologies , Santa Clara , CA , USA ) . Primers; S310A PS1: 5’-GCTCAAAGGAGAGTAGCCAAAAATTCCAAGTAT-3’ and 5’-ATACTTGGAATTTTTGGCTACTCTCCTTTGAGC-3’ , S319A/T320A PS1: 5’-AAGTATAATGCAGAAGCCGCAGAAAGGGAGTCA-3’ and 5’-TGACTCCCTTTCTGCGGCTTCTGCATTATACTT-3’ , S346A PS1: 5’-GAAGCCCAGAGGGACGCTCATCTAGGGCCTCATCGC-3’ and 5’-GCGATGAGGCCCTAGATGAGCGTCCCTCTGGGCTTC-3’ , S353A PS1: 5’-CTAGGGCCTCATCGCGCTACACCTGAGTCACGA-3’ and 5’-TCGTGACTCAGGTGTAGCGCGATGAGGCCCTAG-3’ , T354A PS1: 5’-CCTCATCGCTCTGCACCTGAGTCACGA-3’ and 5’-TCGTGACTCAGGTGCAGAGCGATGAGG-3’ , S357A PS1: 5’-CTACACCTGAGGCACGAGCTGCTGTCC-3’ and 5’-GGACAGCAGCTCGTGCCTCAGGTGTAG-3’ , S353A/S357A PS1:5’-CCTCATCGCGCTACACCTGAGGCACGAGCTGCT-3’ and 5’-AGCAGCTCGTGCCTCAGGTGTAGCGCGATGAGG-3’ . The phosphorylation-inhibited PS1 mutants; T74A GFP-PS1-RFP ( G-PS1-R ) , S310A/S313A G-PS1-R and S365A/S366A/S367A G-PS1-R , and the phosphorylation-mimicking PS1 mutants; T74D G-PS1-R , S310D/S313D G-PS1-R , S365D/S366D/S367D G-PS1-R , S365D G-PS1-R , S366D G-PS1-R and S367D G-PS1-R were generated from the PS1 conformation sensitive FRET probe , G-PS1-R , encoding WT PS1 with enhanced green fluorescent protein ( GFP ) fused to the NT and red fluorescent protein ( RFP ) inserted into the cytoplasmic loop ( Uemura et al . , 2009 ) by site-directed mutagenesis using the QuikChange Site-directed mutagenesis kit . Primers; T74A G-PS1-R: 5’-GAAGATGAGGAGCTGGCATTGAAATATGGCGCC-3’ and 5’-GGCGCCATATTTCAATGCCAGCTCCTCATCTTC-3’ , S310A/S313A G-PS1-R: 5’-AGGAGAGTAGCCAAAAATGCCAAGTATAATGCA-3’ and 5’-TGCATTATACTTGGCATTTTTGGCTACTCTCCT-3’ , S365A/S366A/S367A G-PS1-R: 5’-GTCCAGGAACTTGCCGCCGCTATCCTCGCTGGT-3’ and 5’-ACCAGCGAGGATAGCGGCGGCAAGTTCCTGGAC-3’ . Primers; T74D G-PS1-R: 5’-GAAGATGAGGAGCTGGACTTGAAATATGGCGCC-3’ and 5’-GGCGCCATATTTCAAGTCCAGCTCCTCATCTTC-3’ , S310D/S313D G-PS1-R: 5’-GCTCAAAGGAGAGTAGACAAAAATGACAAGTAT-3’ and 5’-ATACTTGTCATTTTTGTCTACTCTCCTTTGAGC-3’ , S365D/S366D/S367D G-PS1-R: 5’-GTCCAGGAACTTGACGACGATATCCTCGCTGGT-3’ and 5’-ACCAGCGAGGATATCGTCGTCAAGTTCCTGGAC-3’ . S365D G-PS1-R: 5’- GCTGTCCAGGAACTTGACAGCAGTATCCTCGCT-3’ and 5’- AGCGAGGATACTGCTGTCAAGTTCCTGGACAGC-3’ . S366D G-PS1-R: 5’- GTCCAGGAACTTTCCGACAGTATCCTCGCTGGT-3’ and 5’- ACCAGCGAGGATACTGTCGGAAAGTTCCTGGAC-3’ . S367D G-PS1-R: 5’- CAGGAACTTTCCAGCGATATCCTCGCTGGTGAA-3’ and 5’- TTCACCAGCGAGGATATCGCTGGAAAGTTCCTG-3’ . T74D/S310D/S313D G-PS1-R was generated from S310D/S313D G-PS1-R , T74D/S365D/S366D/S367D G-PS1-R was from S365D/S366D/S367D G-PS1-R , S310D/S313D/S365D/S366D/S367D G-PS1-R was from S365D/S366D/S367D G-PS1-R and T74D/S310D/S313D/S365D/S366D/S367D G-PS1-R was from S310D/S313D/S365D/S366D/S367D G-PS1-R using primer sets shown above . The constructs with GFP fused to the NT of PS1 ( G-PS1 , donor only ) , RFP inserted into the large hydrophilic loop domain of PS1 ( PS1-R acceptor only ) , and RFP-GFP fusion ( R-G fusion ) plasmid , in which RFP is fused to the NT of GFP with a short linker , were used as controls for the FRET assays ( Uemura et al . , 2009 ) . R-G324D- ( Kin-8 ) , a dominant negative form of the PKA regulatory subunit α , was obtained from Addgene ( plasmid # 28179 ) ( Olson et al . , 1993 ) . The mouse monoclonal anti-PS1 NT ( RRID:AB_301867 ) , rabbit monoclonal anti-PS1 loop ( RRID:AB_1310605 ) , rabbit monoclonal anti-phosphorylated PS1 at S310 ( RRID:AB_1267317 ) , and mouse monoclonal anti-phosphoserine ( RRID:AB_305611 ) antibodies were purchased from Abcam ( Cambridge , MA , USA ) . The rabbit polyclonal anti-PS1 CT ( RRID:AB_261178 ) and mouse monoclonal anti-FLAG ( RRID:AB_259529 ) antibodies were from Sigma-Aldrich ( St . Louis , MO , USA ) . The mouse monoclonal anti-PS1 loop ( RRID:AB_95175 ) , rabbit polyclonal anti-phosphoserine ( RRID:AB_390205 ) and rabbit polyclonal anti-phosphorylated CREB at S133 ( RRID:AB_310153 ) antibodies were from Millipore ( Temecula , CA , USA ) . Rabbit monoclonal CREB antibody ( RRID:AB_309979 ) was from Upstate Biotechnology ( Lake Placid , NY , USA ) . Mouse ( RRID:AB_737182 ) and rabbit ( RRID:AB_737197 ) normal IgG were from Santa Cruz Biotechnology , Inc . ( Dallas , TX , USA ) . Alexa Fluor 488 ( A488 ) , and Cy3-labeled corresponding secondary antibodies were from Life Technologies ( Grand Island , NY , USA ) and horseradish peroxidase ( HRP ) -conjugated secondary antibodies were from Pierce ( Rockford , IL , USA ) . The Ca2+ ionophore A23187 , PKA inhibitors H-89 and KT5720 , PKA activators forskolin and 8-Bromoadenosine 3’:5’-cyclic monophosphate ( 8-Bromo-cAMP ) , JNK inhibitor SP600125 , PKC inhibitor Ro 31–8220 , and GSK3β inhibitor TDZD-8 were from Sigma-Aldrich ( St . Louis , MO , USA ) . CHO cell lines stably overexpressing WT APP ( 7 W cells ) ( Koo and Squazzo , 1994 ) and stably overexpressing WT APP and WT PS1 ( PS70 cells ) ( Xia et al . , 1997 ) were a kind gift from Dr . Dennis Selkoe ( Brigham and Women’s Hospital , Harvard Medical School , USA ) . PS1/PS2 double knockout mouse embryonic fibroblasts ( PS1/2 dKO MEF ) were a kind gift from Dr . Bart De Strooper ( Katholieke Universiteit Leuven , Leuven , Belgium ) . These cells were authenticated using STR profiling , monitored for mycoplasma contamination every 2 months , and maintained as described previously ( Uemura et al . , 2010 ) . Lipofectamine 3000 ( Life technologies , Carlsbad , CA , USA ) was used for transient transfection according to the manufacturer’s instructions . Primary neuronal cultures were obtained from cerebral cortex and hippocampus of mouse embryos at gestation day 14–16 ( Charles River Laboratories , Wilmington , MA ) . The neurons were dissociated using Papain Dissociation System ( Worthington Biochemical Corporation , Lakewood , NJ , USA ) and were maintained for 13–15 days in vitro ( DIV ) in Neurobasal medium containing 2% B27 supplement , 1% GlutaMax , and 1% Pen/Strep mix ( Life Technologies ) . All experimental procedures using mice were in compliance with the NIH guidelines for the use of animals in experiments and were approved by the Massachusetts General Hospital Animal Care and Use Committee . In this study , we developed five constructs that were packaged into viruses: WT G-PS1-R ( pAAV8-human Synapsin1 ( hSyn1 ) -WT G-PS1-R ) , S365A/S366A/S367A G-PS1-R ( pAAV8-hSyn1-S365A/S366A/S367A G-PS1-R ) , G-PS1 ( pAAV8-hSyn1-G-PS1 ) , PS1-R ( pAAV8-hSyn1-PS1-R ) and RFP-GFP fusion ( pAAV8-hSyn1-R-G ) . These plasmids were constructed by cloning corresponding PS1 constructs into an AAV8 vector containing the hSyn1 promoter and the woodchuck hepatitis virus posttranscriptional regulatory element ( WPRE ) sequences , using NheI/HindIII or HindIII/EcoRI restriction sites . All plasmids were packaged in AAV serotype 8 at the University of Pennsylvania Vector Core . Viral titers for all constructs were between 4 . 9 × 1012 GC/ml to 1 . 5 × 1013 GC/ml . The construction of pAAV2-CBA-Yellow Cameleon 3 . 6 ( YC3 . 6 ) was described previously ( Kuchibhotla et al . , 2008 ) . Experimental procedures for viral injection have been described previously ( Kuchibhotla et al . , 2008 ) . Briefly , five to seven months old male C57BL/6 mice ( Charles River Laboratories , Wilmington , MA , USA ) were used for intra-cortical injection of the viruses . Mice were anesthetized with 1–3% isoflurane and placed in a stereotaxic apparatus . The body temperature of mice was maintained with a heating system throughout the procedures . The surgical region was sterilized with isopropyl alcohol and betadyne , and burr holes were drilled in the skull . Viruses were injected into the somatosensory cortex ( 2 mm lateral , 1 . 5 mm posterior to Bregma , 0 . 7 mm deep ) using a Hamilton syringe from Sigma-Aldrich ( St . Louis , MO , USA ) at a speed of 130 nl/min . After injection , the skin was sutured and treated with Fougera Triple Antibiotic Ointment ( E . FOUGERA and CO . , Melville , NY , USA ) . One month after viral injection , mice were anesthetized with 1–3% isoflurane and placed in a stereotaxic apparatus . A circular area of skull was carefully drilled , removed , and then dura was carefully removed . A circular cranial window soaked in PBS was implanted to the brain surface , and fixed to the remaining skull with dental cement mixed with Krazy glue . For topical KCl application , a burr hole was created in the fixed dental cement to inject KCl with a Hamilton syringe into the space between the cranial window and brain surface . Five to seven months old male C57BL/6 mice were anesthetized as described above , placed into the stereotax , and injected with 1 µl of 100 mM 8-Bromo-cAMP ( right hemisphere ) or vehicle ( left hemisphere ) into the somatosensory cortex . 5 min post-injection , mice were sacrificed using CO2 asphyxiation , perfused intracardially with PBS followed by 4% PFA . For the rescue experiments , 1 µl of 100 µM KT5720 ( right hemisphere ) or vehicle ( left hemisphere ) was injected 75 min prior to 8-Bromo-cAMP treatment . The brains were dissected , post-fixed by immersion in 4% paraformaldehyde ( PFA ) with 15% glycerol ( Sigma-Aldrich , St , Louis , MO ) in PBS , and cryoprotected using 30% glycerol in PBS . Prior to immunostaining , the brains were cut on a freezing sliding microtome ( Leica SM 2000R , Bannockburn , IL ) into 35 μm-thick coronal sections . Brain tissue sections were permeabilized using 0 . 4% TX-100 . For 4% PFA fixed 7 W cells and primary neurons 0 . 1% TX-100 was used . Non-specific binding of the antibodies was blocked by incubation of the cultured cells or free-floating brain sections with 1 . 5% normal donkey serum ( Jackson ImmunoResearch Labs , West Grove , PA ) . The primary antibodies were applied overnight at 4°C , followed by 1 hr incubation at room temperature with corresponding Alexa Fluor 488- or Cy3-conjugated secondary antibodies . The samples were mounted in VectaShield mounting medium ( Vector Laboratories , Inc . , Burlingame , CA ) . The FLIM analysis of PS1 conformation was performed as described previously ( Berezovska et al . , 2005 ) . Briefly , a mode-locked Chameleon Ti:Sapphire laser ( Coherent Inc . , Santa Clara , CA ) set at 850 nm or 770 nm was used to excite GFP or Alexa Fluor 488 donor fluorophores , respectively . A Zeiss LSM510 microscope and x63 oil immersion objective was used for the imaging . The donor fluorophore lifetimes were recorded using a high-speed photomultiplier tube ( MCP R3809; Hamamatsu , Bridgewater , NJ ) and a time-correlated single-photon counting acquisition board ( SPC-830; Becker and Hickl , Berlin , Germany ) . The baseline lifetime ( t1 ) of the donor fluorophore was measured in the absence of the acceptor fluorophore ( negative control , FRET absent ) . In the presence of the acceptor fluorophore ( RFP or Cy3 ) , excitation of the donor fluorophore results in reduced donor emission energy if the donor and acceptor are less than 5–10 nm apart ( FRET present ) . This yields characteristic shortening of the donor fluorophore lifetime ( t2 ) . The acquired FLIM data were analyzed using SPC Image software ( Becker and Hickl , Berlin , Germany ) by fitting the data to one ( negative control ) or two ( acceptor present ) exponential decay curves . In the latter case , t1 of the non-FRETing population was ‘fixed’ and thus excluded from the analysis , and only shorter , t2 , values were analyzed . The FRET efficiency ( %EFRET ) was calculated using the following equation: %EFRET = 100* ( t1-t2 ) /t1 . Higher %EFRET reflects closer proximity between fluorophores labeling the PS1 domains . The spectral FRET assay with single photon excitation for the experiments using cultured cells and immunostained mouse brain sections was conducted as described previously ( Uemura et al . , 2009 ) . Briefly , an Argon laser at 488 nm was used to excite GFP or Alexa 488 , and emitted fluorescence was detected by seven channels of the Zeiss Metadetector within the 502–651 nm or 511–682 nm wavelength range ( 21 . 4 nm spectral bandwidth for each channel ) on a Zeiss LSM510 microscope . Average pixel fluorescence intensity for the whole cell after subtraction of the background fluorescence was measured using Image J . The ratio of fluorescence intensity in the 598 nm channel ( for RFP ) to that in the 513 nm channel ( for GFP ) or 565 nm ( Cy3 ) to 522 nm ( Alexa 488 ) was used as a readout of the FRET efficiency , which reflects the relative proximity between the donor and acceptor . The spectral FRET assay for monitoring PS1 conformation in living mouse brain using two-photon excitation is newly established . First , to determine the excitation wavelength that preferentially excites GFP , the G-PS1-R probe was excited at different wavelengths from 750 nm to 975 nm with a mode-locked titanium/sapphire laser ( MaiTai; Spectra-Physics , Fremont , CA ) . The 900 nm wavelength was chosen to selectively excite GFP , and emitted fluorescence was detected by two emission channels: 495–540 nm range for channel 1 ( for GFP ) and 575–630 nm for channel 2 ( for RFP ) , on an Olympus Fluoview 1000 MPE microscope ( x20 objective , water immersion , NA = 1 . 05 ) ( Olympus Corporation , Tokyo , Japan ) . Time-lapse images were obtained every 10 s for a duration of 2 min . The average pixel fluorescence intensity after subtraction of the background fluorescence for the whole cell was measured using ImageJ in each channel . The R/G ratio was used as readout of the FRET efficiency . Pseudo-colored images were generated in MATLAB . Intracellular Ca2+ levels in 7 W cells were determined using the ratiometric Ca2+-sensitive dye Indo-1 ( Grynkiewicz et al . , 1985 ) . Briefly , Indo-1/AM ( Thermo Fisher Scientific , Inc . , Cambridge , MA ) was dissolved with 20% pluronic F-127 ( Thermo Fisher Scientific , Inc . ) in DMSO and added to the culture dishes at a final concentration of 1 μM Indo-1/AM and 0 . 02% pluronic F-27 for 45 min . Images were obtained using a Zeiss LSM510 microscope ( x25 water immersion objective , Ca2+/Mg2+ containing PBS , 37°C , 5% CO2 ) . A Chameleon Ti:Sapphire laser was used for excitation at 750 nm , and the emitted fluorescence was detected in two channels: 390–465 nm and 500–550 nm . Intraneuronal Ca2+ levels in the somatosensory cortex of living mice was measured using the FRET-based ratiometric probe , Yellow Cameleon 3 . 6 ( YC3 . 6 ) ( Nagai et al . , 2004 ) , as described previously ( Kuchibhotla et al . , 2008 ) . Briefly , a mode-locked titanium/sapphire laser ( MaiTai; Spectra-Physics , Fremont , CA ) at 860 nm was used for excitation of CFP , and the emitted fluorescence was detected in two channels: 460–500 nm ( for CFP ) and 520–560 nm ( for YFP ) ( x20 objective ) . Time-lapse images were obtained every 10 s for 2 min . Pseudo-colored images were generated in MATLAB . The ratiometric analysis of the YFP/CFP ratio and quantitative determination of the [Ca2+] concentrations was conducted as described previously ( Kuchibhotla et al . , 2008 ) . The levels of secreted Aβ40 and Aβ42 were measured using ELISA kits according to the manufacturer’s instructions ( Wako , Osaka , Japan ) . Primary neurons were lysed in radio-immunoprecipitation assay ( RIPA ) buffer with protease and phosphatase inhibitor cocktail ( Fisher Scientific , Pittsburg , PA , USA ) . The cells were homogenized and incubated for 30 min on ice . Each sample was then centrifuged , and the supernatants were collected . Protein concentrations were determined using a Pierce BCA Protein Assay Kit ( Thermo Fisher Scientific , Inc . ) . Aliquots with equal amount of total protein were treated with protein G-Sepharose ( GE Healthcare Bio-Sciences , Uppsala , Sweden ) for 1 hr . After removing protein G-Sepharose by centrifugation , primary antibodies were added to the supernatants . Each sample was rotated for 1 hr and then treated with protein G-Sepharose for 1 hr . The immunoprecipitates were washed with RIPA buffer and resuspended in 2 x LDS sample buffer ( Life Technologies ) . After boiling and centrifugation , the supernatants were subjected to SDS-PAGE on NuPAGE Bis-Tris gels ( Life Technologies ) . Frontal cortex of frozen human brains from neuropathologically verified AD cases were obtained from the brain bank of the Alzheimer’s Disease Research Centre ( ADRC ) at Massachusetts General Hospital , and were lysed by RIPA buffer with protease and phosphatase inhibitor cocktail ( Fisher Scientific ) and calyculin A ( Cell Signaling Technology , Danvers , MA , USA ) . Control cases were non-demented individuals who did not meet pathological diagnostic criteria of AD or any other neurodegenerative diseases ( Table 2 ) . All the subjects or their next of kin gave informed consent for the brain donation . Statistical analysis was performed using StatView for Windows , Version 5 . 0 . 1 or GraphPad Prism 5 . All values are given as means ± SEM . Comparisons were performed using One sample t-test , Student’s t-test , or one-way factorial ANOVA followed by Bonferroni’s post-hoc analysis . For the in vivo time course experiments monitoring PS1 conformation and Ca2+ concentration , two-way repeated-measures ANOVA and Bonferroni post-hoc analysis were used . For the correlation analysis in the human brain-based experiments , Spearman's nonparametric correlation analysis was used . An appropriate sample size was computed when the study was being designed and each experiment was performed using at least three independent trials . p<0 . 05 was considered to indicate a significant difference . | Alzheimer’s disease is a widely recognised disorder caused by the progressive deterioration and death of brain cells . A key feature of the disease is the formation of structures called plaques in the brain . Plaques occur when many copies of a molecule known as amyloid beta stick together outside of the brain cells . Healthy brains also produce amyloid beta but it is in a different form , which cannot form plaques . One in twenty people with Alzheimer’s disease have a family history of the disease . Of these , many are linked to changes in a gene that produces a protein called Presenilin 1 ( or PS1 for short ) . Cells need PS1 to make amyloid beta and the altered versions of PS1 produce the type of amyloid beta that causes Alzheimer’s disease . Yet , in cases that do not run in families , the gene for PS1 is unchanged but the PS1 protein still produces the form of amyloid beta that is linked to Alzheimer’s disease . Maesako , Horlacher et al . wanted to find out how seemingly healthy PS1 proteins can be made to produce plaque-forming amyloid betas . Studies of PS1 from mice revealed that small chemical modifications , called phosphate groups , could be attached to PS1 in a process called phosphorylation . Modified PS1 proteins produce harmful amyloid betas and removing the modifications was enough to make PS1 behave normally again . Maesako , Horlacher et al . found three points in the PS1 protein where phosphorylation could change the behaviour of the protein , the most important one is a site called Ser367 . Further investigation showed that an enzyme called Protein Kinase A ( PKA ) phosphorylates PS1; this enzyme is also able to attach phosphate groups to many different proteins . Maesako , Horlacher et al . went on to show that PS1 is phosphorylated in samples from people with Alzheimer’s disease , suggesting that this is a plausible cause for some cases of the disease . Finding a way to prevent phosphorylation or remove phosphate groups from PS1 could be the first step towards treating these cases of Alzheimer’s disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"neuroscience"
] | 2017 | Pathogenic PS1 phosphorylation at Ser367 |
We test the hypothesis that the posterior parietal cortex ( PPC ) contributes to the control of visually guided locomotor gait modifications by constructing an estimation of object location relative to body state , and in particular the changing gap between them . To test this hypothesis , we recorded neuronal activity from areas 5b and 7 of the PPC of cats walking on a treadmill and stepping over a moving obstacle whose speed of advance was varied ( slowed or accelerated with respect to the speed of the cat ) . We found distinct populations of neurons in the PPC , primarily in area 5b , that signaled distance- or time-to-contact with the obstacle , regardless of which limb was the first to step over the obstacle . We propose that these cells are involved in a sensorimotor transformation whereby information on the location of an obstacle with respect to the body is used to initiate the gait modification .
Navigation in cluttered environments dictates that animals and humans determine their relationship to stationary and moving objects for the purposes of avoidance or interception; these behaviors are essential for survival . Everyday examples of such activities range from the simple , such as stepping over a stationary obstacle and stepping up or down from a curb , to the more complex , such as adjusting one’s gait to kick a moving soccer ball or stepping on or off a moving conveyor belt at the airport . Inherent in this process is the requirement to detect the presence of an obstacle , estimate its location with respect to the body , take into account the rate of gap closure ( between body and object ) , and then use this information to appropriately modify the gait pattern . We suggest that the posterior parietal cortex ( PPC ) makes an essential contribution to this process . Gibson , in his seminal work ( Gibson , 1958 ) , argued that animals could use optic flow to gauge distance and location to an obstacle , and thus modify gait to avoid it . Several studies have since confirmed this premise ( Prokop et al . , 1997; Sun et al . , 1992; Warren et al . , 2001 ) . Later , Lee ( 1976 ) suggested that the brain extracts information about the time-to-contact ( TTC ) with an object from optic flow , a variable he called tau , and that this could be used to control gait . Again , multiple studies on locomotion ( Shankar and Ellard , 2000; Sun et al . , 1992 ) and other movements involving interception of moving targets ( Merchant et al . , 2004; Merchant and Georgopoulos , 2006 ) support this theory . However , it is important to note that tau is only one of several variables available to the brain to avoid an obstacle; distance and other time-related optical variables may also contribute ( Sun and Frost , 1998; Tresilian , 1999 ) . In agreement with these behavioral studies , there is evidence from invertebrates to suggest that neurons can process optic flow for motor activities such as flight control , distance travelled , and landing ( Baird et al . , 2013; Egelhaaf and Kern , 2002; Fotowat and Gabbiani , 2011; Srinivasan and Zhang , 2004 ) . Similarly , the pigeon has neurons in the nucleus rotundus that can extract optic flow and other variables , such as tau , from visual stimuli and which could be used for object avoidance ( Sun and Frost , 1998; Wang and Frost , 1992 ) . In mammals , multiple cortical structures analyze optic flow , including the middle temporal ( MT/V5 ) and medial superior temporal ( MST ) cortices ( Duffy and Wurtz , 1995 , 1997; Orban , 2008 ) , as well as the PPC , the premotor cortex , and even the motor cortex ( Merchant et al . , 2001 , 2003; Schaafsma and Duysens , 1996; Siegel and Read , 1997 ) . However , the important question of how the mammalian nervous system uses optic flow information to guide movement has been studied in only a few behaviors ( Merchant and Georgopoulos , 2006 ) , and then only for arm movements . In this manuscript , we extend these studies by determining how neural structures treat visual information for the control of gait . Our previous work demonstrates the presence of neuronal activity in the PPC that begins several steps before the step over the obstacle and that could contribute to planning ( Andujar et al . , 2010; Drew and Marigold , 2015; Marigold et al . , 2011 ) . In this manuscript , we test the hypothesis that the PPC contributes to obstacle avoidance by constructing an estimation of an approaching object’s location relative to the body’s current state , and in particular the diminishing gap between them and its relation to the ongoing step cycle of each limb ( gap closure ) . We show the presence of neurons in the PPC that code either distance-to-contact ( DTC ) or TTC , and we suggest that this discharge represents the starting point of a complex sensorimotor transformation involved in planning the gait modification .
We trained cats to step over moving obstacles attached to a treadmill . The cats performed the task in a relatively stereotypical manner ( Lajoie and Drew , 2007 ) . Measurements of the DTC and of the TTC with the obstacle both decreased monotonically to a value of ~0 at the onset of the step over the obstacle ( Figure 1A ) . To dissociate between cells potentially related to either DTC or TTC , we recorded cell activity in two complementary locomotor tasks . In one , the obstacle advanced toward the cat at the same speed as the treadmill belt on which the cat walked ( matched task ) while in the other , the obstacle advanced at a slower speed ( visual dissociation task: Drew et al . , 2008; Lajoie and Drew , 2007 ) . As the speed of the treadmill on which the cat walked was the same in both tasks ( 0 . 45 m . s−1 in these experiments ) , DTC and TTC are a function of the speed of the advancing obstacle ( Figure 1B ) . For example , in the matched task , when DTC = 45 cm , TTC = 1000 ms . However , in the visual dissociation task ( obstacle speed slowed to 0 . 3 m . s−1 ) , when DTC = 45 ms , TTC = 1500 ms . This dissociation of DTC and TTC is a fundamental part of our analysis of the contribution of cells in the PPC to the estimation of gap closure . In brief , a cell in which the onset of activity is determined by TTC will discharge at the same time relative to the onset of the step over the obstacle in both the matched and the visual dissociation tasks ( red vertical line at 1000 ms in Figure 1C , D ) . In contrast , during the visual dissociation task , a cell related to DTC would discharge relatively earlier , indicative of the longer time required to cover the fixed distance ( green vertical line at 45 cm in Figure 1C , D ) . The present report is based on 67 cells recorded from two cats ( 37 from cat PCM7 and 30 from PCM9 ) , selected from a much larger database on the basis of the criteria provided in the Materials and methods . These neurons were primarily recorded from the posterior bank of the ansate sulcus , the adjoining lateral bank of the lateral sulcus and the adjacent gyrus between these two sulci , corresponding to area 5b of the PPC; some cells were also recorded from the border region of area 5/7 ( see Figure 2 ) . Some of these cells ( 42/67 ) were included in a previous report ( Marigold and Drew , 2011 ) . Note that we recorded all cells from the right PPC and that the left limb is therefore contralateral to the recording site . As indicated in the preceding section , a cell in which the change in discharge activity is related to the distance between the obstacle and the cat would be expected to become active earlier in the visual dissociation task than in the matched task . An example of such a cell is illustrated in Figure 3 . In this example , cell discharge was low and tonic during unobstructed locomotion ( blue traces ) and then showed a distinct increase in discharge in the two steps before the obstacle both in the left limb leads condition and in the right limb leads condition . This discharge peaked just prior to the onset of the flexor muscle activity during the step over the obstacle ( represented by the black vertical line ) before decreasing to , or below , control levels ( blue trace ) , as the lead limb stepped over the obstacle . This general behavior occurred in both the matched task ( red traces ) and in the visual dissociation task ( green traces ) . The increase in cell discharge , as calculated from the average of the onsets in the individual trials ( see Figure 3—figure supplement 1 ) , began 514 ± 124 ms before the step over the obstacle in the left lead condition of the matched task and 511 ± 231 ms in the right lead condition of the same task ( red vertical lines ) . In the visual dissociation task , the respective values were 745 ± 161 ms and 723 ± 187 ms ( green vertical lines ) . A t-test showed a significant difference in the time of the onset of cell discharge between the matched and visual dissociation task , both for the left limb leads condition ( p<0 . 001 ) and for the right limb leads condition ( p=0 . 002 ) . The activity pattern illustrated in Figure 3 is compatible with the hypothesis that DTC determined the onset of the discharge activity ( see Figure 1B–D ) . Importantly , we also verified that this cell discharge was not step-related . As shown in Figure 3D ( left ) , there is no relationship between the onset of cell discharge and the onset of the activity in the coBr muscle in the step before the obstacle ( a and c in Figure 3C ) . Rather , as expected on the basis of the similarity in the time of onset of cell discharge independent of which limb leads , the relationship to coBr forms two populations , separated by ~500 ms ( ~1 step or half the duration of the step cycle ) . Conversely , discharge onset in the cell overlaps with the onset of activity in the coBr when the left limb leads but with the iBr when the right limb leads ( Figure 3D , middle ) . Overall , we found no constant relationship between the onset of discharge activity and step-related activity in a given limb in this cell . In contrast , we found a significant relationship between the time of the end of cell discharge and the onset of activity in the flexor muscle of the lead limb during the step over the obstacle ( Figure 3D right ) as would be expected for a limb-independent cell ( see Materials and methods ) . We observed similar relationships to those illustrated in Figure 3D in all cells in our database . The cell illustrated in Figure 3 , putatively identified as related to distance based on its relatively earlier discharge onset in the visual dissociation task , is displayed in Figure 4A together with the averaged calculated traces of DTC and TTC . The vertical green and red lines intersect these traces at the average time of onset of the cell discharge as calculated from the individual trials ( see Figure 3 ) . Because the obstacle advanced more slowly in the visual dissociation task , the slope of the DTC trace in this task ( diagonal green trace ) is less than that in the matched task ( diagonal red trace ) . As a consequence , cell discharge begins at approximately the same distance in the two tasks ( horizontal black line ) . In contrast , the interception with the bottom pair of traces shows that cell onset begins at different TTCs in the two tasks . We quantified this relationship with a one-way ANOVA across the four situations , which demonstrated a significant effect for TTC ( Figure 4B bottom ) . Post-hoc t-tests with Bonferroni corrections showed significant differences with task ( matched versus visual dissociation ) but no significant differences with condition ( left or right lead ) . In contrast , the ANOVA showed a non-significant effect with DTC ( Figure 4B , top ) , effectively indicating that the onset of cell discharge occurred when the obstacle was a constant distance from the cat , regardless of task or lead limb . The opposite situation is shown for the cell in Figure 4C . In this case , we found a significant effect of DTC ( Figure 4D , top ) on the onset of cell discharge . Post hoc t-tests showed significant differences ( asterisks ) between the matched right lead situation and both conditions in the visual dissociation task . However , we found no significant effect of TTC on cell discharge ( Figure 4D , bottom ) , indicating that the onset of cell discharge occurred when the obstacle was a constant TTC regardless of task or condition . Overall , we identified two populations of cells related to gap closure: one group in which the onset of discharge activity occurred at a constant DTC , and a second group in which the onset of discharge occurred at a constant TTC . Figure 5 summarizes the results of the ANOVAs calculated from a population of 51/67 cells that we recorded in all four situations ( matched and visual dissociation task , left and right lead ) and for which we could measure the onset of the period of discharge activity from individual trials during the step over the obstacle ( see Figure 3—figure supplement 1 ) . To identify DTC-related cells , we determined those cells—similar to the example in Figure 4A—in which cell discharge varied significantly with respect to TTC ( p<0 . 01 ) but non-significantly with respect to DTC ( p>0 . 01 ) . A total of 14/51 cells fulfilled this criterion . These cells are plotted as the cyan traces in Figure 5A and are encompassed by the cyan square in Figure 5C . A further 15/51 cells fulfilled the inverse criterion ( red traces in Figure 5B and red square in Figure 5C ) and are identified as TTC-related cells . A further 6/51 cells showed significant effects ( p<0 . 01 ) of both distance and time ( green lines in Figure 5A , B left plots and green square in Figure 5C ) and 16/51 cells showed non-significant effects of both time and distance ( gray lines in Figure 5A , B , middle and right , and gray square in Figure 5C ) . As indicated in Figure 2 , cells defined as DTC- and TTC-related were found throughout the explored region of area 5b of the PPC as well as in the border area between areas 5 and 7 . No clustering of categories was observed . Cells of both categories were recorded from each cat ( 9 TTC-related cells from cat PCM7 and 6 from cat PCM9; 6 DTC-related cells from cat PCM7 and 8 from cat PCM9: see Figure 2 for distribution ) . To determine the extent to which these two populations are distinct , we calculated an index ( see Materials and methods ) , based on the difference in standardized discharge rate between the visual dissociation and the matched task , for all cells showing a significant relationship to TTC or DTC . As illustrated in Figure 5D , the two populations ( with the exception of 1 cell ) were clearly separated one from the other , reflecting the difference in their discharge profiles with respect to DTC and TTC . Cells showing a significant relationship with both TTC and DTC ( green circles ) divided into one group or the other . A bootstrapping exercise performed on the DTC- and TTC-related cells supports our contention that such cells form two distinct categories ( Figure 5—figure supplement 1 ) . Of the cells showing a constant relationship to TTC across conditions and task , most began to discharge when the obstacle was between 300 and 1000 ms from the cat ( Figure 5B , right and 5E , top ) . The cells related to DTC discharged when the obstacle was 20 to 40 cm away from the cat ( Figure 5A , right and Figure 5E , bottom ) . These values are compatible with cell discharge beginning one or two steps before the step over the obstacle . The fact that cell onset occurred at varying DTC and TTC ( Figure 5E ) implies that there is a sequential activation of cells included within each of our two major populations . This is illustrated in Figure 6A , D for the matched task for five representative DTC-related cells ( A ) and five TTC-related cells ( D ) . When considering the overall discharge activity of both the population of DTC-related cells ( Figure 6B , C ) and of the TTC-related cells ( Figure 6E , F ) , this staggered onset , together with the progressive increase in discharge observed within individual cells ( Figures 3 and 4 ) , resulted in a prolonged and progressive increase in discharge activity , for both left and right limb lead conditions . This ramp increase begins two to three steps before the step over the obstacle for the DTC-related population and slightly later for the TTC-related population . Moreover , as expected on the basis of the individual examples , the onset of this increase in activity occurs earlier in the visual dissociation task ( green traces ) than in the matched task ( red traces ) for the DTC-related cells , but at the same time for the population of TTC-related cells . The peak of the discharge activity , for both populations , occurs just before or after the onset of the gait modification . It is also noticeable in Figure 6B , C that the onset of the change in activity in the DTC-related population activity in the visual dissociation condition actually begins five to six steps before the step over the obstacle , well before the more prominent burst of activity on which we concentrate . This early increase in discharge is not simply an effect of smearing but represents a propensity for some of this population of cells to show a more tonic increase in the visual dissociation task ( see Figure 6—figure supplement 1 ) . Using all the cells included in the analysis for Figure 5 in the population averages did not change the general form of the discharge ( see Figure 6—figure supplement 1 ) . The populations still showed a clear ramp discharge that peaked at around the time of the onset of the gait modification . This suggests that even those cells without a significant relationship to DTC or TTC participate in the planning of the gait modification . To further probe the relationship between cell activity and gap closure , we also manipulated the relationship between DTC and TTC by accelerating the obstacle several steps prior to the step over it . This acceleration , which we always applied during the visual dissociation task , produced major changes in the organization of the sequence of steps prior to the step over the obstacle , as illustrated in Figure 7A , B . In particular , in all cases , the acceleration produced a change in the sequence of steps such that the limb that stepped over the obstacle was the opposite of that predicted on the basis of the unperturbed sequence . As an example , in Figure 7A , the top illustration represents the step sequence during the unperturbed situation in the visual dissociation task ( right limb lead ) . The sequence of steps is regular , and the cat places the left paw just in front of the obstacle before stepping over it with the right forelimb ( green curved arrow ) . The next sequence down ( condition 1L ) shows the situation when we applied the obstacle acceleration at the onset of the left stance of the left limb , three steps before the predicted step over the obstacle , as indicated by the filled orange box . This accelerated the obstacle quickly toward the cat so that instead of lifting up the left limb in step −1 and placing it in front of the obstacle , as in the unperturbed situation , it instead stepped over the obstacle ( see also Figure 7—figure supplement 1 ) . In the third sequence ( condition 2L ) , we initiated the acceleration two steps earlier ( −5 , filled cyan box ) . As in the preceding sequence , the acceleration of the obstacle reduced the distance between the cat and the obstacle and reset the step sequence , again resulting in the cat stepping over the obstacle with the left limb . Note that the distance of the obstacle from the cat in the right limb in step −4 is similar in all three sequences ( vertical orange line ) while in step −1 the sequence is reversed with respect to that seen in the unperturbed situation ( green vertical line ) , supporting the assertion that the acceleration reversed the sequence of steps over the obstacle . We observed a complementary situation for step sequences in which the cat would normally step over the obstacle with the left limb in the absence of the acceleration ( Figure 7B ) . For example , in the second trace down ( condition 1R ) , the sudden acceleration of the obstacle in step −2 resulted in the cat lifting the right limb ( step −1 ) over the obstacle instead of placing it down and taking an extra step as it did in the unperturbed situation . In the 2R condition , an acceleration applied in step −4 ( filled orange box ) likewise resulted in the loss of a step and a reversal of the expected pattern of activity . As in Figure 7A , the orange and green vertical lines demonstrate the reversal of the sequence . One of the major effects of the acceleration was to decrease the time taken to close the gap between the cat and the obstacle for a given DTC or TTC . In the example illustrated in Figure 7C ( same DTC-related cell as in Figure 3 ) , cell discharge in the unperturbed visual dissociation task ( green trace ) began 745 ms before the step over the obstacle and at a DTC of 30 . 6 cm . In the acceleration task , we applied the acceleration in step −3 of the coBr ( orange box ) , when the obstacle was 39 . 4 cm from the cat . As the obstacle accelerated toward the cat , the cell started to discharge at a DTC of 30 . 2 cm . However , because of the acceleration , this discharge occurred only 403 ms before the cat stepped over the obstacle , resulting in the relative delay of the onset of cell discharge in the acceleration task ( purple trace ) compared to the visual dissociation task ( see Figure 7—figure supplement 2 ) . However , the projection of the average cell onset in the three tasks ( vertical lines ) onto the DTC traces confirms that the discharge in all three tasks occurred at the same DTC ( see values at top left of Figure 7C: DTC ) . In contrast , the projection onto the TTC trace shows that cell onset varied between the three tasks . The constant relationship with DTC for this neuron held also for the 2R acceleration condition as illustrated in Figure 7D . In this condition , cell discharge in the visual dissociation condition began 723 ms before the step over the obstacle at a DTC of 29 . 7 cm . We applied the acceleration in step −4 , when the obstacle was still 56 cm from the cat and , as a result , obstacle velocity had almost returned to its pre-acceleration speed when the cell began to discharge ( cyan trace ) at a DTC of 36 . 4 cm and 733 ms before the step over the obstacle ( see Figure 7—figure supplement 3 ) . Therefore , an acceleration occurring prior to the predicted time of onset and the predicted DTC did not modify the onset of the cell discharge . Most cells displayed similar changes in activity to those illustrated in Figure 7 in response to acceleration of the obstacle . Figure 8A , B illustrate two other DTC-related cells in which acceleration modified the onset and the slope of the onset of the cell discharge . An acceleration just before the step over the obstacle ( 1L condition , purple traces in Figure 8A , B ) produced similar changes to those observed in Figure 7C , in that both cells showed a relatively later onset during the acceleration than during the unperturbed visual dissociation task . In both cells , the DTC at which the cell discharged during the acceleration was similar to that obtained in the matched and the visual dissociation tasks ( upper left of Figure 8A , B ) . A similar , constant , relationship for a TTC-related cell is illustrated in Figure 8C . In this example , TTC remained almost constant for the matched and visual dissociation tasks , as well as during the 1L and the 2L acceleration conditions . The onset of cell discharge during the 1L condition of the acceleration task was delayed with respect to onset during the visual dissociation task for the vast majority ( 33/36 ) of cells tested in this condition ( Figure 8D ) . We found a significant delay in 19/36 of these cells ( t-tests , p<0 . 05 ) . Cell onset was also relatively delayed for most ( 8/9 ) cells following acceleration in the 1R condition and was significant in 2/9 cells ( Figure 8D ) . However , the onset of discharge activity was less frequently delayed in the 2L and the 2R conditions , in which we found significant differences in only 1/9 and 2/26 cells , respectively ( Figure 8E ) . In general , acceleration was less likely to modify the onset of cell discharge the earlier we applied it . To determine whether all cells maintained the same constant relationship with distance or time in trials with accelerations , we repeated the ANOVA analysis described for Figures 4–5 , with the addition of each type of acceleration in turn to the calculation . We found that in the majority of cases , the onset of cell discharge maintained a constant relationship to either distance ( e . g . Figure 5A ) or time ( e . g . Figure 5B ) . We could test 20 acceleration conditions for 8/14 cells showing a relationship to DTC , and in most of these ( 18/20 ) , the relationship with DTC was maintained during the different acceleration conditions ( 7/9 , 1L; 2/2 , 1R; 3/3 , 2L; 6/6 , 2R ) . Similarly , we tested 25 acceleration conditions for 8/15 cells with a constant relationship to TTC and most ( 16/25 ) equally maintained this relationship with the accelerations ( 6/11 , 1L; 5/5 , 1R; 1/1 , 2L; 4/8 , 2R ) . Importantly , most cells showed a marked increase in the slope of the increase in discharge frequency during the acceleration , particularly during the 1L condition ( Figure 8F ) . For example , in the example illustrated in Figure 7C , the slope increased from a value of 35 . 5 spk . s-2 in the unperturbed visual dissociation task to 170 . 7 spk . s−2 during the acceleration ( cyan symbol in Figure 8F ) , while in Figure 7D the increase went from 46 . 2 sps . s−2 to 90 . 3 spk . s−2 . Similarly , we found a clear increase in slope for the examples illustrated in Figure 8A , B ( purple and green symbols , respectively ) . Altogether , three quarters of our examples ( 49/67 , 73% ) showed an increase in the slope of the activity during the acceleration . We observed the largest changes in slope in the 1L and 2R conditions . The increase in the slope of the cell discharge for the 1L and the 2R condition is well illustrated in the population averages of Figure 9A , B , respectively . These population plots show that the slope of the discharge during the acceleration was 3 . 3 times greater than during the unperturbed visual dissociation task for the 1L condition and 2 . 9 times greater in the 2R condition . An increased slope is also very clear for the 1R condition in which the late onset of the acceleration initiated a rapid change in the limb sequence ( see Figure 9—figure supplement 1A ) that we consider more of an online correction than a planned gait modification . However , we did not observe any noticeable change in the slope of the discharge for the 2L condition ( see Figure 9—figure supplement 1B ) , in which the acceleration occurred earlier ( average of 1608 ms ) than the onset of discharge activity in most cells . It is probable that the increased slope of the discharge in the conditions in which the acceleration relatively delayed cell onset provides information on the rate of change of gap closure ( see Discussion ) .
In this manuscript , we demonstrate two distinct neuronal populations in the PPC ( primarily in area 5b ) whose properties support a role in signaling the relationship between body position and object location during walking . One population’s discharge activity increased in relation to particular distances-to-contact with an obstacle . The second population’s discharge activity increased in relation to particular times-to-contact with an obstacle . We propose that walking animals use this information to appropriately modify the spatial and temporal parameters of the gait modification required to negotiate a moving obstacle . These results emphasize a contribution of populations of neurons in the PPC to the control of locomotion that goes beyond the control of limb-specific activity related to limb trajectory or the EMG patterns required to execute the step over the obstacle . Instead , we suggest that this pattern of activity is intricately implicated in the transformation of information obtained from vision into an appropriate motor plan that can be used for obstacle avoidance . The presence of cells discharging to an advancing object is compatible with the existence of cells in multiple areas , including the PPC , that respond to optic flow stimuli ( see Introduction ) . However , the cells in our study did not discharge purely to the visual stimulus , in which case they might have been expected to discharge throughout the period that the obstacle was visible ( 10–12 steps before the step over the obstacle ) . Rather , they began to discharge only when the obstacle was at a fixed DTC or TTC from the cat . Moreover , most of the cells maintained this relationship even when we accelerated the obstacle toward the cat . This suggests that these cells are tuned to respond to objects only when they are within a limited range of DTC or TTC . A similar tuning of cell responses to distance in response to looming stimuli is found in the ventral intraparietal area ( VIP ) of the PPC in non-human primates ( Colby et al . , 1993; Graziano and Cooke , 2006; see also Hadjidimitrakis et al . , 2015 ) , as well as in the premotor cortex ( PMC , Graziano et al . , 1997 ) . Graziano ( Graziano and Cooke , 2006 ) has proposed that such cells may play a role in defensive and avoidance behavior . A similar function might be ascribed to the responses recorded in our task in which the cat must interact with the advancing obstacle by modifying its gait pattern to step over it . The discharge activity should therefore not be considered simply in terms of the visual stimulus but rather in the context of a coordinated and planned motor activity . In this respect , cells responding at fixed DTC and TTC might be considered as providing important context-dependent information that is used to plan the upcoming gait modification . Moreover , as in the experiments referenced above with respect to the VIP and PMC , the cells in our study discharged in a staggered manner over a limited range of DTC and TTC . We believe that this feature of the discharge activity would provide a means for animals to continually monitor the rate of gap closure over the range in which a gait modification needs to be planned . The ability to continually monitor obstacle location over time would facilitate the detection of any non-linearities in the rate of gap closure and might be particularly important in helping the cats negotiate the obstacle when it is accelerated . In a more natural situation , the sequential activation of both DTC and TTC-related cells would be necessary for estimating the gap with a prey moving at unpredictable speeds , different from those of the predator . It is important to emphasize that although the cells discharged at fixed DTC/TTC before the gait modification , cell discharge continued until the step over the obstacle . As such , we believe that the increased discharge prior to the step over the obstacle is not a pure visual response but rather represents the starting point of a complex sensorimotor transformation involved in planning the gait modification . In this view , visual input is essential for initiating the sensorimotor transformation , but once initiated planning can continue in the absence of continual visual input . The results of our earlier lesion studies also support a role for the PPC cells in the sensorimotor transformation required to step over the obstacle . Lesion of the PPC region in which we recorded these cells results in a marked locomotor deficit defined by an inability to appropriately place the plant limb in front of the obstacle ( Lajoie and Drew , 2007 ) . We have previously discussed the reasons that we believe that this deficit is indicative of an error in planning rather than one of perception or action . We propose that in the absence of information about the relative location of the obstacle and of the rate of gap closure provided by the cells in the PPC , the cat is unable to determine where to position its leg in front of the obstacle and when to start the gait modification . The ensemble activity of the population of cells demonstrates a progressive increase in discharge rate up to the time of the gait modification . A similar ramp increase in cell discharge as TTC progressively decreases has been observed in the motor cortex in monkeys trained to intercept a simulated object with their arm ( Merchant and Georgopoulos , 2006 ) . In our task , we propose that this ramp increase in the population discharge provides a signal that indicates the imminent requirement to make the gait modification . This ramp discharge is reminiscent of that observed in several structures and in many tasks in which motor activity is self-initiated ( Lebedev et al . , 2008; Maimon and Assad , 2006a , 2006b; Merchant and Georgopoulos , 2006 ) and is particularly prevalent in tasks in which a decision to move on the basis of ambiguous or delayed information is required ( Cisek and Kalaska , 2005; Cisek and Kalaska , 2010; Roitman and Shadlen , 2002; Thura and Cisek , 2014 ) . Thura and Cisek ( 2014 ) refer to the time at which such a decision is made as the time of commitment . We consider that the peak of activity in our population of cells also indicates a time of commitment , at which point the cat initiates the gait modification . In this respect , it is pertinent that when we accelerated the obstacle , we found a greater slope of the activity between the onset of cell discharge and the onset of the gait modification than in the absence of acceleration . This increased slope allowed the cells to reach a similar peak value in the shorter time available to the cat to make the gait modification . Although we believe that the activity that we observed in this task forms part of the sensorimotor transformation that leads to the gait modification , its function has to be discussed in light of the fact that the discharge is limb-independent , that is , it is identical regardless of whether the left or the right limb is the first to step over the obstacle . This is contrary to one study ( Bernier et al . , 2012 ) that suggests that activity in the PPC is not expressed until the effector limb has been specified , and then only for the contralateral limb ( although see Chang et al . , 2008 ) While this disparity could relate to species or area-specific differences , we believe that the nature of the task requirements is probably the main determinant . In the experiments of Bernier et al . ( 2012 ) , subjects were static and an external cue specified the arm to move . In our task , the cat is walking and the leg that will step over the obstacle is not pre-defined . The cat must process the incoming information on gap closure and must integrate the planned gait modification into its natural rhythm . As such , we suggest that the decision as to which limb to use to step over the obstacle should be viewed as an emergent property of the task rather than as an instructed movement or a decision made in advance of the movement as in the tasks referenced above . One possible manner in which information on gap closure could be used to initiate a gait modification is illustrated in Figure 10 . In this conceptual model , we presume that there is an integration of the gap closure signal with a second signal that provides information on the state of the limb . The integration with limb state ensures that the gait modification will only be initiated when the limb is in the appropriate state , that is , at the end of the stance phase and ready to initiate the transfer of the limb over the obstacle . We propose that this integration proceeds bilaterally and the limb selected to be the first to step over the obstacle depends on which side wins the competition . Although we do not wish to unduly speculate on where this integration would occur , we would note that all of the cells recorded in this study were located in layer V and therefore project to subcortical structures , including the basal ganglia and the cerebellum . In conclusion , our results provide new insights , at the single cell level , into the sensorimotor transformations that underlie the control of visually guided walking . The demonstration of populations of cells that can serve to provide information on gap closure and potentially initiate precise gait changes is a novel contribution to our understanding of the control mechanisms used to guide locomotion . Taken together with the results from studies in various species that show a contribution of the PPC to spatial navigation ( Calton and Taube , 2009; Harvey et al . , 2012; Whitlock , 2014 ) , it is possible that the PPC may have a privileged position in contributing to our ability to plan the gait adjustments needed to negotiate a complex environment .
We performed experiments on the same two cats ( PCM7 and PCM9 ) that we previously used in other experiments ( Marigold and Drew , 2011 ) . We initially trained the cats to walk on a treadmill at a constant speed of 0 . 45 m . s−1 ( unobstructed locomotion ) and then trained them to step over obstacles attached to a second belt that moved at the same speed as the treadmill ( matched task ) . Subsequently , we trained each cat to step over the obstacles that were advanced at a slower speed ( 0 . 3 m . s−1 ) than the treadmill belt on which the cat walked ( visual dissociation task: Drew et al . , 2008; Lajoie and Drew , 2007 ) . Finally , we habituated the cats to a third task in which the obstacle accelerated toward them several steps prior to the step over the obstacle ( acceleration task ) . Accelerations consisted of a ramp increase from 0 . 3 m . s−1 to ~0 . 65 m . s−1 over a period of 450 ms and a symmetrical decrease back to baseline levels . Note that cats very rarely , if ever , hit the obstacle , even in conditions in which the acceleration occurred just before the planned gait modification . Two obstacles ( cylindrical in shape and each of 10 cm cross-section ) , separated by 3 m , were attached to the treadmill belt . Depending on the speed of the treadmill , the cat generally made 12–14 steps ( 6–7 step cycles ) between each obstacle . The obstacle became visible to the cat ~2 m before the step over the obstacle . The obstacle was therefore visible to the cat for 5–7 s before the gait modification , during which time the cat made 10–12 steps . All handling and surgical procedures followed the recommendations of the Canadian Council for the Protection of Animals , and the animal ethics committee at the Université de Montréal approved the experimental protocols ( #12_082 ) . Once trained , we prepared the cats for surgery in aseptic conditions as described in previous papers ( Andujar et al . , 2010; Drew , 1993; Marigold and Drew , 2011 ) . In brief , based on an MRI taken 1–2 weeks before the surgery , we placed a stainless steel baseplate ( internal dimensions = 12 by 6 mm ) over the right PPC and then formed a recording chamber around it with dental acrylic . We implanted pairs of Teflon-insulated , stainless steel braided wires into selected fore- and hindlimb muscles to record electromyographic ( EMG ) activity . The wires ran subcutaneously to a connector on the cranium . To allow for antidromic identification of projection neurons in layer V of the PPC , we inserted microwires into the cerebral peduncle by using a harpoon assembly ( Drew , 1993; Palmer , 1978 ) . Following the surgery , we administered buprenorphine ( 5 µg/kg ) for a period of 72 hr , and antibiotics for a period of 10 days . Experiments started 1 week after the surgery . In each recording session , we introduced a conventional glass-insulated tungsten microelectrode ( 0 . 5–1 . 5 MΩ ) into the PPC using a custom-made microdrive . We advanced the electrode until stimulation of the electrodes in the cerebral peduncle produced antidromic discharges either in an isolated unit or in smaller units in the background . This provided evidence that the electrode had reached layer V . We recorded cell activity from well-isolated single units while the cat walked on the treadmill in the matched task until approximately 10 steps over the obstacle with each leg as the lead limb had been made . After slowing the obstacle ( visual dissociation task ) , locomotion continued until we collected approximately five steps with each leg . In selected subsequent steps , we accelerated the obstacle toward the cat several steps before the step over the obstacle . In these steps , acceleration always occurred 200 ms after the onset of activity in the left brachialis ( Br ) or cleidobrachialis ( ClB ) contralateral to the recording site . Steps in which the obstacle accelerated were interspersed unpredictably with steps where the obstacle continued at its pre-set speed . The entire data collection period occupied 15–20 min , although we sometimes lost cell stability before we could complete the recording session . We band-pass filtered EMG activity at 100–475 Hz and digitized it online at a frequency of 1 KHz . To discriminate cells offline , we digitized cell activity at a frequency of 100 KHz . In all experiments , a six-camera Vicon motion analysis system recorded , at 100 Hz , the position of light-reflecting points placed on a rod attached to the head of the cat and along the length of each obstacle . We synchronized cell , EMG , and motion data for later analyses . We discriminated single units offline on the basis of waveform amplitude and shape . For sections of data with stable action potentials and with stable locomotion , we marked the onset and offset of activity in the left , contralateral ( co ) and right , ipsilateral ( i ) Br or ClB EMG for every step during the entire locomotor sequence . This allowed us to identify each step as a step over the obstacle , one or more steps before the obstacle , or the step after the obstacle . We further identified steps as to whether the left , contralateral forelimb ( left limb leads condition ) or the right , ipsilateral forelimb ( right limb leads condition ) stepped over the obstacle first — in previous publications from this laboratory , these are referred to as lead and trail conditions , respectively . Cell activity during unobstructed locomotion was calculated on the basis of the discharge activity five steps before the step over the obstacle , combining all tasks . As such , activity during unobstructed locomotion is interspersed with the data obtained during the steps over the obstacle and was obtained from the entire recording period . To determine the temporal relationships between cell discharge activity and different behavioral events on a trial-by-trial basis , we transformed cell discharge for each trial into an instantaneous frequency ( 1000/interspike interval ) and filtered it at 50 Hz ( fourth order Butterworth algorithm ) . By using interactive software , we calculated the onset of cell activity relative to the step over the obstacle for each trial as an increase or decrease of activity that exceeded 2SD of the cell discharge that occurred between 2 . 5 and 3 s before the step over the obstacle ( see Figure 3—figure supplement 1 ) . We then calculated the distance of the obstacle from the cat ( distance-to-contact; DTC ) and the time-to-contact ( TTC ) with the obstacle at the moment of the onset of change in cell discharge , for each individual trial , by measuring the relative distance of the rod on the cat’s head from the obstacle . To make box plots of the time of cell onset for each task ( matched , visual dissociation , and acceleration ) and condition ( left and right limb lead ) , we calculated median and interquartile ranges ( IQR ) . We removed data values that exceeded 1 . 5 * IQR from the analysis . On average , this removed 2 . 5% of the trials ( we recorded an average of 62 trials/cell ) . After removal of outliers , we calculated average discharge rates of cell activity ( bin width = 2 ms ) by synchronizing activity to the onset of the Br or ClB . We used these average displays to determine if cells were step-related or step-advanced and whether they were limb-independent or limb-dependent ( Andujar et al . , 2010 ) . A change in discharge activity beginning >200 ms before the onset of activity in the Br or ClB differentiated step-advanced cells from step-related cells . We defined a limb-independent cell as one in which discharge activity ( as determined from the averaged displays ) ended at approximately the same time ( <200 ms difference ) with respect to the coBr ( coClB ) during the left , contralateral forelimb , lead condition and with respect to the iBr ( iClB ) during the right , ipsilateral forelimb , lead condition ( see Andujar et al . , 2010; Marigold and Drew , 2011 ) . We used Systat V13 ( Systat Software Inc . ) for all statistical analyses . One-way ANOVAs determined the effect of distance and time on the time of onset of cell discharge ( significant effects determined at the alpha = 0 . 01 level ) . Significant differences between pairs of values using t-tests were determined using an alpha of 0 . 05 . When multiple comparisons were made we used a Bonferroni correction . To create an index of the difference in discharge with respect to DTC and TTC , we calculated the difference between the averaged standardized discharge rate ( Z score ) during the visual dissociation task ( left + right/2 ) and that during the matched task ( left +right/2 ) . In this index , cells that showed constant activity in both tasks would have TTC and DTC indexes close to zero . Those cells that show a constant relationship to DTC will have a DTC index centered around 0 . 0 and a TTC index close to 1 . 0 , while cells with a constant relationship to TTC will have a TTC index centred around 0 and a DTC index close to −1 . 0 . To determine the extent to which this index succeeded in identifying the two categories of cell , we also performed a bootstrapping exercise in which we used a replacement protocol to create new datasets for each cell ( see legend to Figure 5—figure supplement 1 ) . Calculations were performed 1000 times for each cell . We included cells in the analysis on the following bases: ( 1 ) The cells were located within the caudal bank of the ansate sulcus or the adjoining caudal gyrus , corresponding to area 5b and the border with area 7 . ( 2 ) The cells were located within cortical layer V , as determined on the basis of the antidromic stimulation ( 43/67 cells were antidromically activated , as determined by collision with spontaneous action potentials , and the other 24/67 cells were adjacent to identified cells ) . ( 3 ) All cells discharged >200 ms before the step over the obstacle ( step-advanced cells , Andujar et al . , 2010 ) . ( 4 ) All cells manifested a limb-independent pattern of activity , allowing us to compare activity when the left and right limbs stepped over the obstacle . ( 5 ) We only included cells in the analysis if we recorded at least five steps over the obstacle during unperturbed walking and three steps for each condition during the acceleration task . At the end of the series of experiments , we made small lesions ( 30–50 µA ) in selected locations within the recording chamber and perfused the cat per cardia with formalin . We sectioned the brain in the sagittal plane ( 40 µm sections ) and stained it with cresyl violet . The depth of layer V ( as determined during the recordings ) and the terminal lesions were used to determine the location of the electrode penetrations . | Imagine crossing the street and having to step up onto a sidewalk , or running up to kick a moving soccer ball . How does the brain allow you to accomplish these deceptively simple tasks ? You might say that you look at the target and then adjust where you place your feet in order to achieve your goal . That would be correct , but to make that adjustment you have to determine where you are with respect to the curb or the soccer ball . A key aspect of both of these activities is the ability to determine where your target is with respect to your current location , even if that target is moving . One way to do that is to determine the distance or the time required to reach that target . The brain can then use this information to adjust your foot placement and limb movement to fulfill your goal . Despite the fact that we constantly use vision to examine our environment as we walk , we have little understanding as to how the brain uses vision to plan where to step next . Marigold and Drew have now determined whether one specific part of the brain called the posterior parietal cortex , which is known to be involved in integrating vision and movement , is involved in this planning . Specifically , can it estimate the relative location of a moving object with respect to the body ? Marigold and Drew recorded from neurons in the posterior parietal cortex of cats while they walked on a treadmill and stepped over an obstacle that moved towards them . On some tests , the obstacle was either slowed or accelerated quickly as it approached the cat . Regardless of these manipulations , some neurons always became active when the obstacle was at a specific distance from the cat . By contrast , other neurons always became active at a specific time before the cat met the obstacle . Animals use this information to adjust their gait to step over an obstacle without hitting it . Overall , the results presented by Marigold and Drew provide new insights into how animals use vision to modify their stepping pattern . This information could potentially be used to devise rehabilitation techniques , perhaps using virtual reality , to aid patients with damage to the posterior parietal cortex . Equally , the results from this research could help to design brain-controlled devices that help patients to walk – or even intelligent walking robots . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | Posterior parietal cortex estimates the relationship between object and body location during locomotion |
Lipid metabolism plays an important role in viral infections . We aimed to assess the causal effect of lipid-lowering drugs ( HMGCR inhibitiors , PCSK9 inhibitiors , and NPC1L1 inhibitior ) on COVID-19 outcomes using two-sample Mendelian randomization ( MR ) study . We used two kinds of genetic instruments to proxy the exposure of lipid-lowering drugs , including expression quantitative trait loci of drugs target genes , and genetic variants within or nearby drugs target genes associated with low-density lipoprotein ( LDL cholesterol from genome-wide association study ) . Summary-data-based MR ( SMR ) and inverse-variance-weighted MR ( IVW-MR ) were used to calculate the effect estimates . SMR analysis found that a higher expression of HMGCR was associated with a higher risk of COVID-19 hospitalization ( odds ratio [OR] = 1 . 38 , 95% confidence interval [CI] = 1 . 06–1 . 81 ) . Similarly , IVW-MR analysis observed a positive association between HMGCR-mediated LDL cholesterol and COVID-19 hospitalization ( OR = 1 . 32 , 95% CI = 1 . 00–1 . 74 ) . No consistent evidence from both analyses was found for other associations . This two-sample MR study suggested a potential causal relationship between HMGCR inhibition and the reduced risk of COVID-19 hospitalization . Start-up Fund for high-level talents of Fujian Medical University .
The COVID-19 pandemic has caused millions of infections and deaths , which is caused by a novel coronavirus , severe acute respiratory syndrome coronavirus 2 ( SARS-CoV-2 ) . Lacking drugs specifically targeted to SARS-CoV-2 infection has led to a great interest to identify drugs that can be repurposed to reduce the infection and mortality of the disease . Available studies have suggested an important role of lipid metabolism in viral infections , including in the pathogenesis of SARS-CoV-2 infection ( Proto et al . , 2021 ) . The plausible mechanisms include the involvement of host lipids in the virus life cycle , the influence of cholesterol on the immune cell functions , interfering with the mevalonate pathway , and so on Proto et al . , 2021 . Such evidence indicates the potential protective effect of lipid-lowering drugs against COVID-19 . HMG-CoA reductase ( HMGCR ) inhibitors , known as statins , are the most commonly used class of lipid-lowering drugs , which have a couple of predominant merits , such as the well-proven safety , low cost , and pleiotropic effects . Proprotein convertase subtilisin/kexin type 9 ( PCSK9 ) and Niemann–Pick C1-Like 1 ( NPC1L1 ) are proteins playing a crucial role for the circulating level of low-density lipoprotein cholesterol ( LDL-C ) ( Sabatine , 2019; Williams et al . , 2020 ) . Both PCSK9 inhibitors ( i . e . , evolocumab and alirocumab ) and NPC1L1 inhibitors ( i . e . , ezetimibe ) are FDA-approved lipid-lowering agents ( Sabatine , 2019; Williams et al . , 2020 ) . A number of observational studies have investigated the association between lipid-lowering drugs and COVID-19 outcomes , but generated mixed results ( Butt et al . , 2020; Hariyanto and Kurniawan , 2020; Kow and Hasan , 2020; Zhang et al . , 2020; Gupta et al . , 2021 ) . What’s more , confounding bias and reverse causation cannot be avoided in most of these studies . Mendelian randomization ( MR ) study uses genetic variants as an instrument to perform causal inference between an exposure and an outcome , which could indicate whether an observational association is consistent with a causal effect ( Davies et al . , 2018 ) . Confounding bias can be minimized in MR study because genetic variants are randomly assigned to the individual at birth . Similarly , reverse causation can be avoided because genetic variants are assigned prior to the development of disease . Therefore , we performed two-sample MR analysis in this study to test the association of lipid‐lowering drugs ( HMGCR inhibitiors , PCSK9 inhibitiors , and NPC1L1 inhibitior ) with COVID-19 outcomes ( susceptibility , hospitalization and very severe disease ) .
This two-sample MR study is based on publicly available summary-level data from genome-wide association studies ( GWASs ) and expression quantitative trait loci ( eQTLs ) studies ( Supplementary file 1—Table 1 ) . All these studies had been approved by the relevant institutional review boards and participants had provided informed consents . Three classes of FDA-approved lipid-lowering drugs were included as exposures in this study: HMGCR inhibitors , PCSK9 inhibitors , and NPC1L1 inhibitor . As shown in Table 1 , we used available eQTLs for drugs target genes ( i . e . , HMGCR , PCSK9 , and NPC1L1 ) as the proxy of exposure to each lipid-lowering drug . The eQTLs summary-level data were obtained from eQTLGen Consortium ( https://www . eqtlgen . org/ ) or GTEx Consortium V8 ( https://gtexportal . org/ ) , the details of which are presented in Supplementary file 1—Table 1 . We identified common ( minor allele frequency [MAF] >1% ) eQTLs single-nucleotide polymorphisms ( SNPs ) significantly ( p < 5 . 0 × 10−8 ) associated with the expression of HMGCR or PCSK9 in blood , and the expression of NPC1L1 in adipose subcutaneous tissue as there are no eQTLs in blood or other tissues available at a significance level for NPC1L1 . Only cis-eQTLs were included to generate genetic instruments in this study , which were defined as eQTLs within 1 Mb on either side of the encoded gene . Secondly , to validate the observed association using the eQTLs as an instrument , we additionally proposed an instrument by selecting SNPs within 100 kb windows from target gene of each drug that was associated with LDL cholesterol level at a genome-wide significance level ( p < 5 . 0 × 10−8 ) to proxy the exposure of lipid-lowering drugs . A GWAS summary data of LDL cholesterol levels from the Global Lipids Genetics Consortium ( GLGC ) with a sample size of 173 , 082 were used to identify these SNPs , where only common SNPs ( MAF >1% ) were included ( Willer et al . , 2013; Supplementary file 1—Table 1 ) . Seven SNPs within 100 kb windows from HMGCR gene were selected for proxying HMGCR inhibitors , 12 SNPs from PCSK9 gene identified for PCSK9 inhibitors , and 3 SNPs from NPC1L1 gene selected for NPC1L1 inhibitor . To maximize the strength of the instrument for each drug , SNPs used as instruments were allowed to be in low weak linkage disequilibrium ( r2 < 0 . 30 ) with each other . GWAS summary-level data for COVID-19 outcomes were obtained from the COVID-19 Host Genetics Initiative V4 with a sample size of 1 , 299 , 010 for COVID-19 susceptibility , 908 , 494 for COVID-19 hospitalization , and 626 , 151 for COVID-19 severe disease , respectively ( https://www . covid19hg . org/; COVID-19 Host Genetics Initiative , 2020; Supplementary file 1—Table 1 ) . The study population was restricted to individuals with European ancestry , including meta-analyses of GWASs containing up to 22 cohorts from 11 countries . GWAS from these cohorts used a model adjusted for age , sex , age × age , age × sex , genetic principal components , and study-specific covariates . A COVID-19 case was confirmed by lab or self-reported infections , or electronic health records of infections . The susceptibility outcome was measured by comparing COVID-19 cases and controls who did not have a history of COVID-19 . The hospitalized outcome was measured by comparing COVID-19 hospitalized cases and controls who were never admitted to the hospital due to COVID-19 , including individuals without COVID-19 . The severe disease outcome was measured by comparing COVID-19 cases who died or required respiratory support and controls without severe COVID-19 , including individuals without COVID-19 . We included individuals without COVID-19 as controls for all outcomes to decrease collider bias and allow for population-level comparisons ( Griffith et al . , 2020; Butler-Laporte et al . , 2021 ) .
A total of 921 , 24 , and 11 cis-eQTLs were identified from eQTLGen or GTEx Consortium for drugs target gene HMGCR , PCSK9 , and NPC1L1 , respectively , and the most significant cis-eQTL SNP was selected as a genetic instrument for the target gene of each drug ( Table 1 , Supplementary file 1—Table 2 ) . A total of 7 , 12 , and 3 SNPs within or nearby gene HMGCR , PCSK9 , and NPC1L1 were selected from a GWAS summary data of LDL cholesterol levels in the Global Lipids Genetics Consortium , respectively ( Table 1 , Supplementary file 1—Table 3 ) . F-Statistics for all instrument variants were over 30 , suggesting that weak instrument bias can be minimized in our study ( Supplementary file 1-Tables 2 and 3 ) . Positive control study showed significant associations between exposure to each drug and LDL cholesterol when using eQTLs-proposed instruments ( Supplementary file 1—Table 5 ) , as well as between exposure to each drug and coronary heart disease when using LDL cholesterol GWAS-proposed instruments ( Supplementary file 1—Table 6 ) , further ensuring the efficacy of the selected genetic instruments . From COVID-19 GWASs , a total of 14 , 134 cases and 1 , 284 , 876 controls were used to explore the association with COVID-19 susceptibility , 6406 cases and 902 , 088 controls for COVID-19 hospitalization , and 3886 cases and 622 , 265 controls for COVID-19 severe disease ( Supplementary file 1—Table 1 ) . In Figure 1 and Supplementary file 1—Table 2 , results from SMR analysis found a suggestive evidence for the association of the increased expression of HMGCR gene in blood ( equivalent to a one standard deviation increase ) with the higher risk of COVID-19 susceptibility ( odds ratio [OR] = 1 . 30 , 95% confidence interval [CI] = 1 . 05–1 . 61; p = 0 . 017 ) and COVID-19 hospitalization ( OR = 1 . 38 , 95% CI = 1 . 06–1 . 81; p = 0 . 019 ) , indicating that HMGCR inhibitors might lower the risk of COVID-19 susceptibility and hospitalization . Suggestive evidence was observed regarding the negative association between PCSK9 expression and risk of COVID-19 susceptibility ( OR = 0 . 84 , 95% CI = 0 . 73–0 . 97; p = 0 . 02 ) . No significant association was found between the expression of NPC1L1 and COVID-19 outcomes . In Figure 2 and Supplementary file 1—Table 4 , IVW-MR analysis also found a suggestive evidence for the association between HMGCR-mediated LDL cholesterol ( equivalent to a 1 mmol/l increase ) and the risk of COVID-19 hospitalization ( OR = 1 . 32 , 95% CI = 1 . 00–1 . 74; p = 0 . 049 ) , further supporting a possible protective effect of HMGCR inhibitors against COVID-19 hospitalization . Strong evidence was observed between NPC1L1-mediated LDL cholesterol and the risk of COVID-19 susceptibility ( OR = 2 . 02 , 95% CI = 1 . 30–3 . 13; p = 0 . 002 ) . IVW-MR analysis did not provide any evidence for the association between PCSK9-mediated LDL cholesterol and COVID-19 outcomes . For SMR analysis , HEIDI test suggested that all observed associations were not due to a linkage ( p > 0 . 01 ) , except for the association between HMGCR expression and COVID-19 susceptibility ( p = 0 . 009 ) ( Supplementary file 1—Table 2 ) . We further examine if horizontal pleiotropy was present in the association between HMGCR expression and COVID-19 outcomes by investigating if there was an association between the expression of nearby genes which are significantly associated with the top eQTL SNP ( instrument variant ) of HMGCR and COVID-19 outcomes . We identified six genes , including HMGCR , the expression of which were associated with the instrument variant ( Supplementary file 1—Table 7 ) . Only four genes have available eQTLs at a genome-wide significance level ( p < 5 . 0 × 10−8 ) . Among these four genes , only HMGCR expression was significantly related to COVID-19 susceptibility and COVID-19 hospitalization , suggesting a small role of horizontal pleiotropy in the observed associations ( Supplementary file 1—Table 8 ) . For IVW-MR analysis , Cochran Q test did not find evidence of heterogeneity for all reported results ( all p > 0 . 05; Supplementary file 1—Table 4 ) . Both the intercept term in MR-Egger regression and MR-PRESSO analysis suggested no significant overall horizontal pleiotropy ( all p > 0 . 05; Supplementary file 1—Table 4 ) . A multivariable MR study suggested that BMI and diabetes might play a role in the association between HMGCR-mediated LDL cholesterol level and COVID19 hospitalization ( Supplementary file 1—Tables 9 and 10 ) .
The main strength of our study is the use of genetic instruments to proxy drug exposure , which could minimize confounding bias and avoid reverse causation . Besides , we used two different kinds of genetic instruments to proxy the studied drug , which contributes to validating the effect estimates from each other . A number of sensitivity analyses have been performed to test the efficacy of genetic instruments and the assumptions of MR study . This study has several limitations . Firstly , there are no available eQTLs in blood for NPC1L1 , so we were not able to explore the association between NPC1L1 expression in blood and COVID-19 outcomes . Besides , there are no available eQTLs in liver ( the main tissue related to lipid metabolism ) for these target genes , which might provide more convincing evidence of the observed association . The sample size of eQTL study for PCSK9 and NPC1L1 in GTEx is relatively smaller , which may affect the statistical power for the results of PCSK9 or NPC1L1 inhibition . Secondly , the effect of statins probably varies between subgroups , for example , it may be more effective in patients with chronic diseases ( e . g . , coronary heart disease ) . However , the use of summary-level data did not allow us to perform subgroup analyses , so further MR study with individual-level data is needed to provide more detailed information . Thirdly , the Bonferroni correction for multiple tests suggests that we cannot rule out the false-positive possibility for the finding of the protective effect of statins on COVID hospitalization . Fourthly , confounding bias and/or horizontal pleiotropy cannot be completely excluded although we have performed various sensitivity analyses to test the assumptions of MR study . Fifthly , both eQTLs and GWAS data used in this study were predominantly obtained from European population ancestry , thus these findings should be interpreted with caution when generalizing to other populations . In conclusion , this MR study suggested a causal relationship between HMGCR inhibition and the reduced risk of COVID-19 hospitalization . Clninical trials are called to examine if statins have the protective effect against COVID-19 and further researches are needed to explore the underlying mechanisms . | The virus SARS-CoV-2 has caused millions of infections and deaths during the COVID-19 pandemic , but as of December 2021 , no new drugs targeted to SARS-CoV-2 specifically exist . Thus , it is important to identify existing drugs that can reduce the infection and mortality of this virus , since repurposing old drugs is faster and cheaper than developing new ones . Fats , such as cholesterol , can play an important role in viral infections , meaning that drugs intended to lower the levels of fats in the blood could have a protective effect against SARS-CoV-2 . To test this hypothesis , Huang , Xiao , et al . carried out a Mendelian randomization study to investigate if there is a link between drugs that lower fats and outcomes of SARS-CoV-2 infection , including susceptibility , hospitalization , and severe disease . This approach consists on grouping people according to their version of a particular gene , which minimizes the effect of variables that can cause spurious associations , something known as confounding bias . Thus , Mendelian randomization studies allow scientists to disentangle cause and effect . Using this method , Huang , Xiao , et al . found an association between statins ( a type of drug that decreases the levels of bad cholesterol ) and a reduced risk of being hospitalized after being infected with SARS-CoV-2 . These findings suggest that statins could benefit patients infected with SARS-CoV-2 , and indicate that they should be prioritized in future clinical trials for treating COVID-19 . | [
"Abstract",
"Introduction",
"Materials",
"and",
"methods",
"Results",
"Discussion"
] | [
"epidemiology",
"and",
"global",
"health",
"medicine"
] | 2021 | Association of lipid-lowering drugs with COVID-19 outcomes from a Mendelian randomization study |
RNA-binding proteins ( RBPs ) control multiple aspects of post-transcriptional gene regulation and function during various biological processes in the nervous system . To further reveal the functional significance of RBPs during neural development , we carried out an in vivo RNAi screen in the dorsal spinal cord interneurons , including the commissural neurons . We found that the NOVA family of RBPs play a key role in neuronal migration , axon outgrowth , and axon guidance . Interestingly , Nova mutants display similar defects as the knockout of the Dcc transmembrane receptor . We show here that Nova deficiency disrupts the alternative splicing of Dcc , and that restoring Dcc splicing in Nova knockouts is able to rescue the defects . Together , our results demonstrate that the production of DCC splice variants controlled by NOVA has a crucial function during many stages of commissural neuron development .
Alternative splicing generates gene function complexity in many neural developmental processes , including neuronal differentiation , neuronal migration , axon growth and guidance , and synapse formation and function ( Grabowski and Black , 2001; Lee and Irizarry , 2003; Li et al . , 2007 ) . A large number of axon guidance molecules undergo alternative splicing , including Dscam ( Schmucker et al . , 2000 ) , Netrin ligands and their Dcc and Neogenin receptors ( Keeling et al . , 1997; Reale et al . , 1994; Zhang et al . , 2004 ) , Slit ligands and Robo receptors ( Camurri et al . , 2005; Chen et al . , 2008; Clark et al . , 2002; Dalkic et al . , 2006; Little et al . , 2002; Tanno et al . , 2004 ) , Semaphorin ligands and Neuropilin and Plexin receptors ( Cackowski et al . , 2004; Correa et al . , 2001; Qu et al . , 2002; Takahashi et al . , 2009; Tamagnone et al . , 1999 ) , and Ephrins and Ephs ( Holmberg et al . , 2000; Lai et al . , 1999; Sajjadi et al . , 1991 ) . However , the functional significance of these alternative splicing events and the splicing factors responsible for generating protein variants for these molecules remain largely uncharacterized . Netrin and Dcc ( deleted in colorectal carcinoma ) function during cell migration , neurite specification and growth , axon guidance , synaptogenesis , and tumorigenesis ( Cooper et al . , 1999; Killeen , 2009; Mehlen and Tauszig-Delamasure , 2014; Moore et al . , 2007 ) . Within the spinal cord commissural neurons , DCC is required for Netrin-stimulated axon outgrowth and for attracting the axons to the Netrin-secreting midline ( Dickson , 2002; Dickson and Zou , 2010; Evans and Bashaw , 2010 ) . Mammalian Dcc undergoes alternative splicing to generate two isoforms that differ in the extracellular domain , in the linker sequence between the fourth and fifth fibronectin repeats ( FN4 and FN5 ) ( Reale et al . , 1994 ) . This alternative splicing was first reported in neuroblastoma cells and was found to be disrupted in these tumor cells ( Reale et al . , 1994 ) , but its physiological significance was completely unknown . We refer to the splice variants hereafter as DCClong and DCCshort , with DCClong containing extra 20 amino acids in the FN4-FN5 linker . Interestingly , a recent structure study shows that the two isoforms bind Netrin-1 with comparable affinities , but are likely to adopt distinct conformations upon ligand binding ( Xu et al . , 2014 ) . To reveal the factors that control the alternative splicing of axon guidance genes including Dcc , we carried out an in vivo RNAi screen against candidate RNA-binding proteins ( RBPs ) in cultured mouse embryos , and found that Nova1 and Nova2 knockdown leads to severe defects in the dorsal spinal cord interneurons , including the commissural interneurons . The NOVA ( neuro-oncological ventral antigen ) proteins were first identified as the autoimmune antigens in the neurodegenerative disease POMA ( paraneoplastic opsoclonus myoclonus ataxia; Darnell , 1996 ) . NOVA1/2 are neural-specific KH ( hnRNP K homology ) -type of RBPs that can directly regulate alternative splicing ( Buckanovich et al . , 1996; Lewis et al . , 1999; Ule et al . , 2006 ) . Genome-wide studies have identified many potential NOVA targets that are involved in various neural developmental processes ( Licatalosi et al . , 2008; Ule et al . , 2003; Zhang et al . , 2010 ) . In vivo studies using Nova knockout mice have demonstrated defects in synapse formation and function , and in neuronal migration ( Huang et al . , 2005; Jensen et al . , 2000; Ruggiu et al . , 2009; Yano et al . , 2010 ) . We show here that Nova1/2 loss of function reduces the migration of the spinal cord interneurons and their progenitors , and disturbs the axon outgrowth and guidance of the commissural interneurons . Interestingly , these defects resemble those seen in Dcc knockouts . Consistently , Dcc alternative splicing is perturbed by Nova deficiency in vivo . Through rescue experiments , we show that restoring Dcclong , the diminished isoform in Nova knockouts , is able to reverse the defects . Furthermore , NOVA1/2 regulate Dcc pre-mRNA splicing in in vitro assays . Together , our results demonstrate that Dcc alternative splicing is important for the gene function and is controlled by the NOVA splicing factors .
Using the whole mouse embryo culture technique ( Chen et al . , 2008 ) , we transiently knocked down candidate RBPs with small interference RNAs ( siRNAs ) in the spinal cord and examined the resulting phenotype in commissural axon guidance . We selected RBPs that have neural-specific expression and have been implicated in alternative splicing ( Table 1 ) . The list of candidates is not exhaustive , and the screening against additional RBPs is ongoing . Some of the RBPs , including NOVA , FOX , and PTBP2/nPTB , directly regulate splicing by recognizing specific sequences in pre-mRNAs and controlling spliceosome assembly ( Agnès and Perron , 2004; Li et al . , 2007 ) . Some , such as CELF , UPF , and IGF2BP1 , play an indirect role , often by regulating the stability of selective pre-mRNAs and thus influencing their splicing ( Agnès and Perron , 2004; Ladd , 2013; Li et al . , 2007; Yap and Makeyev , 2013 ) . Others , such as ELAVL/HU , can have both direct and indirect effects on alternative splicing ( Agnès and Perron , 2004; Li et al . , 2007; Scheckel et al . , 2016; Ince-Dunn et al . , 2012 ) . 10 . 7554/eLife . 14264 . 003Table 1 . RNA-binding proteins targeted in the RNAi screenDOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 003GeneProtein familyPhenotypeNoteNova1/2 Neuro-oncological ventral antigen Neuronal migration , axon outgrowth , and axon guidance defects Ptbp2 Polypyrimidinetract-binding protein - Sfpq Splicing factor proline/glutamine rich ( polypyrimidine tract-binding protein associated ) - Fmr1 Fragile X mental retardation protein - Nufip1/2 Nuclear FMRP interacting protein - Elavl1/2/3/4 ELAV ( embryonically lethal abnormal vision ) homolog; Hu syndrome protein Partial midline crossing defect caused by Elavl2 single RNAi Rbfox1/2/3 RNA-binding protein , Fox ( feminizing locus on X ) homolog - Msi1 Musashi RNA-binding protein - Msi1-/- displays neuronal migration and axon guidance defects in precerebellar neurons but not in spinal commissural neurons ( Kuwako et al . , 2010 ) Upf1 UPF1 regulator of nonsense transcripts homolog Neuronal migration , axon outgrowth , and axon guidance defects Celf1/2/3/4/5/6 CUG binding protein , Elav-like family member - Khsrp KH-type splicing regulatory protein - Pabpc1 Poly ( A ) binding protein , cytoplasmic - Igf2bp1 Insulin-like growth factor 2 mRNA binding protein Partial midline crossing defect Regulates beta-actin mRNA transport and translation ( Leung et al . , 2006; Yao et al . , 2006 ) Srpk1/2 Serine/arginine-rich protein specific kinase - - , no phenotype in spinal cord commissural neurons , when family members were targeted individually . Using electroporation , we introduced siRNAs and gfp into the neural progenitors adjacent to the lumen of the neural tube ( Figure 1A ) . We observed that the dorsal interneuron progenitors are mostly targeted , whereas the ventral motor neuron progenitors are not ( Chen et al . , 2008; Figure 1B ) . This is likely due to the fact that the dorsal progenitors are more actively proliferating and differentiating during the culture period . We used Actb-gfp ( aka βactin-gfp ) in order to label the highly heterogeneous populations of commissural neurons , which are mostly descended from the dorsal progenitors . Following electroporation , the mitotic neuroprogenitors migrate laterally away from the ventricle ( Caspary and Anderson , 2003; Helms and Johnson , 2003; Lu et al . , 2015 ) . Upon neural differentiation around E10 in mice , the progenitors exit the cell cycle and migrate out of the ventricular zone ( VZ ) to reach the lateral spinal cord . Differentiated interneurons , including commissural and ipsilateral-projecting neurons continue to migrate laterally and different subpopulations also migrate dorsally or ventrally to reach their final positions . Mature commissural neurons then grow out axons and project them toward and across the ventral midline ( Caspary and Anderson , 2003; Helms and Johnson , 2003; Lu et al . , 2015; Figure 1A ) . 10 . 7554/eLife . 14264 . 004Figure 1 . Nova1/2 loss of function disrupts the development of dorsal spinal cord interneurons . ( A ) Schematic of neural development in the dorsal spinal cord during whole embryo culture . Nucleotides are microinjected and electroporated into neural progenitors at the superficial layer of the ventricular zone ( VZ , outlined by dashed lines ) . We observed that the dorsal interneuron progenitors are mostly targeted whereas the ventral motor neuron progenitors are not . The dorsal interneuron progenitors migrate laterally and exit the VZ to differentiate into commissural and ipsilateral interneurons . These interneurons continue to migrate laterally as well as along the dorsoventral axis . Mature commissural neurons extend axons to and across the ventral midline . ( B ) Transverse sections of spinal cords electroporated with control or pan-Nova siRNAs , and sections from wildtype ( WT ) , Nova double knockout ( dKO ) , and Dcc KO . Actb-gfp was used to label the highly heterogeneous interneuron populations . Nova and Dcc deficiency leads to an increased number of neuroprogenitors in the VZ ( asterisk ) , and fewer and shorter ventrally-projecting axons ( arrow ) . Bracket , the ventral midline . ( C ) Quantification of phenotypes in B . Neuroprogenitors in the VZ is quantified as the ratio between the signal from the medial spinal cord ( boxed area 1 ) to that from the lateral spinal cord ( area 2 ) . Axon ventral projection is quantified as the ratio between the signal from axons that have reached the ventral margin ( area 3 ) and that from the beginning of the axon shaft ( area 4 ) . For quantification of phenotypes in all experiments , if Nova WT and dKO were not littermates , they were first normalized to the respective double heterozygous ( dHet ) littermates , and were then compared with each other . Dcc KOs were compared with WT littermate controls . Also see Figure 1—source data 1 for additional quantification . Data are represented as the mean ± SEM ( Student’s t-test , *p<0 . 05 , **p<0 . 001 ) . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 00410 . 7554/eLife . 14264 . 005Figure 1—source data 1 . Quantification of neuronal migration and axon projection phenotypes in cultured embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 00510 . 7554/eLife . 14264 . 006Figure 1—figure supplement 1 . RNAi phenotypes of candidate RNA-binding proteins . ( A ) Transverse sections of spinal cords treated with control , pan-Nova , and Upf1 siRNAs . The latter two siRNA treatments cause defects in progenitor migration and commissural axon ventral projection ( indicated with asterisk and arrow , respectively ) . Also see Figure 1 for Nova RNAi effect . ( B ) Openbook preparation of spinal cords treated with control , Elavl2 , and Igf2bp1 siRNAs . Neurons electroporated with gfp and siRNAs are orientated to the left , and their axons project to and across the midline ( bracket ) . Elavl2 and Igf2bp1 knockdown causes some commissural axons to abnormally remain on the ipsilateral side ( indicated with arrow ) . Scale bars , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 00610 . 7554/eLife . 14264 . 007Figure 1—figure supplement 2 . Nova and Dcc KOs cause axon projection defect . ( A ) Atoh1-gfp was used to label the dorsal most dI1 population of interneurons and their progenitors . Nova and Dcc KO axons are shorter and fail to reach the midline ( bracket ) . ( B ) Quantification of ventral axon projection in A . The distance from the dorsal margin of the spinal cord to the ventral most axons is compared to the total height of the spinal cord ( 1/2 ratio ) . Data are normalized to WT and are represented as the mean ± SEM ( Student’s t-test , **p<0 . 001 ) . Also see Figure 1—source data 1 for additional quantification . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 007 Among the candidates , knockdown of Nova1/2 , Upf1 , Elavl2 , and Igf2bp1 caused commissural neuron defects ( Figure 1—figure supplement 1 ) . Nova1/2 and Upf1 knockdown disrupted both neuronal migration and axon projection ( Figure 1—figure supplement 1A and see below ) . Elavl2 and Igf2bp1 knockdown partially blocked commissural axons from crossing the midline ( Figure 1—figure supplement 1B ) . Since many RBPs have multiple family members , the lack of effect from other candidates may result from functional redundancy between homologues . We found that single knockdown of either Nova1 or Nova2 had no effect compared with control siRNAs , but a pan-Nova siRNA that targets both homologues caused severe defects ( Figure 1B , C ) . First , there were an increased number of GFP-positive neurons within the VZ , indicating an abnormality in the interneuron progenitors . Second , there were fewer ventrally projecting axons , which often failed to reach the midline , suggesting a defect in commissural axon growth and/or guidance ( Figure 1B , C , Figure 1—source data 1 ) . To corroborate the RNAi effect , we examined Nova1/2 knockout ( KO ) embryos labeled with Actb-gfp . We found that neither Nova1 nor Nova2 single KOs displayed any defect . However , Nova1 KO; Nova2 KO ( referred to hereafter as Nova dKO ) caused the same defects as Nova1/2 double knockdown ( Figure 1B , C , Figure 1—source data 1 ) . In addition , Nova1 Het; Nova2 KO displayed the same defects , whereas other genotype combinations including Nova1 KO; Nova2 Het are phenotypically normal ( see below ) . Thus , Nova1 and Nova2 have redundant functions in regulating dorsal interneuron development and Nova2 appears to play a major role . The Netrin-DCC signaling is required to stimulate commissural axon outgrowth and to attract the axons to the ventral midline , we thus compared cultured Dcc KOs with Nova dKOs . Interestingly , we observed the same defects in Dcc KOs , namely more GFP+ neurons within the VZ , and shorter and fewer ventrally projecting axons ( Figure 1B , C , Figure 1—source data 1 ) . We also labeled Nova and Dcc KOs with Atoh1-gfp ( aka Math1-gfp ) , which is expressed in the dorsal most dI1 subpopulation of interneurons and their progenitors ( Lumpkin et al . , 2003 ) . Similarly , we observed a reduction in axon ventral projection in both mutants ( Figure 1—figure supplement 2 , Figure 1—source data 1 ) . Due to the small number of neurons being labeled , we could not determine if there are more Atoh1-gfp expressing neurons in the VZ in the mutants . To determine the identity of the ACTB-GFP+ neurons and axons , we immunostained the cultured embryos with neuronal and axonal markers . We used PAX3/7 antibodies to label the dorsal interneuron progenitors that give rise to different populations of commissural and ipsilateral neurons ( Caspary and Anderson , 2003; Helms and Johnson , 2003 ) . Pax3 and Pax7 are required to specify the majority of commissural neurons , as their double knockout greatly reduces the ventral commissure formed by commissural axons crossing the midline ( Mansouri and Gruss , 1998 ) . We labeled differentiated interneurons with antibodies against BARHL2 and LHX5 , two transcription factors among many others that are expressed by post-mitotic interneurons . Combinations of these transcription factors define the different lineages of commissural and ipsilateral neurons ( Caspary and Anderson , 2003; Helms and Johnson , 2003 ) . We also labeled commissural axons specifically with anti-ROBO3 ( Sabatier et al . , 2004 ) . Our studies show that the GFP+ neurons within the VZ are dorsal interneuron progenitors and the ones at the lateral spinal cord are differentiated interneurons . In addition , the GFP+ axons that fail to project to the midline are commissural axons ( Figure 2 ) . Furthermore , using fluorescent in situ hybridization , we found that the GFP+ neuroprogenitors and interneurons also express Dcc ( Figure 2—figure supplement 1 ) , consistent with previous reports that Dcc is expressed in these neuronal populations ( Keino-Masu et al . , 1996; Phan et al . , 2011 ) . Taken together , Nova deficiency appears to disturb commissural axon projection and also interferes with earlier stages of commissural neuron development in the progenitors . 10 . 7554/eLife . 14264 . 008Figure 2 . Expression of neuronal and axonal markers in cultured Nova WT and dKO embryos electroporated with Actb-gfp . PAX3/7 immunostaining labels dorsal interneuron progenitors , which reside within the ventricular zone and give rise to different populations of interneurons . BARHL2 and LHX5 stainings label differentiated interneurons at the lateral spinal cord . ROBO3 staining specifically labels the commissural axons . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 00810 . 7554/eLife . 14264 . 009Figure 2—figure supplement 1 . Fluorescent in situ hybridization of Dcc in cultured embryos electroporated with Actb-gfp . Nova dKO was used as there are more GFP+ neurons in the VZ . Dcc expression is seen in GFP+ neuroprogenitors and interneurons . Scale bars , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 009 The increase in GFP+ neuroprogenitors in the VZ could result from slowed migration , increased proliferation , or reduced neuronal differentiation . To follow cell migration , we electroporated the progenitors with Actb-gfp and cultured the embryos for 20 hrs . In WT embryos , we observed that many progenitors have reached the lateral spinal cord after 20 hrs . However , in both Nova and Dcc KOs , almost all GFP+ neurons are still positioned within the VZ ( Figure 3 ) . This supports the idea that neuronal migration is reduced . We examined cell proliferation in Nova and Dcc KOs using cell cycle markers , including phospho-histone H3 ( pH3 ) , a mitotic marker , and Ki-67 , a cell proliferation marker . We found that at E10 . 5 , the number of neural stem cells and progenitors is normal in both mutants ( Figure 3—figure supplement 1 ) . In addition , we labeled the neural progenitors in the whole spinal cord with SOX2 and the dorsal progenitors with PAX3/7 , and found no change in the localization or organization of these progenitor populations ( Figure 3—figure supplement 1 ) . We also examined interneuron differentiation using the BARHL2 , ISL1/2 , and LHX5 markers , and found that a normal number of interneurons are born in Nova and Dcc KOs at E10 . 5 ( Figure 3—figure supplement 2; Figure 4 ) . Therefore , both Nova and Dcc deficiency reduces neuroprogenitor migration , but does not affect cell proliferation or interneuron differentiation . 10 . 7554/eLife . 14264 . 010Figure 3 . Nova and Dcc knockouts delay the lateral migration of dorsal interneuron progenitors . Actb-gfp was electroporated into neural progenitors and embryos were cultured for 20 hrs . The bottom panel shows the closeup images of the boxed area from the middle panel . Many WT neurons have migrated out of the VZ ( demarcated by PAX3/7 staining ) after 20 hrs , whereas Nova and Dcc KO neurons are mostly located within the VZ . Scale bars , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 01010 . 7554/eLife . 14264 . 011Figure 3—figure supplement 1 . Neural stem cells and progenitors appear normal in Nova and Dcc KOs . ( A ) Immunohistochemistry of phospho-histone H3 , a mitotic marker , Ki-67 , a cell proliferation marker , SOX2 , a neural progenitor marker , and PAX3/7 , a dorsal interneuron progenitor marker , in E10 . 5 spinal cords . ( B ) Quantification of phenotypes in A . pH3+ cells were counted and normalized to WT . The ratio between the SOX2+ VZ area and the total spinal cord area was quantified and normalized to WT . Data are represented as the mean ± SEM ( Student’s t-test , ns , not significant ) . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 01110 . 7554/eLife . 14264 . 012Figure 3—figure supplement 2 . Dorsal interneuron differentiation is normal in Nova and Dcc KOs . ( A ) Immunohistochemistry of ISL1/2 and LHX5 in Nova WT , Nova dKO , Dcc WT , and Dcc KO spinal cords at E10 . 5 . The markers are expressed by different subpopulations of interneurons in the dorsal spinal cord . ( B ) Quantification of ISL1/2+ neurons located in the dorsal half of the spinal cord . Data are normalized to WT and are represented as the mean ± SEM ( Student’s t-test , ns , not significant ) . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 01210 . 7554/eLife . 14264 . 013Figure 4 . Nova and Dcc knockouts delay neuronal migration but do not disturb neuronal differentiation . ( A ) Immunohistochemistry of BARHL2 in Nova WT , Nova dKO , Dcc WT , and Dcc KO spinal cords . A normal number of BARHL2+ neurons are generated in Nova and Dcc mutants at E10 . 5 . At E11 . 75 , BARHL2+ neurons fail to arrive at the lateral margin of the spinal cord in the KOs ( indicated by arrow ) . Their ventral migration is also delayed . At E12 . 5 , both dI1i ( ipsilateral ) and dI1c ( commissural ) populations appear normal in the mutants . ( B ) Quantification of phenotypes in A . For quantification of the lateral migration , the distance between the lateral margin of the spinal cord and the medial most BARHL2+ neurons are divided into two . The percentage of neurons within each half is shown . For quantification of the ventral migration , the distance from the dorsal margin of the spinal cord to the ventral most BARHL2+ neurons are divided into four . The percentage of neurons within each quarter ( 1 to 4 from dorsal to ventral ) is shown . Three embryos from each genotype and at least five sections from each embryo were quantified . Data are normalized to WT for total neuron number and dI1c/dI1i ratio . Data are represented as the mean ± SEM ( Student’s t-test for total neuron number , dI1c/dI1i ratio , and lateral migration , **p<0 . 001 , ns , not significant . Two way ANOVA and Bonferroni post test for ventral migration , *p<0 . 05 ) . Scale bars , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 013 We further examined the migration of post-mitotic interneurons using the BARHL2 marker . Barhl2 acts downstream of Atoh1 to specify the dI1 interneurons , including the dI1i ( ipsilateral ) and dI1c ( commissural ) populations ( Ding et al . , 2012; Figure 4A ) . DI1 neurons migrate both ventrally and laterally , and dI1i migrates even further laterally than dI1c ( Ding et al . , 2012 ) . As discussed above , a normal number of BARHL2+ neurons are generated at E10 . 5 in Nova and Dcc KOs ( Figure 4 ) . By E11 . 75 , most neurons in WT have migrated ventrally and some start to arrive at the lateral margin of the spinal cord ( Ding et al . , 2012; Figure 4 ) . In contrast , there are fewer BARHL2+ neurons at the ventral most and lateral most positions in Nova and Dcc KOs ( Figure 4 ) , suggesting a reduction in both dorsoventral and mediolateral migration . Later at E12 . 5 , the dI1c and dI1i populations can be well discerned in WT ( Ding et al . , 2012; Figure 4A ) . Similarly , Nova and Dcc KOs also had two distinct populations of BARHL2+ neurons and the ratio between them is normal ( Figure 4 ) . Therefore , despite a transient delay in the migration , both commissural and ipsilateral neurons are properly specified . To understand the defect in commissural axon ventral projection , we examined axon outgrowth in Nova dKO using explants of the dorsal spinal cord ( DSC ) . In this assay , commissural axons extend out of the explant in the presence of Netrin-1 after a culture period of 16 hrs ( Serafini et al . , 1994 ) . In the absence of Netrin-1 , commissural axons are also able to grow out , but only after an extended period of culturing ( after 40 hrs; Keino-Masu et al . , 1996 ) . We isolated E11 . 5 DSC from Nova dKOs and first tested if commissural neurons are defective in Netrin-stimulated axon outgrowth . Even after 24 hrs in the presence of Netrin-1 , when control explants had robust axon outgrowth , Nova dKO axons had very little growth ( Figure 5 ) . The same outgrowth defect was also seen in Dcc KOs , as previously reported ( Xu et al . , 2014; Figure 5 ) . 10 . 7554/eLife . 14264 . 014Figure 5 . Nova dKO disrupts Netrin-1 induced axon outgrowth in dorsal spinal cord ( DSC ) explant assays . ( A ) DSC axon outgrowth in Nova WT , Nova dKO , Dcc WT , and Dcc KO in the presence or absence of 250 ng/ml Netrin-1 . Axons were visualized by rhodamine-phalloidin staining . Nova dKO and Dcc KO display a drastic reduction in axon growth in response to Netrin . However , both mutants are able to extend axons after an extended culture period in the absence of Netrin . The insets in the bottom panel show no axon outgrowth after 24 hrs of culturing for all genotypes in the absence of Netrin . ( B ) Quantification of axon outgrowth in A . Axon outgrowth is represented as the ratio between the signal from all axons extending out of the explant and that from the cell bodies within the explant , after background extraction . Three embryos from each genotype and at least five explants from each embryo were quantified . Data are normalized to WT and are represented as the mean ± SEM ( Student’s t-test , **p<0 . 001 , ns , not significant ) . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 01410 . 7554/eLife . 14264 . 015Figure 5—figure supplement 1 . Commissural axon outgrowth in Nova embryos . ( A ) Cultured dorsal spinal cord explants from nine genotypes . Explants were cultured for 24 hrs in the presence of 250 ng/ml Netrin-1 and axons were labeled with rhodamine-phalloidin . ( B ) Quantification of axon growth in A ( normalized to WT ) . Data are represented as the mean ± SEM ( one way ANOVA and Bonferroni post test , **p<0 . 001 , ns , not significant ) . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 015 We also examined Netrin-independent axon outgrowth by culturing the explants in the absence of Netrin-1 for 40 hrs ( Figure 5 ) . As expected , we did not observe any axon outgrowth after 24 hrs . However after 40 hrs , there was robust axon outgrowth from both control and Nova dKO , and the degree of growth is comparable between the two . In addition , Dcc KOs also displayed the same degree of Netrin-independent axon outgrowth ( Figure 5 ) . Therefore , like Dcc KOs , Nova dKO commissural axons do not have a general growth defect . Rather , they fail to grow in response to Netrin stimulation . Loss of Netrin-DCC mediated attraction in vivo reduces the number of commissural axons that are able to reach the midline ( Fazeli et al . , 1997; Serafini et al . , 1996; Xu et al . , 2014 ) . We thus examined commissural axon guidance to the midline using immunostaining of the axonal markers ROBO3 and TAG-1 . We found that at E10 . 5 and E11 . 5 , the intensity of these axonal markers from the ventral half of the spinal cord was significantly reduced in Nova dKOs , suggesting a reduction of ventral axon projection ( Figure 6 ) . At E11 . 5 , there are usually two main commissural axon bundles . Both are reduced in Nova dKOs and the more lateral one is more profoundly affected . The size of the ventral commissure , formed by axons crossing the midline , is also reduced ( Figure 6 ) . By E12 . 5 , the reduction in the commissure size is still significant but is somewhat alleviated than at earlier stages . A similar reduction in axons reaching and crossing the ventral midline is also seen in Dcc KOs ( Fazeli et al . , 1997; Xu et al . , 2014 ) . The severity of the defect is similar between Nova and Dcc mutants ( Figure 6 ) . One distinction between the two mutants is that Dcc KO axons are defasciculated and often invade the motor column ( Xu et al . , 2014 ) , whereas such a defect was not observed in Nova dKOs ( Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 14264 . 016Figure 6 . Nova dKO disrupts commissural axon ventral projection . ( A ) and ( B ) Immunohistochemistry of commissural axonal markers , ROBO3 and TAG-1 respectively , in transverse sections of the spinal cord . Two main axon bundles can be observed at E11 . 5 ( arrows ) and the more lateral one is severely reduced in Nova dKOs . In the top panel of B , arrows indicate commissural axon projection , and the staining in the ventral lateral spinal cord is from motor axons . ( C ) Quantification of axon guidance phenotypes in A . To quantify axon ventral projection , the signal from the ventral half of the spinal cord was first normalized to the signal from the whole spinal cord , and then normalized to WT controls . The commissure size is represented as the ratio between the thickness of the axon bundle at the midline and that of the floorplate . Data are normalized to WT . Three embryos from each genotype and at least five sections from each embryo were quantified . Data are represented as the mean ± SEM ( Student’s t-test , *p<0 . 05 , **p<0 . 001 ) . Scale bars , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 01610 . 7554/eLife . 14264 . 017Figure 6—figure supplement 1 . Nova dKO does not cause commissural axons to invade the motor column . ( A ) Immunohistochemistry of ROBO3 in Nova WT , Nova dKO , Dcc WT , and Dcc KO spinal cords . Only the ventral half is shown ( MC , motor column ) . Some commissural axons in Dcc KO , but not in Nova dKO , are defasciculated and invade the motor column ( arrows ) . ( B ) Quantification of phenotypes in A . The amount of commissural axons in the motor column is represented as the ratio between the ventral area covered by commissural axons and that of the total ventral area . Data are normalized to WT and are represented as the mean ± SEM ( Student’s t-test , *p<0 . 05 , ns , not significant ) . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 017 Besides Netrin , SHH ( sonic hedgehog ) and VEGF ( vascular endothelial growth factor ) have also been shown to attract commissural axons ( Charron et al . , 2003; Ruiz de Almodovar et al . , 2011 ) . However , the loss of SHH or VEGF and their receptors in vivo does not lead to a significant reduction of the ventral commissure ( Charron et al . , 2003; Okada et al . , 2006; Ruiz de Almodovar et al . , 2011 ) . SHH and VEGF also do not induce commissural axon outgrowth in the DSC assay , and their deficiency does not affect commissural axon outgrowth in vivo . Taken together , our results show that Nova loss of function disrupts Netrin-DCC signaling during commissural axon outgrowth and guidance . To determine if the Netrin-DCC pathway is affected on the molecular level , we set out to examine the expression and alternative splicing of the pathway components . Using in situ hybridization , we found that Nova1 and Nova2 are highly expressed in commissural neurons , but not in the floorplate , where Netrin is expressed ( Figure 7—figure supplement 1 ) . This expression pattern as well as the fact that isolated Nova dKO neurons fail to extend axons in response to Netrin suggest that Nova1/2 most likely function cell autonomously within commissural neurons . Using quantitative RT-PCR and western blotting , we first examined the total mRNA and protein levels of Dcc . We found that they were not significantly altered in Nova dKOs , although there was a slight increase in both levels ( Figure 7—figure supplement 2A , B ) . Both human and mouse Dcc undergoes alternative splicing at exon 17 to give rise to Dcclong and Dccshort , with Dcclong containing extra 60 bp ( Reale et al . , 1994; Figure 7A ) . Human Dcc undergoes additional alternative splicing in multiple regions in tumor cells ( Reale et al . , 1994; Huerta et al . , 2001 ) . By RT-PCR using spinal cord tissues from both wildtype and Nova dKO , we found that mouse Dcc does not produce alterative mRNAs in these additional regions . To determine if there are previously unknown alternative splicing events in mouse Dcc , we amplified and sequenced overlapping cDNA fragments from CD-1 mice , and did not identify any additional alternative splicing . 10 . 7554/eLife . 14264 . 018Figure 7 . Nova dKO disrupts the alternative splicing of Dcc . ( A ) Schematic of Dcc alternative splicing and the resulting isoforms . The alternative 60 bp in exon 17 is shaded grey . ( B ) Dcc isoform expression in nine Nova1/2 genotypes , as measured by quantitative RT-PCR using E11 . 5 dorsal spinal cord . The isoform expression level is normalized to a common region in Dcc cDNA ( total Dcc; see materials and methods ) . Note that although the changes in Nova2 single KOs are statistically significant , the in vivo phenotype of Nova2 KOs is not significantly different from controls ( see Figure 5—figure supplement 1 ) . ( C ) Dcc isoform expression from Nova embryos detected by semi-quantitative PCR that amplify both isoforms . Dorsal and ventral spinal cords are separated to distinguish commissural neurons ( dorsally located ) and motor neurons ( ventrally located ) . Nova dKO affects Dcc alternative splicing in both populations . Three animals from each genotype were quantified . Data are represented as the mean ± SEM ( one way ANOVA and Bonferroni post test , **p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 01810 . 7554/eLife . 14264 . 019Figure 7—figure supplement 1 . Nova1 and Nova2 expression in E11 . 5 spinal cord detected by in situ hybridization . Both genes are expressed in interneurons at the lateral spinal cord and are highly expressed in the commissural neurons ( C ) . M , motor neurons . FP , the floorplate . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 01910 . 7554/eLife . 14264 . 020Figure 7—figure supplement 2 . Nova dKO does not affect the total level of Dcc , Neo1 , and Robo3 expression . ( A ) Total Dcc mRNA level in the spinal cords from nine genotypes of Nova1/2 , measured by quantitative RT-PCR . There is a slight but not significant increase . ( B ) DCC protein level detected by SDS-PAGE and western blotting in Nova WT and dKO spinal cords . The anti-DCC ( AF5 ) recognizes both isoforms . There is also a slight but not significant increase . ( C ) Total Neo1 mRNA level in Nova WT and dKO . ( D ) Total Robo3 mRNA level in Nova WT and dKO . Data are represented as the mean ± SEM ( one way ANOVA and Bonferroni post test in A . Student’s t-test in B , C and D . ns , not significant ) . For all three genes , multiple regions of the cDNAs were examined and one is shown as an example . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 02010 . 7554/eLife . 14264 . 021Figure 7—figure supplement 3 . Nova dKO does not affect the alternative splicing of Neo1 , Robo3 , or Epha5 . ( A ) Neo1 alternative splicing in Nova mutants . Two alternative regions are shown , one at the homologous region to Dcc exon 17 and the other at Neo1 exon 27 . Quantitative RT-PCR results are shown in the chart on the left and semi-quantitative RT-PCRs of the same regions ( two animals for each genotype ) are shown in the electrophoresis image on the right . ( B ) Robo3 alternative splicing in Nova mutants . Six alternative regions are shown as examples and additional regions were examined . No change is seen in any region . ( C ) Epha5 alternative splicing at exon 7 in Nova mutants . Three animals from each genotype were quantified . Data are normalized to WT and are represented as the mean ± SEM ( Student’s t-test , ns , not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 021 Using quantitative RT-PCR , we specifically examined Dcc alternative splicing at exon 17 and found that it was significantly disturbed in Nova dKOs . Dcclong was greatly diminished , whereas Dccshort was upregulated ( Figure 7B , C ) . In addition , the changes appear to be sensitive to Nova1/2 gene copy number . When Nova2 is reduced from two copies ( wildtype ) to one ( heterozygote ) and further to zero ( knockout ) , Dcclong decreases and Dccshort increases stepwise ( Figure 7B ) . Nova1 KO; Nova2 KO has the most significant changes among all genotypes . Consistently , the defects seen in Nova1 KO; Nova2 KO is the most severe ( Figure 5—figure supplement 1 ) . Together , these results suggest that Nova normally regulates Dcc alternative splicing by promoting Dcclong in a dose-sensitive manner . It is interesting to note that even though the changes in Nova2 single KOs are statistically significant , no defect was seen in this genotype ( Figure 5—figure supplement 1 ) . This suggests that there may be a threshold for the level of Dcc isoforms for normal function . Alternatively , the somewhat elevated total Dcc expression ( Figure 7—figure supplement 2A , B ) may be able to compensate for a small disruption of activity . Consistent with our results , accompanying paper by Saito et al . also found that the alternative splicing at Dcc exon 17 is significantly altered in E18 . 5 Nova2 KO cortex , which may contribute to additional axon guidance defects in the brain . Several molecules have been suggested to function as Netrin co-receptors within commissural neurons , including Neogenin ( NEO1 ) , ROBO3 , DSCAM , and APP ( Ly et al . , 2008; Rama et al . , 2012; Xu et al . , 2014; Zelina et al . , 2014 ) . Loss of function in Neo1 , a Dcc homolog and a NOVA target ( see accompanying paper by Saito et al . ) , does not affect commissural axon outgrowth or guidance by itself , but can enhance the defects of Dcc KOs ( Xu et al . , 2014 ) . The ROBO3 receptor , specifically the ROBO3A . 1 isoform , represses premature repulsion before axons reach the midline and also forms a complex with DCC to mediate Netrin attraction ( Chen et al . , 2008; Sabatier et al . , 2004; Zelina et al . , 2014 ) . DSCAM and APP have not been shown to be required for Netrin activity in vivo ( Palmesino et al . , 2012; Rama et al . , 2012 ) . Using primers that detect different regions of the cDNAs , we found that the total mRNA level of Neo1 and Robo3 was not affected ( Figure 7—figure supplement 2C , D ) . We also examined Neo1 alternative splicing in a homologous region to Dcc exon 17 , and found that it was not significantly altered in the dorsal spinal cord in Nova dKOs ( Figure 7—figure supplement 3A ) . Other alternative regions of Neo1 ( Keeling et al . , 1997 ) were not significantly altered either ( Figure 7—figure supplement 3A ) . Robo3 pre-mRNA undergoes alternative splicing in multiple regions ( Camurri et al . , 2005; Chen et al . , 2008; Yuan et al . , 1999 ) . We measured the expression levels of these alternative areas , as well as the adjacent constant exons , and found no change in any of them in Nova dKOs compared with controls ( Figure 7—figure supplement 3B ) . For further comparison , we also examined the alternative splicing of the Epha5 receptor , which is a known NOVA target but has not been shown to play a role in the development of the dorsal spinal cord interneurons . We did not observe any change in the alternative splicing of Epha5 exon 7 ( Figure 7—figure supplement 3C ) , which is found to be altered in E18 . 5 Nova2 KO cortex ( see accompanying paper by Saito et al . ) . Together , these results show that Dcc is the most likely target of Nova1/2 that is affected within commissural neurons , while other receptors are unlikely to be responsible for the Nova dKO defects . If the Nova dKO defects are indeed caused by disturbed Dcc alternative splicing , with Dcclong greatly reduced , restoring Dcclong in the mutants should be able to reverse the defects . We first used the axon outgrowth assay to compare the rescuing abilities of Dcc isoforms . We electroporated either isoform into the spinal cords at E10 . 5 , cultured the embryos for one day to allow exogenous protein expression , and then carried out the DSC assay . As shown above , Nova dKO axons fail to grow in response to Netrin-1 ( Figure 5; Figure 8A , B ) . When Dcclong was introduced back into Nova dKOs , many axons extended out of the explant , whereas Dccshort was unable to rescue the defect ( Figure 8A , B ) . For comparison , we also overexpressed Robo3A . 1 in Nova dKOs , and found that it was unable to restore commissural axon outgrowth in Nova dKOs ( Figure 8A , B ) . 10 . 7554/eLife . 14264 . 022Figure 8 . Expression of DCClong , but not DCCshort or Robo3A . 1 , is able to rescue Nova dKO defects . ( A ) Dorsal spinal cord assays using Nova WT and dKO neurons electroporated with gfp , Dcclong , Dccshort , or Robo3A . 1 . The embryos were electroporated at E10 . 5 and cultured for one day to allow protein expression . DSC assays were then carried out and axons were directly visualized by GFP fluorescence . Explants were cultured for 24 hrs with 250 ng/ml Netrin-1 . Only DCClong expression can rescue the outgrowth defect in Nova dKOs . ( B ) Quantification of DSC axon length and number in A . Three embryos from each treatment/genotype and at least five explants from each embryo were quantified . ( C ) Transverse sections of spinal cords from Nova WT and dKO electroporated with gfp , Dcclong , Dccshort , or Robo3A . 1 . Only Dcclong is able to rescue the neuronal migration and axon projection defects . Bracket , ventral midline . ( D ) Quantification of phenotypes in C ( see description in Figure 1C ) . Also see Figure 1—source data 1 for additional quantification . ( E ) Expression of DCClong , DCCshort , and ROBO3A . 1 proteins in electroporated embryos ( two embryos are shown ) . All three proteins were immunoprecipitated by a C-terminal HA tag and detected by western blotting . ( F ) Expression of DCClong , DCCshort , and ROBO3A . 1 proteins in dissociated dorsal spinal cord neurons , as detected by anti-HA . Data are represented as the mean ± SEM ( one way ANOVA and Bonferroni post test , *p<0 . 05 , **p<0 . 001 , ns , not significant ) . Scale bars , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 02210 . 7554/eLife . 14264 . 023Figure 8—figure supplement 1 . Rescue of Nova dKOs by Dcc isoforms in embryos labeled with Atoh1-gfp . ( A ) Nova dKO was electroporated with gfp , Dcclong , or Dccshort . Only Dcclong expression is able to rescue the axon projection defect . ( B ) Quantification of ventral axon projection in A . The distance from the dorsal margin of the spinal cord to the ventral most axons is compared to the total height of the spinal cord ( 1/2 ratio ) . Data are normalized to WT and are represented as the mean ± SEM ( one way ANOVA and Bonferroni post test , *p<0 . 05 , **p<0 . 001 ) . Also see Figure 1—source data 1 for additional quantification . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 023 We then introduced Dcc isoforms back into cultured embryos to determine if restoring Dcclong could rescue additional defects seen in Nova dKOs . We electroporated the cDNAs and Actb-gfp into embryos at E9 . 5 and cultured the embryos for two days . We found that Dcclong expression in Nova dKOs could indeed ameliorate the neuronal migration and axon ventral projection defects ( Figure 8C , D , Figure 1—source data 1 ) . In contrast , overexpression of Dccshort , the isoform that is abnormally upregulated , did not rescue . In fact , it exacerbated the defects , causing even fewer axons to project ventrally toward the midline ( Figure 8C , D , Figure 1—source data 1 ) . We also overexpressed Robo3A . 1 in Nova dKOs and found that it did not have any effect on neuronal migration or axon projection ( Figure 8C , D , Figure 1—source data 1 ) . To confirm the expression of the exogenous proteins , which are tagged with an HA peptide at the very C-terminus , we performed immunoprecipitation from electroporated embryos and analyzed the protein levels by SDS-PAGE and western blotting . All proteins were expressed at detectable and comparable levels ( Figure 8E ) . We also dissociated electroporated spinal cords and examined the protein expression within GFP+ neurons . Using HA antibodies , we were able to detect all three proteins within axons ( Figure 8F ) , showing that the proteins are properly expressed and localized . Therefore , the rescue of Nova dKOs only by Dcclong suggests that the defects result directly from a loss in Dcc activity , in particular Dcclong activity . The inability of ROBO3A . 1 to rescue Nova dKOs is consistent with the fact that Robo3 expression and alternative splicing are not disrupted by Nova dKO . Furthermore , using the Atoh1-gfp marker , we also found that Dcclong expression , but not Dccshort expression was able to rescue the axon projection defect in Nova dKOs to a large extent ( Figure 8—figure supplement 1 , Figure 1—source data 1 ) . NOVA proteins specifically recognize clusters of YCAY ( Y=C/U ) sequences ( Buckanovich and Darnell , 1997 ) . A genome-wide study using HITS-CLIP ( high-throughput sequencing of RNA isolated by crosslinking immunoprecipitation ) has identified candidate target sites within Dcc pre-mRNA ( Zhang et al . , 2010 ) . To determine if NOVA1/2 directly regulate Dcc alternative splicing , we carried out splicing assays using a Dcc minigene containing the genomic DNA between exons 16 and 17 ( Figure 9A ) . The alternative sequence is located within exon 17 and the candidate YCAY clusters are located within exon 16 , intron 16 , and exon 17 . We coexpressed the minigene with an empty vector or Nova1 , Nova2 , or Ptbp2 ( an unrelated splicing factor that has not been shown to affect Dcc alternative splicing ) . When the vector alone was coexpressed , we detected two RT-PCR products corresponding to Dcclong and Dccshort , respectively ( Figure 9B ) . When Nova1 or Nova2 was coexpressed , Dcclong was upregulated , while Dccshort was reduced . Such changes were not observed when Ptbp2 was coexpressed ( Figure 9B ) . These results suggest that NOVA1/2 promote Dcclong , consistent with the observed reduction of Dcclong in Nova dKOs . Therefore , the in vitro splicing assays recapitulate Dcc alternative splicing pattern in vivo . 10 . 7554/eLife . 14264 . 024Figure 9 . NOVA1/2 regulate Dcc alternative splicing . ( A ) Schematic of a Dcc minigene containing the genomic DNA between exons 16 and 17 . The alternative sequence is shaded gray . Three candidate Nova binding sites , which are YCAY ( Y=C/U ) clusters , are located within exon 16 , intron 16 , and exon 17 , respectively ( dashed lines indicate the number of YCAY repeats ) . ( B ) Alternative splicing of wildtype and mutant minigenes . The two RT-PCR products correspond to Dcclongand Dccshort , respectively . Two additional RT-PCR products are produced by exon 17 mutations ( asterisk ) , from utilizing two downstream cryptic splice sites . Mutations in intron 16 block NOVA1/2 from increasing Dcclong . ( C ) Alternative splicing of Dcc minigenes containing different numbers of mutations in intron 16 . With an increasing number of mutations , NOVA1/2 gradually lose their ability to promote Dcclong . DOI: http://dx . doi . org/10 . 7554/eLife . 14264 . 024 To further determine the specificity of the effect of NOVA , we mutated the YCAY repeats to YAAY , which can no longer be recognized by NOVA ( Buckanovich and Darnell , 1997 ) . The exon 16 cluster mutations did not cause any change in Dcc alternative splicing ( Figure 9B ) . In contrast , when the intron 16 cluster was mutated , NOVA1 or NOVA2 could no longer promote Dcclong , indicating that these sites are essential NOVA binding sites . The exon 17 cluster is in close proximity to the splice acceptor site for Dccshort , and its mutation produced two additional RT-PCR products , generated from utilizing two downstream cryptic splice acceptors . However , NOVA1 or NOVA2 could still increase Dcclong ( Figure 9B ) . Overall , these results show that the intron 16 cluster is the main binding site for NOVA , and the binding promotes the production of Dcclong . The intron 16 cluster contains six YCAY repeats and is conserved in humans , with seven repeats present in human sequence . We next examined if the number of repeats could affect how effectively NOVA regulates Dcc alternative splicing by introducing mutations into a subset of the repeats . We found that mutations in just one repeat had little effect , but mutations in four repeats led to a partial reduction in Dcclong when Nova1/2 were co-expressed ( Figure 9C ) . With five and six mutated repeats , there were even further decreases in Dcclong and corresponding increases in Dccshort ( Figure 9C ) . These results are consistent with the dose-dependent interaction between NOVA and their RNA targets ( Darnell , 2006; Figure 7B ) , and further support that these sequences are bona fide NOVA binding sites .
Neuronal and axonal migration requires dynamic regulation of cell signaling , particularly evident as cells/axons encounter an intermediate target such as the midline of the central nervous system ( Dickson and Zou , 2010; Evans and Bashaw , 2010; Kaprielian et al . , 2001 ) . Alternative splicing has been increasingly implicated as an important means to generate temporal and spatial specific functions for guidance molecules ( Chen et al . , 2008; Park and Graveley , 2007 ) . To further understand the functional significance of alternative splicing , it is important to identify the relevant splicing factors and their targets . Through an in vivo RNAi screen against RBPs , we discovered that Nova1/2 are key regulators of both neuronal and axonal migration in the spinal cord interneurons . We found that Nova deficiency reduces the migration of mitotic progenitors and differentiated interneurons in the spinal cord ( Figure 3 , 4 ) . We observed similar defects in Dcc KOs , which were mostly uncharacterized before . Within the commissural interneurons , Nova dKO disrupts Netrin-induced axon outgrowth and ventral projection to the midline , also resembling the Dcc mutant ( Figure 5 , 6 ) . The phenotypic similarity between Nova and Dcc KOs strongly suggests that Dcc activity is affected by Nova dKO . Indeed , this is confirmed by the observations that Dcc alternative splicing is disrupted by Nova dKO ( Figure 7 ) , that Nova dKO defects can be largely rescued in vitro and in vivo by restoring Dcclong ( Figure 8 , Figure 8—figure supplement 1 , Figure 1—source data 1 ) , and that NOVA controls Dcc alternative splicing in vitro ( Figure 9 ) . Also importantly , our results show that disrupting Dcc alternative splicing without affecting the total Dcc level leads to as severe defects as Dcc KO , underscoring the importance of alternative splicing for the gene function . The defects in neuronal migration , axon outgrowth , and axon guidance in Nova and Dcc KOs appear to share a common feature , which is a temporal delay . The migration is slowed but not completely blocked ( Figure 4 ) . The axon outgrowth is initially defective but becomes normal after an extended culture period ( Figure 5 ) . The axon guidance defect also appears to be somewhat alleviated at E12 . 5 compared to earlier stages ( Figure 6 ) . This delay could be due to an incomplete loss of the gene function in the KOs . Alternatively , NOVA proteins have been shown to regulate many target genes including Dcc in a temporal specific manner , generating different ratios of splice variants at different developmental stages ( Yano et al . , 2010; also see accompanying paper by Saito et al . ) As Nova deficiency changes the ratio between DCC variants , it may alter the developmental state of the neurons . Consequently , this leads to abnormal responsiveness of the neurons to the extracelluar environment and thus defects in neuronal and axonal migration . The fact that the loss of Dcclong in Nova dKOs cannot be compensated by the increase in Dccshort demonstrates that the two isoforms are functionally distinct . The two DCC isoforms differ in the FN4-FN5 linker , with both FN4 and FN5 domains interacting with Netrin ( Xu et al . , 2014 ) . A structure study of Netrin in complex with the DCC or NEO1 receptor found that DCC variants can bind to Netrin-1 with comparable affinities , but are likely to adopt distinct conformations in the ligand-receptor complex ( Xu et al . , 2014 ) . Netrin and DCCshort form a continuous liand:receptor complex , whereas DCClong is likely to form a 2:2 ligand:receptor complex with Netrin ( Xu et al . , 2014 ) . Whether the presence of both isoforms and at different ratios can produce additional conformations of the ligand-receptor complex is completely unknown . How the architecturally distinct complexes can lead to different intracellular signaling also remains an intriguing question . The defects seen in Nova dKOs could result from the loss in Dcclong , the increase in Dccshort , or the combination of both . We cannot yet distinguish between these possibilities . The rescue of Nova dKOs by DCClong , but not by DCCshort , can also be interpreted in different ways . One possibility is that DCClong is fully responsible for the gene function , whereas DCCshort has no activity . Another is that both isoforms are required and each has its specific activity . Only DCClong can rescue Nova KOs because it is reduced in the mutants , while DCCshort is still present . Since the two Dcc isoforms are normally expressed at comparable levels in commissural axons at E11 . 5 ( Figure 7C ) , it is likely that each isoform has its unique activity during neuronal migration and axon guidance . Dcc alternative splicing was first identified using human neuroblastoma cells IMR32 ( Reale et al . , 1994 ) . Compared with normal mouse brain tissues , IMR32 expresses a decreased level of Dcclong and an elevated level of Dccshort ( Reale et al . , 1994 ) . These changes are in the same pattern as those in Nova dKOs . Thus , altered Dcc alternative splicing may also contribute to tumor development . Allelic loss of the 18q21 region encompassing Dcc is identified in about 70% of primary colorectal cancers and is also found in other types of cancers ( Mehlen and Fearon , 2004 ) . It remains to be seen if altered Dcc alternative splicing accounts for additional cases of colorectal cancers and other cancers . Furthermore , Dcc plays additional roles in the nervous system , such as in dendritic growth and guidance ( Furrer et al . , 2003; Nagel et al . , 2015; Suli et al . , 2006; Teichmann and Shen , 2011 ) , and in synapse formation and function ( Colon-Ramos et al . , 2007; Goldman et al . , 2013; Horn et al . , 2013 ) . Therefore , it is important to determine if Dcc alternative splicing is also important for other biological processes .
Nova1 , Nova2 , and Dcc KOs were generated and described previously ( Fazeli et al . , 1997; Huang et al . , 2005; Jensen et al . , 2000; Ruggiu et al . , 2009; accompanying paper by Saito et al . ) . All strains were outcrossed to the CD-1 strain . Nova double heterozygotes ( dHet ) were intercrossed to generate all nine genotypes used in the study . The siRNAs for candidate RNA-binding proteins were designed and synthesized by IDT ( Coralville , IA ) . A pool of three siRNAs were used in the screen . Once phenotypes were seen , individual siRNAs were validated and the most potent siRNA was used for further phenotypic analyses . The sense sequences for the most potent Nova siRNAs are as follows: Nova1 5’tacaacctcagaccaccgttaatcctg3’ , Nova2 5’gaccatcgtgcagctccagaaggagac3’ , and pan-Nova 5’agccaccatcaagctgtctaagtccaa3’ . Dcc , Nova1 , Nova2 , Ptbp2 cDNAs were cloned from wildtype CD-1 mouse spinal cords . Robo3A . 1cDNA was generated previously ( Chen et al . , 2008 ) . Actb-gfp ( egfp in pCAGGS ) and Atoh1-gfp markers were previously described ( Lumpkin et al . , 2003; Matsuda and Cepko , 2004 ) . WEC was carried out as previously described ( Chen et al . , 2008 ) . For the RNAi screen , embryos were electroporated at E9 . 5 with siRNAs and gfp into one side of the spinal cord and were cultured for 40-48 hrs . The embryos were then fixed in 4% paraformaldehyde , cryopreserved in 30% sucrose , and embedded in OCT . 20 μm transverse sections were collected and examined using fluorescent microscopy . Alternatively , openbook preparation of the spinal cord was performed and examined by fluorescent microscopy as previously described ( Chen et al . , 2008 ) . For phenotypic quantification in all experiments , if Nova WT and dKO were not littermates , they were first normalized to the respective dHet littermates , and were then compared with each other . Dcc KOs were compared with WT littermate controls . To minimize developmental variation , we used embryos of comparable sizes and examined spinal cord tissues from the brachial level . For quantification of cultured embryos labeled with Actb-gfp , neuroprogenitors in the VZ is represented as the ratio between the signal from the medial spinal cord and that from the lateral spinal cord . Axon ventral projection is represented as the ratio between the signal from axons at the ventral margin of the spinal and that from the beginning of the axon bundle . The signal intensity was measured using ImageJ ( NIH , Bethesda , MD ) . For quantification of Atoh1-gfp labeled embryos , ventral axon projection is represented as the ratio between the distance from the dorsal margin of the spinal cord to the ventral most axons and the total height of the spinal cord . The distances were measured using ImageJ . In all phenotypic analyses , the defects are consistently seen in all embryos examined . The severity of defects is comparable between animals and between different sections of the same embryo . Representative images are shown in all figures . Numbers of animals and sections examined are listed in Figure 1—source data 1 . IHC was carried out as previously descried ( Xu et al . , 2014 ) . Antibodies used in the study include anti-PAX3/7 ( PA1-107 , Thermo Fisher , Waltham , MA , raised against PAX3 and cross reacts with PAX7 ) , anti-BARHL2 ( NBP2-32013 , Novus Biologicals , Littleton , CO ) , anti-LHX5 ( AF6290 , R&D , Minneapolis , MN ) , anti-ISL1/2 ( 39 . 4D5 , DSHB , Iowa City , IA ) , anti-ROBO3 ( rabbit polyclonal , Chen et al . , 2008 ) , anti-TAG1 ( 4D7 , DSHB ) , anti-pH3 ( 9701 , CST , Danvers , MA ) , anti-Ki-67 ( 12202 , CST ) , and anti-SOX2 ( 3728 , CST ) . DSC assay was performed as previously descried ( Xu et al . , 2014 ) . For DSC taken from embryos grown in vivo , the explants were labeled with rhodamine-phalloidin ( Thermo Fisher ) . The outgrowth was quantified as the ratio between the signal from the axons and that from the cell bodies , after background extraction . The signal intensity was measured using ImageJ . For the rescue experiments using the dorsal spinal cord explants , embryos were electroporated with cDNAs and gfp at E10 . 5 and cultured for 24 hrs . GFP-positive DSCs were microdissected and cultured . The length and total number of GFP positive axons were quantified using ImageJ . At least three embryos from each genotype and five explants from each embryo were quantified , with genotypes and treatments blinded . Spinal cord tissues were microdissected , and the dorsal and ventral halves were separated to distinguish the commissural and motor neuron populations . Total RNA was extracted using Trizol ( Thermo Fisher ) , and reverse transcription was carried out using Maxima RT ( Thermo Fisher ) . Quantitative PCR was performed using a Realplex2 thermocycler ( Eppendorf , Hamburg , Germany ) . Semi-quantitative PCR was performed to generate multiple isoforms in a single reaction and compare the relative expression by electrophoresis . The cycle number used in semi-quantitative PCR was determined by quantitative PCR to obtain products during the exponential amplification phase . ISH was performed as previously described ( Braissant and Wahli , 1998 ) . The antisense probes were in vitro transcribed using T7 polymerase and labeled with DIG ( Digoxigenin , Roche , Basel , Switzerland ) . Sense probes were used as negative controls . For non-fluorescent ISH , AP ( alkaline phosphatase ) -conjugated anti-DIG antibody ( Roche ) was used to detect bound antisense probes , and was visualized using colorimetric AP substrates . For FISH , HRP ( horse radish peroxidase ) -conjugated anti-DIG antibody ( Jackson ImmunoResearch , West Grove , PA ) was used and the signal was visualized using the TSA ( tyramide signal amplication ) system ( Perkin Elmer , Waltham , MA ) . To confirm exogenous protein expression , DSC neurons electroporated with Dcc or Robo3 cDNA and gfp were microdissected , dissociated ( 0 . 05% trypsin , 0 . 5 mM EDTA ) , and cultured for 24 hrs in PDL ( poly-D-Lysine , 100 μg/ml ) coated culture dish in the culture medium ( Neurobasal , 1x B27 , 50 U/ml Pen/Strep , and 250 ng/ml Netrin-1 ) . Cells were then fixed with 4% paraformaldehyde , and stained with anti-HA ( 3F10 , Roche ) and Alexa Fluor 594-conjugated secondary antibodies ( Jackson ImmunoResearch ) . Dcc genomic DNA that spans exons 16 and 17 ( 5 . 6 kb total ) was PCR amplified from mouse spinal cords and cloned into the pDEST26 gateway vector containing a CMV promoter ( Thermo Fisher ) . Dcc minigene was transfected into HEK293T cells together with the splicing factors or an empty vector at a 1:1 ratio . Cells were cultured for 48 hr and the total RNA was collected using Trizol ( Thermo Fisher ) . Reverse transcription was carried out from a T7 promoter ( present in pDEST26 ) using SMARTScribe reverse transcriptase ( Clontech , Mountain View , CA ) , and semi-quantitative PCR was performed to amplify multiple isoforms . Point mutations were introduced by PCR reactions using Pfu polymerase ( Agilent , Santa Clara , CA ) , and were confirmed by DNA sequencing . A V5 tag at the C-terminus of NOVA1 , NOVA2 , and PTBP2 was used to confirm protein expression using western blotting . | 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 . If alternative splicing is important for this process , it is also important to ask: which splicing factors are relevant and which genes do these splicing factors regulate ? Through genetic and molecular studies using mouse embryos , Leggere et al . found that the NOVA family of splicing factors are essential for the development of the nervous system . In particular , the NOVA splicing factors control the alternative splicing of a gene called Dcc . This gene produces proteins that play a number of roles , including helping axons to grow and guiding the axons to the correct location in the developing nervous system . A related study by Saito et al . showed that two forms of NOVA splicing factors – called NOVA1 and NOVA2 – have different roles in the nervous system , and describes the role of NOVA2 in more detail . Leggere et al . will now carry out additional studies to determine the unique role of each protein variant produced from the Dcc gene . Future research will also investigate how NOVA proteins help generate these variants at the right time and in the right place . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2016 | NOVA regulates Dcc alternative splicing during neuronal migration and axon guidance in the spinal cord |
Functional magnetic resonance imaging using blood oxygenation level dependent ( BOLD ) contrast is well established as one of the most powerful methods for mapping human brain function . Numerous studies have measured how low-frequency BOLD signal fluctuations from the brain are correlated between voxels in a resting state , and have exploited these signals to infer functional connectivity within specific neural circuits . However , to date there have been no previous substantiated reports of resting state correlations in the spinal cord . In a cohort of healthy volunteers , we observed robust functional connectivity between left and right ventral ( motor ) horns , and between left and right dorsal ( sensory ) horns . Our results demonstrate that low-frequency BOLD fluctuations are inherent in the spinal cord as well as the brain , and by analogy to cortical circuits , we hypothesize that these correlations may offer insight into the execution and maintenance of sensory and motor functions both locally and within the cerebrum .
Since the early 1990s , thousands of functional magnetic resonance imaging ( fMRI ) studies have offered new insights into the functional architecture of the brain and have significantly increased our understanding of normal and aberrant brain functions . The earliest papers investigated task-based fMRI , where evoked BOLD signal changes were interpreted as hemodynamic responses subsequent to neural activity ( Ogawa et al . , 1990; Bandettini et al . , 1992; Kwong et al . , 1992; Ogawa et al . , 1992 ) and were used to infer which brain regions were activated for a specific task . The range and impact of fMRI methods were expanded in 1995 when Biswal et al . established the existence of correlations between low-frequency ( < 0 . 08 Hz ) BOLD signals from spatially distinct locations when no task was performed ( Biswal et al . , 1995 ) , and over 4000 subsequent papers have documented different aspects of resting state fMRI . Importantly , these correlations have been widely adopted to infer functional connectivity between cortical regions ( Greicius et al . , 2003; Fox et al . , 2005; Smith , 2012 ) . The identification of patterns of highly correlated low-frequency signals in the resting brain provides a powerful approach to delineate and describe neural circuits , and an unprecedented insight into the manner in which distributed regions work together to achieve specific functions ( Pizoli et al . , 2011 ) . In this study , we present the first robust demonstrations that similar phenomena can be detected within the gray matter of the human spinal cord , and we report our preliminary attempts to perform resting state connectivity studies within the cords of normal volunteers . Although the vast majority of fMRI studies have explored function in the cerebrum , there have been a few investigations of function in the human brainstem and spinal cord . fMRI in the spinal cord was first performed ( Yoshizawa et al . , 1996 ) in 1996 , and task-based ( motor and/or sensory ) spinal fMRI has since been demonstrated by a handful of groups worldwide ( Stroman et al . , 1999; Backes et al . , 2001; Madi et al . , 2001; Giove et al . , 2004; Moffitt et al . , 2005; Maieron et al . , 2007; Giulietti et al . , 2008; Agosta et al . , 2009a; Cohen-Adad et al . , 2010; Summers et al . , 2010; Brooks et al . , 2012; Sprenger et al . , 2012 ) . Spinal cord fMRI has primarily been used to study motor and sensory/pain pathways in the healthy spinal cord , but has also been shown to be sensitive to changes in patients with spinal cord injury ( Stroman et al . , 2002 , 2004; Kornelsen and Stroman , 2007 ) and multiple sclerosis ( Agosta et al . , 2008a , 2008b , 2009b; Valsasina et al . , 2010 , 2012 ) . Importantly , these spinal fMRI studies were focused on understanding spinal cord function when performing tasks , and to date only one paper has reported an investigation of resting state BOLD fluctuations in the human spinal cord , from which the results were equivocal ( Wei et al . , 2010 ) . Partly , the lack of positive reports may reflect the relatively poor signal-to-noise ratio of spinal cord images achievable at conventional field strengths ( 1 . 5 Tesla and 3 . 0 Tesla ) and the inherent limitations of low spatial resolution in studying small structures . The advent of ultra-high magnetic fields ( 7 Tesla and above ) and implementation of appropriate multichannel spinal cord coils , along with improved image acquisition and correction protocols , provides new opportunities for high-resolution fMRI of the spinal cord with increased sensitivity to BOLD fluctuations in the small gray matter structures that are typically not well visualized at lower fields . The spinal cord is essentially a long , cylindrical neural structure responsible for relaying motor and sensory information between the brain and body , it sits within a bath of cerebrospinal fluid ( CSF ) , and is surrounded by large vertebral bodies and intervertebral discs ( Figure 1A ) . A butterfly-shaped gray matter structure surrounded by densely packed white matter is found within the cord ( Figure 1B ) . The gray matter is anatomically described by ventral ( anterior ) , lateral , and dorsal ( posterior ) horns , though the lateral and ventral horns in thoracic and lumbar segments are often summarized as anterior ( ventrolateral ) gray matter . The dorsal horn contains neurons that receive sensory information from the extremities while upper motor neurons synapse onto lower motor neurons in the ventral horn , which relay information to the extremities ( Kandel et al . , 2000 ) . 10 . 7554/eLife . 02812 . 003Figure 1 . Resting state spinal cord fMRI at 7 Tesla . ( A ) Mid-sagittal slice from a healthy volunteer showing the complete cervical cord and typical axial slice placement for this resting state study . In all subjects the imaging stack was centered on the C3/C4 junction , providing full coverage of C3 and C4 and partial coverage of C2 and C5 . ( B ) T2*-weighted anatomical image at C4 acquired with 0 . 6 × 0 . 6 × 4 mm3 voxels and interpolated to 0 . 31 × 0 . 31 × 4 mm3 . Excellent contrast permits visualization of the characteristic butterfly-shaped gray matter column . ( C ) A single T2*-weighted functional image of this axial slice ( acquired with 0 . 91 × 0 . 91 × 4 mm3 voxels ) . Functional images are high quality with minimal geometric distortions and T2* blurring and permit adequate spatial delineation between white matter and cerebrospinal fluid . DOI: http://dx . doi . org/10 . 7554/eLife . 02812 . 003 The relatively small number of spinal fMRI studies to date and lack of well-developed investigative tools make it difficult to formulate clear hypotheses of what low-frequency ( < 0 . 08 Hz ) BOLD signal correlations may be expected between sub-regions of spinal cord gray matter in a resting state . However , the formulation of such hypotheses may , in part , be guided by known anatomical connections or spinal cord function . For example , the likely existence of central pattern generators in the human spinal cord that subserve basic locomotion ( Kandel et al . , 2000 , p . 753 ) suggest that functional connectivity may exist between ventral ( motor ) horns . Similarly , reflexes suggest connections between a dorsal ( sensory ) horn and both ipsilateral and contralateral ventral horns ( Kandel et al . , 2000 , Figure 36-2 ) , and are primarily apparent in the presence of noxious stimuli . The ascending sensory pathways and descending motor pathways also suggest that there may be connectivity along the length of the cord , at least within individual dermatomes for dorsal horns ( Kandel et al . , 2000 , p . 445 ) . However , it must be emphasized that the lack of a direct anatomical connection between two sub-regions of spinal gray matter does not preclude the possibility of finding connectivity between these regions because they may be indirectly connected via other pathways . In practice , even if significant low-frequency signal variations related to function are manifest , they may be obscured by cord motion and various other sources of physiological noise . Here we adapt the paradigm used for earlier investigations of connectivity in the brain whereby we define very small regions of interest in anatomically distinct parts of spinal gray matter which in general subserve defined functions , and examine interregional steady-state correlations between them . In addition , we derive the patterns of voxels that show significant temporal correlation with selected single voxels within regions . These approaches have been successfully used , for example , in the cortex to delineate motor circuits . The use of high resolution images at 7 Tesla permit the reliable separation of ventral , dorsal , and bilateral segments of the cord so we can examine functional connectivity between sub-regions guided by known anatomical features . Functional connectivity along the cord may also be examined by considering any subregion in one slice and the same or another subregion in adjacent ( or other ) slices , but here we limit the group analysis of our first report to functional connectivity assessed within axial slices .
Functional images were preprocessed to mitigate rigid-body motion and physiological noise , and spatially interpolated to match the digital resolution of the T2*-weighted anatomical images ( Figure 2 ) . A 14-step standardized analysis protocol ( described in 'Materials and methods' ) was used for each of the 22 subjects studied . In each subject , temporal signal-to-noise ratio ( TSNR ) was measured in spinal gray matter upon completion of the functional-to-anatomical affine registration ( step #9 ) as well as after the application of CSF and white matter ‘regressors of no interest’ ( steps #11 and #12 ) . Across all 22 subjects , we observed a 30% increase in median TSNR ( from 29 . 3 to 38 . 1 ) after the application of these few regressors , demonstrating the importance of characterizing and removing structured noise sources ( Xie et al . , 2012 ) . After band-pass filtering to isolate the frequency range of interest ( 0 . 01–0 . 08 Hz ) , single-subject analyses show that statistically significant correlations are measurable between selected regions and are reproducible across subjects . As an illustrative example with the corresponding time series , an analysis performed on one subject ( female , 23 years old ) demonstrates connectivity with the contralateral ventral horns in the same slice and with adjacent slices when a seed region is selected in the center of the right ventral horn ( Figure 3 ) . A stringent threshold of |z| > 3 . 29 ( a two-tailed 99 . 9% confidence interval ) was selected to show that connectivity is focused in the gray matter horns and not in central gray matter ( connecting left and right sides and largely dominated by the central canal ) nor adjacent white matter , which provides evidence that such gray matter correlations cannot be simply attributed to spatially correlated physiological noise and more likely represent genuine functional connectivity . Further examples of within-slice connectivity analyses in single subjects confirm that reproducible focal connectivity is found between ventral horns ( Figure 4A–F ) and between dorsal horns ( Figure 4G–J ) . There is also evidence of plausible connectivity with central gray matter ( Figure 4K ) and between ventral and dorsal horns ( Figure 4L ) , but these correlations are less consistent across all slices and not statistically significant at the group level . To quantify the occurrence of within-slice correlations between gray matter sub-regions across slices , we averaged time courses within each respective gray matter sub-region ( defined in step #14 ) and considered only positive correlations at a more conventional 95% confidence interval ( z > 1 . 65; one-tailed ) . Across all 264 slices ( 12 slices/subject × 22 subjects ) , we observed that 67% of slices ( 177 of 264 ) exhibit significant correlations between ventral horns and 37% of slices ( 97 of 264 ) exhibit significant correlations between dorsal horns . In comparison , a markedly fewer number of slices ( only 1 in 5 ) exhibited significant correlations between the remaining four pairs: 21% between left ventral and left dorsal horns ( 55 of 264 ) , 21% between left ventral and right dorsal horns ( 55 of 264 ) , 20% between right ventral and left dorsal horns ( 54 of 264 ) , and 23% between right ventral and right dorsal horns ( 62 of 264 ) . 10 . 7554/eLife . 02812 . 004Figure 2 . Functional weighted spinal cord images at 7 Tesla . A single volume of twelve contiguous T2*-weighted slices centered on the C3/C4 junction ( as illustrated in Figure 1A ) in one subject . Each volume was acquired with 0 . 91 × 0 . 91 × 4 mm3 voxels and resampled to 0 . 31 × 0 . 31 × 4 mm3 voxels during the affine functional-to-anatomical registration . Excellent contrast between white matter and cerebrospinal fluid facilitates accurate registration between such functional volumes and high-resolution anatomical images ( Figure 1B ) . The use of a 3D acquisition sequence with relatively short echo time and relatively few k-space lines per radiofrequency pulse provides high-quality images with minimal signal drop-out and geometric distortions , although artifacts caused by fat shift of the nerve root sleeve in the phase-encode direction still affect the dorsal edge in a few slices . DOI: http://dx . doi . org/10 . 7554/eLife . 02812 . 00410 . 7554/eLife . 02812 . 005Figure 3 . A single-subject analysis of resting state functional connectivity with corresponding time series . For clarity , only outlines of the gray matter butterfly and surrounding white matter are shown ( rostro-caudal from left to right ) . Red and yellow represent statistically significant positive correlation with the seed time series ( using a two-sided 99 . 9% confidence interval where red is 3 . 29 < z ≤ 3 . 89 and yellow is z > 3 . 89 ) , and blue represents negative correlation ( z < −3 . 29 ) . The seed voxel is selected in the right ventral horn in C5 , and exhibits functional connectivity with the contralateral ventral horn in the same slice as well as adjacent slices . Such connectivity between ventral horns is observed across all subjects . In each of the four plots , a 3 . 5-min segment of the seed time course is shown in black and the time course of the corresponding region of interest is shown in magenta . The highest correlations are observed in the contralateral ventral horn on the same slice ( z = 4 . 10 ) and on the adjacent slices ( z = 4 . 38 ) . Correlations with central gray matter ( z = 2 . 55 ) and adjacent white matter ( z = 0 . 84 ) are relatively low , which , given the small size of the spinal cord , suggest that such correlations are genuine and not dominated by widespread physiological noise . DOI: http://dx . doi . org/10 . 7554/eLife . 02812 . 00510 . 7554/eLife . 02812 . 006Figure 4 . Examples of within-slice resting state functional connectivity across subjects . These analyses were performed using AFNI's ‘InstaCorr’ with p < 0 . 001 and no minimum cluster size . In each panel , a seed voxel is marked with a green crosshair and resultant correlations are overlaid on the anatomical image . ( A ) – ( F ) Connectivity between ventral horns for subjects 1 , 3 , 8 , 10 , 11 , and 13 , respectively . ( G ) – ( J ) Connectivity between dorsal horns for subjects 5 , 16 , 18 , and 22 , respectively . ( K and L ) Less common correlations within gray matter . In ( K ) ( subject 20 ) , focal connectivity between ventral horns and with central gray matter . In ( L ) ( subject 7 ) , connectivity between ventral horns but also with the contralateral dorsal horn . At the single-subject level , there is some evidence for functional connectivity between ventral and dorsal horns , but such correlations are less common across slices and not statistically significant at the group level . DOI: http://dx . doi . org/10 . 7554/eLife . 02812 . 006 A group-level analysis of functional connectivity between sub-regions of spinal cord gray matter and adjacent white matter confirmed that the most robust correlations are observed between left and right ventral ( motor ) horns ( p < 0 . 01; corrected ) , as well as between left and right dorsal ( sensory ) horns ( p < 0 . 01; corrected ) ( Figure 5A ) . No significant group-level correlations ( p > 0 . 05 ) were observed between other gray matter sub-regions , nor between spinal cord gray and white matter . Weak positive correlations were observed between left and right dorsal column white matter ( p < 0 . 05; corrected ) , and negative correlations were observed between left ventral white matter and right dorsal white matter ( p < 0 . 01; corrected ) , and between right ventral white matter and both left ( p < 0 . 01; corrected ) and right ( p < 0 . 01; corrected ) dorsal white matter . The apparent existence of negative correlations in resting state spinal cord data is not unexpected because anticorrelations are commonly observed in resting state analyses of the brain ( Chang and Glover , 2010 ) and have been a topic of intense discussion for over a decade . The ranges of values within these six statistically significant distributions are presented as box-and-whisker plots ( Figure 5B ) . The lower quartile is above zero in both gray matter plots , demonstrating that positive gray matter connectivity is a robust and reproducible measurement . In comparison , temporal correlations between white matter sub-regions are more variable and exhibit both positive and negative median correlations . The raw data used to generate this figure is provided as Figure 5—source data 1 . Additional analyses were performed ( Figure 5—figure supplement 1 , Figure 5—figure supplement 3 , and Figure 5—figure supplement 5 ) to confirm that positive gray matter correlations are stable across various preprocessing procedures whereas white matter correlations are positive before but negative or non-significant after white matter regression ( step #12 ) . These supplementary analyses also showed that weaker positive correlations between sub-regions of left and right dorsal white matter remained positive and significant across various preprocessing configurations . 10 . 7554/eLife . 02812 . 007Figure 5 . Group-level functional connectivity between sub-regions of spinal gray matter ( GM ) and surrounding white matter ( WM ) within slices . ( A ) In GM , strong positive correlations are observed between left ( LV ) and right ( RV ) ventral horns , as well as left ( LD ) and right ( RD ) dorsal horns ( *p<0 . 05; **p<0 . 01; Bonferroni corrected ) . Weaker positive and negative correlations are observed within WM . No statistically significant correlations are observed between spinal GM and WM ( upper right quadrant ) . ( B ) Box-and-whisker plots showing the median and upper and lower quartiles of the six statistically significant results identified in ( A ) . Whiskers extend out to 1 . 5 times the distance between the upper and lower quartiles , and an outlier ( beyond the whiskers ) is denoted by a plus . Wilcoxon signed rank tests identify the distributions of z-scores ( across all slices and subjects ) that are significantly different from zero ( p < 0 . 05 ) . Functional connectivity between WM sub-regions is more variable and exhibits both positive and negative median correlations . In comparison , median GM correlations between LV-RV and LD-RD are positive across all 22 subjects . Additional analysis permutations ( described in Figure 5—figure supplements 1 , 3 , and 5 with example GM power spectra shown in Figure 5—figure supplements 2 , 4 , and 6 ) reveal that the three negative WM correlations are influenced by WM regression ( step #12 ) and thus are open to more than one interpretation . However , supplementary analyses reveal that the positive GM correlations between ventral horns and between dorsal horns persist across all preprocessing permutations . These additional analyses further support the conclusion that positive correlations between GM horns are not artifactual—possibly created by preprocessing choices or frequency bandwidth selection—and most likely represent genuine functional connectivity . DOI: http://dx . doi . org/10 . 7554/eLife . 02812 . 00710 . 7554/eLife . 02812 . 008Figure 5—source data 1 . Matlab file containing the raw data used to generate Figure 5 . The dimensions of this matrix are 8 × 8 × 22 where the 8 × 8 represents all possible comparisons between the eight sub-regions ( Figure 5A ) and the upper-triangular entries are non-zero . For each of the 28 unique comparisons , the 22 entries in the third dimension correspond to the median measurement of functional connectivity across all 12 slices for each of the 22 subjects . DOI: http://dx . doi . org/10 . 7554/eLife . 02812 . 00810 . 7554/eLife . 02812 . 009Figure 5—figure supplement 1 . Functional connectivity matrices resulting from preprocessing pipeline permutations . As before , functional data were band-pass filtered between 0 . 01 and 0 . 08 Hz ( *p<0 . 05; **p<0 . 01; Bonferroni corrected ) . For clarity the labels are not shown for each column/row but are the same as in Figure 5 . ( A ) Preprocessing was performed as described in the Methods except CSF and WM regressors ( steps #11 and #12 ) were not applied . Each GM sub-region is highly correlated with all other GM sub-regions and similarly each WM sub-region is highly correlated with all other WM sub-regions . Interestingly , there are no significant group-level correlations between GM and WM sub-regions , suggesting that ‘global’ GM fluctuations tend to be constrained to GM and ‘global’ WM fluctuations tend to be constrained to WM . ( B ) Preprocessing was performed as described in ‘Materials and methods’ except a WM regressor ( step #12 ) was not applied . The application of only CSF regressors reduced correlations within both GM and WM , but all inter-region correlations remained and were statistically significant . ( C ) Preprocessing was performed exactly as described in ‘Materials and methods’ ( i . e . , this panel is the same as Figure 5A except for a larger dynamic range for z-scores ) . Regression of the principal eigenvector of all time series in the WM mask significantly altered correlations within both GM and WM . This unintuitive result may be explained if WM masks contain signal contributions from adjacent GM voxels , which is certainly possible given the small size of the GM butterfly and unavoidable partial volume effects , and sub-millimeter functional-to-anatomical warping inaccuracies caused by magnetic field inhomogeneities . ( D ) To investigate the possibility that WM masks contain fractions of GM voxels , and the impact of different regression masks on correlations between sub-regions , preprocessing was performed as described in ‘Materials and methods’ except step #12 extracted the principal eigenvector of all time series within a combined WM and GM mask . Given the small size of each GM mask relative to its neighboring WM mask , the principal eigenvector from the combined mask ( WM&GM ) should be similar to only using WM . Indeed , regression with these modified eigenvectors produces similar group-level correlations between WM sub-regions and introduce weak negative correlations between GM sub-regions . In GM , positive correlations between ventral horns and between dorsal horns still persist after this more invasive regressor , providing further evidence that these strong temporally correlated fluctuations are unlikely to be caused by physiological motion or spatially-correlated noise . To further investigate the spatial extent of global fluctuations within GM , we repeated the preprocessing as described in ‘Materials and methods’ but eroded the WM mask slightly before extracting the principal eigenvector . The results of this tertiary analysis ( not shown ) were slightly different but statistically similar to Figure 5A , demonstrating that voxels on and near the GM-WM boundary in fact characterize physiological fluctuations from both tissue types due to unavoidable partial volume effects and sub-millimeter warping inaccuracies . Overall , these supplementary analyses suggest that the preprocessing performed in the main manuscript ( regression of the principal eigenvector from WM masks without erosion ) is an appropriate strategy for suppressing extraneous fluctuations within both WM and GM without introducing substantial negative correlations between GM sub-regions . Finally , power spectra for WM and GM sub-regions in ( C ) are presented in Figure 5—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02812 . 00910 . 7554/eLife . 02812 . 010Figure 5—figure supplement 2 . Power spectra across gray and white matter sub-regions for data filtered between 0 . 01 and 0 . 08 Hz . Power spectra for ( A ) ventral and ( B ) dorsal GM sub-regions in Figure 5—figure supplement 1C that exhibit significant positive correlations ( z > 1 . 65; one-tailed ) , and all WM sub-regions . For each frequency the plotted power represents median power across slices and subjects . Within a range from 0 . 015 Hz to 0 . 075 Hz , ventral GM exhibits 32% more power than WM whereas dorsal GM exhibits only 5% more power . This difference in power between GM sub-regions may be explained through the preprocessing pipeline permutations presented in Figure 5—figure supplement 1 . Figure 5—figure supplement 1A , B show comparable z-scores between ventral and dorsal GM horns , and this difference does not appear until after WM regression ( Figure 5—figure supplement 1C ) or WM&GM regression ( Figure 5—figure supplement 1D ) . Therefore , regression with this final eigenvector suppresses a larger contribution of signal fluctuations ( and thus power ) from dorsal GM than ventral GM , which may be attributed to the small size of the dorsal horns and partial volume averaging effects ( because the WM mask is not eroded before PCA ) , as well as to unavoidable sub-millimeter registration inaccuracies . The large peak at ∼0 . 75 Hz is likely due to physiological noise , so it could be argued that this noise power significantly affects the measurements of functional connectivity . To investigate this further and discount the possibility that our results are driven by physiological noise around 0 . 75 Hz , the analyses in Figure 5—figure supplement 1 are repeated after data are filtered with a narrower band-pass filter between 0 . 01 Hz and 0 . 07 Hz . These results are presented in Figure 5—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 02812 . 01010 . 7554/eLife . 02812 . 011Figure 5—figure supplement 3 . Functional connectivity matrices resulting from preprocessing pipeline permutations after band-pass filtering between 0 . 01 and 0 . 07 Hz . Functional connectivity matrices resulting from preprocessing pipeline permutations after band-pass filtering between 0 . 01 and 0 . 07 Hz to suppress power from likely physiological noise at ∼0 . 75 Hz ( *p<0 . 05; **p<0 . 01; Bonferroni corrected ) . For clarity the labels are not shown for each column/row but are the same as in Figure 5 . ( A ) Preprocessing was performed as described in the Methods except CSF and WM regressors ( steps #11 and #12 ) were not applied , and step #13 used a different frequency bandwidth . ( B ) Preprocessing was performed as described in ‘Materials and methods’ except a WM regressor ( step #12 ) was not applied and step #13 used a different frequency bandwidth . ( C ) Preprocessing was performed as described in ‘Materials and methods’ except step #13 used a different frequency bandwidth . ( D ) Preprocessing was performed as described in ‘Materials and methods’ except step #12 extracted the principal eigenvector of all time series within a combined WM and GM mask and step #13 used a different frequency bandwidth . Only a couple of minor changes are apparent between these matrices and those presented in Figure 5—figure supplement 1 , suggesting that a slight decrease in the upper filter range from 0 . 08 Hz to 0 . 07 Hz to further suppress physiological noise does not appear to have a significant impact on the group-level connectivity analyses . Power spectra for WM and GM sub-regions in ( C ) are presented in Figure 5—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 02812 . 01110 . 7554/eLife . 02812 . 012Figure 5—figure supplement 4 . Power spectra across gray and white matter sub-regions for data filtered between 0 . 01 and 0 . 07 Hz . Power spectra for ( A ) ventral and ( B ) dorsal GM sub-regions in Figure 5—figure supplement 3C that exhibit significant positive correlations in the original analysis ( Figure 5A; z > 1 . 65; one-tailed ) , and all WM sub-regions . For each frequency the plotted power represents median power across slices and subjects . Within a range from 0 . 015 Hz to 0 . 065 Hz , ventral GM similarly exhibits 31% more power than WM whereas dorsal GM exhibits 5% more power . The narrower bandwidth completely suppresses the noise peak at ∼0 . 75 Hz but the results from functional connectivity analyses are almost unchanged ( Figure 5—figure supplement 1 vs Figure 5—figure supplement 3 ) , confirming that this peak does not significantly affect the overall results . Our original decision to filter resting state spinal cord data between 0 . 01 Hz and 0 . 08 Hz ( step #13 ) was motivated by the approach most commonly used for resting state analyses in the brain ( filtering between 0 . 01 Hz and 0 . 08–0 . 1 Hz ) , although it is not yet clear if this frequency range is optimal for resting state spinal cord analyses . To investigate the possibility that frequencies above 0 . 08 Hz may contribute to inherent functional connectivity , the analyses in Figure 5—figure supplement 1 are once again repeated after data are filtered with a wider band-pass filter between 0 . 01 Hz and 0 . 13 Hz . These results are presented in Figure 5—figure supplement 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 02812 . 01210 . 7554/eLife . 02812 . 013Figure 5—figure supplement 5 . Functional connectivity matrices resulting from preprocessing pipeline permutations after band-pass filtering between 0 . 01 and 0 . 13 Hz . Functional connectivity matrices resulting from preprocessing pipeline permutations after band-pass filtering between 0 . 01 and 0 . 13 Hz ( *p<0 . 05; **p<0 . 01; Bonferroni corrected ) . For clarity the labels are not shown for each column/row but are the same as in Figure 5 . ( A ) Preprocessing was performed as described in the Methods except CSF and WM regressors ( steps #11 and #12 ) were not applied , and step #13 used a different frequency bandwidth . ( B ) Preprocessing was performed as described in ‘Materials and methods’ except a WM regressor ( step #12 ) was not applied and step #13 used a different frequency bandwidth . ( C ) Preprocessing was performed as described in ‘Materials and methods’ except step #13 used a different frequency bandwidth . ( D ) Preprocessing was performed as described in ‘Materials and methods’ except step #12 extracted the principal eigenvector of all time series within a combined WM and GM mask and step #13 used a different frequency bandwidth . The inclusion of frequencies between 0 . 08 Hz and 0 . 13 Hz primarily strengthens GM correlations between ventral horns ( relative to the results obtained using only frequencies between 0 . 01 Hz and 0 . 08 Hz ) but also increases the statistical significance of WM correlations between LD and RD . Power spectra for WM and GM sub-regions in ( C ) are presented in Figure 5—figure supplement 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 02812 . 01310 . 7554/eLife . 02812 . 014Figure 5—figure supplement 6 . Power spectra across gray and white matter sub-regions for data filtered between 0 . 01 and 0 . 13 Hz . Power spectra for ( A ) ventral and ( B ) dorsal GM sub-regions in Figure 5—figure supplement 5C that exhibit significant positive correlations in the original analysis ( Figure 5A; z > 1 . 65; one-tailed ) , and all WM sub-regions . For each frequency the plotted power represents median power across slices and subjects . Within the range of higher frequencies between 0 . 075 Hz and 0 . 125 Hz , ventral GM exhibits 28% more power than WM whereas dorsal GM and WM exhibit comparable total power ( < 0 . 3% difference ) . These data show that the original filter bandwidth between 0 . 01 Hz and 0 . 08 Hz was a sound choice , but also suggest that frequencies slightly above 0 . 08 Hz may contain additional power related to GM connectivity . Future work will acquire resting state spinal cord data with a faster sampling rate to better understand the relative frequency-dependent contributions from BOLD signal fluctuations and physiological noise . DOI: http://dx . doi . org/10 . 7554/eLife . 02812 . 014
We have presented the first conclusive demonstration that ultra-high field fMRI can non-invasively detect and characterize resting state BOLD signals within the gray matter of the human spinal cord . Low-frequency temporal correlations were observed between gray matter horns in all subjects , and examples of these correlations are presented in Figure 3 and Figure 4 . Within a given axial slice , the strongest and most robust correlations were between left and right ventral horns . Correlations were also observed between left and right dorsal horns , and the reproducibility of correlations between ventral horns and between dorsal horns was demonstrated within a cohort of 22 healthy volunteers ( Figure 5 ) . Although correlations between ventral and dorsal gray matter were also observed at the single-subject level ( Figure 4L ) , such findings were less frequent and not statistically significant at the group level . The absence of group-level correlations between spinal gray matter and adjacent white matter showed that these positive gray matter correlations are unlikely to be driven by spatially-correlated physiological noise . In fact , supplementary analyses without CSF or white matter regressors ( Figure 5—figure supplement 1a ) revealed strong correlations within gray matter and within white matter but not between gray and white matter , suggesting that gray and white matter in the spinal cord may exhibit different degrees of physiological fluctuations . The main analyses ( Figure 5 ) also showed anticorrelations between white matter sub-regions . These observations were not unexpected because negative correlations are commonly seen in the brain ( Chang and Glover , 2010 ) , but additional analyses ( Figure 5—figure supplement 1 ) revealed that correlations within white matter were heavily influenced by preprocessing methodology . As a result , the nature of negative white matter correlations remains unclear and requires further investigation . However , predominantly positive correlations between dorsal white matter sub-regions persisted across preprocessing permutations . Although the origins of these positive white matter correlations remain to be determined , and whereas in general BOLD signals from activation have been difficult to detect in brain white matter , we note that Ding et al . recently reported the reliable detection of anisotropic correlations of resting state BOLD signals in brain white matter that appear to mimic white matter tracts identified by diffusion imaging methods ( Ding et al . , 2013 ) . Moreover , the white matter sub-regions within the cord also tend to be close to draining veins . Another observation is that z-scores measured between ventral horns tended to be higher than z-scores measured between dorsal horns—a finding that was highly significant ( p < 0 . 01 using a two-tailed Wilcoxon signed rank test; ‘signrank’ in Matlab ) . There are several possible reasons for this finding . Firstly , the dorsal horns tend to be slightly narrower than the ventral horns , and thus may be more susceptible to registration inaccuracies , residual physiological noise , and partial volume averaging with adjacent white matter . Secondly , as shown in Figure 2 , signal dropout and unavoidable artifacts affect the dorsal horns but not ventral horns in a few slices , which would bias the results in favor of ventral horn connectivity . Finally , even if spatial artifacts , registration inaccuracies , signal dropout , physiological noise , and partial volume averaging effects were minimized , it may be that functional connectivity between dorsal horns is more variable if the automated selection of gray matter sub-regions ( step #14 ) isolate signals from different laminae within dorsal horns ( Ruscheweyh and Sandkühler , 2002 ) . To date , clinical applications of task-based spinal fMRI studies have primarily been targeted to subjects with multiple sclerosis ( Agosta et al . , 2008a , 2008b , 2009b; Valsasina et al . , 2010 , 2012 ) and spinal cord injury ( SCI ) ( Stroman et al . , 2002 , 2004; Kornelsen and Stroman , 2007 ) . We propose that the non-invasive methods of resting state spinal cord functional connectivity developed in this paper may be most readily translatable to clinical investigations characterizing damage due to acute or chronic SCI and monitoring the efficacy of surgical or pharmacological interventions . Functional connectivity and the assessment of plasticity in the human spinal cord and animal models of SCI have been topics of intense research for many years ( Bregman et al . , 1997; Raineteau and Schwab , 2001; Cai et al . , 2006; Freund et al . , 2011 ) because SCI affects 260 , 000 people in the United States ( a prevalence of ∼1 in 1200 ) with 11 , 000 new injuries reported each year ( National Spinal Cord Injury Statistics Center , 2010 ) . Studies have investigated the role of propriospinal neurons in partial recovery from incomplete SCI ( Bareyre , 2008; Flynn et al . , 2011a ) while in vitro analyses of functional connectivity in mouse spinal cord have relied on electrophysiological methods ( Flynn et al . , 2011b ) . A new intervention strategy using epidural stimulation and stand training was recently developed that has to date restored voluntary movement in four patients with complete paralysis , demonstrating that functional connectivity across a lesion may be restored with epidural stimulation ( Angeli et al . , 2014 ) . Such studies would likely benefit from an ability to assess the functional architecture of the spinal cord throughout therapy . The majority of spinal injuries are , however , incomplete , and lost function may eventually return to near-normal levels; however , the progression of functional recovery after incomplete SCI remains poorly understood due to the absence of a non-invasive method to reliably assess spinal cord connectivity in vivo . We propose that resting state acquisitions of the cervical spinal cord will become a valuable tool for characterizing changes in functional connectivity in SCI , and for the prognosis and monitoring of progression of recovery via spontaneous repair and/or surgical intervention . For example , the imaging volume ( shown in Figure 1A ) may be centered on a focal injury to the cord , and functional connectivity above and below the injury may be assessed . This process with identical slice placement could then be repeated serially over time to investigate phenomena of neural plasticity and adaptation of spinal pathways . Moreover , even in normal subjects , the functional organization of the spinal cord is relatively under-explored and remains poorly understood . Observations of altered resting state connectivity in the brain in numerous disorders ( Fox and Greicius , 2010 ) and as a function of behavior or cognitive skills suggest that such correlations reflect an important level of organization and may play a fundamental role in the execution and maintenance of various functions ( Pizoli et al . , 2011 ) . Thus , investigation of resting state spinal cord networks could similarly have widespread applicability in studying central nervous system diseases that affect motor and/or sensory pathways such as cervical spondylotic myelopathy , neuromyelitis optica , acute disseminated encephalomyelitis , arachnoiditis , transverse myelitis , amyotrophic lateral sclerosis , and multiple sclerosis . Our study has three main limitations . Firstly , although we observed statistically significant correlations along the cord in single-subject analyses ( for example , in left ventral gray matter in Figure 3 ) , we chose to constrain the group analysis to investigating connectivity within each axial slice . This was done because incorporating correlations along the cord would have made the analysis significantly more complicated by increasing the number of potential correlations by an order of magnitude . Furthermore , as shown in Figure 2 , some slices exhibited regions of signal dropout at the dorsal edge caused by fat shift of the nerve root sleeve in the phase-encode ( anterior-posterior ) direction . The absence of a statistically significant correlation with a slice impacted by main field ( B0 ) inhomogeneities or an artifact cannot in and of itself be interpreted as proof that connectivity does not exist between particular sub-regions of interest , and ultimately very careful single-subject analyses will need to be performed to reliably characterize functional connectivity along the cord . Secondly , although we have investigated one aspect of reproducibility ( via a group analysis of 22 healthy volunteers ) and established the existence of spinal cord functional connectivity , further investigations of within-subject reproducibility still need to be performed . This could be done by acquiring multiple resting state runs from a single scanning session and/or by scanning the same volunteer on multiple days while ensuring that the imaging volume is consistently placed in the same location ( e . g . , centered on the C3/C4 junction , as shown in Figure 1A ) . The reproducibility of these connectivity measures over months and years will need to be quantified before such techniques can be reliably used to study disease progression or gerontology . Thirdly , this study reports empirical findings of functional connectivity in the spinal cord but does not directly address the physiological origins of these low-frequency BOLD signal fluctuations . However , as a point of reference , observations of functional connectivity were first reported in the brain nearly two decades ago ( Biswal et al . , 1995 ) and yet the origin of these fluctuations is still a topic of intense discussion . Although the nature of resting state correlations is not fully understood , recent studies have combined tractography and functional imaging with techniques such as network analysis and graph theory to explore how these networks may have emerged ( Deco et al . , 2013a , 2013b; van den Heuvel and Sporns , 2013; Mišić et al . , 2014; Goñi et al . , 2014 ) . The origin of resting state networks has also been explored within the context of evolution and the expansion of the cerebral cortex ( Buckner and Krienen , 2013 ) . We therefore propose that because the spinal cord is an integral part of the central nervous system , one simple explanation is that low-frequency BOLD fluctuations in the brain and spinal cord share the same origin . This theory would then suggest that there may be long-range connections between networks in the spinal cord and cerebrum ( and cerebellum ) , and future research should consider the nature of resting state networks not only within the brain but within the entire central nervous system . Previous attempts to detect functional connectivity in the spinal cord have undoubtedly been confronted by significant technical challenges . At conventional field strengths ( 1 . 5 Tesla and 3 . 0 Tesla ) , the lower signal-to-noise ratio and BOLD contrast limit the spatial resolution and sensitivity for detecting BOLD signals , so large voxels are typically used and multiple acquisitions are averaged . At 7 Tesla , high resolution BOLD acquisitions of the spinal cord have not evolved as rapidly as for the brain in part due to the widespread dependence on single-shot echo-planar acquisitions at lower fields and the lack of specialized coils to image the spinal cord . We addressed these challenges and limitations by using a 7 Tesla scanner , novel fMRI image acquisition and data correction protocols , and a dedicated 16-channel radiofrequency coil array designed for cord imaging . Our results are the first demonstrations of functional imaging of the human spinal cord at 7 Tesla ( Barry et al . , 2013a ) and high-resolution resting state functional connectivity in the spinal cord ( Barry et al . , 2013b ) , and are likely to be of significant relevance in understanding basic aspects of spinal cord function both in normal development and in clinical disorders of the central nervous system .
Experiments were performed on a Philips Achieva 7 Tesla scanner with a custom-designed ( Nova Medical Inc . ) quadrature transmit and 16-channel receive coil array for cervical spinal cord imaging . 22 healthy volunteers ( 11 male , 21–63 years; 11 female , 23–34 years; 28 . 4 ± 8 . 8 years ) with no history of spinal cord injury or neurological impairment were recruited and scanned under protocols approved by the Institutional Review Board at Vanderbilt University Medical Center . Female participants of childbearing potential required a negative urine pregnancy test for the scan to proceed . Non-MR study data were collected and managed using REDCap electronic data capture tools hosted at Vanderbilt University ( Harris et al . , 2009 ) . REDCap ( Research Electronic Data Capture ) is a secure , web-based application designed to support data capture for research studies . Anatomical axial images with high spatial resolution and T2*-weighting ( Figure 1B ) were acquired with the following MR parameters: field of view = 160 × 160 mm2 , 12 4-mm slices ( centered on the C3/C4 junction , as shown in Figure 1A ) , nominal voxel size = 0 . 6 × 0 . 6 × 4 mm3 , interpolated voxel size = 0 . 31 × 0 . 31 × 4 mm3 , repetition time = 303 ms , echo time = 8 . 2 ms , flip angle = 25° , sensitivity encoding ( SENSE ) ( Pruessmann et al . , 1999 ) reduction factor = 2 . 0 ( anterior-posterior ) , signal acquisitions = 8 , total acquisition time = 5 min 22 s . Functional images with identical slice placement were acquired with a 3D multi-shot gradient-echo sequence ( van der Meulen et al . , 1988 ) previously shown to minimize T2* blurring and geometric distortions in cortical fMRI at 7 Tesla ( Barry et al . , 2011 ) . The functional MR parameters for the first 11 subjects were: field of view = 160 × 160 mm , twelve 4-mm slices , voxel size = 0 . 91 × 0 . 91 × 4 mm3 , repetition time = 18 ms , echo time = 7 . 8 ms , flip angle = 15° , echo train length = 9 , SENSE reduction factor = 1 . 56 ( anterior-posterior ) , volume acquisition time = 3 . 6 s ( 300 ms/slice ) , number of volumes = 150 ( after 10 ‘dummy’ scans ) , total scan time = 9 min 38 s , max gradient strength = 30 mT/m , max slew rate = 175 mT/m/ms . The functional MR parameters for the last 11 subjects were the same except for the following minor adjustments: repetition time = 17 ms , echo time = 8 . 0 ms , volume acquisition time = 3 . 34 s ( 278 ms/slice ) , total scan time = 9 min . For all subjects , respiratory and cardiac cycles were externally monitored and recorded using a respiratory bellow ( placed on the abdomen ) and pulse oximeter ( placed on the left index finger ) . Functional data were corrected for physiological noise using methods that are commonly used in fMRI of the brain in addition to novel data-driven ‘regressors of no interest’ . The steps applied to all spinal fMRI data are as follows:For each slice of anatomical and functional images , a 2D Gaussian weighting kernel was manually defined with the full-width-at-half-maximum set at the CSF boundaries . Weighting masks defined on anatomical images were used in affine registration ( step #8 ) , and weighting masks defined on functional images were used in affine registration as well as rigid-body motion correction ( step #5 ) . For each slice , a ‘not-spine’ mask was defined by drawing a region around the entire spinal cord and then logically inverting it ( used in step #3 ) . For each slice , data-driven ‘regressors of no interest’ were selected via principal component analysis ( PCA ) of all voxels within the not-spine mask to identify structured noise sources that would similarly affect the spinal cord and external ( neck ) regions . The number of eigenvectors selected reflected up to 80% of the slice-wise cumulative variance or until the difference between two successive eigenvalues was less than 2% ( typically 3–5 per slice ) . These vectors were regressed from the time series of all voxels within a slice , and significantly improved the efficacy of motion correction ( step #5 ) by mitigating widespread intensity fluctuations due to physiological processes ( e . g . , swallowing ) . For each slice , a representative ( target ) volume was automatically selected for motion correction by calculating the median intensity of each voxel ( over time ) and then selecting the volume closest to the median image ( identified via minimal least squares error ) . Rigid-body motion correction was performed on a slice-wise basis ( using 3dWarpDrive in AFNI [Cox , 1996] ) using the target volumes ( identified in step #4 ) . Motion was constrained to be within-plane translation ( i . e . , no rotation of the spinal cord ) . To mitigate the detrimental effects of sporadic artifacts ( e . g . , swallowing ) on motion parameter estimation , translation estimates were filtered with a 5-point median filter and then re-applied ( using 3dAllineate in AFNI [Cox , 1996] ) to the original data before motion correction . The initial registration ( to obtain motion parameter estimates ) used quintic interpolation and the final transformation used sinc interpolation . An established image correction technique called RETROICOR [Glover et al . , 2000] ( implemented in AFNI [Cox , 1996] ) was applied to the entire functional volume to further reduce quasi-periodic intensity variations due to physiological noise . Using the high-resolution anatomical images as a reference , masks defining the boundaries of gray matter , white matter , and CSF were created for each slice . Affine registration of the target volume ( identified in step #4 ) to the anatomical image was performed ( via 3dAllineate ) on a slice-wise basis . The Hellinger metric was selected as the cost function , and the degrees of freedom were constrained to within-plane translation , scaling ( maximum of 1% in the read direction and 5% in the phase-encode direction ) , and shearing ( maximum of 5% ) . The affine transforms ( defined in step #8 ) were applied to all functional volumes ( via 3dAllineate ) , and transformed functional images were resampled ( with sinc interpolation ) to match the final resolution of the anatomical volume ( voxel size = 0 . 31 × 0 . 31 × 4 mm3 ) . The quality of the final functional-to-anatomical alignments was visually verified using MRIcron ( www . mccauslandcenter . sc . edu/mricro/mricron ) . For each slice , additional data-driven ‘regressors of no interest’ were selected via PCA of all functional voxels within the CSF mask ( defined in step #7 ) to identify structured noise sources that would similarly affect gray matter and CSF . The number of eigenvectors selected reflected up to 50% of the slice-wise cumulative variance or until the difference between two successive eigenvalues was less than 2% ( typically 2–6 per slice ) . These vectors were regressed from the time series of all spinal cord voxels within a slice . For each slice , a ‘global’ white matter signal was calculated via PCA of all functional voxels within the white matter mask and extraction of the first eigenvector ( typically representing 10–30% of the variance ) . This was primarily done to mitigate any residual variance due to shifting of the white matter boundary ( caused by motion ) but would also reduce variance caused by residual physiological noise . This vector was regressed from the time series of all gray and white matter voxels within a slice . In preparation for analyses of functional connectivity , resultant functional data were band-pass filtered between 0 . 01 and 0 . 08 Hz using a Chebyshev Type II filter ( ‘cheby2’ and ‘filtfilt’ in Matlab ) to emphasize low-frequency signals of interest . In preparation for group analyses , gray and white matter masks ( defined in step #7 ) were subdivided into quadrants to identify left and right ventral and dorsal horns ( excluding central gray matter connecting left and right sides ) , as well as the four adjacent white matter regions . Each of these eight sub-region masks ( per slice ) was morphologically eroded ( using ‘imerode’ in Matlab ) to remove the outermost voxels and mitigate partial volume effects . For gray matter the morphological eroding object was a disk with a radius of 3 voxels , and for white matter it was a disk with a radius of 11 voxels . If the eroded sub-region did not contain any voxels ( i . e . , the disk was too large ) then the disk size was incrementally decreased by 1 voxel , and the erosion process repeated until the innermost area of each sub-region was extracted . Time series extracted from each eroded sub-region were also used in single-subject analyses ( except for in Figure 3 and Figure 4 , which used single-voxel correlations ) and power spectra calculations ( figure supplements for Figure 5 ) . Figure supplements for Figure 5 investigate modifications to steps #11 , #12 , and/or #13 , and deviations from this standardized preprocessing pipeline are described in the respective figure legends . For the voxel-based analysis shown in Figure 3 , the linear correlation coefficient ( r ) was calculated between a seed voxel and all other voxels in Matlab . These correlation values were converted to z-scores using the Fisher r-to-z transformation z = tanh−1 ( r ) ( dof–3 ) 1/2 where dof is the estimated degrees of freedom for each voxel after correction for first-order autocorrelation ( Rogers and Gore , 2008 ) . A statistical threshold of |z| > 3 . 29 ( a two-tailed 99 . 9% confidence interval ) was selected to clearly show that gray matter correlations tend to be focused in the center of the horns . A minimum cluster threshold of nine contiguous interpolated voxels ( 3 . 49 mm3 , approximately equal to the native functional voxel volume of 3 . 31 mm3 ) was also used to further protect against spurious correlations . For the single-subject analyses presented in Figure 4 , AFNI's ‘InstaCorr’ ( Cox , 1996 ) was used to display correlations between a single gray matter voxel and all other voxels within the spinal cord . The correlation threshold was p < 0 . 001 and in these analyses no cluster thresholding was used to reveal all correlations within gray and white matter . Finally , for the group-level region-of-interest ( ROI ) -based analyses ( Figure 5 ) , the time series of individual voxels were averaged within each eroded mask ( created in step #14 ) for each subject . The linear correlation coefficient was then calculated between the averaged time course from a sub-region and all other sub-regions . These correlation coefficients were converted to z-scores and corrected for first-order autocorrelation ( Rogers and Gore , 2008 ) . For each ROI comparison ( e . g . , left dorsal horn vs right dorsal horn ) , the median z-score ( unthresholded ) was calculated across all 12 slices for each of the 22 subjects , and then a two-tailed Wilcoxon signed rank test ( ‘signrank’ in Matlab ) identified group-level distributions of 22 z-scores that were significantly different from zero ( p < 0 . 05 or p < 0 . 01 with a Bonferroni correction factor of 28 ) . | Brain imaging methods such as functional magnetic resonance imaging ( fMRI ) can provide us with a picture of what the brain is doing when a person is carrying out a specific task . For example , an fMRI scan recorded whilst someone is reading is likely to show activity in regions in the left hemisphere of the brain that are known to be involved in language comprehension . fMRI can also be used to measure patterns of neuronal activity when someone is awake but not engaged in a specific task . This approach , known as resting state fMRI , can be used to examine which regions of the resting brain are active at the same time . Researchers are interested in these patterns of brain activity because they reflect neural circuits that work together to produce different functions and behaviors . Over 4000 papers have used resting state fMRI to study the human brain . However , to date there has been no conclusive investigation of resting state activity in the spinal cord . This is largely because the spinal cord is much smaller than the brain , and most fMRI scanners are not sensitive enough to study it in detail . Consequently , little is known about intrinsic neural circuits in the resting spinal cord . Now Barry et al . have used advances in fMRI technology to show that resting state functional connectivity does indeed exist in the spinal cord . Correlations were found in the resting levels of activity between spatially distinct areas of the cord , specifically between the ventral horns and between the dorsal horns . The ventral horns relay motor signals to the body , whilst the dorsal horns receive sensory signals from the body . These findings also have clinical applications . Some patients with incomplete spinal cord injuries can recover near normal function , but the mechanisms responsible for this recovery are unclear because clinicians have not been able to probe neuronal connections in the spinal cord in a non-invasive manner . The work of Barry et al . should help with efforts to understand the neuronal changes that support recovery from spinal cord injury . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2014 | Resting state functional connectivity in the human spinal cord |
Somitogenesis is regulated by a molecular oscillator that drives dynamic gene expression within the pre-somitic mesoderm . Previous mathematical models of the somitogenesis clock that invoke the mechanism of delayed negative feedback predict that its oscillation period depends on the sum of delays inherent to negative-feedback loops and inhibitor half-lives . We develop a mathematical model that explores the possibility that positive feedback also plays a role in determining the period of clock oscillations . The model predicts that increasing the half-life of the positive regulator , Notch intracellular domain ( NICD ) , can lead to elevated NICD levels and an increase in the oscillation period . To test this hypothesis , we investigate a phenotype induced by various small molecule inhibitors in which the clock is slowed . We observe elevated levels and a prolonged half-life of NICD . Reducing NICD production rescues these effects . These data provide the first indication that tight control of the turnover of positive as well as negative regulators of the clock determines its periodicity .
The formation of the vertebrate skeleton and associated musculature relies on the ordered and timely segmentation of mesodermal tissue during early embryonic development . Segmentation is progressive and occurs in an anterior to posterior direction as the body axis elongates ( reviewed in Oates et al . , 2012 ) . Mesodermal segments , or somites , ‘pinch off’ at the rostral limit of the two rods of pre-somitic mesoderm ( PSM ) that lie on either side of the caudal neural tube , with a distinct , species-specific periodicity; every 90 min in the chicken and every 120 min in the mouse . This periodicity is regulated by a molecular oscillator , known as the segmentation clock , that is defined by cyclic gene expression within PSM tissue with a period that regulates somite formation . These ‘clock’ genes belong to the Notch , WNT and FGF signalling pathways and are expressed in a cyclical pattern in the PSM of a wide variety of vertebrate species ( Krol et al . , 2011; reviewed in Dequéant et al . , 2008; Gibb et al . , 2010; Maroto et al . , 2012 ) . On a single cell level in the vertebrate PSM , oscillatory gene expression is believed to be established through negative feedback loops of unstable clock gene products . Previous mathematical models of the somitogenesis clock that invoke the mechanism of delayed negative feedback predict that the clock period can be approximated as a sum of the delays involved in processes such as transcription , splicing , translation and transport and the half-lives of inhibitors of clock gene expression ( Lewis , 2003; Monk , 2003; Ay et al . , 2013; Hanisch et al . , 2013 ) . Subsequent studies have confirmed that the period can be altered by genetically modifying genes that encode negative regulators ( e . g . , Hes7 , Her6 , Her7 , Nrarp ) ( Herrgen et al . , 2010; Schröter and Oates , 2010; Kim et al . , 2011; Takashima et al . , 2011; Oates et al . , 2012; Harima , et al . , 2013; Hoyle and Ish-Horowicz , 2013 ) . Moreover , it has been shown that regulation of clock gene mRNA turnover and degradation is gene specific and regulated at the level of the 3′UTR ( Hilgers et al . , 2005; Nitanda et al . , 2014 ) . In contrast , there has been relatively little experimental work demonstrating the predicted role of protein stability in regulating the periodicity of clock gene oscillations ( Hirata et al . , 2004 ) . It is notable that models that invoke delayed negative feedback as the mechanism underlying the somitogenesis clock were developed primarily from zebrafish data , where the clock period is relatively short and there are significantly fewer oscillating components than other vertebrate species ( Krol et al . , 2011 ) . Moreover , in zebrafish there is experimental evidence that Notch plays a key role in the synchronisation of neighbouring oscillators ( Delaune et al . , 2012 ) but that it is not necessary for oscillations themselves ( Ozbudak and Lewis , 2008 ) . In contrast , Notch signalling in mouse and chick is thought to be essential for oscillations ( Ferjentsik et al . , 2009 ) . One crucial factor in this regard is the Notch1 intracellular domain ( NICD ) which is cleaved following ligand activation , translocates to the nucleus and activates Notch target gene expression . NICD is unstable and , at least in mouse , its production appears as pulsatile , spatio-temporal waves that traverse the rostro-caudal axis of the PSM in the manner of a clock gene ( Huppert et al . , 2005 ) . Importantly , pulsatile Notch1 protein expression was recently shown to be dependent on Notch signalling in the PSM of both chick and mouse ( Bone et al . , 2014 ) ; this raises the possibility that , in mouse and chick at least , positive feedback , mediated by Notch and NICD , plays a role in the clock mechanism . In this study we develop a mathematical model of the clock that describes both repression and NICD-mediated activation of clock gene transcription . The model predicts that increasing the half-life of NICD results in elevated levels of NICD and a longer clock period . Notably , these effects are not observed in the model upon increasing the half-life of the repressing clock protein . To test the predictions , we use a pharmacological approach that robustly delays the phase patterns of a clock gene in the chick and mouse PSM . Under these conditions , using a custom made antibody that recognises endogenous chick NICD , we show that , concomitant to delaying the pace of oscillations , the inhibitors increase both levels of NICD and its half-life . In agreement with model predictions , we rescue the phase patterns of clock gene expression by reducing NICD production . These data therefore suggest that increasing the half-life of NICD and thus increasing its turnover time increases the period of the somitogenesis clock .
In order to investigate the role of both positive and negative feedback in the circuitry of the somitogenesis clock , we developed a mathematical model in which the key regulatory interactions are that an inhibitory clock protein and NICD negatively and positively regulate gene transcription , respectively ( see ‘Materials and methods’ and Figure 1A , B ) . Hence in the absence of clock protein but presence of NICD , transcription occurs . Conversely , transcription is inhibited when there are high levels of inhibitor and low levels of activator . 10 . 7554/eLife . 05842 . 003Figure 1 . A mathematical model of positive and negative regulation of the somitogenesis clock . ( A ) A schematic illustration of inter-cellular coupling via Delta-Notch signalling . ( B ) A schematic illustration of the reduced model in which the pulsatile production of NICD is down stream of clock gene expression . ( C–E ) Representative numerical solution of the model Equations ( 1 ) - ( 3 ) . Clock protein ( p ( t ) , solid line ) and NICD ( n ( t ) , dashed line ) levels are plotted against time . ( C ) 'Wild-type' oscillations ( k_4 = 0 . 063 ) , ( D ) Decreasing NICD decay rate yields longer period oscillations ( k_4 = 0 . 023 ) , ( E ) Decreasing Hes7 decay rate lowers levels of Hes7 and NICD ( k_2 = 0 . 035 ) . ( F ) The oscillation period ( colour ) is plotted against Hes7 and NICD half-lives . ( G ) NICD level is plotted against Hes7 and NICD half-lives . ( H ) Oscillator period is plotted against NICD production rate and half-life . ( E–G ) Solid lines depict emergence of the non-trivial steady-state . Dashed lines depict points in parameter space where the nontrivial steady state undergoes a Hopf bifurcation . Damped Oscillations ( DO ) , Stable Oscillations ( SO ) , No Oscillations ( NO ) . Unless otherwise stated , parameter values are: k1=123 min−1 , k2=0 . 058 min−1 , k3=134 min−1 , k4=0 . 063 min−1 , T1=44 . 0 min , T2=67 . 0 min , KP=125 , KN=125 . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 003 In Figure 1C we plot a representative numerical solution of the model . In a given cycle of the clock , NICD activates transcription , resulting in the production of cytoplasmic protein after some delay , T1 . Clock protein then represses transcription and after a further delay , T2 , NICD is produced and activates the next cycle of the clock . In the model , perturbing protein half-lives alters the balance between activation and repression in a manner that modifies the clock period . In Figure 1D we show that increasing the NICD half-life increases the period of the oscillation and the levels of both NICD and clock protein . In Figure 1E we show that increasing the clock protein half-life does not greatly affect the oscillator period but reduces levels of NICD . In order to further explore model behaviour we calculated solutions of the model in different regions of parameter space . In Figure 1F , G we show that the oscillator period and levels of NICD increase with NICD half-life but that these effects are not seen upon increasing the clock protein half-life . In Figure 1H we show that the increased period , arising as a consequence of increased NICD half-life , can be rescued by reducing the NICD production rate . These numerical solutions demonstrate that , in the model , the balance between NICD production and degradation rates plays a key role in determining the clock period . Additionally , we note that there is a critical value of the production rate below which the trivial steady-state becomes stable ( solutions tend to zero ) and that perturbing clock protein and NICD production and decay rates can drive the system into stable states ( e . g . , by increasing repressor half-life in Figure 1F or reducing the NICD production rate in Figure 1H ) . It has previously been reported that modulating either the Shh or Wnt pathways can affect the pace of clock gene oscillations in the PSM ( Gibb et al . , 2009; Resende et al . , 2010; Gonzalez et al . , 2013 ) . To experimentally test the predictions of our model , we used a half-embryo assay to investigate whether levels and half-life of NICD were elevated under conditions where clock gene oscillations were robustly delayed using a pharmacological approach ( see ‘Materials and methods’; Gibb et al . , 2009 ) . In the first instance , PSM explants from chick embryos between Hamburger Hamilton ( HH ) stages 10–12 were exposed to XAV939 , a specific Wnt inhibitor that acts by inhibiting Tankyrase1 ( TNK1 ) and TNK2 enzymes , which normally degrade Axin2 . Hence upon treatment with XAV939 , AXIN2 protein is stabilised and maintains phosphorylated β-catenin protein in the destruction complex ( Huang et al . , 2009 ) . Following a titration assay , we determined that 100 μM XAV939 treatment was the lowest concentration that caused robust and reproducible down regulation of the Wnt target genes cAxin2 and cLef1 compared to the corresponding DMSO-treated contralateral explants ( Figure 2—figure supplement 1C , D; n = 61/68; Bone et al . , 2014 ) and led to increased levels of phosphorylated β-catenin at Ser33 , Ser37 , Thr41 , as expected ( Figure 2—figure supplement 1A; n = 8/10 ) . Using the same assay , we investigated the effect of the inhibitor on the dynamic expression of cLfng in the PSM . Following exposure to 100 μM XAV939 , the domain of cLfng expression was at least 1 phase behind that of the control explant ( Figure 2A , M; n = 57/64 ) and occasionally this resulted in the treated explant forming one less somite boundary than the control explant . These data imply that 100 μM XAV939 treatment noticeably lengthens the period of the oscillations . In explant pairs where both sides were DMSO-treated , an asymmetric pattern of cLfng expression was rarely seen ( n = 2/18 , data not shown ) and , as such , the effects of XAV939 on cLfng expression could not be attributed to DMSO , natural variability in expression , or to the assay itself . In order to ensure that the oscillations were delayed and not halted , a fix and culture assay ( see ‘Materials and methods’ ) revealed that cLfng expression was still dynamic in the presence of 100 μM XAV939 ( Figure 2B; n = 8/9 ) . Furthermore , Phospho-histone H3 ( pH3 ) and NucView analyses demonstrated that neither proliferation nor apoptosis , respectively , were significantly affected following drug treatment ( Figure 2—figure supplement 2A , D; n = 4 , p = 0 . 236; n = 4 , p = 0 . 292 ) . These data clearly demonstrate that Wnt inhibition delays the period of the segmentation clock in the chick PSM . Using the same half-embryo inhibitor assay in the mouse embryo , we found that exposure to 100 μM XAV939 delayed the pace of mLfng oscillations as compared to control DMSO-treated E10 . 5 half PSM explants ( Figure 2I; n = 19/26 ) . Moreover , a fix and culture assay revealed that mLfng expression was still dynamic in the presence of 100 μM XAV939 ( Figure 2J; n = 12/20 ) . 10 . 7554/eLife . 05842 . 004Figure 2 . XAV939 , Roscovitine , DRB and PHA767491 treatment delays the pace of the segmentation clock in the chick and mouse PSM . Bissected chick or mouse caudal explant pairs treated ‘−’ or ‘+’ inhibitor ( A , C , E , G ) or treated with inhibitor and subjected to the fix and culture assay ( B , D , F , H ) and then analysed by in situ hybridisation for Lfng mRNA expression : ( A , C , E , G ) Treatment of chick PSM explants in the presence ( + ) or absence ( − ) of XAV939 ( A ) , Roscovitine ( C ) , DRB ( E ) and PHA7667491 ( G ) for 3 hr reveals that ‘+’ explants have lagging expression of cLfng , often with one less somite formed than the ‘−’ explants . ( B , D , F , H ) : After 3 hr treatment in the presence of XAV393 ( B ) , Roscovitine ( D ) , DRB ( F ) and PHA767491 ( H ) , one chick PSM explant was fixed while the other was treated for another 45 min , showing that cLfng expression is still dynamic in the presence of these inhibitors . ( I , K ) : Mouse PSM explants treated in the presence or absence of XAV939 ( I ) or Roscovitine ( K ) for 4 hr revealed a delay in the oscillations of mLfng expression . ( J , L ) : Treatment of one mouse PSM explant for 4 hr , and the other for 5 hr in the presence of XAV939 ( J ) or Roscovitine ( L ) reveals that mLfng mRNA expression is still oscillating in the PSM . The red arrowheads identify the somites that have formed during the in vitro culture period of the assay . ( M ) Schematic representation of the expression domains of Lfng in the PSM in the three different phases of one oscillation cycle . S1 , SII = the most recently formed somite . ( P ) = previous cycle . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 00410 . 7554/eLife . 05842 . 005Figure 2—figure supplement 1 . XAV939 , Roscovitine , DRB treatment down-regulates expression of the Wnt target Axin2 and the Shh target Gli1 in the chick PSM . ( A ) : Western Blot analysis of pooled PSM lysates from half PSM explants treated ‘+’ or ‘−’ the inhibitors using antibodies against phosphorylated β-catenin , reveals levels of this form of β-catenin is increased in the chick PSM after 3 hr treatment with XAV939 ( mean fold-change = 2 . 856 , p = 0 . 016 ) , but unaffected after 3 hr treatment with Roscovitine ( mean fold-change =1 . 134 , p = 0 . 617 ) or DRB ( mean fold-change = 0 . 785 ) . ( B ) : Similar analysis looking at levels of RNA Polymerase which is phosphorylated at Serine 2 ( Ser2 ) reveal this modification appears unaffected by 3 hr treatment with Roscovitine , DRB or PHA767491 , but levels of RNA polymerase phosphorylated at Serine 5 ( Ser5 ) are somewhat decreased following this treatment . ( C ) : Bissected chick caudal explant pairs treated ‘−’ or ‘+’ the inhibitors and then analysed by in situ hybridisation show expression of cLef1 is down-regulated after 3 hr treatment in the presence of XAV939 in the chick PSM . ( D–F ) : Expression of the cAxin2 is similarly down-regulated in the chick PSM after 3 hr treatment in the presence of XAV939 ( D ) , Roscovitine ( E ) , or DRB ( F ) . ( G–I ) : Expression of cGli1 is also down-regulated after 3 hr treatment in the presence of XAV939 ( G ) , Roscovitine ( H ) , or DRB ( I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 00510 . 7554/eLife . 05842 . 006Figure 2—figure supplement 2 . XAV939 , Roscovitine , DRB treatment does not induce apoptosis or affect cell proliferation in the chick PSM . ( A–C ) : Mean numbers of phospho-histone3 ( Ph3 ) -positive cells in the chick PSM are not significantly affected after 3 hr treatment with XAV939 ( A ) , Roscovitine ( B ) , or DRB ( C ) . ( D–F ) : Mean numbers of NucView-positive cells in the in the chick PSM are not significantly affected after 3 hr treatment with XAV939 ( D ) , Roscovitine ( E ) , or DRB ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 00610 . 7554/eLife . 05842 . 007Figure 2—figure supplement 3 . XAV939 , Roscovitine , DRB and PHA-767491 treatment delays dynamic mRNA expression of Notch target clock genes in the chick PSM thereby extending the clock period . ( A–D ) : Bissected chick caudal explant pairs treated ‘−’ or ‘+’ the inhibitors and then analysed by in situ hybridisation for the Notch target gene cHairy2 reveals cHairy2 mRNA expression is delayed in the treated side as compared to the control after 3 hr treatment with XAV939 ( A ) , Roscovitine ( B ) , DRB ( C ) , PHA-767491 ( D ) . ( E–J ) : Bissected chick caudal explant pairs treated for 3 hr ‘−’ or ‘+’ the inhibitors and then subjected to fix and culture where the cultured side is cultured for 90 min ( E , F ) or 2 hr ( G–J ) and then analysed by in situ hybridisation for cLfng mRNA expression . Red arrows indicate newly formed somites from the beginning of treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 007 Similarly , we directly perturbed the Shh pathway in the half-embryo assay using Cyclopamine , a well-established inhibitor of Shh signalling . At 25 μM , Cyclopamine abolished expression of the Shh target gene cGli1 in the chicken half-PSM explant compared to the DMSO control ( Figure 3A; n = 5/5 ) . Surprisingly , and in contrast to published data ( Resende et al . , 2010 ) , cLfng expression in the PSM was not delayed at either 25 μM ( Figure 3B , D; n = 6/9 ) or 50 μM ( Figure 3C , D; n = 3/4 ) . This assay suggests that Shh signalling does not play a direct role in the regulation of the segmentation clock . 10 . 7554/eLife . 05842 . 008Figure 3 . Cyclopamine treatment does not affect the pace of cLfng oscillations in the chick PSM . Bissected chick caudal explant pairs treated ‘−’or ‘+’ cyclopamine and then analysed by in situ hybridisation for Lfng or Gli1 mRNA expression . ( A ) : At 25 µM , cyclopamine inhibits expression of the Shh target gene cGli1 in the chick PSM after 3 hr treatment . ( B , C ) : The expression domains and intensity of cLfng mRNA are the same in the PSM of the ‘−’ or ‘+’ cyclopamine halves of each embryo and there is no difference in the number of somites in each half of a pair thus we conclude the pace of Lfng oscillation is not affected by cyclopamine treatment for 3 hr at either 25 µM ( B ) or 50 µM ( C ) . ( D ) Schematic representation of the expression domains of Lfng in the PSM in the three different phases of one oscillation cycle . S1 , SII = the most recently formed somite . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 008 Given our results with the Wnt inhibitor and that NICD stability is known , in other contexts , to be regulated by GSK3B ( Espinosa et al . , 2003; Jin et al . , 2009; see also Foltz et al . , 2002 ) , we investigated whether the effect upon the period was associated with a change in levels of NICD , as the model predicted . To that end we generated a polyclonal anti-cNICD antibody raised against the chicken N-terminal sequence of the cleaved chicken Notch1 intracellular domain . The epitope is only exposed after gamma secretase cleavage and is not accessible in the un-cleaved form . After confirming specificity by Western Blotting against recombinant cNICD protein ( Figure 4A , lanes 1 , 2 ) , we cultured half-PSM explant pairs in the presence of the γ-secretase inhibitor LY411575 ( Lanz et al . , 2004 ) or DMSO for 3 hr . Western Blot revealed a specific band against endogenous cNICD that ran at the same size as recombinant cNICD protein and at the predicted molecular weight of ∼87 kDa in the DMSO control pool which was abolished in the +LY411575 treated pool , thereby validating the specificity of the antibody ( Figure 4A , lanes 3 , 4; Figure 4D , lanes 1 , 2 ) . We performed immunohistochemistry on sagittal sections of chick PSM tissue to analyse the spatial localisation profile of NICD in this tissue and found it to exhibit a very similar profile to that described for mNICD in the mouse PSM ( Huppert et al . , 2005 ) . Thus cNICD is produced in ( and cleared from ) discrete domains of the PSM in a manner reminiscent of the expression profile of clock genes in different phases of the cycle ( Figure 4B ) . 10 . 7554/eLife . 05842 . 009Figure 4 . cNICD levels in the chick PSM are elevated after exposure to XAV939 , Roscovitine , DRB or PHA767491 . ( A ) : A polyclonal cNICD antibody was raised against the N-term sequence of the cleaved chicken Notch1 intracellular domain . The epitope is only exposed after gamma secretase cleavage and is not accessible in the uncleaved form . By Western Blot analysis the antibody detects a band of protein at around 90KDa ( see arrow ) in an in vitro-generated recombinant cNICD sample , which is not detected in the control IVT sample; this band disappears in half chick PSM pools ( see ‘Materials and methods’ ) after 3 hr treatment with 150 nM LY411575 , a concentration shown previously to have no toxic effects to PSM tissue ( Bone et al . , 2014 ) . ( B ) : Immunohistochemistry for cNICD protein on 16 μm sagittal sections of sucrose agar embedded HH10 chicken embryo tails showing dynamic phases of localisation in the PSM and schematic representations to the right of each panel . S1 - S3 = somite; S0 - S-1 = prospective somite region of anterior PSM . ( C ) : A schematic illustrating the cNICD degradation assay: corresponding pools of 9 PSM explants are incubated in the presence or absence of an inhibitor for 3 ( chick ) or 4 hr ( mouse ) , before treating all pools with the protein synthesis inhibitor cycloheximide ( CHX ) for a further hour . Lysates are then collected for Western Blot analysis . ( D ) : Representative Western Blot showing levels of cNICD protein in lysates of chick PSM pools are increased after treatment with XAV939 , Roscovitine , DRB and PHA767491 , but not with cyclopamine . Lanes 1 and 2 show exposure to 150 nM LY411575 alone for 3 hr removes all NICD and serves as a control for the western . Lanes 3 and 4 show exposure to CHX alone in the last hour of culture severely depletes NICD levels as compared to NICD levels in the pool of contralateral PSM explants cultured in DMSO . ( E ) : Levels of mNICD in the mouse PSM pools are increased following treatment with XAV939 . Lanes 1 and 2 show that NICD protein levels drop drastically after exposure for 1h to CHX as compared to strong NICD signal in lysate from the contralateral half PSMs treated with EtOH . ( F ) : Representative Western Blot showing levels of cNICD protein in lysates of chick PSM pools are increased after treatment with XAV939 , Roscovitine and DRB when LY411575 is added to the culture for the last hour ( in place of CHX ) . Lanes 1 and 2 show exposure to LY411575 alone only in the last hour severely depletes NICD levels as compared to the NICD levels in the pool of contralateral PSM explants cultured in DMSO . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 00910 . 7554/eLife . 05842 . 010Figure 4—figure supplement 1 . XAV939 , Roscovitine , and DRB treatment increases the level of cNICD in the chick PSM . ( A ) : A schematic showing the degradation assay notably without the cycloheximide step described in the assay for Figure 4: pools of 9 explants were treated either in the presence or absence of inhibitors for 3 hr before lysis . ( B–D ) : After the treatment depicted in ( A ) , Western Blot analysis revealed levels of cNICD are elevated in chick PSM explant pools after treatment with XAV939 , p = 0 . 0361 ( B ) , Roscovitine , p = 0 . 00653 ( C ) , or DRB , p = 0 . 00601 ( D ) relative to their respective DMSO control treated pools . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 010 To investigate if exposure to XAV939 affects the levels of endogenous cNICD , chick embryos were bissected as before and contralateral sides pooled . For each reagent , one pool was cultured for 3 hr in DMSO and the other was cultured in the inhibitor . In order to cease production of new NICD , both control and inhibitor treated pools were cultured for an additional hour in the presence of cycloheximide ( Figure 4C ) . Lysates were then analysed by Western Blot for NICD . Strikingly , we observed that the level of cNICD protein was increased significantly in the presence of XAV939 ( Figure 4D , lanes 5 , 6; n = 7/9; Table 1 ) compared to the corresponding DMSO control pool where the resident cNICD had degraded substantially in the presence of cycloheximide . For each assay , additional control explants confirmed that: i ) Lfng was delayed in inhibitor treated explants; and ii ) samples cultured for 3 hr in the absence of inhibitor and then for the last hour in cycloheximide alone showed a reduction in levels of NICD when compared to contralateral explants cultured in Ethanol for the last hour ( Figure 4D , lanes 3 , 4 ) . Notably , even in the absence of cycloheximide ( Figure 4—figure supplement 1A ) , levels of cNICD in the chicken PSM were consistently higher when treated with XAV939 ( Figure 4—figure supplement 1B; n = 5/5; Table 1 ) compared to respective DMSO control lysates , although the mean fold-change increase was slightly less . These assays reveal that exposure to XAV939 elevates the level of cNICD protein in the chicken PSM , just as it delays oscillations of the Notch target gene cLfng . Significantly , exposure to Cyclopamine did not affect the levels of cNICD in treated PSM explants just as it did not affect the pace of cLfng oscillations in this tissue ( Figure 4D , lanes 13 , 14; n = 2/2; Table 1 ) . 10 . 7554/eLife . 05842 . 011Table 1 . Summary table of densitometry quantifications for cNICD and mNICD in inhibitor assaysDOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 011TreatmentNMean adjusted relative fold change of cNICD protein± standard error of the meanStatistical testp-valueXAV93993 . 078±0 . 776One-sample T-test0 . 0280*XAV939 ( no CHX ) 52 . 491±0 . 480One-sample T-test0 . 0361*Roscovitine101 . 94±0 . 255One-sample Signed Rank test0 . 002*Roscovitine ( no CHX ) 71 . 594±0 . 146One-sample T-test0 . 00653*DRB82 . 557±0 . 702One-sample Signed Rank test0 . 008*DRB ( no CHX ) 61 . 369±0 . 081One-sample T-test0 . 00601*Cyclopamine21 . 263±0 . 106One-sample T-test0 . 244PHA76749123 . 479±0 . 451One-sample T-test0 . 115MLN492438 . 831±2 . 734One-sample T-test0 . 103TreatmentNMean adjusted relative fold change of mNICD protein± Standard error of the meanStatistical testp-valueXAV93982 . 221±0 . 396One-sample signed rank test0 . 008**demarcates a statistically significant difference . Densitometry was performed on Western Blots and the fold change of cNICD or mNICD in the + inhibitor treated sample relative to the corresponding DMSO control was adjusted to the relative change in the tubulin loading control . The given statistical test was used to compare the average value for each assay to a fixed fold-change value = 1 ( i . e . no change ) and a p-value obtained . Given the consistent effect of 100 μM XAV939 upon both cLfng and mLfng oscillations , we used the same assay to investigate whether this reagent also affected levels of mNICD in the mouse PSM , with the initial culture period extended to 4 hr to allow for the longer clock period in the mouse embryo . Strikingly , we observed that the level of mNICD protein in the PSM was increased significantly in the presence of XAV939 ( Figure 4E , lanes 3 , 4; n = 8 ) compared to the corresponding DMSO control pool of half PSM explants , where mNICD levels had dropped substantially in the presence of cycloheximide ( Figure 4E , lanes 1 , 2; Table 1 ) . Thus , the effect of XAV939 to delay the pace of Lfng oscillations across the PSM and to increase the level of NICD in the PSM is conserved in both chick and mouse . Previous studies have shown that the stability of NICD in other contexts can be regulated by cyclin dependent kinase ( CDK ) mediated phosphorylation ( Fryer et al . , 2004 ) . Given the prediction of the mathematical model that modification of NICD stability can influence the period of the clock , we investigated whether perturbing CDK signalling would affect either clock gene oscillations and/or levels of NICD in the chick PSM . To that end , we directly perturbed the CDK pathway in the half-embryo assay using the CDK inhibitor Roscovitine . Roscovitine is a 2 , 6 , 9-substituted purine analogue that competes with ATP for the active binding site on CDKs ( MacCallum et al . , 2005 ) . We found that Roscovitine treatment at 10 µM clearly delayed oscillation of cLfng expression relative to the DMSO control ( Figure 2C; n = 59/62 ) whilst the fix and culture assay confirmed that cLfng expression was still oscillatory ( Figure 2D; n = 8/10 ) . Neither proliferation ( Figure 2—figure supplement 2B; n = 5 , p = 0 . 204 ) nor apoptosis ( Figure 2—figure supplement 2E; n = 4 , p = 0 . 886 ) were significantly affected in the PSM . However , exposure to 100 or 300 μM Roscovitine severely impacted both proliferation and apoptosis and strongly down-regulated or abolished cLfng expression ( data not shown ) . These findings reveal that the oscillations of cLfng expression in the chicken PSM are slowed by Roscovitine treatment at a concentration that does not appear to affect cell proliferation . Moreover , we found that exposure to 10 μM Roscovitine delayed the pace of mLfng oscillations as compared to control DMSO-treated E10 . 5 mouse half PSM explants ( Figure 2K; n = 20/22 ) . The fix and culture assay revealed that mLfng expression was still dynamic in the presence of 10 μM Roscovitine ( Figure 2L; n = 8/10 ) . Roscovitine is a broad spectrum CDK inhibitor . We used an additional two CDK inhibitors reported to specifically reduce the kinase activities of CDK7/CyclinH and CDK9/CyclinT towards the C-terminal domain ( CTD ) of RNA pol II: 5 , 6-Dichloro-1-beta-D-ribofuranosylbenzimidazole ( DRB; a nucleoside analogue ) and PHA-767491 . We found that exposure of chick half-PSM explants to 10 μM DRB delayed cLfng oscillation phase compared to the control explant ( Figure 2E; n = 67/73 ) and the fix and culture assay confirmed that cLfng expression was still oscillatory in the PSM at this concentration ( Figure 2F; n = 15/18 ) . Western Blotting analysis confirmed that phosphorylation of Serine 5 on the CTD of RNA pol II was reduced by 10 μM DRB treatment ( Figure 2—figure supplement 1B; n = 3/3; mean fold change = 0 . 709 ) . Neither proliferation ( Figure 2—figure supplement 2C; n = 5 , p = 0 . 884 ) nor apoptosis ( Figure 2—figure supplement 2F; n = 3 , p = 0 . 844 ) were significantly affected . Similarly , exposure to 1 μM PHA-767491 reduced phosphorylation of Serine 5 on the CTD of RNA pol II ( Figure 2—figure supplement 1B; n = 2; mean fold change = 0 . 709 ) , delayed cLfng oscillation phase compared to the control half-PSM ( Figure 2G; n = 13 ) and the fix and culture assay confirmed that cLfng expression was still oscillatory in the PSM at this concentration ( Figure 2H; n = 6 ) . Another cycling Notch target , cHairy2 , was also consistently delayed compared to the respective DMSO control when treated with either XAV939 ( Figure 2—figure supplement 3A; n = 4/6 ) , Roscovitine ( Figure 2—figure supplement 3B; n = 6/7 ) , DRB ( Figure 2—figure supplement 3C; n = 5/6 ) or PHA-767491 ( n = 8/8; Figure 2—figure supplement 3D ) . These data demonstrate that a panel of CDK inhibitors robustly delays Notch target clock gene oscillations in the PSM in a similar way to that seen following Wnt inhibition and that these effects are separable from the cell cycle . In order to investigate whether the CDK inhibitors all act through the Wnt pathway to elicit their effect on the clock period , we assayed transcription of the Wnt target cAxin2 in the caudal chicken PSM . We found that 100 μM XAV939 ( Figure 2—figure supplement 1D ) , 10 μM Roscovitine ( Figure 2—figure supplement 1E; n = 31/41 ) and 10 μM DRB ( Figure 2—figure supplement 1F; n = 29/42 ) robustly down-regulated transcription of cAxin2 in the caudal chicken PSM , compared to the DMSO control . These data raise the possibility that the effects of these reagents on the segmentation clock are via inhibition of Wnt activity . However , the increased phosphorylation of β-catenin at Ser33 , Ser37 , Thr41 observed with XAV939 , which serves as a readout of Wnt inhibition , was not observed following exposure to 10 μM Roscovitine ( Figure 2—figure supplement 1A; n = 5/6 ) or 10 μM DRB ( Figure 2—figure supplement 1A; n = 4/4 ) , suggesting that the mechanism by which these CDK inhibitors inhibit transcription of cAxin2 is not through the inhibition of canonical Wnt signalling . However , alternative mechanisms of Wnt inhibition exist that do not lead to an increase in β-catenin phosphorylation and thus we cannot rule out an effect of CDK inhibition via changes in Wnt signalling . Having shown that Wnt inhibition or exposure to a panel of CDK inhibitors robustly delays the pace of the segmentation clock , a key question is whether or not there is a common mechanism underlying these observations . We thus investigated whether levels of NICD were perturbed by the CDK inhibitor treatments . Using the same assay as described above , one pool of PSMs was cultured for 3 hr in DMSO and the contralateral halves were cultured in the inhibitor . Both pools were then cultured for an hour in the presence of cycloheximide before Western Blot analysis ( Figure 4C ) . Remarkably , and in agreement with model predictions , we observed that the level of cNICD protein was again increased significantly in the presence of Roscovitine ( Figure 4D , lanes 7 , 8; n = 9/10 ) , DRB ( Figure 4D , lanes 9 , 10; n = 6/8 ) and PHA-767491 ( Figure 4D , lanes 11 , 12; n = 2/2 ) compared to the corresponding DMSO control pool ( See Table 1 ) . Again , even in the absence of cycloheximide ( Figure 4—figure supplement 1A ) , levels of cNICD in the chicken PSM were consistently higher when treated with Roscovitine ( Figure 4—figure supplement 1C; n = 5/7 ) , and DRB ( Figure 4—figure supplement 1D; n = 4/6 ) compared to respective DMSO control lysates ( See Table 1 ) . Thus , Roscovitine , DRB and PHA-767491 treatments all increase the level of cNICD protein in the chicken PSM , just as they all delay oscillations of the Notch target gene cLfng . Hence , taken together the data with the Wnt and CDK inhibitors fulfil the predictions of the model that increased levels of NICD will be associated with an increase to the periodicity of clock gene oscillations . To verify the extent to which the period had been extended in the presence of each of the reagents , we performed a time course of the fix and culture assay . We determined the time needed for one full oscillation by extending the time in culture for the cultured half until the expression profiles matched in the two explants , with an extra somite having formed in the cultured explant . Untreated control pairs completed one full oscillation and made an extra somite in 90 min whereas exposure to the reagents resulted in an oscillation period of approximately 120 min ( n = 8 for each inhibitor and for DMSO; Figure 2—figure supplement 3E-J ) . A prediction of the clock and wave-front model of somitogenesis is that a longer clock period will result , assuming a constant wave-front speed , in the formation of fewer , larger somites in a given time . Hence we analysed somite number in drug-treated explants cultured for an extended period of 6 hr and found that fewer somites formed in the presence of Roscovitine ( n = 9 , mean number = 2 . 11 ) , DRB ( n = 9 , mean number = 1 . 94 ) , PHA-767491 ( n = 10 , mean number = 2 . 5 ) and XAV939 ( n = 7 , mean number = 3 . 14 ) compared to DMSO controls ( n = 15 , mean number = 4 . 25 ) . Additionally , using the fix and culture assay to confirm that the clock is still operating under treatment conditions , we observed that Lfng expression is still dynamic after the 6 hr culture period ( Figure 5D–H; n = 5/5 for each reagent and for DMSO ) . Moreover , once the somites from the determined region of the PSM had segmented we observed a significant stepwise increase in somite size along the anteroposterior axis in the embryos cultured in the presence of XAV939 , Roscovitine , PHA-767491 or DRB as compared to DMSO control treated embryos ( Figure 5A–C , I–R ) . Taken together , these data demonstrate that XAV939 , Roscovitine , PHA-767491 and DRB treatments all result in an extension of the clock period and the formation of fewer , larger somites . 10 . 7554/eLife . 05842 . 012Figure 5 . Length of newly formed somites are increased following long-term treatment with inhibitors . ( A–C ) : Graphs showing log base 2 ( fold ) changes in 2 consecutive somite length ratios between the average somite lengths for the first ( +1 ) , second ( +2 ) , and third ( +3 ) formed somites following treatment initiation in the presence of DMSO , XAV939 , Roscovitine , PHA767491 or DRB . T-tests were performed to test the null hypothesis that the mean somite length ratio is the same as in DMSO controls . The null hypothesis was rejected in the starred cases ( p < 0 . 005 ) with the following p values XAV = 0 . 0211; Rosc = 0 . 0002; Pha = 0 . 0001; DRB = 0 . 0007 ( D–H ) Bissected chick caudal explant pairs subjected to the fix and culture assay after 6 hr treatment in the presence of DMSO , XAV939 , Roscovitine , PHA767491 or DRB . In each case the cultured side showed cLfng expression had advanced . ( I–M ) In situ hybridisation of cTbx6 on whole embryos following 6 hr treatment in the presence of either DMSO , XAV939 , Roscovitine , PHA767491 or DRB . Asterix = newly formed somite ( s ) during culture , line = length of most recently formed somite , scale bar = 100 μm . ( N–R ) High magnification image of the samples in ( I–M ) . XAV939 treated embryos do not form as clearly defined boundaries as the other inhibitor treated embryos . ( F ) = following cycle . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 012 Given the delays inherent to Notch and Delta-like1 ( DLL1 ) trafficking and their intercellular interaction , it is possible that new NICD could be produced even in the presence of CHX , based on signalling through receptors that were made prior to the initiation of CHX treatment . Thus , if XAV939 , Roscovitine or DRB increased the production of DLL1 or Notch1 , the observed increase in NICD might be secondary to increased signalling , rather than a result of changes in protein turnover . To address this possibility , we replaced the 1h CHX treatment with a 1h LY411575 treatment to specifically block production of new NICD , repeated the assay and measured levels of cNICD by Western Blot analysis . Thus , one pool of PSMs was cultured for 3 hr in DMSO and the contralateral halves were cultured in XAV939 , Roscovitine or DRB . Both pools were then cultured for an hour in the presence of LY411575 before Western Blot analysis ( Figure 4F ) . Remarkably , we observed that the level of cNICD protein was again increased in the presence of XAV939 ( Figure 4F , lane 4 ) , Roscovitine ( Figure 4F , lane 6 ) and DRB ( Figure 4F , lane 8 ) compared to the corresponding DMSO control pools ( Figure 4F , lanes 3 , 5 , 7 ) where NICD levels were severely diminished due to endogenous NICD degradation plus exposure to LY411575 for an hour . As a control , explants cultured in explant media for 3h followed by 1h LY411575 showed severe depletion of NICD compared to the corresponding DMSO pool of contralateral halves ( Figure 4F , lanes 1 , 2 ) . These data demonstrate that the higher levels of cNICD observed following exposure to the inhibitors is not due to increased production of NICD . In order to test the hypothesis that the observed increased level of NICD following inhibitor treatment is due to modification of the half-life of NICD , we sought to quantify the half-life of endogenous cNICD in the PSM . Pools of 9 explants from corresponding embryos were prepared as above but cultured in the presence or absence of a given inhibitor . Following an initial 30 min exposure to cycloheximide in both pools to inhibit new protein synthesis , the decay of cNICD protein was monitored as follows: in one pool protein lysate was immediately prepared whilst in the other , culture for an additional 30–60 min was performed before lysate was prepared ( Figure 6—figure supplement 1 and data not shown ) . Thus , the temporal decay of cNICD was quantified ( see Figure 6A ) . Notably the measured profiles are of the form predicted by Equation ( 4 ) . 10 . 7554/eLife . 05842 . 013Figure 6 . cNICD half-life in the PSM is increased after exposure to Roscovitine , DRB and PHA-767491 . ( A ) : Endogenous cNICD levels analysed by Western Blot ( after normalisation to alpha tubulin loading control ) in PSM explants cultured initially in the presence of each reagent for 3 hr and subsequently in the presence of inhibitor plus cycloheximide for an additional 30 min before the time-course . Panels show log transformation on repeated data . The half life is proportional to the inverse of the inferred slope ( Equation [4] ) . ( B ) Bar chart showing the half-life of cNICD in each condition . Asterix represents statistically significant differences p < 0 . 05 . ( C ) Left Panel: Treatment in the presence ( + ) or absence ( − ) of MLN4924 for 3 hr reveals that the inhibitor treated explant has lagging expression of cLfng compared to the control contralateral explant . Right panel: After 3 hr in the presence of MLN4924 one PSM explant was fixed while the other was treated for another 45 min , showing cLfng expression is still dynamic in the presence of this inhibitor . ( D ) Following the assay described in Figure 4 , levels of cNICD , normalised to levels of alpha tubulin , in the chick PSM pools are highly increased in the presence of MLN4924 . Western Blot analysis of pooled PSM lysates from half PSM explants treated ‘+’ or ‘−’ MLN4924 reveals levels of phosphorylated β-catenin are increased in the chick PSM after 3 hr treatment with MLN4924 . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 01310 . 7554/eLife . 05842 . 014Figure 6—figure supplement 1 . Roscovitine , PHA-767491 and DRB treatment increases the level of cNICD in the chick PSM . Figure 6—figure supplement 1 ( A–B ) : Western Blot analysis to monitor the decay of cNICD protein: after 3 hr treatment with the reagents the pools of 9 explants were exposed to cycloheximide for 30 min to inhibit new protein synthesis , and subsequently the decay of cNICD protein was monitored whereby one pool was removed from culture and protein lysate prepared while the corresponding pool was cultured for an additional 30–60 min before lysate also prepared . The +/− lanes with Cycloheximide ( CHX ) show that levels of cNICD fall dramatically in the absence of de novo protein synthesis in the presence of this reagent . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 014 From these data we infer the NICD half-life and find that it takes the value 11 . 87 ( +/− 4 . 97 ) minutes in untreated control embryos ( Figure 6A; n = 5 ) . Strikingly , in the DRB , Roscovitine and PHA-767491 treated explants , the half-life of cNICD protein is 29 . 59 ( +/− 9 . 29 ) , 22 . 01 ( +/− 7 . 04 ) min , and 36 . 41 ( +/− 8 . 88 ) min , respectively ( Figure 6A , B; n = 4 for each inhibitor ) . These data show that DRB , Roscovitine and PHA-767491 treatments all significantly increase the half-life of cNICD protein in the chick PSM . Thus , the data support the model prediction that an increased NICD half-life results in a longer clock period . Whilst the measurements of NICD half-life are consistent with the model prediction that increased NICD half-life will result in longer period oscillations and increased levels of NICD , we reasoned that if NICD half-life is controlling the clock period , we ought to be able to increase the clock period by targeting NICD turnover . NICD stability is known to be regulated by the Skp , Cullin , F-boxSel10/FBXW7 ( SCF Sel10/FBXW7 ) E3 ubiquitin ligase complex ( Gupta-Rossi et al . , 2001 ) . We tested whether turnover of NICD within the PSM affects the clock pace by repeating the half-embryo assay in the presence or absence of MLN4924 which inactivates SCF E3 ubiquitin ligase complexes by suppressing neddylation of Cullin-1 ( Soucy et al . , 2009 ) . As a control , we analysed levels of phosphorylated β-catenin which is also targeted for ubiquitination and degradation by the SCF E3 ubiquitin ligase complex . As expected , exposure of PSM explants to 1 μM MLN4924 led to increased levels of phosphorylated β-catenin ( Figure 6D ) . Notably , exposure of PSM explants to 1 μM MLN4924 caused a delay in the pace of Lfng dynamic expression across the PSM ( Figure 6C; n = 16 ) and increased levels of cNICD ( Figure 6D; n = 3; Table 1 ) . The fix and culture assay confirmed that cLfng expression was still oscillatory in the PSM at this concentration ( Figure 6C; n = 16 ) . Exposure to higher concentrations of MLN4924 ( 10–100 μM ) resulted in progressively greater increases in the level of cNICD but abolished dynamic Lfng expression across the PSM ( data not shown ) . Notably , the mathematical model predicts that oscillations are only supported within a limited range of NICD ( and Hes7 ) half-lives ( See Figure 1 ) . These data support the central thesis provided in this paper: a balance between positive and negative regulators is necessary for oscillations of the somitogenesis clock . Taken together , the data presented thus far support the hypothesis that increasing the half-life of NICD elevates levels of NICD and lengthens the period of oscillations . In order to directly address the causality between increased NICD stability and the change in pace of oscillations , we attempted to rescue the effect of the inhibitors by modifying NICD production in the PSM . We performed a titration curve with the gamma secretase inhibitor LY411575 to reduce but not remove NICD production . PSM pools were cultured for 3 hr in DMSO and their contralateral halves cultured in different concentrations of LY411575 . In the presence of 5 nM LY411575 , cNICD levels decreased significantly compared to the corresponding DMSO pool ( Figure 7A , lanes 1 , 2 ) , while in the presence of 1 nM or 0 . 1 nM LY411575 , the levels of cNICD protein were similar to control DMSO treated contralateral PSMs ( Figure 7B , lanes 1 , 2 , Figure 7C , lanes 1 , 2 ) . Hence , pools of PSMs were cultured for 3 hr in DMSO and the contralateral halves cultured in 10 μM Roscovitine together with 5 nM , 1 nM or 0 . 1 nM LY411575 prior to Western Blot analysis . The higher levels of NICD usually seen in the presence of Roscovitine alone ( Figure 7D ) are rescued by 1 nM LY411575 such that they resemble the levels of NICD in the corresponding DMSO pool ( Figure 7B lanes 3 , 4 ) . These findings concur with the predictions clearly proposed by our model ( Figure 1H ) . In contrast , 10 μM Roscovitine together with 5 nM LY411575 showed severely depleted levels of cNICD compared to the corresponding DMSO pool ( Figure 7A lanes 3 , 4 ) , resembling the effect of 5 nM LY411575 alone ( compared to Figure 7A , lanes 1 , 2 ) . 10uM Roscovitine together with 0 . 1 nM LY411575 showed increased levels of cNICD compared to the corresponding DMSO pool , resembling the effect of 10 μM Roscovitine alone ( Figure 7C , lanes 3 , 4 , compared to Figure 7D ) . 10 . 7554/eLife . 05842 . 016Figure 7 . Exposure to 1 nM LY411575 rescues the increased levels of cNICD caused by 10 μM Roscovitine . Endogenous cNICD levels analysed by Western Blot ( after normalisation to alpha tubulin loading control ) in PSM explant pools cultured for 3 hr in the presence ( + ) or absence ( − ) of ( A ) 5 nM LY411575 ( lane 1 , 2 ) ; 10 μM Roscovitine together with 5 nM LY411575 ( lane 3 , 4 ) or ( B ) 1 nM LY411575 ( lane 1 , 2 ) ; 10 μM Roscovitine together with 1 nM LY411575 ( lane 3 , 4 ) or ( C ) 0 . 1 nM LY411575 ( lane 1 , 2 ) ; 10 μM Roscovitine together with 0 . 1 nM LY411575 ( lane 3 , 4 ) ( D ) or 10μM Roscovitine alone . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 016 We repeated these experiments and analysed the effect upon the mRNA expression of the clock gene cLfng . 5 nM LY411575 severely reduced cLfng expression in the PSM ( Figure 8A; n = 4/4 ) whereas cLfng appeared largely unaffected in the presence of 1 nM or 0 . 1 nM LY411575 , compared to the control explant ( Figure 8B , C; n = 22/25; n = 11/13 , respectively ) . We then assessed whether we could rescue the changes in phase patterns brought about by Roscovitine , DRB or XAV939 by reducing NICD production through exposure to 1 nM or 0 . 1 nM LY411575 . Strikingly , we observed a rescue of the phase pattern in a significant number of explants cultured in the presence of Roscovitine , DRB or XAV939 together with 1 nM LY411575 ( Figure 8G–I; n = 15/39 , n = 8/20 , n = 6/15 , respectively ) compared to explants cultured in the presence of Roscovitine , DRB or XAV939 alone which show a delay to the phase pattern ( Figure 8D–F; n = 14/15 , n = 9/10 , n = 7/8 , respectively ) . Explants cultured in the presence of Roscovitine or XAV939 together with 0 . 1 nM LY411575 did not rescue the delay caused by these inhibitors ( Figure 8—figure supplement 1 and data not shown; n = 0/9; n = 0/4 ) . 10 . 7554/eLife . 05842 . 017Figure 8 . Exposure to 1 nM LY411575 rescues the delay in clock gene oscillations caused by Roscovitine , DRB and XAV939 . Bissected chick or mouse caudal explant pairs treated ‘−’ or ‘+’ inhibitor and then analysed by insitu hybridisation for Lfng mRNA expression: ( A–C ) Treatment of chick PSM explants in the presence ( + ) or absence ( − ) of 5 nM LY411575 ( A ) , 1 nM LY411575 ( B ) , 0 . 1 nM LY411575 ( C ) for 3 hr reveals that 5 nM LY411575 severely downregulates cLfng expression whereas 1 nM and 0 . 1 nM do not appear to change levels or domain of cLfng expression . ( D–F ) : Treatment of chick PSM explants in the presence ( + ) or absence ( − ) of Roscovitine ( D ) , DRB ( E ) or XAV939 ( F ) for 3 hr reveals that ‘+’ explants have lagging expression of cLfng , with one less somite formed than the ‘−’ explants . ( G–I ) : Treatment of chick PSM explants in the presence ( + ) or absence ( − ) of 1 nM LY411575 together with Roscovitine ( G ) , DRB ( H ) or XAV939 ( I ) for 3 hr reveals that 1 nM LY411575 rescues the delay in clock gene oscillations caused by these three inhibitors such that the cLfng expression domains in the ‘−’ and ‘+’ explants are very similar . The red arrowheads identify the somites formed during the in vitro culture period of the assay . ( P ) = previous cycle . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 01710 . 7554/eLife . 05842 . 018Figure 8—figure supplement 1 . 0 . 1 nM LY411575 does not rescue the delay in oscillation caused by Roscovitine . Bissected chick caudal explant pair treated ‘−’ or ‘+’ 10μM roscovitine and 0 . 1 nM LY411575 and then analysed by in situ hybridisation for the Notch target gene cLfng reveals cLfng mRNA expression is delayed in the treated side as compared to the control after 3 hr treatment . Black arrows indicate somites . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 018 These very striking data showing a rescue of both NICD levels and the delay to the clock oscillations provide strong evidence that increased half-life and higher levels of NICD brought about by Roscovitine , DRB or XAV939 underlie the delay to clock gene oscillations observed on exposure to the inhibitors ( Figure 9 ) . 10 . 7554/eLife . 05842 . 015Figure 9 . A schematic illustration depicting how modification of NICD stability may cause a delay to the period of the oscillations . ( A ) Schematic showing the competition between molecules of Hes7 and NICD for binding to the promoter of a clock gene . Exposure to Roscovitine , XAV939 or DRB causes increased stability and levels of NICD and thus NICD occupies the promoter for longer than in the wild-type situation , thereby prolonging the oscillation period . Exposure to 1 nM LY411575 together with Roscovitine , XAV939 or DRB reduces the level of NICD production offsetting the increased stability which rescues the delay . t1 = delay due to transcription of Hes7 and Notch; t2 = the delay due to translation of Hes7; t3 = delay due to translation of Notch and subsequent ligand dependent production of NICD . ( B ) Graphical model depicting how modifying the stability of NICD may cause a delay to the period of the oscillations by delaying the time-point at which proportional levels of NICD and Hes7 allow switching of promoter occupancy of these two factors . DOI: http://dx . doi . org/10 . 7554/eLife . 05842 . 015
Previous reports have suggested that inhibition of Shh signalling slows the segmentation clock in the chick PSM ( Resende et al . , 2010 ) . In this study , we clearly show that , in our hands , inhibiting the Shh pathway does not affect clock pace or the level of cNICD in the chick PSM within the timeframe of our assay . A possible reason for this discrepancy with Resende et al . is that we culture explant pairs for 3 hr whereas they observed delays to somite formation , in the absence of Shh signalling , only after 7 . 5 hr culture and alterations to clock gene expression after 4 . 5 hr culture . One possible explanation for the later arising effects they observe on the pace of the clock and somite formation is that they that are indirect secondary effects arising from Shh inhibition . In contrast , we report Wnt inhibition , within a 3 hr time frame , elicited by exposure to XAV939 , increases the half-life of NICD in the chick and mouse PSM , leading to elevated NICD levels and a clear and robust prolonged period of clock gene oscillations and somite formation . Intriguingly , Gonzalez et al . previously reported exposure of mouse PSM explants to LiCl , which acts to inhibit GSK3B , lengthens the period of Hes7 oscillations in the mouse PSM . However , the authors report this effect may rather be via LiCl activation of the MapK pathway which is important for Hes expression in the PSM ( Gonzalez et al . , 2013 ) , since more specific approaches to elevate Wnt signalling have been shown not to change the endogenous period of the segmentation clock ( Aulehla et al . , 2008; Gibb et al . , 2009 ) . Moreover , Gonzalez et al . found the Wnt inhibitor CKI-7 delayed clock gene oscillations in the mouse PSM , in line with our own observations ( Gibb et al . , 2009; Gonzalez et al . , 2013 ) . We demonstrate that three CDK inhibitors , Roscovitine , DRB and PHA-767491 , also increase the half-life of cNICD and delay the pace of Lfng oscillations at concentrations that have no significant effect on cell proliferation . At higher concentrations of inhibitors that did affect proliferation , the oscillations of Lfng expression were abolished . Importantly , these data separate the effects of the inhibitors on the segmentation clock from those on the cell cycle . This finding is consistent with the ‘clock and wavefront’ model of somitogenesis , rather than a cell cycle based model ( Cooke and Zeeman , 1976; Stern et al . , 1988 ) . A striking finding is the fact that these very different pharmacological reagents share distinctive phenotypic outputs: increased levels and half-life of NICD and extended period of the clock gene oscillations . This raises the compelling prospect that these effects are causally linked . Recent data showing that Notch1 protein is dynamically and periodically expressed in the chick and mouse PSM in a Notch dependent fashion ( Bone et al . , 2014 ) demonstrate that Notch1/NICD is itself part of a positive feedback loop . We experimentally validate the prediction that under conditions where half-life of NICD is increased , the period of the clock is extended . Our data thus expand upon previous models by highlighting the necessity for an added level of regulation , namely the strict regulation of the half-life for the positive input signal , NICD . Hence , it appears equally important to balance levels of both positive and negative input to correctly regulate the periodicity of clock gene oscillations . This result is similar to observations made in Nrarp-null mice , which display increased NICD levels in the PSM and a decrease in somite and vertebral body number although the pace of clock gene oscillations and somite size was not assessed ( Kim et al . , 2011 ) . Our data with the neddylation inhibitor MLN4924 further substantiate the hypothesis that extended NICD half-life is causally linked to the extended clock period , given that this reagent , which acts via a completely distinct molecular mechanism , nevertheless leads to the same phenotypic output as the Wnt and CDK inhibitors . Thus , in addition to the expected higher levels of NICD in the PSM of MLN4924 treated explants , we saw a corresponding increase in the period of Lfng oscillations , as seen for XAV939 , Roscovitine , DRB and PHA-767491 treated explants . In addition to elevating levels of NICD , exposure to MLN4924 elevates levels of phosphorylated β-catenin within PSM lysates . Since we show that Wnt inhibition can lead to elevated NICD levels and a delay to clock oscillations , this raises the possibility that the effects of MLN4924 are mediated via Wnt inhibition rather than directly via NICD turnover . However , we propose this is unlikely since in the absence of a degradation pathway the elevated levels of phosphorylated β catenin may in fact continue to participate in target gene transcription . Indeed , APC mutant lines which allow β catenin phosphorylation but are defective in ubiquitination are oncogenic , consistent with the idea that phosphorylated β catenin can indeed activate target genes ( Yang et al . , 2006 ) . Thus , we propose that the extended period of the clock seen following exposure to MLN4924 is due to the elevated levels of NICD brought about via inhibition of NICD degradation . Exposure to higher concentrations of this inhibitor leads to correspondingly higher levels of NICD in PSM explants and this accompanies a loss of dynamic Lfng expression . A similar correlation is seen in PSM explants exposed to increasing concentrations of Roscovitine , DRB , PHA-767491 . It has previously been shown that overexpression or complete loss of Notch signalling abolishes clock gene oscillations and causes severe segmentation defects ( Niwa et al . , 2007; Feller et al . , 2008; Ferjentsik et al . , 2009 ) . These data , together with previous reports investigating the effect of modulating protein levels of key clock components ( Hirata et al . , 2004; Kim et al . , 2011 ) , indicate that there is a tolerance window for the degree of variance in levels of positive and negative input which will still allow for oscillatory gene expression . These experimental findings are consistent with the predictions of our model which also clearly predict that there is a window within which variations in half-life of both Hes7 and NICD will allow for oscillations . Taken together , our computational and experimental data show that slowing , as opposed to stopping , of the segmentation clock gene oscillations may be brought about through transient and/or partial stabilisation of Notch activity . The rescue of both phase pattern changes and NICD levels elicited by exposure to both low concentrations of LY411575 together with Roscovitine , DRB or XAV939 provides compelling evidence that increased half-life and higher levels of NICD brought about by these inhibitors underlies the delay to clock gene oscillations ( Figure 9 ) . Following exposure to Roscovitine , DRB , PHA-767491 and XAV939 , the Western Blot data reveal an additional band for NICD at a slightly lower molecular weight in some of the inhibitor treated samples as compared to the control treated samples . This observation is characteristic of an increase in abundance of a variant of a protein which lacks small post-translational modifications such as phosphate groups . This shift may therefore represent the loss of phosphorylation marks which have been shown to be important for NICD regulation and degradation ( Fryer et al . , 2004; Jin et al . , 2009 ) . These data support the idea that post translational modification events may regulate the stability of NICD that could then directly affect the pace of the segmentation clock . As alluded to above , post translational modifications may underlie the effects of the 4 reagents that each modulate NICD half-life and clock period . We demonstrate that XAV939 treatment elevates levels of GSK3B-mediated phosphorylated β catenin in the PSM and it is possible that it might equally facilitate GSK3B mediated phosphorylation of NICD which has previously been reported to affect NICD stability ( Espinosa et al . , 2003; Jin et al . , 2009 ) . It is noteworthy that the effect of this phosphorylation event on NICD stability is likely to be context dependent since different groups have reported this modification can increase or decrease NICD stability and it is not known if any of these interactions occur in the PSM ( Foltz et al . , 2002; Espinosa et al . , 2003; Jin et al . , 2009 ) . Similarly , it is possible that the CDK inhibitors modulate CDK-mediated phosphorylation of NICD which has also been reported to affect NICD stability ( Fryer et al . , 2004 ) . Interestingly , exposure to specific CDK7 , CDK8 or CDK9 inhibitors did not yield the delayed oscillation phenotype or increased levels of NICD and half-life of NICD in the chick PSM ( data not shown ) . These data are somewhat surprising given that there are published reports describing a role for CDK8 in phosphorylating and thereby destabilising NICD . Indeed , CDK8-mediated phosphorylation in the PEST domain of NICD is required to target this protein for ubiquitination and subsequent degradation by the E3 ligase FBW7 ( Fryer et al . , 2004 ) . Whilst it is possible that this mode of regulation of NICD stability by this specific CDK does not occur in the PSM , we acknowledge that it is also possible that the effect on pace and cNICD stability are only made apparent when inhibiting a combination of these activities as seen with Roscovitine , DRB and PHA-76749 . It is important to note that an alternative explanation for the lack of effect of specific CDK7 , CDK8 , or CDK9 inhibitors is that the effect of Roscovitine , DRB and PHA-767491 may be through inhibition of a different set of kinases . The precise identification of the kinase ( s ) involved in destabilising cNICD will require further investigation . It is also formally possible that the common outcome of exposure to these four reagents to influence the pace of the clock involves regulation at the transcriptional level . Roscovitine is known to have highest efficacy against CDK2/CyclinE , CDK5 , CDK7/CyclinH and CDK9/CyclinT ( McLue et al . , 2002 ) whereas DRB and PHA-767491 reduce the kinase activities of CDK7/CyclinH and CDK9/CyclinT towards Serine 2 and Serine 5 on the C-terminal domain ( CTD ) of RNA pol II . Exposure to DRB , Roscovitine , or PHA-767491 leads to a reduction in Ser5 RNA pol II phosphorylation ( Figure 2—figure supplement 1B ) which is required to promote initiation of transcription as RNA pol II moves away from the gene promoter ( Kormarnitsky et al . , 2000 ) but it is also associated with ‘poised’ genes bound by both Ser5 phosphorylated RNA pol II and Polycomb repressor complexes ( Brookes and Pombo , 2012 ) . Thus , one possibility is that reduced Ser5 phosphorylation of RNA pol II may leave some ‘poised’ genes in the PSM in a more repressed state . Importantly , certain gene transcripts are more susceptible to DRB treatment than others ( Zandomeni and Weinmann , 1984; Chodosh et al . , 1989; Wada et al . , 1998 ) and indeed we find not all gene transcripts in the PSM are down-regulated by these inhibitors ( e . g . , cLfng and cHairy2 expression is delayed in the PSM but not down-regulated ) . Thus , the effects of these inhibitors may be to modify certain ‘poised’ genes to a more repressed state that specifically include genes encoding factors required for destabilising NICD . Given that XAV939 elicits the same phenotypic output as exposure to the CDK inhibitors it is conceivable that all four reagents affect the transcription of some common genes; some involved in regulating the clock while others not; indeed we identified one gene commonly affected by all 4 reagents namely Gli1 ( known to be regulated by Wnt signalling in the neural tissue and somites ( Munsterberg et al . , 1995 ) , Figure 2—figure supplement 1E–G ) . For this particular gene we show it is unlikely to be involved in clock regulation ( Figure 3 ) . However , a transcriptomic analysis of common target genes would allow this possibility to be fully explored to identify genes co-regulated by these reagents which potentially encode regulators of NICD stability in the PSM . Finally , a recent report has beautifully demonstrated that in zebrafish embryos the rate of tissue shortening , in addition to the periodicity of clock gene oscillations , is important in determining the periodicity of segmentation ( Soroldoni et al . , 2014 ) . Further analyses are required to investigate if this is a conserved feature in all vertebrate species . That analysis notwithstanding , our data nevertheless clearly show that the pace of the segmentation clock itself has a strong influence upon the pace of somite formation since significantly fewer somites are formed within a given time-frame when the pace of clock gene oscillations and NICD stability are affected . In conclusion , our data suggest that the stability and turnover of NICD is inextricably linked to the regulation of the pace of clock gene expression across the PSM . | During embryo development , animals with backbones ( also called vertebrates ) repeatedly lay down pairs of segments along the axis that runs from the head to the tail of the embryo . These segments , known as somites , eventually form part of the skeleton , as well as the associated muscle , cartilage , tendons and some skin . Importantly , the segments in some species take longer to form than those in other species , and they also form in proportion to the overall size of the animal . A ‘segmentation clock’ regulates the timing of somite formation via cycles in which genes are repeatedly switched on and then off again . Some aspects of this process are well understood . Firstly , many ‘clock genes’ are known to produce proteins that can inhibit their own production . However , this ‘negative feedback’ is typically delayed because it takes time to produce and transport protein within a cell . The inhibitory proteins are also unstable and their breakdown leads to an end of their inhibitiory effect . It is also known that: some proteins send signals to neighbouring cells while others , including one called Notch , receive them; and the received signals activate the expression of clock genes . However , until now , no one had studied how the turnover ( that is , the production and breakdown ) of the proteins that activate clock gene expression could regulate the pace of the clock . Wiedermann , Bone et al . used a two-pronged approach to investigate this question . First , they developed a computational model that accounted for both inhibition and activation of clock gene expression . The model predicts that the clock slows down when the levels of a positive regulator called Notch intracellular domain ( or NICD for short ) are high . This is because the negative regulators would have to overcome the increased positive regulators to switch off the clock genes . A slower segmentation clock would be expected to give rise to fewer , larger somites in a given length of time when compared to a similar clock with a faster pace . To test these predictions , Wiedermann , Bone et al . next conducted experiments on chicken embryos , which are commonly used in studies of animal development . The experiments agreed with the model predictions . That is , when treated with a variety of drugs that affected NICD turnover and thereby increased the levels of NICD , the clock slowed and these chicken embryos developed fewer , but larger somites . As predicted by the mathematical model , these effects were rescued when Wiedermann , Bone et al . reduced the production of NICD . These findings show that a balance of positive and negative regulators determines the pace of the segmentation clock . | [
"Abstract",
"Introduction",
"Results",
"Discussion"
] | [
"developmental",
"biology",
"computational",
"and",
"systems",
"biology"
] | 2015 | A balance of positive and negative regulators determines the pace of the segmentation clock |
Homeostatic renewal of many adult tissues requires balanced self-renewal and differentiation of local stem cells , but the underlying mechanisms are poorly understood . Here we identified a novel feedback mechanism in controlling intestinal regeneration and tumorigenesis in Drosophila . Sox21a , a group B Sox protein , is preferentially expressed in the committed progenitor named enteroblast ( EB ) to promote enterocyte differentiation . In Sox21a mutants , EBs do not divide , but cannot differentiate properly and have increased expression of mitogens , which then act as paracrine signals to promote intestinal stem cell ( ISC ) proliferation . This leads to a feedback amplification loop for rapid production of differentiation-defective EBs and tumorigenesis . Notably , in normal intestine following damage , Sox21a is temporally downregulated in EBs to allow the activation of the ISC-EB amplification loop for epithelial repair . We propose that executing a feedback amplification loop between stem cells and their progeny could be a common mechanism underlying tissue regeneration and tumorigenesis .
Adult stem cells have important roles in maintaining tissue and organ homeostasis by their prolonged ability to produce progenitor cells that differentiate into multiple types of mature cells . Production of the progenitor cells from stem cells must be coordinated with cell differentiation and the overall tissue demand , as disruption of this coordination could lead to tissue degeneration , if cell production is not sufficient , or hyperplasia/ tumorigenesis , if the cell production is unrestricted and exceeds the pace of cell differentiation . However , the molecular mechanism that coordinates progenitor cell proliferation with cell differentiation is largely unknown . The adult Drosophila midgut has been established as a simple and useful system for the study of the stem cell behavior during homeostatic tissue renewal and in response to environmental changes ( Biteau et al . , 2011; Jiang and Edgar , 2012 ) . Like mammalian intestine , the Drosophila midgut epithelium is constantly replenished by adult intestinal stem cells ( ISCs ) ( Micchelli and Perrimon , 2006; Ohlstein and Spradling , 2006 ) , although at a relatively slower pace . In addition , signaling pathways that regulate mammalian ISC activity , such as Wnt , JAK/STAT , EGFR/Ras , Hippo , BMP and Notch , also play important roles in regulating Drosophila ISC activity during normal homeostasis and/or stress conditions ( reviewed by ) ( Biteau et al . , 2011; Pasco et al . , 2015 ) . The Drosophila ISC , which generates a relatively simple stem cell lineage , can be specifically marked by Delta ( Dl ) , the Notch ligand . After each asymmetric division , an ISC will produce a new ISC and a committed progenitor cell named enteroblast ( EB ) , which will further differentiate into either an enterocyte or an enteroendocrine cell , depending on the levels of Notch activation it received from ISCs ( Ohlstein and Spradling , 2007 ) . Enterocyte differentiation from EB requires high levels of Notch activation , and JAK/STAT signaling activity is required for both enterocyte and enteroendocrine cell differentiation from EB ( Beebe et al . , 2010; Jiang et al . , 2009; Lin et al . , 2010 ) . Aside from the signaling pathways , many transcription factors have been identified as important regulators of cell differentiation . Enterocyte differentiation from EB requires downregulation of Escargot ( Esg ) and activation of Pdm1 ( Korzelius et al . , 2014; Loza-Coll et al . , 2014 ) , whereas enteroendocrine cell differentiation from EB requires release of the inhibition by the transcriptional repressor Tramtrack and activation of acheate-scute complex ( AS-C ) genes and Prospero ( Pros ) , the enteroendocrine cell determination factor ( Bardin et al . , 2010; Wang et al . , 2015; Zeng and Hou , 2015 ) . It is largely unclear how these signaling pathways and transcription factors are coordinately regulated for balanced self-renewal of ISCs and differentiation of EBs to maintain intestinal homeostasis . Sox family transcription factors , which share a DNA binding high-mobility-group domain , are known as important regulators of cell fate decisions during development and in adult tissue homeostasis ( Kamachi and Kondoh , 2013; Sarkar and Hochedlinger , 2013 ) . In mouse small intestine , Sox2 is expressed in ISCs and progenitor cells and is critical for ISC maintenance and differentiation of Paneth cells ( Furuyama et al . , 2011; Sato et al . , 2011 ) . Several Sox family proteins have been identified in Drosophila ( McKimmie et al . , 2005 ) , but their potential roles in the ISC lineage are unclear . Here we characterized the function of a Drosophila Sox gene , Sox21a , in the ISC lineage . We find that Sox21a is expressed in EBs and acts as a tumor suppressor in the midgut epithelium . One important function of Sox21a is to promote enterocyte differentiation by inducing Pdm1 expression . By studying its tumor suppressing function , we identified a novel feedback amplification loop between ISC and EB , which is normally suppressed by Sox21a . Temporal activation of this loop is essential for damage-induced intestinal regeneration , whereas sustained activation of this loop leads to tumorigenesis . Therefore , we revealed a novel mechanism that coordinates stem cell activity with progenitor cell differentiation , and connects regeneration with tumorigenesis .
From cell-type specific gene expression analysis of Drosophila midgut cells ( Dutta et al . , 2015 and unpublished data ) , we noticed a Sox family gene , Sox21a , whose RNA expression was detected only in ISCs and EBs . Interestingly , Sox21a seems to be mainly expressed in midgut , but not other organs in larva and adult Drosophila ( Chintapalli et al . , 2007 ) . To characterize its expression pattern in vivo , we first generated polyclonal antibodies against Sox21a , and demonstrated that this antisera could specifically mark Sox21a antigen in the midgut epithelium ( Figure 1E and Figure 1—figure supplement 1 ) . Immunostaining of the wild type midgut with this antisera revealed that Sox21a was largely undetectable in the midgut of newly eclosed and young flies of two to three days old ( Figure 1A ) . Its expression began to appear with age and at 4–5 days old , weak Sox21a expression appeared specifically in Dl+ ISCs and Notch-activated EBs that can be marked by a Notch activation reporter , Su ( H ) Gbe>GFP ( NRE>GFP ) ( Figure 1B ) . At 7 days old , its expression could also be detected in early ECs , which display increased cell ploidy ( Figure 1C–D ) . The Dl+ ISC and its immediate daughter EB ( marked by NRE>GFP ) are usually adjacent to each other , forming an ISC-EB pair ( Ohlstein and Spradling , 2007 ) . In each ISC-EB pair , the level of Sox21a expression was usually higher in EB than in ISC ( Figure 1C , F ) . In each progenitor cell nest where the early EC still retained NRE>GFP expression , the early EC usually displayed a higher Sox21a expression level than the EB or the ISC in the same ISC-EB-early-EC cell nest ( Figure 1D , F ) , suggesting that Sox21a is up-regulated in differentiating EBs . Consistent with this notion , A GFP reporter driven by an enhancer Gal4 line for Sox21a ( GMR43E09-Gal4 ) showed specific expression of GFP in ISCs and EBs . Again , the intensity of GFP was much higher in EBs than in ISCs ( Figure 1G ) . Previous studies suggest that EB differentiation requires mesenchymal-epithelial transition through downregulation of Esg ( Antonello et al . , 2015; Korzelius et al . , 2014 ) . Interestingly , in fixed midgut where Sox21a was not expressed , EBs commonly displayed triangle-shaped morphology ( Figure 1A ) . We propose that these EBs are probably the dormant EBs that remain in the undifferentiated state ( Figure 1H ) . By contrast , in EBs where Sox21a began to be expressed , EBs instead displayed oval-shaped morphology ( Figure 1C , D ) . These EBs are likely the activated EBs that are undergoing mesenchymal-epithelial transition ( Figure 1H ) . Taken together , these observations suggest that Sox21a is preferentially expressed in differentiating EBs in the intestinal epithelium . This unique expression pattern indicates that Sox21a might have a role during the process of EB differentiation into enterocyte . 10 . 7554/eLife . 14330 . 003Figure 1 . Sox21a is preferentially expressed in differentiating EBs . ( A–D ) Sox21a ( red ) expression in midguts of 2 to 7 days old females co-stained with Dl ( white ) and NRE>GFP ( green ) . Separate color channels for the insets were enlarged and displayed on the right side of each corresponding images . ISCs and EBs were indicated by white or yellow arrowheads , respectively . Sox21a expression was largely undetectable in the midgut of newly eclosed females ( A ) Weak Sox21a expression was detectable in ISCs and EBs in midgut of 5 days old females ( B ) . Sox21a expression increased with age , but EB showed a relatively higher expression level than ISC ( C ) . In each ISC-EB-early-EC cell nests , the early EC displayed the highest expression level than the EB or the ISC ( D ) . ( E ) As a control , no signal was detected by anti-Sox21a staining ( red ) in Sox21a null midgut . ( F ) The relative fluorescence intensity of Sox21a expression in each ISC-EB pairs and ISC-EB-Early EC cell nests ( also see method ) . Error bars represent s . e . m . n is as indicated . ** denotes student’s t test p<0 . 01 . ( G ) Expression of GMR43E09>GFP ( green ) and Dl ( red ) in midgut of 5-day-old females . GFP levels were higher in EB ( yellow arrowhead ) than in adjacent ISC ( black arrowhead ) . ( H ) A schematic model for dynamic Sox21a expression pattern: the Sox21a expression level may distinguish different state of progenitors . Sox21a expression is at the minimum in differentiation-arrested EBs which typically display a triangle shape with cell protrusions . We refer to this state of EBs as 'dormant' . Sox21a expression is elevated in EBs when EBs begin to differentiate and display oval-shaped morphology . We refer to this state of EBs as 'active' . Sox21a expression reaches its highest level in differentiating EC , and then gradually disappears as EC matures . Scale bars in A , B , 10 μm; in E , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14330 . 00310 . 7554/eLife . 14330 . 004Figure 1—figure supplement 1 . The Sox21a antibody is highly specific . ( A ) Staining with anti-Sox21a antibody ( red ) in Sox21a mutant clones ( green , dashed lines ) . The signal was always absent in Sox21a mutant clones . ( B ) Staining with anti-Sox21a antibody ( red ) in Sox21a overexpression ( OE ) clones ( green , dashed lines ) . The signal was greatly upregulated in Sox21a OE clones . Scale bars: 10 μmDOI: http://dx . doi . org/10 . 7554/eLife . 14330 . 004 To study the function of Sox21a , we generated two gene deletion lines using the CRISPR/Cas9 system , and named these two alleles as Sox21a [JC1] and Sox21a [JC2] , which carries approximately 2 . 15 kb and 2 . 65 kb DNA fragment deletion , respectively , in the gene locus ( Figure 2A ) . Because the coding region of Sox21a is almost entirely deleted in both alleles , they can be regarded as null alleles for Sox21a . Consistent with the midgut-restricted expression during development , Sox21a homozygous mutant flies are viable and fertile . Staining midgut of newly eclosed mutants showed that the general gut morphology and the cellular makeup of the monolayer midgut epithelium was largely normal , although the epithelial cells appeared to be more sparsely distributed in the epithelium compared to the wild type intestinal epithelium ( Figure 2B ) . This indicates that Sox21a could have a role in regulating the proliferation of midgut progenitor cells during development . Strikingly , when the mutant flies got older , multilayered cell clusters composed of 50 or more cells ( termed tumors hereafter ) were frequently observed in both anterior and posterior midgut , and tumor incidence steadily increased with age ( Figure 2C , D ) . At day 6 after eclosion , approximately 49 . 1% and 57 . 9% of [JC1] and [JC2] homozygous female flies had at least one tumor in the midgut , respectively , and at day 11 , approximately 77 . 3% and 94 . 1% of [JC1] and [JC2] flies had gut tumors , respectively ( Figure 2D ) . Tumor size also grew with time , and in aged mutants , the tumor masses often filled up the entire lumen space , suggesting that tumors grow rapidly once it is initiated . 10 . 7554/eLife . 14330 . 005Figure 2 . Sox21a mutation causes the development of intestinal tumors composed of differentiation-defective cells . ( A ) A diagram describing the molecular lesions of two Sox21a mutant alleles . Both alleles carry a large DNA fragment deletion spanning the coding region of Sox21a . ( B–C ) Cross-section of midgut from 2 day old ( B ) and 7 day old ( C ) Sox21a mutant flies . With age , Sox21a mutants developed tumors with multilayered structure ( C ) . ( D ) Quantitative analysis of the tumor incidence at different regions of midgut with age . Genotypes of flies analyzed: wild type ( WT ) , Sox21a-/- [JC1] and Sox21a-/- [JC2] . n = 40–50 guts . ( E–G’ ) Staining of NRE-lacZ ( green ) and Dl + Pros ( red ) in the midgut of WT ( E ) , Sox21a mutant flies at day 2 ( F ) and day 7 ( G , G’ ) after eclosion . Compared to WT guts , Sox21a mutant guts did not show EB accumulation at day 2 ( E , F ) , but showed dramatic EB accumulation at day 7 ( G , G’ ) . Accumulated lacZ+ cells were negative for Dl or Pros expression ( G’ ) . ( H , H’ ) Staining of NRE-lacZ ( red ) in Sox21a mutant clones on day 7 after clone induction . LacZ was cell-autonomously activated . Non-cell autonomous LacZ+ cell clusters were also observed ( see text ) . ( I–I’ ) Staining of Dl ( white ) , Pros ( red ) in Sox21a mutant clones . The Dl+ cells were scatteredly distributed in the clones , and Pros+ cell was rarely found within the mutant clones . ( J , J’ ) Staining of Pdm1 ( red ) in Sox21a mutant clones . . Pdm1 expression was absent in the entire mutant clones . Scale bars: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14330 . 005 To study the cellular outcome of the mutant cells , we examined the mutant midgut with several cell fate markers at day 2 and day 7 after eclosion . Like all other cells , the Su ( H ) Gbe-lacZ ( NRE-lacZ ) + EBs are more sparsely distributed in the epithelium of the mutant intestine than of the wild type intestine at day 2 after eclosion ( Figure 2E , F ) . However , the NRE-lacZ+ cells rapidly accumulated in the epithelium of the mutant intestine at day 7 ( Figure 2G , G’ ) . Thus , Sox21a mutants postnatally develop intestinal tumors . Co-staining Dl with NRE-lacZ revealed that ISCs and EBs were intermingled with each other in EB-accumulated regions ( Figure 2G , G’ ) . Those observations suggest that the tumors in Sox21a mutant intestine are mainly composed of ISCs and EBs , and indicate that Sox21a might be required for EB differentiation . To test the differentiation capability of Sox21a mutant cells , we generated Sox21a homozygous mutant clones by mitotic recombination using the MARCM system . As expected , the mutant clones cell-autonomously gave rise to intestinal tumors which was mainly composed of NRE-lacZ+ cells intermingled with Dl+ cells ( Figure 2H , H’ ) . We also observed clusters of NRE-lacZ+ cells outside of the clones ( data not shown ) , which could be due to a non-cell autonomous effect by the mutant clones . There was also mild increase of Dl+ cells that were scatteredly distributed within the mutant clones ( Figure 2I , I’ ) . Rarely but occasionally Pros+ cells could be found in the mutant clones ( Figure 2I and data not shown ) , indicating that Sox21a is not absolutely required for enteroendocrine cell differentiation . Polyploid cells were also found in the clones but their size was relatively smaller than ECs ( Figure 2I–J ) . Interestingly , Pdm1 , an enterocyte marker ( Lee et al . , 2009 ) , failed to be expressed in Sox21a mutant cells in a cell-autonomous manner ( Figure 2I , I’ ) . These observations suggest that Sox21a is cell-autonomously required for EB differentiation into enterocyte . The above results demonstrate that Sox21a is required for the expression of Pdm1 . Because Pdm1 expression is sufficient to induce differentiation of progenitor cell into enterocyte ( Korzelius et al . , 2014 ) , Sox21a may function to promote enterocyte differentiation by inducing Pdm1 expression . To test this hypothesis , we generated a UAS-Sox21a transgene and forcibly expressed it in progenitor cells using esgGal4 , an ISC- and EB-specific Gal4 driver , and examined the consequences . Tub-Gal80ts transgene was also added to the system to allow temporal control of transgene expression . Under normal conditions , virtually all Esg>GFP+ cells were negative for Pdm1 expression ( Figure 3A , A’ ) . However , when Sox21a was forcibly expressed for 5 days , approximately 34 . 2 ± 1 . 6% of Esg>GFP+ cells became positive for Pdm1 expression ( Figure 3B , B’ , E ) . The Pdm1+ cells were always one of the GFP+ ISC-EB pairs , indicating that expression of Sox21a could not induce Pdm1 expression in ISCs , but EBs ( Figure 3B , B’ ) . To test this hypothesis , we forcibly expressed Sox21a specifically in EBs using the Su ( H ) Gbe-Gal4 ( NRE-Gal4 ) driver . Indeed , transient induction of Sox21 expression for 5 days induced approximately 72 . 3 ± 3 . 8% of NRE>GFP+ cells to turn on Pdm1 expression ( Figure 3D , D’ , E ) , and these cells began to show polyploidy , indicating that they are in the process of enterocyte differentiation ( Figure 3D , D’ ) . Next , we generated MARCM clones in which all GFP+ cells had constitutive Sox21a overexpression . The ISC-containing clones grew normally in size and were able to produce both polyploidy enterocytes and Pros+ enteroendocrine cells ( Figure 3F ) , suggesting that forced Sox21a expression does not inhibit ISC activity and does not block enteroendocrine cell differentiation . Taken together , these data suggest that Sox21a overexpression is not sufficient to induce differentiation of ISCs , but is able to induce precocious differentiation of EBs by inducing Pdm1 expression . 10 . 7554/eLife . 14330 . 006Figure 3 . Sox21a functions to promote EB differentiation into EC . ( A–B’ ) Expression of Esg>GFP ( green ) and Pdm1 ( red ) in control ( A , A’ ) and Sox21a overexpressed guts ( B , B’ ) . Conditional expression of Sox21a in Esg+ cells for 7 days induced a part of Esg+ cells to turn on Pdm1 expression . ( C–D’ ) Expression of NRE>GFP ( green ) and Pdm1 ( red ) in control ( C , C’ ) and Sox21a overexpressed guts ( D , D’ ) . Conditional expression of Sox21a in NRE+ cells for 5 days induced most of the NRE+ cells to turn on Pdm1 expression . ( E ) Quantification of the percentage of Pdm1+ cells in GFP+ cells . Error bars represent s . e . m . n is as indicated . *** denotes student’s t test p<0 . 001 . ( F ) Staining of Dl ( white ) and Pros ( red ) in MARCM clones with Sox21a overexpression ( Sox21a OE ) on day 5 after clone induction . Dl+ ISCs and Pros+ enteroendocrine cells could be detected in Sox21a OE clones . Scale bars in ( A–D ) , 20 μm; in F , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14330 . 006 The requirement for Sox21a in EB differentiation explains the EB accumulation phenotype in Sox21a mutant intestine . Rapid EB-like tumor development indicates that the mutant EBs might have re-entered the cell cycle to replicate themselves , although normally Notch-activated EBs are non-mitotic and primed for enterocyte differentiation . BrdU incorporation assay revealed that the tumor cells had strikingly high ability to incorporate BrdU ( Figure 4A , B ) . As expected , the mutant intestine had exponentially increased number of mitotic cells with age ( Figure 4C ) . In addition , significantly increased number of mitotic cells was found in the tumors compared to non-tumor areas ( not shown ) . These observations indicate that the increased mitotic activity drives tumor growth . To determine the identity of mitotic cells in Sox21 mutant tumors , we co-stained the mitotic marker phospho-histone 3 ( PH3 ) with various cell fate markers . Surprisingly , virtually all mitotic cells in the tumor were NRE-lacZ- ( 353 out of 354 ) ( Figure 4D–E ) . Thus , the proliferative cells that drive tumor growth are not EBs but ISCs . 10 . 7554/eLife . 14330 . 007Figure 4 . Sox21a mutant EBs are non-mitotic but are tumor-initiating cells . ( A–B’ ) BrdU ( green ) incorporation assay in WT control and Sox21a-/- gut . Compared to WT gut , Sox21a mutant gut showed increased BrdU incorporation in tumor regions . ( C ) Mitotic index of WT and Sox21a-/- midgut at day 5 and day 10 after eclosion . Error bars represent s . e . m . n = 40–50 guts . * denotes student’s t test p<0 . 05 . *** denotes student’s t test p<0 . 001 . ( D–E ) pH3 ( red ) staining in the tumor region of Sox21a mutant midgut ( D , D’ ) . Virtually all pH3+ cells were NRE>GFP- cells ( D’ , E ) . ( F , G ) Sox21a-RNAi driven by NRE-Gal4ts led to tumorous accumulation of GFP+ EBs . ( H ) Quantitative analysis of tumor incidence in the midgut of WT control , Sox21a-/- , Sox21a-/- ; NRE>Sox21a and Sox21a-/- ; Dl>Sox21a flies . Expression of Sox21a transgene in EB , but not in ISC , could effectively suppress tumor development in Sox21a mutant midgut . Total number of guts examined is as indicated . Scale bars: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14330 . 007 One possible explanation for the increased ISC proliferation in the tumor is that , in addition to a function in EB differentiation , Sox21a also functions in ISCs to inhibit cell proliferation . However , Specifically knocking down Sox21a by RNAi ( HMJ21395 or JF02191 ) in EBs using Su ( H ) Gbe-Gal4 , UAS-GFP; Tub-Gal80ts ( referred to as NRE-Gal4ts ) was sufficient to induce rapid accumulation of EB-like cells in the intestinal epithelium ( Figure 4F , G ) , a phenotype largely similar to that observed in Sox21a mutant intestine . Therefore , loss of Sox21a in EB is sufficient to induce tumor development , indicating that Sox21a mutant EBs are tumor-initiating cells , although they are non-mitotic cells . If the non-mitotic EBs are indeed tumor-initiating cells in Sox21a mutants , we would expect to see that EB-specific transgene expression of Sox21a should be able to prevent tumor development in Sox21a mutant intestine . We therefore generated Sox21a mutant flies carrying NRE-GAL4ts and UAS-Sox21a transgenes , in which the expression and therefore the function of Sox21a will be restricted in EBs in the intestinal epithelium . In another experiment , we also generated Sox21a mutant flies carrying Dl-GAL4ts and UAS-Sox21a transgenes , in which the expression and therefore the function of Sox21a will be restricted in ISCs in the intestinal epithelium . Indeed , EB-specific expression of Sox21a largely suppressed intestinal tumor formation in both anterior and posterior midgut of Sox21a mutants ( Figure 4H ) . By contrast , ISC-specific expression of Sox21a failed to suppress intestinal tumor formation in Sox21a mutants ( Figure 4H ) . Taken together , these data suggest that EB is the tumor-initiating cell that is necessary and sufficient for intestinal tumor development in Sox21a mutant intestine . Because Sox21a mutant EBs are non-mitotic yet are responsible for tumor initiation , the mutant EBs must non-cell autonomously induce ISC proliferation to promote tumor formation . To understand the underlying mechanisms , we compared the transcriptome of Esg+ progenitor cells with and without Sox21a depletion . Multiple secreted growth factors were found to be up-regulated by two fold or more in Sox21a-depleted cells , including Spitz ( Spi ) , a ligand for EGFR/Ras/MAPK signaling , Upd3 , a ligand for JAK/STAT signaling , and Pvf2 and Pvf1 , which also induces Ras/MAPK signaling via PVR ( Figure 5A ) . Interestingly , these factors all function as positive regulators of ISC proliferation during normal homeostasis and/ or in response to injury or infection ( Beebe et al . , 2010; Biteau et al . , 2011; Bond and Foley , 2012; Buchon et al . , 2009; 2010; Choi et al . , 2008; Jiang et al . , 2009; 2011; Lin et al . , 2010; Xu et al . , 2011 ) . In addition , increased Spi signaling promotes tumor development from Notch mutant ISCs ( Patel et al . , 2015 ) , and increased Upd3 signaling in EBs following the disruption of the Misshapen-Warts-Yorkie pathway also induce ISC proliferation and intestinal hyperplasia ( Li et al . , 2014 ) . This raises an interesting hypothesis that Sox21a mutant EBs may induce ISC proliferation via paracrine Ras/MAPK and/or JAK/STAT signaling . In vivo validation of gene expression was subsequently conducted with available enhancer trap lines . Spi-lacZ is an enhancer trap line for Spi , and its lacZ expression was barely detectable in the wild type midgut epithelium , including ISCs and EBs , probably due to low levels of baseline expression in healthy midgut ( Figure 5B , B’ ) . However , lacZ expression became readily detectable in EBs following Sox21a depletion by RNAi ( Figure 5C , C’ ) . By contrast , we did not observe significant upregulation of Upd3-lacZ , a reporter for Upd3 expression ( Figure 5—figure supplement 1 ) . Interestingly , staining with phospho-ERK ( pERK ) , a direct readout of MAPK signaling activity , revealed that the pERK signal was strongly enhanced in the diploid cells adjacent to the EB cells with Sox21a depletion ( Figure 5D , E ) . Importantly , the increase of pERK activity in Sox21a mutant epithelium was completely suppressed when Sox21a was re-expressed in EBs ( Figure 5F , F’ ) , suggesting that Sox21a depletion in EB is both sufficient and necessary for enhanced pERK signaling in Sox21a mutant intestine . We found that conditional overexpression of Spi in EBs also produced a similar EB-like tumor phenotype , indicating that Spi upregulation could be one of the reasons responsible for the enhanced pERK signaling in ISCs ( Figure 5G , G’ ) . Consistent with this hypothesis , EB-specific knockdown of spi significantly reduced , although not completely eliminated , the ISC overproliferation phenotype in Sox21a mutant intestine ( Figure 5H ) . As a comparison , EB-specific expression of Sox21a transgene produced a much stronger inhibitory effect on ISC proliferation in Sox21a mutant intestine , and the mitotic index was reduced to wildtype levels ( Figure 5H ) . Taken together , these observations suggest that in Sox21a mutant intestine , loss of Sox21a in EBs causes derepression of multiple mitogens , including Spi , and others , such as Pvf2 and Pvf1 , that all act as paracrine signals to induce Ras/MAPK signaling activity in ISCs and promote ISC proliferation . As a result , more mutant EBs are produced , which further propel ISC proliferation . This leads to the establishment of a positive amplification loop between ISCs and their progeny , which eventually leads to uncontrolled production of differentiation defective EBs from ISCs and consequently tumorigenesis . 10 . 7554/eLife . 14330 . 008Figure 5 . Spi/MAPK signaling is required for tumorigenesis driven by Sox21a mutant EBs . ( A ) A scatter plot shows the comparison of gene expression profiles of Esg-GFP+ cells in wild type and Sox21a-/- midgut . Green dots depict genes of receptor binding proteins with 2-fold and higher changes , and red dots depict genes of transcription factors with 2-fold and higher changes in the mutant midgut . ( B–C’ ) Spi-lacZ ( red ) and NRE>GFP ( green ) expression in control and NRE>Sox21a RNAi midguts of 7-day-old females . Spi-lacZ expression was undetectable in the wild type midgut ( B , B’ ) , but was elevated in EBs of NRE>Sox21a-RNAi midgut ( C , C’ ) . ( D–F ) pERK ( red ) staining in midguts of the following genotypes: NRE>GFP ( control , D , D’ ) , NRE>Sox21a RNAi ( E , E’ ) and NRE>Sox21a; Sox21a-/- ( F , F’ ) . EB- specific Sox21a-RNAi caused dramatic enhancement of pERK signal in the diploid cells adjacent to EBs , and EB-specific expression of Sox21a suppressed pERK upregulation in Sox21a mutant midgut . ( G , G’ ) Transgenic Spi expression driven by NRE-gal4ts led to tumorous accumulation of EBs . Those accumulated EBs were all negative for Dl expression ( G’ ) . ( H ) A plot showing that either EB- specific knockdown of spi or EB- specific expression of Sox21a transgene significantly reduced ISC overproliferation in Sox21a mutant intestine , and sox21a transgene expression had a much stronger effect . Error bars represent s . e . m . n is as indicated . * denotes student’s t test p<0 . 05 . ***p<0 . 001 . Scale bars: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14330 . 00810 . 7554/eLife . 14330 . 009Figure 5—figure supplement 1 . Upd3-lacZ was not significantly upregulated in Sox21a-depleted EBs . ( A ) Staining of Upd3-lacZ ( red ) in midgut of NRE-Gal4; UAS-GFP ( green ) . ( B ) Staining of Upd3-lacZ ( red ) in midgut of NRE-Gal4; UAS-GFP; UAS-Sox21a RNAi . GFP was in green . Upd3-lacZ was not obviously upregulated in Sox21a-depleted EBs . Weak upregulation of Upd3-lacZ was sometimes observed in the nearby GFP- cells . Scale bars: 10 μmDOI: http://dx . doi . org/10 . 7554/eLife . 14330 . 00910 . 7554/eLife . 14330 . 010Figure 5—figure supplement 2 . Sox21a was still expressed in JAK/STAT-compromised clones . Staining of Sox21a ( red ) in dome468 clones , in which the cells are defective in differentiation . Expression of Sox21a could still be detected in dome mutant cells . Scale bar: 10 μmDOI: http://dx . doi . org/10 . 7554/eLife . 14330 . 010 The regulatory role for Sox21a in Spi signaling indicates that Sox21a not only functions to promote EB differentiation , but may also provide a feedback mechanism from differentiating EBs for controlling ISC activity , and thereby intestinal homeostasis . We therefore tested whether the expression of Sox21a could be regulated upon tissue damage . Feeding flies with Dextran sulfate sodium ( DSS ) has been established as an effective mean to induce intestinal damage and inflammation , which then triggers ISC activation for accelerated epithelial repair ( Amcheslavsky et al . , 2009 ) . A relatively rapid damage and repair process was set for this experiment ( Figure 6A ) . Interestingly , following DSS treatment , Sox21a expression was rapidly downregulated in EBs ( Figure 6B , B’ , C , C’ , I ) . This downregulation was accompanied by massive EB accumulation in the epithelium resulted from increased ISC proliferation ( Figure 6C , C’ ) . However , during the recovery phase following the DSS treatment , Sox21a was dramatically upregulated in the accumulated EBs ( Figure 6D , D’ , I ) , and differentiating ECs ( Figure 6E , E’ ) . Interestingly , spi-lacZ showed exactly the reciprocal expression pattern to Sox21a during the damage and regeneration process , as its expression was activated during the DSS treatment , and diminished at the recovery phase ( Figure 6F–H” ) . We also examined the expression of the Sox21a expression reporter GMR43E09>GFP during the damage and repair process . Consistent with the changes in Sox21a expression , GMR43E09>GFP was transiently downregulated right after the damage and then upregulated 1 . 5–2 days after the damage ( Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 14330 . 011Figure 6 . Sox21a-Spi- mediated ISC-EB amplification loop participates in damage-induced intestinal regeneration . ( A ) The scheme of damage induction and recovery . ( B ) ( B–E’ ) Sox21a ( red ) expression in the midgut of flies fed with sucrose-soaked diet ( B , B’ ) and DSS-soaked diet ( C , C’ ) as well as the flies of 2 day recovery after DSS feeding ( D , D’ , E , E’ ) . Compared with sucrose-treated control gut ( B , B’ ) , DSS treatment showed rapid decline of Sox21a expression in EBs ( C , C’ ) . During the recovery phase , Sox21a was dramatically upregulated in EBs ( D , D’ ) and differentiating ECs ( E , E’ ) . ( F–H” ) Spi-lacZ ( red ) expression in the midgut of flies treated with sucrose ( F–F” ) or DSS ( G–G” ) as well as the flies at the recovery phase after DSS treatment ( H–H” ) . Compared with sucrose-treated midgut , in which Spi-lacZ expression was undetectable ( F–F” ) , DSS treatment induced Spi-lacZ expression ( G–G” ) . Spi-lacZ expression was shut down again at the recovery phase ( H–H” ) . ( I ) Fluorescence intensity of Sox21a expression in EBs relative to background in the midgut of flies fed with sucrose , DSS and flies at the recovery phase after DSS treatment . Sox21a expression in EB was virtually reduced to background levels in DSS-induced damage phase , and then massively upregulated during the recovery phase ( REC ) . Error bars represent s . e . m . n is as indicated . **** denotes student’s t test p<0 . 0001 ( J ) Quantification of pH3+ cells in midguts of indicated genotypes . EB- specific depletion of Spi could partially reduce DSS- induced mitosis , while EB-specific transgene expression of Sox21a could strongly inhibit DSS- induced mitosis . Error bars represent s . e . m . n is as indicated . * denotes student’s t test p<0 . 05 . ***p<0 . 001 . Scale bars: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14330 . 01110 . 7554/eLife . 14330 . 012Figure 6—figure supplement 1 . GMR43E09>GFP ( green ) expression in the midgut of flies treated with sucrose ( A–A’ ) or DSS ( B , B’ ) as well as flies at the recovery phase after DSS treatment ( C–C’ ) . Compared with sucrose-treated midgut ( A , A’ ) , DSS treatment caused rapid downregulation of GFP expression in EBs ( B , B’ ) . A temporal elevation of GFP expression was observed in the intestinal epithelium during the recovery phase ( C , C’ ) . Scale bar: 20 μmDOI: http://dx . doi . org/10 . 7554/eLife . 14330 . 012 These observations indicate that the Sox21a-Spi regulatory pathway could mediate damage-induced intestinal regeneration . To functionally test this hypothesis , we asked whether reversing the expression of Sox21a or Spi in EBs could affect intestinal regeneration following damage . Specific depletion of Spi in EBs significantly reduced the proliferative response of ISCs following DSS treatment ( Figure 6J ) . On the other hand , overexpression of Sox21a in EBs almost completely eliminated the proliferative response of ISCs ( Figure 6J ) . These results suggest that dynamic regulation of the Sox21-Spi pathway is critical for effective intestinal regeneration after damage . Compared to the effect by EB>Sox21a , incomplete suppression of ISC proliferation by EB>Spi-RNAi may indicate that Spi is merely one of the factors downstream of Sox21a that mediate damage-induced ISC proliferation . Taken together , our data suggest a model in which an ISC-EB feedback amplification loop controlled by dynamic expression of Sox21a regulates intestinal homeostasis , regeneration and tumorigenesis ( Figure 7 ) . Normally , Sox21a is expressed in differentiating EBs to promote enterocyte differentiation . Following tissue damage , Sox21a is temporally downregulated in EBs to allow the production of mitogens , which then act back on ISCs to promote ISC proliferation . This leads transient activation of ISC-EB amplification loop for accelerated production of progenitor cells prepared for epithelial differentiation . At the recovery phase , Sox21a is upregulated to shut down the ISC-EB amplification loop and promote cell differentiation and homeostasis re-establishment . In Sox21a mutant intestine , the ISC-EB feedback amplification loop is continuously active , which drives massive production of differentiation-defective cells and tumorigenesis ( Figure 7 ) . 10 . 7554/eLife . 14330 . 013Figure 7 . Schematic models of the ISC-EB feedback loop during normal homeostasis , regeneration and tumorigenesis . ( A ) During normal homeostasis ( left panel ) , ISC and EB typically exist as a progenitor pair . Sox21a is expressed in EB to promote EB differentiation and at the same time to restrict ISC activity by inhibiting paracrine signals , such as Spi , etc . . In Sox21a mutant midgut ( right panel ) , continuous paracrine signals drives continuous activation of the ISC-EB loop , leading to massive production of differentiation-defective cells and tumorigenesis . ( B ) Sox21a-mediated amplification loop is employed in damage-induced intestinal regeneration . Tissue damage causes temporal downregulation of Sox21a in EB and consequently derepression of mitogenic factors , which act in a paracrine manner to promote ISC proliferation . This leads to a temporal activation of the ISC-EB amplification loop for rapid production of progenitor cells prepared for epithelial differentiation . During the recovery after the damage is withdrawn , Sox21a is temporally upregulated in EBs to promote epithelial differentiation and homeostasis reestablishment . DOI: http://dx . doi . org/10 . 7554/eLife . 14330 . 013
Sox family proteins in metazoan are divided into different groups based on their similarity in biochemical properties , and Sox21a belongs to the SoxB2 subgroup whose function is relatively less studied compared to the most related SoxB1 subgroup proteins , such as the pluripotency factor Sox2 ( McKimmie et al . , 2005; Sarkar and Hochedlinger , 2013 ) . By cellular and genetic analysis , here we have characterized the functions of Sox21a , a member of the SoxB2 group in Drosophila and revealed two major roles . First , Sox21a is essential for EC differentiation from EB . Sox21a protein is mainly expressed in differentiating EBs , a pattern that is consistent with its role in EB differentiation . Loss of Sox21a causes accumulation of undifferentiated cells that fail to express Pdm1 , the EC marker . The majority of these mutant cells remain as diploid EBs and some begin to show polyploidy , indicating that mechanisms controlling EC differentiation and cell ploidy can be uncoupled . On the other hand , forced transgene expression of Sox21a in EBs accelerates their differentiation into ECs as evidenced by precocious expression of Pdm1 . However , forced expression of Sox21a in ISCs does not induce their differentiation , suggesting that Sox21a is necessary but not sufficient for inducing EC differentiation from ISCs . One possible explanation for this is that EC differentiation requires both Sox21a and Notch activity . Indeed , although Notch activity is known to be both necessary and sufficient for inducing EC differentiation from ISCs , Notch activated EBs that have already been presented along the length of midgut in young flies remain dormant for several days until there is a need for cell replacement ( Antonello et al . , 2015; Hochmuth et al . , 2011 ) . These observations indicate that activation of Notch alone is not sufficient to induce EB differentiation , but only primes EB for EC differentiation . Notch-activated EBs will stay in undifferentiated state until Sox21a is activated , which then promotes differentiation of the primed EBs into ECs . Another surprising role for Sox21a revealed in this study is that it provides a feedback regulation of ISC activity by suppressing mitogenic signals from differentiating EBs . This function is important for controlling the strength of the ISC-EB amplification loop for balanced self-renewal of ISCs and differentiation of EBs . Because Sox21a is also required for EB differentiation , disruption of this function will cause sustained activation of the ISC-EB amplification loop as well as blocked EB differentiation , leading to the formation of EB-like tumors . Importantly , following epithelial damage , Sox21a is quickly downregulated in EBs . This allows temporal activation of the ISC-EB loop for rapid production of progenitor cells prepared for epithelial repair . During recovery , Sox21a is then temporally upregulated in EBs . This not only stops the ISC-EB amplification loop to avoid excessive EB production , but also accelerates cell differentiation for epithelial repair . Therefore , Sox21a does not simply act as a tumor suppressor in intestine . It is dynamically regulated to control the process of epithelial regeneration in response to various environmental changes via regulating the strength of the ISC-EB amplification loop . Because Sox21a expression in intestine is dynamic during normal adult development , it is conceivable that its expression could possibly be influenced by physiological changes , such as food intake and activity of symbiotic bacteria , and fine-turning Sox21a activity could be important for maintaining regular epithelial turnover . How Sox21a expression is regulated is unclear , but signaling pathways that are implicated in regulating intestinal regeneration , such as JAK/STAT or EGFR/Ras pathways are potential candidate regulators , especially JAK/STAT , which is known to be essential for EB differentiation . During the preparation of this manuscript , two groups have reported the function of Sox21a in Drosophila midgut ( Meng and Biteau , 2015; Zhai et al . , 2015 ) . Meng and Biteau did not observe the tumor suppressive function of Sox21a , possibly because they used a weak mutant allele of Sox21a in their study . Notably , Zhai et al . , observed a similar tumor suppressive role for Sox21a as reported here . Interestingly , they suggest that Sox21a is regulated by JAK/STAT signaling , as Sox21a transgene expression is able to rescue the differentiation defects caused by disrupted JAK/STAT signaling . However , we found that Sox21a was still expressed in JAK/STAT compromised EBs ( Figure 5—figure supplement 2 ) . Therefore , how Sox21a is regulated during normal homeostasis and regeneration remains to be further explored , and it is possible that Sox21a could be controlled by a combination of regulators in a cell type-specific manner , with different mechanisms in ISCs and EBs . Because Sox proteins commonly function together with other cell type-specific co-factors in regulating gene transcription ( Kamachi and Kondoh , 2013; Sarkar and Hochedlinger , 2013 ) , Sox21a could function with different co-factors in ISCs and EBs . These are interesting questions worthy of further investigation . Cells in a given tumor are usually heterogeneous and based on the ability to initiate tumors , tumor cells can be divided into tumor-initiating cells and non-tumor-initiating cells . Here in this case , Sox21 mutant EBs can be regarded as the tumor-initiating cells in vivo . Depleting Sox21a specifically in EBs is sufficient to initiate EB-like tumors . Conversely , restoring Sox21a function specifically in EBs is sufficient to prevent tumor development in Sox21a mutant intestine . However , unlike typical tumor-initiating cells , Sox21a mutant EBs are post-mitotic cells . In addition , their ability to initiate tumors depends on the activity of local ISCs . Therefore , this study also reveals a novel example of tumor-initiating cells in vivo that do not divide themselves , but can “propagate” themselves by utilizing local stem cells . In short , by studying the function of Sox21a in Drosophila ISC lineages , we identified a novel feedback amplification loop between stem cells and their progeny that mediates epithelial regeneration and tumorigenesis . It has long been suggested that tissue regeneration and tumorigenesis are intimately associated , although the mechanistic connection is still obscure ( Oviedo and Beane , 2009 ) . The Sox21a-Spi mediated- ISC-EB amplification loop revealed here may provide a simple example of potential mechanisms that could connect tissue regeneration with tumorigenesis: transient activation of the stem cell- progeny amplification loop promotes regeneration , whereas sustained or irreversible activation of the amplification loop promotes tumorigenesis . We propose that this could be a general mechanism underlying tissue regeneration and tumorigenesis in other tissues , including that in mammals and humans .
The following stocks were used in this study: UAS-Sox21a-RNAi ( BDSC , #53991 , #31902 ) ( Ni et al . , 2011 ) ; esg-Gal4 , UAS-GFP ( gift from Shigeo Hayashi ) ; Su ( H ) -Gbe-lacZ ( gift from Sarah Bray ) ; Su ( H ) -Gbe-Gal4 and Dl-Gal4 ( gift from Xiankun Zeng and Steven Hou ) ( Zeng et al . , 2010 ) ; GMR43E09-Gal4 ( BDSC , #46247 ) ; domeG0468 FRT19A ( Lin et al . , 2010 ) ; UAS-sSpi ( gift from Talila Volk ) ; UAS-Spi-RNAi ( VDRC , #v3922 ) ; Spi-lacZ ( gift from Henry Sun ) . Esg-GFP ( GFP trap , gift from Lynn Cooley ) . These stocks were generated in this study: Sox21aJC1; Sox21aJC2; FRT2A Sox21aJC1; UAS-Sox21a; Upd3-lacZ was generated as previously described ( Jiang et al . , 2009 ) . Two Sox21a mutants were generated by Cas9 mediated gene knock out described before ( Kondo and Ueda , 2013 ) . The following guide RNAs were used: For Sox21aJC1: 5’-GGAGGCGCGCCTGTAGGTCC-3’ and 5’- GAATGGACGCTTCTGTCCTT-3’ For Sox21aJC2: 5’-GGAGGCGCGCCTGTAGGTCC-3’ and 5’- GATGCCGGGCGCGGAGTCAA-3’ To construct the Sox21a transgenic flies , the following primers were used to amplify the coding region of Sox21a from cDNA: 5’- ACGGTGAATCGGCAGTCTAA-3’ and 5’- AGCCATTTGTTTGGGTTCCAG-3’ The 2 . 2 kb PCR product was cloned into pUAST vector . Transgenic fly strains were generated by P element-mediated germline transformation . Polyclonal antibody against Sox21a was generated from rabbit using the following synthetic peptide: CHPHHVQLAAATLSAKYGFGS . The cysteine residual that was added at the N terminal end of the peptide was used to conjugate keyhole limpet hemocyanin ( KLH ) . Serum obtained from immunized rabbit was purified by antigen affinity chromatography . Purified anti-serum at final dilution of 1:1000 was used . The binary GAL4/UAS system was used for spatial and temporal control of transgene expression in intestine ( Brand and Perrimon , 1993; McGuire et al . , 2003 ) . MARCM system was used to generate mitotic clones in intestinal epithelium ( Lee and Luo , 1999 ) . MARCM clones were induced in 2–3 day old female progenies by 37°C heat shock for 1 hr . For the binary expression system , crosses were made at 18°C . 2–3 day old female progenies with desired genotype were then transferred to 29°C , cultured with regular corn meal with yeast paste and transferred every two days prior to dissection and analysis . Immunostaining of Drosophila midgut was performed as previously described ( Lin et al . , 2008 ) . The following primary antibodies were used in this study: mouse anti-Dl ( DSHB , 1:100 dilution ) ; mouse anti-Pros ( DSHB , 1:300 ) ; mouse anti-phospho-Histone H3 antibody ( Cell Signaling Technology , 1:500 ) ; rabbit anti-pERK ( Cell Signaling Technology , 1:200 ) ; rabbit polyclonal anti-lacZ antibody ( Cappel , 1:6000 ) ; rabbit anti-Pdm1 ( gift from Xiaohang Yang , 1:1000 ) ; rabbit anti-Pros ( gift from Yuh-Nung Jan , 1:1000 ) . Secondary antibodies used in this study: goat anti-rabbit or anti-mouse IgGs-conjugated to Alexa ( 568 or Cy5 ) ( Molecular Probes , 1:300 ) . Images were captured using a Zeiss LSM510 confocal microscope . All images were adjusted in Adobe Photoshop and assembled in Illustrator . The Image J software ( National Institutes of Health , Bethesda , MD , USA , http://rsb . info . nih . gov/ij/ ) was used to measure the fluorescence intensity of Sox21a staining . The relative fluorescence intensity was calculated as the corrected fluorescence intensity of EB or Early EC divided by the corrected fluorescence intensity of ISC in the same cell nests . The corrected fluorescence intensity is the average of fluorescence intensity of each cell’s nuclear region minus the average of background fluorescence intensity . 2 day old female flies were cultured on standard corn meal supplemented with 2 mg/mL BrdU ( Sigma-Aldrich ) for 5 days , transferred every day . Female guts were dissected and fixed in 4% formaldehyde for 30 min and followed by DNase I treatment for 15 min at 37°C . The samples were then washed 3 times with PBT and incubated with a rat anti-BrdU ( Abcam ) overnight at 4°C . 7 days old female flies were collected and starved for 10 hr and then were cultured with 5% sucrose solution with or without 5% DSS ( MP Biomedicals ) soaked in kimwipe paper at 29°C for one and half days . Half of the flies were then dissected for analysis ( damage phase ) , and another half were continuously cultured on regular corn meal supplied with yeast paste for one to two days at 29°C before dissection and analysis ( recovery phase ) . Profiling of intestinal progenitor cells were performed according to the method described previously ( Dutta et al . , 2013 ) . The lines of Esg-GFP and Esg-GFP; Sox21a-/- were used to harvest intestinal progenitor cells . Hundreds of guts were dissected in DEPC-PBS on ice within 1 hr and digested with 1 mg ml−1 Elastase ( Sigma , cat . no . E0258 ) in 1 hr at 25°C . Dissociated cells were pelleted at 400g for 20 min , resuspended in DEPC-PBS , filtered with 70 μm filters ( BD Falcon ) and sorted using a FACS Aria II sorter ( BD Biosciences ) . Esg-GFP fusion was used to express green fluorescent protein ( GFP ) to sort each cell population , using w1118 midgut to set fluorescence gate . For each of the three biological replicates , about 250 , 000 sorted cells were used to harvest total RNA with the Direct-zol™ RNA MiniPrep kit ( Zymo research , cat . no . E0258 ) , as described in detail at Bio-protocol ( Chen et al . , 2016 ) . Illumina TruSeq RNA Sample Prep Kit ( Cat#FC-122-1001 ) was used with 50 ng of total RNA for the construction of sequencing libraries . Single-end reads were mapped to the Drosophila melanogaster genome ( Release 5 ) using TopHat ( v2 . 0 . 10 ) . The GTF annotations from the Ensembl release ( BDGP5 ) were also supplied to tophat using the -G flag to allow Tophat to utilize known splice junctions . Each sequencing experiment generated an average 11 . 18 million raw reads , and 86% was successfully mapped for each experiment . Gene expression was quantified by the number of reads that fall into the exons . The results are normalized to RPKM ( reads per kilobase of exon model per million mapped reads ) using Cufflinks ( v2 . 2 . 1 ) . Differentially expressed genes were identified by using the following criteria: P value <= 0 . 05 , Fold change >= 2 . | Within our bodies we have stem cells that are responsible for maintaining many of our tissues in a healthy state and healing wounds after an injury . When these adult stem cells divide , they can produce daughter cells . Through a process called differentiation , these daughter cells can become mature cells that replace old or damaged cells . However , the stem cells also produce copies of themselves – in a process known as self-renewal – to ensure that an organ does not run out of stem cells . The rates of differentiation and self-renewal must be carefully balanced: too much differentiation can eventually lead to the degeneration of the tissue , whereas too much self-renewal can cause tumors to develop . One of the main questions in the stem cell field is how tissues and organs balance these opposing processes . The fruit fly mid-gut is a model system for investigating this question , and is similar to the intestine of mammals . The mid-gut is composed of three main cell types: intestinal stem cells , enteroblasts ( immediate daughter cells of the stem cells ) and enterocytes ( fully differentiated , mature cells ) . Previously published data showed that a protein called Sox21a is present in the intestinal stem cells and enteroblasts , but not in the mature enterocytes . To investigate the role of Sox21a in more detail , Chen et al . deleted the gene that produces Sox21a in fruit flies . These mutant flies developed tumors in their guts , indicating that Sox21a is a tumor suppressor . Further experiments revealed that Sox21a normally drives enteroblasts to differentiate into enterocytes , and also prevents the enteroblasts from communicating with the stem cells to indicate that more enteroblasts are needed . In further experiments , Chen et al . gave otherwise healthy fruit flies a drug that injured their guts . This caused Sox21a activity to decrease temporarily , allowing more enteroblasts to be produced from intestinal stem cells to repair the damage . The mid-gut therefore has an intricate “feedback amplification” system that maintains an appropriate number of each type of cell . In future , other experiments will be needed to determine whether similar feedback amplification systems are found in other tissues , and to investigate the extent to which these systems are found in mammals . Furthermore , understanding this process in more depth could increase our knowledge about how cancerous tumors grow . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"cell",
"biology"
] | 2016 | A feedback amplification loop between stem cells and their progeny promotes tissue regeneration and tumorigenesis |
Vocal development is the adaptive coordination of the vocal apparatus , muscles , the nervous system , and social interaction . Here , we use a quantitative framework based on optimal control theory and Waddington’s landscape metaphor to provide an integrated view of this process . With a biomechanical model of the marmoset monkey vocal apparatus and behavioral developmental data , we show that only the combination of the developing vocal tract , vocal apparatus muscles and nervous system can fully account for the patterns of vocal development . Together , these elements influence the shape of the monkeys’ vocal developmental landscape , tilting , rotating or shifting it in different ways . We can thus use this framework to make quantitative predictions regarding how interfering factors or experimental perturbations can change the landscape within a species , or to explain comparative differences in vocal development across species DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 001
In our study , the specific vocal behavior under investigation is the production of mature contact ( ‘phee’ ) calls . Adult marmoset monkeys produce these vocalizations when alone and out of sight of others ( undirected context ) ( Borjon et al . , 2016 ) . If another marmoset is within earshot , then the pair will begin taking turns exchanging these calls ( directed context ) ( Takahashi et al . , 2013 ) . Very young infants are only gradually able to produce mature sounding contact calls ( Takahashi et al . , 2015; Zhang and Ghazanfar , 2016 ) , and contingent vocal interactions with parents appears to accelerate this process ( Takahashi et al . , 2015 , 2016 ) . Here , we use optimal control theory to construct a Waddington-like developmental landscape to model this process . Optimal control approaches have long been used in studies of motor behaviors and their application requires four specifications: ( 1 ) well-defined behaviors , ( 2 ) a biomechanical model of the system , ( 3 ) a cost function , and ( 4 ) an optimization criterion that describes the probabilities of those behaviors ( Wolpert and Landy , 2012 ) . The theory posits that the probability of producing a specific motor action can be calculated by knowing the cost that a given behavior demands ( Wolpert and Landy , 2012 ) . If the cost to produce an action is high , that action should be less probable than another whose cost is lower . In the current study , the four specifications are the following: ( 1 ) Immature and mature contact calls are the behaviors; ( 2 ) The biomechanical model is one established for songbird vocalizations ( Amador and Mindlin , 2008; Perl et al . , 2011; Amador et al . , 2013 ) that we have shown is also appropriate for marmoset monkeys ( Takahashi et al . , 2015 ) ; ( 3 ) The cost function is the amount of effort required to produce contact calls; and ( 4 ) The optimization criterion is the maximum entropy principle . Maximizing the entropy allows us to identify the probability distribution that is most consistent with the cost function and makes the fewest assumptions . In essence , the goal of our study is to understand how each of the elements of vocal behavior – the vocal apparatus , muscles , nervous system , social interaction – modifies this cost function over postnatal days . The overall pattern of vocal ( contact call ) development consists of a change in dominant frequency , a rapid transition from immature to mature calls , and a correlation between the amount of parental feedback and the rate of this transition ( Figure 1c–e ) . We will use the optimal control approach to take the following inferential steps in order to explain this pattern of vocal development ( Figure 2 ) . First , we will use the biomechanical model to simulate growth of the vocal apparatus ( specifically , the vocal tract length ) ( Figure 2a , b ) . We will then fit the model’s parameters so that it can reproduce the dominant frequency changes observed in marmoset monkey vocal development ( Figures 1c and 2c ) . Second , we will test whether these changes in the vocal tract length can also account for the rapid transition from immature to mature contact calls ( phee/cry ratio; Figure 1d ) . To do this , we combine the cost function ( Figure 2d ) with the optimization criterion which together generate a probability distribution for the production of immature and mature calls ( Figure 2e ) . Third , the prediction is either falsified or supported by comparing the model-based phee/cry ratio with the real marmoset phee/cry ratio ( Figure 2f ) . If the prediction is falsified , we add a new element to the cost function ( e . g , change in muscle control ) which changes it shape and thus changes the probability distribution of call types produced in the emerging landscape . We then repeat the inferential steps using the vocalization data , cost function , and optimal control theory ( Figure 2g ) . To distinguish when a statement is about the model or about the real data , we will always indicate the corresponding model parameter when discussing the model . With the intent to make the main message of the article as clear as possible , we postpone most of the mathematical content to Materials and methods and the Appendix . The reader interested in the mathematical aspects of the modeling will find callouts in relevant places of the main text . 10 . 7554/eLife . 20782 . 004Figure 2 . Illustration of the inferential process used in the study . ( a , b ) A biomechanical model is made of the infant marmoset monkey vocal apparatus . ( c ) The model is used to simulate how the growth of the vocal tract lowers the dominant frequency of calls . Model data ( yellow line ) can be fitted to the real data ( purple line ) . ( d , e ) Optimal control theory is used to generate a cost function for producing different call types and the maximum entropy principle is used to calculate a probability distribution . ( f ) Using the probability distribution , we can calculate the phee/cry ratios produced by the simulated vocal tract growth ( gray line ) and compare with the real marmoset phee/cry ratio data ( purple line ) . ( g ) The contributions of other individual elements ( see Figure 1a ) are gradually added to the framework using a sequential inferential approach together with mathematical modeling . DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 004 In what follows , we first present the biomechanical model of the marmoset vocal apparatus as this serves as the foundation of our optimal control approach . We then present our findings related to the growth of the vocal tract and the successive additions of muscle , nervous system and social interaction to the developmental landscape .
Establishing a biomechanical model for the vocalizations produced by developing marmoset monkeys is required for the optimal control approach . Briefly , we use a model that is a second order ordinary differential equation with two possible time-varying parameters: α ( t ) , representing the air pressure produced by the lungs and β ( t ) , representing vocal fold tension ( Figure 3a ) . Different values of α and β generate different combinations of air pressure and laryngeal tension , resulting in distinct acoustic signals . The third parameter γ is a fixed inverse time scale that sets the upper frequency range of glottal ( vocal fold ) oscillations . The glottal air flow ( Pglottal ) is then filtered by the vocal tract to produce the final vocal output ( Psound ) . The vocal tract is modeled as a cylinder in which the filtering property depends on its length L and reflection coefficient r ( Figure 3a ) . Details of the model are described in Materials and methods: The vocal fold model and From vocal vibrations to calls , Equations ( 14–17 ) ; parameter values are given in Table 1 and further mathematical details appear in the Appendix . 10 . 7554/eLife . 20782 . 005Figure 3 . A biomechanical model of marmoset vocal apparatus . ( a ) Representation of the biomechanical model of the vocal production apparatus . In our one-mass model x ( t ) , y ( t ) are displacement and velocity of vocal folds; nondimensional lung air pressure , vocal fold tension and overall inverse timescale are represented by parameters α ( t ) , β ( t ) and γ . Glottal exit air flow Pglottal is filtered by the vocal tract , modeled as a cylinder of length L with reflection coefficient r at the mouth , to produce vocal output Psound T/2=L/csound . T/2=L/csound is the one way travel time with sound speed csound . ( b–d ) Examples of real infant calls ( top ) and model simulation of the same calls ( bottom ) . ( e ) Example of a sequence of infant calls ( top ) and model simulation ( bottom ) . ( f ) Different values of air pressure and vocal fold tension produce distinct types of calls . Gray region represents parameter values that do not produce vocalization ( i . e . , self-sustained oscillation ) . ( g ) Isofrequency curves . Lines show air pressure and vocal fold tension values that produce glottal air flow that oscillates at the same frequencies; parameters in the gray region do not produce self-sustained oscillations . ( h ) Iso-amplitude curves . Lines show air pressure and vocal fold tension values that produce glottal air flow with same amplitudes . ( i ) Plot showing gains: the ratios between sound produced after the resonance ( vocal output ) and before the resonance ( glottal air flow ) ; warmer colors indicate higher ratios . The diagonal line ( α=β ) is parametrized by θ . au = arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 005 By varying the air pressure α and vocal fold tension β , the model produces immature and mature contact calls ( cries , subharmonic-phees and phees ) with nearly identical acoustic features to those produced by infant marmosets ( Figure 3b–d ) ; it can also simulate sequences of calls ( Figure 3e ) . Respiration α and vocal fold tension β can change in time to produce the different call types . To obtain the results in Figure 3b–e , α and β were varied as increasing and/or decreasing linear ramps . Figure 3f shows the parameter regions that result in each call type . Lower respiratory power α and vocal fold tension β produce cries , whereas higher values produce phees . When α and β are small ( gray region , Figure 3f ) there is no vocal production . Physiological respiratory data support the predictions of the model ( Takahashi et al . , 2015 ) . By varying the parameters , the fundamental frequencies and amplitude of vocal sounds can be changed . Higher fundamental frequencies are obtained when the air pressure α and/or the laryngeal muscle tension β increases ( Monsen et al . , 1978; Hollien , 2014 ) . Consistent with this , Figure 3g shows that the model has isofrequency ( same frequency ) lines for glottal airflow that increase with higher air pressure α and/or muscle tension β . Vocal amplitude is mainly controlled by the air pressure ( Sundberg et al . , 1993 ) , which the model expresses as nearly vertical iso-amplitude ( same amplitude ) curves in Figure 3h . The glottal air flow is then filtered by the resonant vocal tract . The gain g is measured as the ratio between the amplitudes of vocal output ( after vocal tract resonance ) and of glottal air flow ( before vocal tract resonance ) ( g ( α , β ) =maxtPsound ( t ) /maxtPglottal ( t ) ) . Glottal air pressures that oscillate at the resonance frequencies produce higher gains than those that do not ( Ghazanfar and Rendall , 2008 ) . Figure 3i shows the effect on the gain produced by different values of air pressure and muscle tension . The highest gains are obtained for glottal airflow at approximately 9–10 kHz ( Figure 3g , i ) . In humans and other primates , vocal development includes a lowering of the dominant frequency of calls ( Hammerschmidt et al . , 2000 , 2001; Kent and Murray , 1982; Scheiner et al . , 2002; Pistorio et al . , 2006; Takahashi et al . , 2015 ) ( Figure 1c ) . Such changes in frequency in early vocal acoustics are typically associated with increases in the size of the vocal folds: as they get bigger they naturally oscillate more slowly , producing lower frequency sounds . Some early vocalizations are also noisy ( see the cry in Figure 1b ) . Noisiness in vocal acoustic features in general are typically associated with instabilities in the vocal fold movements ( Kent and Murray , 1982; Fitch et al . , 2002; Tokuda et al . , 2002 ) . Our initial modeling study of the biomechanics of marmoset monkey vocal development revealed that , unexpectedly , the vocal tract may additionally play an important role in generating the acoustic features present in both immature and mature vocalizations ( Takahashi et al . , 2015 ) . Thus , in this study , we explore the role of vocal tract growth on shaping the developmental landscape . When an animal’s body size increases during development , so does the length of its vocal tract ( Fitch and Giedd , 1999 ) . Since longer vocal tracts have lower main and subharmonic resonance frequencies f0 , f0/2 , f0/3 , etc . , we expect the resonance frequency to decrease over development . To test this , we fitted the developmental change in dominant frequency observed in the undirected context data ( Figure 4a ) and estimated the developmental change in this feature due to the changing length of the vocal tract L ( Figure 4b ) . As expected , the increase in L and the associated changes in resonance frequencies during development can explain the observed reduction in the dominant frequency of vocalizations . Thus , the change in dominant frequency is a developmental feature that can be associated with changes in vocal tract length . Having established that , we can now use optimal control theory to determine if vocal tract length L can also explain other features of the infant marmoset vocal development . In particular , we will examine if the change in L can explain the rapid transition from producing mostly immature vocalizations like cries and subharmonic-phees to mostly adult-like contact phee calls ( Figure 1d ) ( Takahashi et al . , 2015; Zhang and Ghazanfar , 2016 ) . To do so , we will need to calculate the probability to produce immature and mature calls . Optimal control theory will allow us to do this , but first we must define an ethologically relevent 'cost' of producing vocalizations . 10 . 7554/eLife . 20782 . 006Figure 4 . Growth of the vocal tract . ( a ) Change in dominant frequency of infant marmoset calls during development . Yellow curve shows the value of resonant frequency fitted by the biomechanical model . Red dots are the mean dominant frequency of each postnatal day for all 10 infants ( n=301 sessions ) . ( b ) Vocal tract length estimated by the model assuming a closed-closed cylindrical tube ( brown curve ) ; shaded region indicates 95% confidence interval . ( c ) Infant marmosets produce calls that maximize distance and efficiency . Therefore , the cost C ( θ ) of producing a call is inversely related to the gain g ( θ ) . ( d ) Cost function to produce calls at different air pressure and vocal fold tension values ( θ ) . Blue , yellow , and green dots indicate parameter regions for cry , subharmonic-phee , and phee production , respectively . Minimal cost is achieved for phees , which have glottal air flow oscillating at the natural frequency of the vocal cavity; θ-axis is in log-scale . ( e ) Probability density to produce calls at different θ values; color code is the same as in ( d ) . Increasing η concentrates probability in the parameter range that produces phees . ( f ) Population and model phee/cry ratios . Purple line is the population value of phee/cry ratio for the real marmoset infant data; shaded region indicates 95% confidence interval ( n=195 sessions ) . Gray lines indicate phee/cry ratios predicted by the model for different values of η . ( g ) Growth ( lengthening ) of the vocal tract can explain the lowering of the dominant frequency , but not the transition from cries to phees . DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 006 Based on what we know about the ethology of infant marmoset monkeys , there are benefits to producing vocalizations with higher gains ( i . e . , vocalizations that are louder , longer and more tonal ) ( Figure 4c ) . Marmoset infant cries , subharmonic-phees , and phees are produced when they are separated from the parents ( Takahashi et al . , 2015 ) . These vocalizations are louder compared to other infant calls and result in parents approaching the infant , and so are considered contact calls ( Newman , 1985 ) . However , infant marmoset calls that are more tonal ( or 'phee' like; [Figure 1b , right panel] ) are more likely to elicit parental responses ( Takahashi et al . , 2016 ) . Hence , we model the cost of producing a call at different air pressure and vocal fold tension as inversely related to the gain g ( α , β ) =maxtPsound ( t ) /maxtPglottal ( t ) . We can therefore write the cost to produce a vocalization with a given air pressure ( α ) and vocal fold tension ( β ) as ( 1 ) C ( α , β ) =-logg ( α , β ) , where α , β∈[0 , 1 . 1] remain in the region of viable calls ( see Figure 3f ) . The higher the gain for this function , the lower the cost . The logarithm is used to make the unit of gain proportional to decibels ( dB ) . To simplify our analysis and allow visualization , in what follows we will consider only the diagonal section α=β of the parameter space , labeled θ , that passes through the region of cries , subharmonic-phees , and phees . Other choices of α and β that include these three calls yield similar results . The cost function Equation ( 1 ) becomes ( 2 ) C ( θ ) =-logg ( θ ) , where θ∈[0 , 1 . 1] . Figure 4d illustrates our first ‘landscape’: the cost function with troughs indicating where glottal air pressure oscillates at the vocal tract’s resonance frequency and subharmonics f0 , f0/2 , etc . This cost function describes one section of the developmental landscape related to respiration and vocal fold tension . We can now describe the effect of developmental changes in vocal tract length L on the shape of this landscape . An increase in L causes a decrease in the location of the troughs with respect to θ , and vice-versa ( Figure 4d ) . The different color regions indicate the different types of calls produced by the model for a given θ . Minimal cost is obtained when the infant produces mature contact phee calls because the frequency of glottal oscillations match f0 . Given the cost function , we want to predict the probability that the infant will produce a call with a given air pressure and vocal fold tension . This is achieved by using the maximum entropy principle , via application of the softmax action selection rule ( Jaynes , 1982; Wilson et al . , 2014 ) . This will give the probability of producing different calls that is consistent with the cost function and makes the fewest possible assumptions ( see Materials and methods: Softmax action selection rule for details ) . This rule implies that the probability to produce a call with a given θ is proportional to the exponential of the negative of the cost: ( 3 ) Prob ( θ ) =exp ( −ηC ( θ ) ) /Z . Here η is a non-negative parameter that controls the concentration of the probability distribution and that can be estimated from the data . Z=∫exp ( −ηC ( θ ) ) dθ is the normalizing constant such that the total probability is one . Figure 4e shows that increasing η increases the probability to produce phees . When η is zero , all parameter values are equally likely and we obtain the minimum possible proportion of phees . Now we can ask a key question . Is a developmental landscape that only incorporates changes in vocal tract growth sufficient to explain not only lowering of the dominant frequency ( Figure 1c ) , but also the other features of marmoset monkey vocal development ? If so , then it should be able to explain the rapid transition from immature to mature calls during development ( Figure 1d; [Takahashi et al . , 2015] ) . To test this hypothesis , we calculated the phee/cry ratio , defined as ( 4 ) phee/cry ratio=Prob ( phee ) −Prob ( cry ) Prob ( phee ) +Prob ( cry ) , for the data and the model . Using the model , we can calculate the probability to produce a specific type of call by integrating the probability density for the air pressure and vocal fold tension that produce each type of call . Specifically , if Acry is the set of parameters θ for which the model produces cries ( Figure 4e , blue region ) , we have ( 5 ) Prob ( cry ) =∫AcryProb ( θ ) dθ . Similarly , if Aphee is the set of parameters for which the model produces phees ( Figure 4e , green region ) , we have ( 6 ) Prob ( phee ) =∫ApheeProb ( θ ) dθ . Figure 4f ( gray lines ) shows that during development , changes in vocal tract length L have only a small influence on the phee/cry ratio and increasing η only increases the probability of phees . But the phee/cry ratio in the marmoset data is negative for early postnatal days , showing more cries , and exhibiting a fast transition to mostly phee production after 20-30 postnatal days . Therefore , there are no values of η and L that can fit the data and the cost function that includes only the change in vocal tract length cannot predict the cries-to-phees transition observed in development ( Figure 4f , g ) . In other words , the changes in the position of troughs in the landscape due to vocal tract length increases are insufficient to explain other features of vocal development beyond lowering of the dominant frequency . Therefore , we will next consider the development of muscular control in the vocal apparatus . Laryngeal and respiratory muscle size , strength , and dynamics significantly change through postnatal development in humans ( Moore , 2004; Sasaki , 2006 ) . We expect the control of respiratory and laryngeal muscles to change similarly during development in marmoset monkeys . Based on this assumption , one possibility is that the larger proportion of cries that occurs in the early postnatal period is due to very young infants having difficulty producing higher air pressures and vocal fold tensions required to generate mature ( phee ) calls ( Figure 3f ) . Producing higher values requires stronger respiratory and laryngeal muscles and greater coordination ( Takahashi et al . , 2015 ) . Our aim , therefore , is to estimate a new cost function and hence developmental landscape based on both vocal tract growth and the development of muscular control . We will model the cost of muscular control by modeling the required muscular effort as λθ: a linear function of θ with a parameter λ whose values define how steep is the change in muscular effort for larger values of air pressure and vocal fold tension . Figure 5a shows the muscular effort for different values of θ and λ . In this second function , the total cost to produce a call for a given value of θ is the sum of the cost of the vocal tract change Equation ( 2 ) and muscular effort: ( 7 ) C ( θ ) =-logg ( θ ) +λθ , 10 . 7554/eLife . 20782 . 007Figure 5 . Development of muscular control in the vocal apparatus . ( a ) Muscular control necessary to produce different air pressure and vocal fold tension; higher values of λ imply a greater effort to produce given air pressure and vocal fold tension . Blue , yellow , and green dots indicate parameter regions for cry , subharmonic-phee , and phee production , respectively . ( b ) Cost functions for different values of λ . ( c ) Probability to produce calls at different air pressure and vocal fold tension . For higher values of λ , probability to produce phee diminishes and the probability to produce cries increases . ( d ) Phee/cry ratio fitted by the model ( white curve ) . Colors indicate the probability density of the phee/cry ratio for the marmoset population ( n=195 sessions ) ; warmer colors indicate higher probability densities . ( e ) Estimated muscle effort coefficient ( λ ) during development ( brown curve ) ; shaded region indicates 95% confidence interval ( n=195 sessions ) . ( f ) Relationships between the probability of contingent parental responses and zero-crossing day for real data ( purple line ) and the model ( gray line ) ; shaded region indicates 95% confidence interval ( n=10 infants ) . ( g ) Changes in muscular control can explain the population change in the phee/cry ratio , but not the social feedback-influenced the individual timing of this transition . DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 007 for θ∈[0 , 1 . 1] . Figure 5b shows this cost function for different values of θ and λ . Higher values of λ increase the cost to produce phees ( green ) more rapidly than the cost to produce cries ( blue ) . Therefore , the effect of adding λθ to the cost function is to rotate the developmental landscape counter-clockwise , increasing the cost of producing phees . Using the maximum entropy principle as before ( softmax action selection rule Equation ( 3 ) ) , we can calculate the probability to produce calls for a given θ . As expected from the effect of λ on the cost function ( Figure 5b ) , Figure 5c shows that higher values of λ imply a lower probability to produce phees and higher probability to produce cries . This indicates that the developmental transition from cries to phees can be a consequence of a decrease in λ ( i . e . , an increase in muscular control ) during development . To test this possibility , we fitted the phee/cry ratio data using the cost function Equation ( 7 ) ( Figure 5d ) . The fit follows the phee/cry ratio curve obtained from the directed context data obtained from infant marmosets ( Figure 4f ) . Figure 5e shows the values of λ estimated by applying the model to these real data . As expected , we find that λ decreases during development ( i . e . , muscular control increases ) . Thus , a two-element developmental landscape that includes vocal tract growth and the development of muscle control of the vocal apparatus can account for two key features of vocal development: lowering of the dominant frequency as calls become more mature and the rapid transition from early immature calls to mature ones . Our next question is whether this two-element landscape can also explain individual variability in the timing of the rapid transition . This timing is represented by the zero-crossing day ( Figure 1d , e ) when the number of immature and mature calls produced is equal ( Takahashi et al . , 2015 ) . Our prior work demonstrated that the individual timing of the zero-crossing day appears to depend upon the number of contingent responses provided by parents when they hear the infant’s contact calls ( Takahashi et al . , 2015 ) . Thus , to answer this question , we calculated the correlation between the zero-crossing day and the probability of contingent parental responses to infant calls ( Takahashi et al . , 2015 ) . We observe that there are clear correlations between the amount of parental feedback and the rate of the cry-to-phee transition ( Figure 5f , purple line ) but these cannot be explained by the cost function that only includes the elements of vocal tract growth and muscular control improvements ( Figure 5f , gray line ) . Therefore , an additional factor is needed , one that can control the vocal apparatus and is influenced by social feedback – the nervous system ( Figure 5g ) . As in songbirds ( West and King , 1988; Chen et al . , 2016 ) and humans ( Kuhl et al . , 2003; Goldstein et al . , 2003; Goldstein and Schwade , 2008 ) , contingent parental responses appear to influence vocal development in marmoset monkeys ( Takahashi et al . , 2015 , 2016 ) . The timing of transition from a cry-dominated early developmental period to phee-dominated later period is correlated with the amount of contingent parental vocal feedback that each infant receives ( Figure 1e ) ( Takahashi et al . , 2015 ) . Contingent parental responses are those that are produced within 5 s of an infant call . Infants that receive a higher proportion of contingent parental calls exhibit earlier transitions from cries to phees . This , of course , is social feedback-based reinforcement learning mediated by large-scale networks in the nervous system ( Syal and Finlay , 2011 ) . Given that increasing muscular control ( i . e . , decreasing λ ) increases the phee/cry ratio , we hypothesize that the change in the nervous system driven by social feedback affects the daily rate at which λ decreases during development . In light of this , the amount of change in λ would be proportional to the amount of parental feedback that the infant receives: a larger proportion of parental feedback will decrease λ by a larger amount . Therefore , we propose the following relationship between the value of λ as a function of time , λt , indexed by postnatal day , and the average proportion of contingent parental feedback , represented by F: ( 8 ) λt=λt-1-κF-δ . Here κ is a parameter that models the effect of learning and can be calculated from the data . δ models the neuromuscular development that is independent of contingent parental calls . Like human infant babbling ( Koopmans-van Beinum et al . , 2001 ) , infant marmosets will eventually produce adult-like calls with little or no parental feedback ( Takahashi et al . , 2015; Gultekin and Hage , 2017 ) . Thus , the daily change in λ decomposes into two parts: one ( κF ) that depends on parental feedback and another ( δ ) that is independent of such feedback . Equation ( 8 ) implies that λ decreases linearly with t: ( 9 ) λt=λ0- ( κF+δ ) t , where λ0 is the starting value at postnatal day 0 . The new cost function for each postnatal day which includes vocal tract growth , muscular control and nervous system development is ( 10 ) Ct ( θ ) =-loggt ( θ ) +λ0θ-δtθ-κFtθ , where the subscript t indicates dependence on time . Equation ( 10 ) derives from Equation ( 7 ) with λ replaced by λt=λ0- ( κF+δ ) from Equation ( 9 ) . Figure 6a shows the effect of different proportions of contingent parental calls on the development λ of as predicted by this cost function . If there is no parental vocal feedback ( F=0 ) , e . g . , the infant is deaf or raised in social isolation , λ still decreases , but at a slower rate determined by δ ( black line ) . Figure 6b shows that the proportion of parental feedback is negatively correlated to the timing of transition from cries to phees . Therefore , learning in the developing nervous system facilitated by social feedback tilts the developmental landscape , so that the transition from cries to phees happens sooner and faster . Figure 6c ( blue dots ) shows the relationship between the proportion of contingent parental calls and the zero-crossing day in the data and the same relationship fitted using the cost function Equation ( 10 ) ( yellow curve , see Materials and methods: The full cost function and more parameter choices for further details ) . The fitting shows that the relationship between the proportion of contingent parental responses and the rate of transition from cries to phees can be explained by the development of the nervous system facilitated by parental feedback . Nevertheless , this cost function does not explain why parents produce different amounts of contingent calls . Therefore , we have to consider how the social interaction with parents may depend on other variables of infant vocal development ( Figure 6d ) . 10 . 7554/eLife . 20782 . 008Figure 6 . Learning in the developing nervous system . ( a ) Developmental change of λ for different values of the probability of contingent parental response , F , with constant learning parameter κ=0 . 2126 ( see Materials and methods: The full cost function and more parameter choices ) . Higher values of parental feedback cause faster decay of λ . ( b ) Predicted phee/cry ratios for different values of the probability of contingent parental responses . Higher values of parental feedback cause earlier and faster transitions from cries to phees . Color code is the same as in ( a ) . ( c ) Relationship between the probability of contingent parental response and zero-crossing day; blue dots represent real data ( n=10 infants ) and yellow line is the model fit . ( d ) Changes in the nervous system can explain the relation between the rate of transition from cries to phees and the probability of contingent parental feedback , but not the amount of parental feedback . DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 008 The interactions between parents and an infant are predictive of overall health , quality of attachment and the subsequent communication skills of the child . Unhealthy infants who do not vocalize a lot tend to be fed and held less by mothers , and are slowed in their speech development and thus adversely affect the probability of interactions with parents ( Zeskind , 2013; Lester , 1985 ) . Differences in such vocal output can be related to differences in growth ( Zeskind and Lester , 1981 ) . Therefore , one hypothesis is that infant marmoset monkeys with faster growth rates call more and , as a result , receive more contingent feedback from parents which would accelerate the transition from immature to more mature calls . If this is true , then the higher frequency of parental feedback should be a consequence of parents responding to healthier , more vocal infants . If the hypothesis is falsified , it would suggest that the direct effect of parental feedback is to change the infant’s developing nervous system , thereby affecting the rate of this vocal transition independently of overall growth rates . To model these relationships , let W and N respectively be the weight change over development ( a measure of growth ) and the call rate of the infant marmosets . We can write the frequency of parental feedback F as a simple linear function: ( 11 ) F=b0+b1W+b2N+ϵ , where ϵ is noise independent of W and N , b0 is the intercept , and b1 , b2 are coefficients relating W and N to F . If b1 or b2 is different from zero , we have evidence of an indirect effect . To test this hypothesis , we fitted Equation ( 11 ) to the infant marmoset vocalization data collected in the directed context . We find that no coefficient bi is significantly different from zero ( n=10 , b0=0 . 083p=0 . 675 , p=0 . 675 , b1=0 . 290 , p=0 . 361 , b2=−0 . 051 p=0 . 678 ) . We also tested whether W and N are separately correlated to F ( Figure 7a , b ) . Again , both correlations are not significantly different from zero ( respectively , p=0 . 378 and 0 . 896 ) . This corroborates the alternative hypothesis that parental feedback has a direct effect on the infant nervous system that cannot be accounted for by the growth or call rates of infants . 10 . 7554/eLife . 20782 . 009Figure 7 . Relationship between parental feedback and infant growth . ( a ) Relationship between rate of infant weight change W and the probability of parental responses F . Red circles represent data ( n=10 infants ) . Line indicates linear fit; r= Pearson correlation . ( b ) Relationship between rate of infant phee call production N and probability of parental responses F; plot convention as in ( a ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 009 What makes an infant marmoset transition from immature to mature-sounding vocalizations ? By combining the influences of the developing vocal tract , muscles of the vocal apparatus and the nervous system , we can now present an integrated landscape of vocal development in the manner envisioned by Waddington ( Waddington , 1957 ) . Figure 8a summarizes the relationships between these different elements of vocal production and the corresponding changes in vocal development . In our framework , these elements define the dynamics of the cost function , i . e . , the shape of the developmental landscape . Figure 8b illustrates the landscape plotted over ( θ , t ) -space . Its interpretation is as follows: ( 1 ) Development of vocal tract length L changes the resonance frequency by shifting the troughs/valleys of the landscape represented by the shape of gt ( θ ) ( Figure 8c ) ; ( 2 ) neuromuscular maturation increases the probability to produce phees by reducing the cost function by an amount δθ per day , i . e . rotating the landscape from one postnatal day to the next ( Figure 8d ) ; and ( 3 ) nervous system development driven by social feedback further increases the probability to produce phees by tilting the entire landscape by an amount that is the product of the learning rate κ and the proportion of parental feedback F ( Figure 8e ) . 10 . 7554/eLife . 20782 . 010Figure 8 . Waddington landscape for vocal development . ( a ) Developmental changes associated with each vocal component: vocal tract length L , neuromuscular maturation δ , learning rate κ , and parental feedback F . ( b ) Different components of vocal behavior change distinct features of the developmental landscape . Similar colors indicate regions with the same cost values; darker colors indicate lower costs . The blue solid line shows the natural frequency of the vocal tract , which depends upon its length L . Neuromuscular maturation parameter δ changes the shape of the landscape . The nervous system , influenced by parental feedback κF , changes the slope of the landscape , speeding up development as t increases; θ-axis represents values in logarithmic scale . ( c ) Change in landscape as vocal tract length L increases for fixed δ , κF ( left to right ) . ( d ) Change in landscape as neuromuscular maturation δ increases for fixed L , κF ( left to right ) . ( e ) Change in landscape as learning rate κ times amount of parental feedback F increases for fixed L , δ ( left to right ) . See Table 2 for parameter values . DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 010 To better visualize the dynamics of the landscape as it applies to an individual marmoset infant’s vocal development , we can associate a diffusion process to it ( Video 1 ) . The video shows the states of a particle driven by the gradient of the cost function Ct ( θ ) of Equation ( 10 ) and white noise , on 11 postnatal days separated by 6-day intervals . The position of the particle indicates the call types produced on that postnatal day and the amount of time spent producing each call . Much like the basins of attraction proposed for cell differentiation ( Wang et al . , 2011 ) , the deeper the valley , the longer the diffusion process spends in it each day . As time elapses , the cost function Ct ( θ ) deforms so that the probability of observing cries decreases and phees become more likely , with a zero crossing day in the third or fourth week , depending on the individual . See Materials and methods: Softmax action selection rule for more information . 10 . 7554/eLife . 20782 . 011Video 1 . Animation showing a typical realization of a diffusion process with cost function C as described in Materials and methods: Softmax action selection rule . The particle travels through a developmental landscape that changes its shape due to changes in vocal apparatus , muscle strength , nervous system , and social interaction . The particle’s location represents the behavior of a marmoset infant . In early postnatal days , it stays mostly in the parameter region ( θ ) producing immature calls , whereas in later postnatal days , it stays mostly in the region producing more mature calls . Diffusion dynamics are shown at intervals of six days . Lower left panel shows the numbers of cries and phees produced in each simulated postanatal day; lower right panel shows the phee/cry ratio for the same postnatal days . DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 011
The key to our optimal control-based elaboration of the vocal development landscape was the biomechanical model for vocal production . The model was originally developed to describe bird song production ( Amador and Mindlin , 2008; Perl et al . , 2011; Amador et al . , 2013 ) and then adapted to model infant marmoset vocal production ( Takahashi et al . , 2015 ) . The main advantage of the model is its ability to produce all infant marmoset calls by varying only two parameters: air pressure and vocal fold tension; continuous changes in these parameters can produce spectrally distinct cries , subharmonic-phees , and phees . These are sufficiently distinct that they were previously considered to be different types of calls ( Pistorio et al . , 2006; Bezerra and Souto , 2008 ) . The ability of our biomechanical model to generate such acoustic diversity contrasts with previous models . For example , the origin of cries in nonhuman primates has been attributed to turbulent or chaotic dynamics of the vocal folds ( Fitch et al . , 2002 ) , perhaps as a consequence of vocal fold asymmetry ( Herzel , 1993 ) and/or source-vocal tract interactions ( Hatzikirou et al . , 2006 ) . Our model produces cries simply through a mismatch between the low frequency periodic glottal air flow and the higher frequency resonance of the infant’s upper vocal tract; no chaotic dynamics occurs . The primary difference between cries and phee calls is that the frequency of glottal oscillations is lower in the former ( see Figure 9 ( left ) ) . This result provides direct biomechanical support for the hypothesis that cries are the scaffolding for vocal maturation in both marmosets ( Takahashi et al . , 2015 ) and humans ( Kent and Murray , 1982 ) . 10 . 7554/eLife . 20782 . 012Figure 9 . Producing marmoset cries and phees with the model . ( a ) Trajectories of x plotted vs . y for Equation ( 14 ) for a cry ( left ) and a phee ( right ) . Parameter values ( α , β ) = ( 0 . 09364 , 0 . 088 ) for cry and ( 0 . 151 , 0 . 895 ) for phee respectively . ( b ) Glottal air flows Pglottal produced by the model and ( c ) vocalizations Psound produced after resonance in the vocal tract for a cry and a phee . ( d ) Cry and phee waveforms for calls recorded from infant marmosets; compare with model waveforms shown in ( c ) . Note different vertical scales on left and right columns , indicating that phees are substantially louder than cries . DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 012 In our study , vocal development is understood as a transformation of the cost function through time as a consequence of changes in the vocal apparatus , muscles , nervous system , and social interaction . To calculate the probability that an infant marmoset produces cries , subharmonic-phees , or phees , we first defined the cost of producing a call with a given air pressure and vocal fold tension . We then calculated the probability of producing each type of call using the maximum entropy principle . The idea of a cost that changes in time to describe development goes back at least to Waddington’s epigenetic landscape metaphor ( Waddington , 1957 ) , but in Waddington’s formulation the metaphorical landscape is static and the paths that phenotypical differentiation might take are genetically determined . Modern perspectives using Waddington’s landscape metaphor ( including the current study ) think of development as probabilistic and allow the landscape to change shape over time ( Thelen and Smith , 2006; Wang et al . , 2011; Ferrell , 2012; Sasahara et al . , 2015 ) . For example , Sasahara et al . investigated the development of rhythmic structure in the songs of Bengalese finches using a landscape perspective . They showed that rhythm development exhibits branching and new trajectories along which early , simple vocalizations developed into diverse note types followed by specific silent gaps . The trajectory patterns differed considerably among individual birds , but rhythm proficiency progressed exponentially in all birds ( Sasahara et al . , 2015 ) . In our probabilistic landscape , we inferred the role of vocal tract growth , muscular control and the influence of social feedback on nervous system development . This allowed us to explain – in an integrative manner – the role these elements together play in the transformation of immature to mature contact calls in developing marmosets . We used these somewhat generic elements to most clearly illustrate ( in our view ) the developmental phenomena , as there is no prior study of this kind . However , a more detailed landscape could certainly be generated by at least three means . First , more elements could be added . For example , lowering of the dominant frequency may also be due to growth-related increases in the size of the vocal folds ( Hammerschmidt et al . , 2000 , 2001 ) , but we only considered the vocal tract . Similarly , 'muscular control' and 'nervous system' in our landscape could be more specifically represented by separating the development of individual muscles and neural connections , respectively , related to vocal apparatus control . Second , other infant behaviors may act as scaffolding or otherwise constrain or facilitate vocal development ( Iverson , 2010 ) . In the case of infant marmosets , the ability to self-monitor ( and thus to take turns vocalizing with parents ) matures in an experience-independent manner at the same time as they transform immature contact calls into mature versions ( Takahashi et al . , 2016 ) . The current study did not incorporate how such changes in self-monitoring could also shape the developmental landscape for this vocal transformation . Third , we made assumptions about the developmental trajectory of the elements . For example , we assumed that the development of muscular control and learning in the nervous system were linear processes . This simplification has the benefit of making clear the main phenomena in our framework , but more precise data on the developmental trajectories of muscles or learning-related neuronal activity would provide more accurate predictions . Our framework is general enough to incorporate such details for a deeper understanding . For example , if the linear functions can be replaced by more realistic , perhaps non-linear , functions relating air pressure , vocal fold tension and muscular control , they could be incorporated . Finally , one part of our inferential sequence was that increased muscular control was due to learning-related changes in the nervous system via social reinforcement . An alternative inferential sequence could have been adopted . For example , improvements in muscular control independent of learning could have resulted in more mature-sounding infant calls and thereby increased the rate of parental vocal feedback . This would lead to a different explanation of the correlation between the rate of transition from cries to phees and the amount of parental feedback . We did not test this possibility in our inferential sequence because this hypothesis would be valid only if the change in social interaction were incorporated in the model before changes in the nervous system . The behavioral data do not support this alternative sequence of events: parental call rate , and strength of the dynamic interaction between infants and parents , remain constant throughout development ( Takahashi et al . , 2015 , 2016 ) . Thus , a change in social interaction driven by muscle development ( before learning-related neural changes ) cannot explain the relationship between parental feedback and rate of transition to more mature calls . An integrative understanding of vocal development is important for a variety of reasons , because while we know that many communication disorders originate in problems early in life , we lack any clear grasp of the initial problems . By the time a child is diagnosed with a disorder , the symptoms represent a build up of earlier developmental events . For example , the early vocalizations of infants elicit attention , care and vocal responses from parents ( Lester , 1985; Zeskind , 2013 ) . Infants who do not vocalize much tend to be fed and held less by mothers , and are slowed in their vocal development . The lack of adequate early vocal output by infants may be due to many factors , including abnormal growth of the vocal apparatus , weak laryngeal and respiratory muscles , and/or problems related to nervous system function , such as arousal dysregulation or deficits in motor control and learning . Understanding the mechanisms for human communication , and how it may go awry , requires the use of model animals that naturally exhibit at least a subset of similar communicative behaviors . The early vocal development of marmosets shares a number of parallels with prelinguistic vocal development in humans ( Ghazanfar and Zhang , 2016 ) , perhaps due to convergent evolution of a cooperative breeding strategy ( Borjon and Ghazanfar , 2014; Ghazanfar and Takahashi , 2017 ) . Moreover , we are gaining knowledge of the genetics of this species ( Harris et al . , 2014 ) and , more specifically , the sensorimotor physiology related to its vocal production ( Eliades and Wang , 2008; Miller et al . , 2015; Zhang and Ghazanfar , 2016; Borjon et al . , 2016; Roy et al . , 2016 ) . Recent innovations establishing genetically-modified marmosets ( Sasaki et al . , 2009; Okano et al . , 2016 ) will allow for any number of experimental routes needed to gain novel insights into vocal development . The landscape framework in the current study could be used to make quantitative predictions on the effects of genetic or other types of experimental manipulations . For example , the landscape framework combined with genetic engineering could be used to make predictions regarding the influences of communication- or connectivity-related genes expressed during neuroembryological development in marmosets ( Matsunaga et al . , 2013; Kato et al . , 2014 ) . Naturally , marmosets do not share with humans every aspect of postnatal vocal development . Songbirds , for example , are much better suited to investigate the shared mechanistic basis for more sophisticated forms of vocal learning ( Lipkind et al . , 2013 ) , though such learning occurs at different life-history stages . The vocal development landscape may be used to illuminate why there are species differences in both the degree to which vocalizations can be learned and the life history-timing of such learning . For example , vocal development data from songbirds and humans could be used to generate landscapes for comparison with the marmoset landscape . Closely related species which differ radically in their vocal behavior could also be compared in this manner . For instance , the landscapes of New World squirrel monkeys , whose vocalizations change very little during development ( Hammerschmidt et al . , 2001 ) , could be quantitatively compared to each other and with the marmoset landscape . Similarly , evolutionary insights could be gained by comparing vocal development landscapes of the white-rumped munia and its domesticated counterpart , the Bengalese finch , whose song behaviors and biologies differ considerably ( Katahira et al . , 2013; Suzuki et al . , 2014 ) . Moreover , as the evolution of a phenotype in essence defines its developmental trajectory , providing the developmental parameters for different species could illuminate how changes in their respective landscapes lead to similarities or differences in their adult vocal behaviors . Overall , we believe that the integrated systems view provided by the vocal development landscape not only eschews the incorrect view that there are privileged levels of understanding behavior and its development ( Noble , 2012; Krakauer et al . , 2017 ) , but also enables us to make predictions regarding how natural or experimental perturbations ( e . g . , changes in social feedback , weakening of muscles , disruptions of neural circuits , genetic engineering , etc . ) will affect the development of vocal behavior , and why species differ in their capacity to learn communication signals .
All experiments were approved by the Princeton University Institutional Animal Care and Use Committee . The data analyzed in this work is a subset of the dataset that was previously published ( Takahashi et al . , 2015 ) and can be found at http://science . sciencemag . org/content/suppl/2015/08/13/349 . 6249 . 734 . DC1 . The subjects used in the study were 10 infants and six adults ( three male-female pairs , >2 years old ) , captive common marmosets ( Callithrix jacchus ) housed at Princeton University . The colony room is maintained at a temperature of approximately 27°C and 50–60% relative humidity , with a 12L:12D light cycle . Marmosets live in family groups; all were born in captivity . They had ad libitum access to water and were fed daily with standard commercial chow supplemented with fruits and vegetables . Additional treats ( peanuts , cereal , fruits and marshmallows ) were used prior to each session to transfer the animals from their home-cage into a transfer cage . The vocalizations of marmoset monkey infants were recorded starting on the first postnatal day in two different contexts: undirected ( i . e . , social isolation ) and directed ( with auditory , but not visual , contact with their mother or father ) . The details of the full experiments were described previously ( Takahashi et al . , 2015 ) . Here , the experimental procedures are described in brief for the convenience of the reader . Early in life , infants are always carried by a parent . Thus , the parent carrying the infant ( s ) was first lured from the home cage into a transfer cage using treats . The infant marmoset was then gently separated from the adult and taken to the experiment room where it was placed in a second transfer cage on a flat piece of foam . The testing corner was counterbalanced across sessions . A speaker was placed at a third corner equidistant from both testing corners and pink noise ( amplitude decaying inversely proportional to frequency ) was broadcast at 45 dB ( at 0 . 88 m from speaker ) in order to mask occasional noises produced external to the testing room . An opaque curtain of black cloth divided the room to visually occlude the subject from the other corner . A digital recorder ( ZOOM H4n Handy Recorder ) was placed directly in front of the transfer cage at a distance of . 76m . Audio signals were acquired at a sampling frequency of 96 kHz . Every session typically consisted of two consecutive undirected experiments ( one twin followed by the other ) and one directed experiment ( just one of the twins on a given day ) . Each session started with the undirected experiments lasting 5 min each . The order of the infants was counterbalanced . As soon as the undirected experiment was finished , one of the parents was brought to the experiment room and put into the opposing corner of the room . A second digital recorder ( ZOOM H4n Handy Recorder ) was placed directly in front of the parent at a distance of 0 . 76m from the transfer cage . During this setup procedure and throughout the directed experiment , the opaque curtain prevented the infant and the parent from having visual contact . The directed experiment lasted for ≈15 min . The order of which parent participated in the interaction was counterbalanced . If the parent took more than 15 min to be lured for the directed calls experiment , the experiment was aborted to avoid any excessive separation stress on infants and parents . The number of undirected experiments with at least one call production was 40 , 38 , 38 , 38 , 37 , 39 , 19 , 15 , 16 , 21 ( 10 infants , 301 sessions , 73 , 421 utterances ) . The number of directed experiments for each infant was 17 , 13 , 13 , 18 , 24 , 24 , 22 , 21 , 21 , 22 ( 10 infants , 195 sessions ) . The number of subjects used in this study is based on a previous cross-sectional developmental study of marmoset vocalization that studied nine marmosets ( Pistorio et al . , 2006 ) . A post hoc power analysis using G*Power 3 . 1 showed an achieved power of 0 . 818 for the correlation in Figure 5f ( n=10 , Pearson’s r=-0 . 771 , Type I error =0 . 05 , H0:r=0 ) . All the experimental data used in this article is documented and can be found at http://science . sciencemag . org/content/suppl/2015/08/13/349 . 6249 . 734 . DC1 . To determine the onset and offset of a syllable , a custom made MATLAB routine automatically detected the onset and offset of any signal that differed from background noise over a specific frequency range . To detect the differences , the full recording signal was first bandpass filtered between 6 and 10 kHz . Second , the signal was resampled to a 1 kHz sampling rate , a Hilbert transform was applied and its absolute value was calculated to obtain the amplitude envelope of the signal . The amplitude envelope was further low pass filtered to 50 Hz . A segment of the recording without any call ( silent ) was chosen as a comparison baseline . The 99th percentile of the amplitude value in the silent period was used as the detection threshold . Sounds with an amplitude envelope higher than the threshold were considered as possible vocalizations . Finally , to ensure that sounds other than call syllables were excluded , a researcher verified whether each detected sound was a vocalization or not , based on the spectrogram . After detecting the onset and offset of calls , a custom made MATLAB routine calculated the dominant frequency of each syllable . The dominant frequency of a syllable was calculated as the average frequency at which the spectrogram had maximum power . A cubic spline curve was fitted to the population data using the MATLAB csaps function . Each automatically detected call was manually classified as phee , phee-cry , subharmonic-phee , cry , twitter , and trill , based on the spectro-temporal profile measured by the spectrogram . To ensure validity of our classification procedure , 10 sessions chosen at random were classified by two different individuals and compared . The classification matched in more than 99 . 9% of the call syllables . The six call types show very distinct spectro-temporal profiles and can be easily classified by eye ( Pistorio et al . , 2006; Bezerra and Souto , 2008 ) . For the directed calls experiments , a whole ( i . e . , multisyllabic ) call was defined as any uninterrupted sequence of utterances of the same type ( phee or cry ) with previous offset to next onset separated by less than 500 ms ( DiMattina and Wang , 2006; Takahashi et al . , 2013 ) . To quantify the developmental transition from cry to phee , for each session and subject , the ratio between the number of phees minus cries and the number of phees plus cries was calculated , i . e . , ( 12 ) phee/cry ratio= ( # of infant phee calls produced−# of infant cry calls produced ) ( # of infant phee calls produced+# of infant cry calls produced ) . A cubic spline curve was fitted to the phee/cry ratio data to obtain the phee/cry ratio curve . The zero-crossing day was defined as the first point at which the phee/cry ratio curve crossed zero , transitioning from a negative to a positive value . The zero-crossing day quantifies how quickly each infant transitioned from the cry-abundant initial period to phee-dominated later period . A parental call was classified as a contingent response to an infant call if the parental call onset was separated by less than 5 s from the infant call offset with no other call between them ( Takahashi et al . , 2015 ) . To test if the contingent parental responses were related to how fast infants transition from cries to phees , we calculated the Pearson’s correlation and the linear regression between the proportion of infant phees to which the parents responded before the zero-crossing day ( total number of contingent parental responses before the zero-crossing day divided by the total number of infant phees in the period ) and the zero-crossing day . To calculate the correlation , only the proportion of contingent parental responses that occurred before the zero-crossing day were included to be consistent with the causal ordering in which the possible cause ( contingent parental response ) happens before the effect ( zero-crossing day ) . We used MATLAB csaps function to calculate the correlation and significance test . To investigate how nonlinearities in infant marmoset calls arise , and why they decline throughout development , we extended previous biomechanical models of the human speech production system . The resulting biomechanical model of the larynx and upper vocal tract is based on the one-mass model of Titze ( Titze , 1988 ) , which is simpler than earlier two-mass models ( Ishizaka and Flanagan , 1972; Herzel , 1993; Lucero , 1993 ) and can produce a wide range of birdsong ( Amador and Mindlin , 2008; Perl et al . , 2011; Amador et al . , 2013 ) . In the next two sections we describe the model; further technical details are provided in the Appendix . Titze ( 1988 ) approximates vocal fold dynamics using two modes of vibration: lateral displacement of the tissues in the form of a mucosal wave , and a flapping motion due to out-of-phase oscillations at the entry and exit of the glottis ( Perl et al . , 2011 ) . Titze’s model uses the body-cover hypothesis , which proposes that laryngeal vibrations are governed by muscles and cartilage that determine its geometry , and by its covering of soft tissue that allows waves to propagate in the direction of air flow . Bilateral symmetry in vocal fold oscillations is assumed , simplifying the system to a single degree of freedom oscillator of the form ( 13 ) mx¨ ( t ) +b ( x ( t ) , x˙ ( t ) ) x˙ ( t ) +k ( x ( t ) , t ) x ( t ) =f ( x ( t ) , x˙ ( t ) , t ) , where m is the mass of the vocal folds and x , x˙ and x¨ respectively their lateral displacement , velocity and acceleration; b ( x , x˙ ) x˙ and k ( x , t ) x are nonlinear damping and stiffness forces , f ( x , x˙ , t ) is the driving force due to lung air pressure , and t denotes time . As we shall see , the functions b ( x , x˙ ) and k ( x ) determine the kinds of dynamics produced , and they are typically written as power series . Even truncating these series at third order leaves many coefficients to be determined , and we therefore make a nonlinear change of coordinates that transforms Equation ( 13 ) to its normal form that appears in Figure 3a: ( 14a ) x˙=y , ( 14b ) y˙=−α ( t ) γ2−β ( t ) γ2x+γ2x2−γxy−γ2x3−γx2y . Here the number of coefficients or control parameters is reduced to 3 . Normal forms preserve all qualitative aspects of the dynamics of the original system in the neighborhoods of critical parameter values where bifurcations ( Guckenheimer and Holmes , 1983 ) occur and different dynamical behaviors appear . That this could be done for Equation ( 13 ) was first realized by Perl et al . ( 2011 ) . In this case the parameters α ( t ) and β ( t ) ( which may vary with time ) represent lung air pressure and vocal fold tension , and γ is a time constant . Details on the derivations of Equations ( 13 ) and ( 14 ) are provided in the Appendix . 10 . 7554/eLife . 20782 . 013Figure 10 . Bifurcation set and phase portraits of the model ( Equation ( 14 ) ) . Top left panel shows the bifurcation set in the parameter space spanned by air pressure and muscle tension ( α , β ) . Solid curves indicate saddle-node bifurcations in which pairs of fixed points disappear leaving regions II , III and IV , and Hopf bifurcations in which a stable limit cycle appears entering region I from region V and region III from region IV . Phase portraits in ( x , y ) -space illustrate vocal fold dynamics in regions I-V . Sustained oscillations surrounding a source produce calls in region I; a source , sink and saddle coexist with a small limit cycle in region III , but viable calls are not produced . A unique sink exists in region V , two sinks and a saddle in region IV , and a sink , saddle and source in region II; no sustained oscillations appear in these regions . Solid part of the line labeled θ starting at the Takens-Bogdanov point indicates the axis used in evaluating cost functions . Note that region of ( α , β ) -parameter space is smaller than that in Figure 3f–i . DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 013 Models such as Equations ( 13 ) and ( 14 ) have been fitted to experimental data and model simulations have been compared with human vocalization and bird song ( Mergell et al . , 2000; Sitt et al . , 2008; Zañartu et al . , 2011; Amador et al . , 2013 ) . However , vocal production in marmosets has not been extensively studied and detailed measurements of lung pressure and muscle activity are lacking . As a proxy for this data , recordings of different marmoset calls were used to fit model parameters in the present work . The relative simplicity of the normal form ( 14 ) is helpful in this regard . Equipped with a simple model of laryngeal dynamics , we next derive the resulting sound pressure signals emitted from the mouth . Again seeking simplicity , we appeal to source-filter theory , which assumes that the vocal fold dynamics are independent of filtering within the upper vocal tract ( Titze , 1994 ) . The derivation of Titze ( Titze , 1988 ) , outlined in Appendix §§1 . 1 . 1-1 . 1 . 2 , shows that the pressure Pglottal at entry to the upper vocal tract is proportional to x ( t ) at the midpoint vocal fold position . Figure 9a shows phase space plots of x and y for air pressure and vocal fold tension corresponding to a cry ( left ) and a phee ( right ) . Figure 9b shows the corresponding time histories of x ( t ) . The resulting pressure changes propagate through the upper vocal tract and mouth cavity , which we model as a uniform cylinder . At the exit from the cylinder , part of the wave is reflected back towards the entrance ( glottis ) and the rest is transmitted as sound . Letting T/2 be the time for sound to travel the length , L , of the cylinder , the supraglottal pressure , Pin , at the inlet to the upper vocal tract has the following form: ( 15 ) Pin ( t ) =f ( x ( t ) ) -rPin ( t-T ) , where f ( x ( t ) ) is a function of x ( t ) ( Appendix §1 . 1 . 2 ) and r∈[0 , 1] is the reflection coefficient . Near any given point x ( t ) the time-dependent function f ( x ( t ) ) may be approximated by a Taylor series , and ignoring second and higher order terms we obtain ( 16 ) Pin ( t ) =cx ( t ) -rPin ( t-T ) , where cx ( t ) =Pglottal , and c is a nonnegative constant . In Takahashi et al . ( 2015 ) a third order approximation was used ( see Appendix 1 . 1 . 2 , Equation ( 34 ) ) , but given that the higher order terms are small and produce only minor effects , here we use only the first order term . Note that the vocal fold dynamics x ( t ) determined by Equation ( 14 ) are independent of sound pressure in the vocal tract , but the incoming pressure Pin ( t ) is affected by the reflection rPin ( t-T ) . Finally , the emitted sound is the part not reflected back towards the vocal folds: ( 17 ) Psound ( t ) = ( 1−r ) Pin ( t−T/2 ) Figure 9c shows the signals Psound which result from the effect of resonance on Pglottal in Figure 9b for comparison with waveforms from examples of a cry and phee recorded from an infant marmoset in Figure 9d . Unlike the zebra finch song model of Amador et al . ( Amador et al . , 2013 ) , we do not model the mouth cavity separately; our model can reproduce typical marmoset calls well without this refinement , as shown in Figure 3b–e . Thus , the mathematical model is defined by Equations ( 14 ) and ( 16–17 ) . The components of the model are summarized in Figure 3a in relation to those of the marmoset’s vocal apparatus in panel ( a ) , and Table 1 lists parameter values and ranges used in simulations . To verify that the model could reproduce realistic marmoset calls , the parameters were fit manually to match the spectrotemporal data of calls as shown in Figure 3b–e . 10 . 7554/eLife . 20782 . 014Table 1 . Parameter values used for simulations to fit marmoset calls . The notation [0 , 1 . 1] means that values are chosen in the range 0 to 1 . 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 014ParameterDescriptionValue ( s ) dtTime step size ( μs ) 5αNondimensional pressure[0 , 1 . 1]βNondimensional muscle tension[0 , 1 . 1]γTime constant ( 1/ms ) 45cPressure coefficient1rPressure reflection coefficient0 . 8T/2Time for one way sound travel in vocal tract ( μs ) 50 Numerical simulations of Equations ( 14 ) and ( 16–17 ) were carried out using Euler’s method in custom written MATLAB codes . Parameter values are given in Table 1 . To generate the simulated calls , we varied α ( t ) and β ( t ) within the range [0 , 1 . 1] and matched the frequency spectra and temporal profiles of the simulated sound to the corresponding vocalizations . To improve the fit between the model and recordings , pink noise was added to the simulation to match its presence in the background of the exemplar vocalizations in Figure 3b–e , using the MATLAB pinknoise function ( file exchange #42919 by Hristo Zhivomirov [Kasdin , 1995] ) . The parameter β was held fixed for the cry , while α ( t ) was ramped up and down in a piecewise-linear manner; for the other calls , both α ( t ) and β ( t ) were ramped up and down to produce the varying fundamental and harmonic frequencies of calls such as those in Figure 3c–e . High pass filtering of Psound ( t ) was done using MATLAB eegfilt . Below , we provide the MATLAB code used to solve Equations ( 14 ) and ( 16 ) . function [x , y , p_in] = funcamador ( gamma , a , b , r , T , c , x1 , y1 , dt ) %FUNCAMADOR . M This function will use the Euler method to simulate the %motion of the vocal folds . % This simulation will run for 1 s with time step dt and with % initial conditions x1 and y1 . %% Initializing system t = 1000; N = floor ( t/dt ) ; x = zeros ( 1 , N + 1 ) ; y = zeros ( 1 , N + 1 ) ; x ( 1 ) = x1; y ( 1 ) = y1; p_in = zeros ( 1 , N + 1 ) ; %% Simulating using Euler for n = 1:N x ( n + 1 ) = x ( n ) + dt*y ( n ) ; y ( n + 1 ) = y ( n ) + dt* ( -a*gamma^2 - b*gamma^2*x ( n ) + gamma^2*x ( n ) ^2 . . . - gamma*x ( n ) *y ( n ) - gamma^2*x ( n ) ^3 - gamma*x ( n ) ^2*y ( n ) ) ; if n < T + 1 p_in ( n + 1 ) = c*x ( n ) ; else p_in ( n + 1 ) = c*x ( n ) - r*p_in ( n-T ) ; end end Combination calls like that of Figure 3e suggest that infants can dynamically modulate their vocal output by relatively small muscular changes , since switches between the call types occur very rapidly ( Zhang and Ghazanfar , 2016 ) . We show that small changes in air pressure ( α ) and laryngeal muscle tension ( β ) can switch our model’s output from cries to phees . Figure 10 illustrates the vocal fold dynamics produced by Equation ( 14 ) over a range of values of air pressure α and muscle tension β . The top left panel shows curves in ( α , β ) -space on which bifurcations of fixed points ( equilibria ) of this equation occur . As parameter values ( α , β ) cross these curves , equilibria appear or disappear , their stability types change , and limit cycles representing sustained periodic oscillations in x ( t ) can appear , as illustrated in the phase portraits corresponding to regions I-V . Appendix §1 . 1 . 4 details the calculations that yield the bifurcation curves . Only in the shaded region I , lying above the upper saddle-node bifurcation curve and to the right of the Hopf bifurcation curve , do robust stable limit cycles and hence calls exist , to which almost all solutions converge . Each passage around the cycle corresponds to the vocal folds opening and closing once . Moreover , as ( α , β ) approach the saddle-node bifurcation curve from above , the period of oscillations grows to infinity , so that small changes in these parameters can produce large changes in waveform and hence spectral content . In this region x ( t ) varies rapidly and slowly in different parts of the cycle , implying a broad frequency content ( see Figure 9a , b ( left ) above ) . This extreme sensitivity is responsible for the rapid switches from cries to phees as lung pressure and/or vocal tension increases . No other regions reliably yield calls . Values of α<0 cannot produce sustained oscillations , because in region V there is a single stable equilibrium and in region IV there two stable equilibria and a saddle point; in both cases all solutions converge on equilibria and no sound is produced , consistent with the biological intuition that low driving pressure produces no sound . In region II one stable and one unstable equilibrium coexist with a saddle , again without limit cycles , and all solutions approach equilibria . A small stable limit cycle surrounds the unstable equilibrium in region III , but random perturbations due to noise typically drive the system to the stable equilibrium , thus quenching the oscillations . We therefore focus on parameter values in region I . Since Equation . ( 14 ) only captures the behavior of the original vocal fold model of Titze ( 1988 ) locally ( Figure 11; see Materials and methods: The vocal fold model ) , we restrict the control parameters to 0<α<1 . 1 and 0<β<1 . 1 and use numerical simulations to fit values of α , β that produce the spectrograms and waveforms of calls of interest . The remaining parameters γ , c , r and T were chosen to reproduce observed resonant frequencies and sound pressure levels , as described in Appendix §1 . 1 . 2 , and were fixed at the values listed in Table 1 , unless otherwise specified . 10 . 7554/eLife . 20782 . 015Figure 11 . The larynx and glottis model . The coordinate system is shown with fixed depth l , lateral displacement x ( t ) at midpoint , cross sectional areas a1 , a2 at larynx entry and exit , ag at midpoint , air pressures P1 , P2 , and prephonatory widths x01 , x02 at entry and exit . Adapted from Titze ( 1988 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 015 Within the parameter ranges that stably produce calls , we investigated the relationship between parameter values and characteristics of the resulting model call . To obtain these , we iterated over many parameter values , recorded the natural frequency and amplitude of the calls produced , and computed their gains g ( α , β ) and g ( θ ) , as shown in Figure 3i . For a closed-closed tube , the fundamental frequency is given by f0=csound2L=1T Hz ( Kinsler and Frey , 1962 ) , which provides the relationship L=csoundT/2 . We used csound=350 m/s . We then calculated the resonance frequencies of the biomechanical model for upper vocal tract lengths L=7 . 9 , 8 . 7 , 9 . 6 , 10 mm and interpolated the frequency over this range with a cubic spline curve , thus relating L∈[7 . 9 , 10] to the resonance frequencies 1/T . Using a second cubic spline curve fitted to the marmoset data and the relationship between L and the resonance frequencies obtained previously , we calculated the corresponding vocal tract lengths L . For the data we used the dominant ( highest amplitude ) frequencies as surrogates of resonance frequency . The 95% confidence interval for the resulting estimated vocal tract lengths were calculated from the dominant frequency data by resampling with replacement 1000 times and repeating the estimation method on each of the resampled data sets . Initially , the calls produced by the model were classified manually as had been done with the infant recordings ( Takahashi et al . , 2015 ) . To facilitate analysis of the model , an automatic classifier was developed and manually validated on a smaller sample . For each pair ( α , β ) , the call was simulated for one second . The envelope was calculated using the Hilbert transform of the call lowpass filtered at 4 kHz . Then , the power spectrum of the modeled call and that of its envelope were calculated . Cries are generated by a combination of slow vocal fold vibrations and resonance ( Figure 9 ( left ) ) . Therefore , we expect the power spectrum of the amplitude modulation to contain a peak for the cry but not for the subharmonics or phee . To differentiate between the latter we compared the relative power of the first and the second peak of the power spectrum of the call itself . For the phee , we expect the first peak to be f0 , and hence the largest peak in the spectrum . In contrast , for subharmonics we expect peaks of the power spectrum to occur below the resonance frequency , so the largest peak will not be the first . As a result , if the first peak was larger than the second peak , then the call was classified as a phee and otherwise as a subharmonic . The softmax action selection rule is obtained by applying the maximum entropy principle . We state the principle in a simplified form that suffices for this article . Let C:Θ→𝐑 be a cost function which may also be called an ‘observable’ or ‘utility’ of the system . Assume that C has expected value ∫ΘC ( θ ) p ( θ ) dθ=E . Given a cost function C , a natural question is to know what is the probability distribution p ( θ ) that the animal will execute a specific action . In our case , knowing the cost C of producing a vocalization , we ask what is the probability that a marmoset will produce a call with air pressure and laryngeal tension θ . The maximum entropy principle specifies that the probability density p associated with the cost function C should be the one that maximizes the entropy H ( p ) =-∫Θp ( θ ) logp ( θ ) dθ and satisfies the expectation constraint ∫ΘC ( θ ) p ( θ ) dθ=E . In other words , the maximum entropy principle states that given a cost function and a constraint , we must choose a probability distribution that makes the fewest possible assumptions ( because maximal ignorance equates to maximal entropy ) . Such a probability distribution is said to follow the softmax action selection rule and can be written as p ( θ ) =exp ( -ηC ( θ ) ) /Z , where Z is a normalizing constant so that ∫Θp ( θ ) dθ=1 and η is chosen to satisfy the constraint on the expected value of C . Probabilities were computed using a right Riemann sum approximation . A complementary way to understand the softmax action selection rule is to introduce gradient dynamics with a potential derived from the cost function C ( Video 1 ) . More precisely , consider the diffusion equation ( 18 ) dθt=∂C∂θ ( θt ) dt+2/ηdWt , where ∂C/∂θ is the gradient of C and Wt is a Wiener process . The equilibrium probability distribution for the dynamics θt of Equation ( 18 ) is given by Equation ( 3 ) : ( 19 ) p ( θ ) =exp ( -ηC ( θ ) ) /Z . The resulting diffusion process therefore has an equilibrium measure given by the probability distribution predicted by the softmax action selection rule . If vocalizations are produced at periodic intervals ( approximately once per second in infant marmosets [Zhang and Ghazanfar , 2016] ) , with the parameter defined by the stochastic process θt , then the probability of producing each call type is given by the time that θt spends in each valley of the cost function Ct ( θ ) . This probability is found by integrating p ( θ ) over the parameter region defining each call type , as in Equations ( 5 ) and ( 6 ) . To generate the simulations for Video 1 , we approximated the diffusion process ( 18 ) by a random walk with a potential that is a discretization of C . To allow rapid visualization of the typical diffusion dynamics , we arbitrarily sped up the timescale . The cost function that only includes the contribution of vocal apparatus and muscle control is given by ( 20 ) C ( θ ) =-logg ( θ ) +λθ . To estimate the values of λ for each postnatal day , we first fitted a cubic spline curve to the marmosets' phee/cry ratio data . Then we calculated the value of λ that best approximated the phee/cry ratio curve for each postnatal day . The exact values of λ depend on the choice of η , but the difference is only in the scaling factor and the result in Figure 5e is representative for any choice of η . In Figure 5c , e we used η=5 . Larger values of η have a similar effect , but since the probability densities are more concentrated on the peaks it is harder to display the effects of different λ’s in analogues of Figure 5c , e . The final time-varying cost function with all its parameters can be written as ( 21 ) Ct ( θ ) =-loggt ( θ ) +λ0θ-δtθ-κFtθ . We can decompose Ct ( θ ) as follows . The biomechanical contribution is represented by gt ( θ ) , where the dependency on time t comes from the change in the vocal tract length L or equivalently from the time T/2=L/csound for sound to traverse the cylinder . The change in muscle development is represented by δ , the change in the nervous system by κ , and F represents the contribution of social feedback . To obtain the sharp transition from low to high phee/cry ratio , a good value was η=300 ( Figure 6b ) . With this parameter , the zero-crossing day ( z0 ) occurred when λz=0 . 7927 , implying that ( 22 ) λ0-λzz0=κF+δ . We chose λ0=3 so Equation ( 22 ) yields z0=2 . 2073κF+δ ( see Figure 6c ) . Any value of λ0>λz would give the same curve fitting as we need only to rescale κ and δ accordingly . λ0 is the only parameter that cannot be estimated from the data . Fitting the function to the data relating the amount of parental feedback ( F ) and zero-crossing day z0 , we get κ=0 . 2126 and δ=0 . 0654 . Table 2 lists the parameters used to produce Figure 8b–e . 10 . 7554/eLife . 20782 . 016Table 2 . Parameter values used to plot the developmental landscapes in Figure 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 20782 . 016ParameterFigure 8aFigure 8bFigure 8cFigure 8dT/2 ( μs ) 5040 , 45 , 505050η300300300300λ03333κ ( days-1 ) 0 . 21260 . 21260 . 21260 . 2126δ ( days-1 ) 0 . 065400 . 0250 , 0 . 0333 , 0 . 04170 . 0417F0 . 122000 . 1176 , 0 . 1566 , 0 . 1961 We tested if the rate of weight change W and the rate of infant phee production N before the zero-crossing day could predict the frequency of parental feedback F . To calculate the weight change we first calculated the difference between two consecutive weight measurements and divided by the number of days between them to obtain the local rate of weight change . The overall rate of weight change was calculated as the average of local rates of weight change before the zero-crossing day . If there were a linear relationship between the weight change and the frequency of parental feedback , we would expect a significant Pearson correlation ( r ) between these parameters . We also fitted a multiple linear regression between the explanatory variables W and N , and the dependent variable F . We applied the two-sided t-test to verify the nullity of regression coefficients ( n=10 infants ) . We concluded that neither infant weight increases nor changes in phee call numbers could predict the frequency of parental feedback . | As infants develop they learn new behaviors and refine existing ones . For example , human infants progress from crying to babbling to producing speech-like sounds . A complex sequence of changes in muscles , the nervous system and in patterns of interactions with other individuals all contribute to these emerging behaviors . Despite this complexity , most studies of vocal development have only considered single factors in isolation . A study of speech development , for example , might examine how changes in the brain enable infants to imitate sounds . However , that same study will probably ignore how changes in the structure of the vocal cords , or in the behavior of the parents , also promote imitation . Young marmoset monkeys , like human infants , gradually develop from producing immature cries to adult-like calls . Teramoto , Takahashi et al . built a computational model of this process and compared the model to data from real animals . The first version of the model focused solely on how the marmosets’ vocal cords grow , and did not fully reproduce how adult-like calls emerge in real marmosets . Teramoto , Takahashi et al . therefore added factors to the model that simulate improvements in muscle control , learning in the nervous system and in the behavior of other animals . These findings show that , to reflect how adult-like calls emerge in real marmosets , the model needs to include all of these factors . The model developed by Teramoto , Takahashi et al . may also provide insights into why vocal learning and some other behaviors emerge in some species and not others . It may also be used to predict the consequences of disrupting individual processes in young animals at particular points in time and how such disruptions shape the way an animal develops on its way to adulthood . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | Vocal development in a Waddington landscape |
Efficient transportation is crucial for urban mobility , cell function and the survival of animal groups . From humans driving on the highway , to ants running on a trail , the main challenge faced by all collective systems is how to prevent traffic jams in crowded environments . Here , we show that ants , despite their behavioral simplicity , have managed the tour de force of avoiding the formation of traffic jams at high density . At the macroscopic level , we demonstrated that ant traffic is best described by a two-phase flow function . At low densities there is a clear linear relationship between ant density and the flow , while at large density , the flow remains constant and no congestion occurs . From a microscopic perspective , the individual tracking of ants under varying densities revealed that ants adjust their speed and avoid time consuming interactions at large densities . Our results point to strategies by which ant colonies solve the main challenge of transportation by self-regulating their behavior .
Many organisms such as herds of migrating wildebeests , swarms of insects and bacteria , starling flocks , fish shoals or pedestrian crowds take part in flow-like collective movements ( Ball , 2004; Berdahl et al . , 2013; Berdahl et al . , 2018; Buhl et al . , 2006; Chowdhury et al . , 2005; Fourcassié et al . , 2010; Giardina , 2008; John et al . , 2009; Moussaïd et al . , 2011; Sumpter , 2010; Vicsek and Zafeiris , 2012; Zhou et al . , 2008 ) . In most cases , all individuals cruise along the same path in a unique direction , which facilitates coordination . The task of maintaining a smooth and efficient movement becomes more challenging when individuals travel in opposite directions and are bound to collide ( Fourcassié et al . , 2010 ) . Along with humans ( Helbing et al . , 2005; Moussaïd et al . , 2011 ) , ants are one of the rare animals in which collective movements are bidirectional . Ants are central-place foragers , which entails a succession of journeys between their nest and their foraging site . When exploiting large food sources , many species lay chemical trails along which individuals commute back and forth ( Czaczkes et al . , 2015; Gordon , 2014 ) . The flow of individuals on these trails can reach several hundred ants per minute ( Couzin and Franks , 2003 ) . Yet , ants seem to fare better than us when it comes to traffic management ( Burd et al . , 2002; Dussutour et al . , 2004; Fourcassié et al . , 2010; Hönicke et al . , 2015 ) . However , we lack direct experimental evidence showing that ants do not get stuck in traffic jam at high density ( Burd et al . , 2002; Hönicke et al . , 2015; John et al . , 2009 ) . In traffic engineering , the relation between the density of individuals k and the flow q ( the speed v times the density k ) is often described via the fundamental diagrams ( Chowdhury , 2000; Greenshields et al . , 1935; Helbing , 2009; Helbing , 2001; Helbing and Huberman , 1998; Hoogendoorn and Bovy , 2001; Pipes , 1953; Underwood , 1960 ) ( Figure 1A ) . The speed-density and flow-density diagrams vary depending on the system under scrutiny but share similar properties . First , the flow q increases with the density k from zero to a maximum value and then decays until it goes back to zero at the so-called maximum jam density kj . The flow-density curves are usually concave with an optimal value for k on the path at which maximum flow or capacity is reached ( Helbing and Huberman , 1998 ) . Second , the speed is maximum when an individual is traveling alone ( free flow speed vf ) and decreases with the density k . The speed becomes zero and individuals stop at jam density that is v ( kj ) =0 ( Nagatani , 2002; Treiber et al . , 2000 ) . Despite their apparent efficiency in traffic management , few studies have investigated the relation between density , speed and flow in ants ( Burd et al . , 2002; Gravish et al . , 2015; Hönicke et al . , 2015; John et al . , 2009 ) . In leaf-cutting ants ( Burd et al . , 2002 ) and fire ants ( Gravish et al . , 2015 ) speed decreases when density increases while for wood ants ( Hönicke et al . , 2015 ) and mass raiding ants ( John et al . , 2009 ) the speed remains constant when density increases . However , the range of densities tested was large enough to observe traffic jams only in the study conducted with fire ants traveling in tunnels ( Gravish et al . , 2015 ) . The highest densities as well as the estimated occupancy ( fraction of area covered by ants ) , recorded in leaf-cutting ants ( Burd et al . , 2002 ) , wood ants ( Hönicke et al . , 2015 ) and mass raiding ants ( John et al . , 2009 ) were relatively low: 0 . 8 ants . cm−2 ( occupancy 0 . 20 ) , 0 . 6 ants . cm−2 ( 0 . 13 ) and 0 . 3 ants . cm−2 ( 0 . 10 ) and not sufficiently high to generate a traffic jam as ants never exceeded the capacity of the trail , that is the maximum value of the flow allowed by the trail width . Here , we investigated if ants succeed in maintaining a smooth traffic flow and avoid traffic congestion under the widest possible range of densities . We used the European supercolony of Argentine ants Linepithema humile , which is a major pest around the world and the largest recorded society of multicellular organisms ( Giraud et al . , 2002 ) . In our experiment , a colony was connected to a food source using a bridge ( Figure 1B ) . To manipulate density , we used a combination of bridges of different widths ( 5 , 10 and 20 mm ) and experimental colonies of different sizes ( from 400 to 25 , 600 ants ) . We conducted a total of 170 experiments . The flow q and the density k were recorded on each experiment every second for one hour giving us 612 , 000 flow/density ( non-independent ) observations pairs . We succeeded in generating large variations of density ( from 0 to 18 ants . cm−2 ) and occupancy ( from 0 to 0 . 8 ) .
We first studied ant traffic at a macroscopic level . The flow of ants q heading in both directions was plotted as a function of density in Figure 2A . The flow q increased with the density k to a certain point and then it remained constant . We analyzed the relationship between k and q using three different macroscopic traffic functions ( Greenshields et al . , 1935; Pipes , 1953; Underwood , 1960 ) ( Figure 2B ) . All the parameters of the functions were fitted using a nonlinear least squares fit procedure ( Figure 2B , Supplementary file 1 ) . All the statistical models performed similarly well but failed at predicting the data at intermediate and large densities . Thus , based on experimental data , we introduced a two-phase flow function to describe the relationship q-k as a piecewise linear function , with first a linear increase of the flow , followed by a constant value when the jamming density was reached . ( 1 ) Two−phaseflowfunction:q ( k ) =k⋅vifk≤kjandq ( k ) =kj⋅v=qjifk>kj Then , we conducted a model selection analysis using Akaike weights to assign conditional probabilities to all statistical models . Thanks to our large dataset , the result was unequivocal: the two-phase statistical model was selected ( Figure 2C , Figure 2—figure supplement 1 , Supplementary file 1 ) . Why did the ants not jam or clog , as one would expect in usual traffic situations ? This could result from a spatiotemporal organization of the flow at high density ( Fourcassié et al . , 2010 ) . The traffic is considered as spatially organized when the flows of inbound and outbound ants are not completely intermingled and lane segregation occurs ( Couzin and Franks , 2003; Fourcassié et al . , 2010 ) . Temporal organization arises when oscillatory changes in the flow direction are observed and the traffic becomes intermittently unidirectional that is alternating clusters of inbound and outbound ants occurs ( Fourcassié et al . , 2010 ) . Both type of organizations limits the rate of time-consuming contacts ( collisions ) and allows ants to maintain a smooth traffic ( Fourcassié et al . , 2010 ) . However , in our experiments , no clear evidence of such organizations was observed . In contrast , when ant density reached a critical threshold , inbound and outbound flows became intermingled temporally ( Figure 3—figure supplement 1 ) and spatially ( Figure 3A , Figure 3—figure supplement 2 , Video 1 and Video 2 ) . In addition , contrary to pedestrian traffic ( Helbing et al . , 2005; Moussaïd et al . , 2011 ) , the relationship between density k and flow q was only marginally influenced by the degree of asymmetry in the flows ( Figure 3B , Figure 3—figure supplement 3 ) . Simply put , it did not increase faster with the density k when traffic was mostly unidirectional ( i . e . proportion of outbound flow qo close to 1 or 0 , traffic considered as asymmetric ) than when it was entirely bidirectional ( i . e . proportion of outbound flow qo close to 0 . 5 , traffic considered as symmetric ) . This is illustrated in Figure 3B by the tendency for the isoclines to run parallel to the y-axis ( Figure 3B ) . Therefore , we focused our next analyses on individual behavior to understand how ants maintained a constant flow despite the increasing density . From an individual behavior perspective , most traffic-flow functions ( Greenshields et al . , 1935; Pipes , 1953; Underwood , 1960 ) suggest that individual speed decreases non-linearly with the density due to friction between individuals . However , the two-phase flow traffic function suggested that there was no evidence of such friction between ants when the density was below eight ants . cm−2 that is the flow increased linearly , whereas above eight ants . cm−2 frictions appeared but increased only linearly with the density that is the flow remained constant over a wide range of densities . How can we quantify such friction at the individual level ? A key factor determining ant speed is the number of contacts ( i . e . collisions ) experienced with nestmates , which makes ants stop and thus decays their overall speed ( Burd and Aranwela , 2003; Dussutour et al . , 2005; Gravish et al . , 2015; Wang et al . , 2018 ) . To test if the number of contacts played the role of a hidden variable linking density and speed , we measured the number of contacts C , the density k and the traveling time T for a sample of 7900 ants individually tracked on a 2 cm section of the bridge . As density k grew , the number of contacts C increased linearly ( C = 0 . 61 k Figure 4A ) that is the larger the density , the higher the number of contacts . We observed a linear effect of the number of contacts C on traveling time T , that is each contact slowed down the ants . ( T = T0 + C · ∆T with T0 = 0 . 95 s and ∆T = 0 . 24 s , Figure 4B ) . T0 can be interpreted as the traveling time to cross the bridge without contacts ( free flow traveling time ) , whereas ∆T is the time lost per contact . So far , density k only had a negative impact on the speed v: it increased the number of contacts C which in return increased traveling time T . However , in our two-phase diagram , density k had no or minor effect on speed v in the phase 1 . Therefore , there had to be a positive effect of density k on speed v in this regime . Thus , the interplay between T , k and C had to be more subtle . To combine multiple effects , we estimated the expected traveling time T depending on both density k and number of contacts C . For a given number of contacts C , we visualized the average traveling time T ( see Figure 5A ) for various values of density k , using local regression fitting . The vertical distance between two neighbors’ curves was given by ∆T . As expected , traveling time T increased with the number of contacts C , but the key information is that density k actually made traveling time T decay initially ( up to k ≈ 5 ) . For a given number of contacts C , ants moved faster on the bridge at intermediate density k ( i . e . k ≈ 5 ) . To further visualize this positive effect of the density , we estimated the free flow speed vf , that is speed without contact vf = L/ ( T-C·ΔT ) , where L=2cm is the length of the recording section on the bridge ( Figure 5B ) . vf is first increasing with density k up to k≈5 ants . cm−2 and then decays back to its initial value . This phenomenon might be explained by a pheromone effect . It is well known that Argentine ants deposit pheromones both when leaving and when returning to the nest and travel faster on a well-marked trail ( Deneubourg et al . , 1990 ) . To incorporate both effects of the density , we proposed the following formula for the speed: ( 2 ) vk=LT0+ΔT⋅Ck⋅α+β⋅k⋅e-γ⋅kwhere C ( k ) =0 . 61·k is the average number of collisions , the three other parameters α , β and γ model the pheromone effect: α corresponds to the intrinsic attractiveness of the unmarked bridge , β represents the positive effect of k and γ measures the range where the pheromone effect might occur . These three parameters were estimated using a non-linear regression algorithm ( α=0 . 812 ± . 009 , β=0 . 160 ± . 010 , γ=0 . 156 ± . 007 ) . We observed a decay of the speed v when k increased ( Figure 6A ) . From this estimation of the speed , we deduced the following formula for the flow: ( 3 ) qk=k⋅vk We plotted in Figure 6B the predicted flow q and the experimental observations of the sub-sample dataset ( N = 7900 observations ) used for the estimation . We noticed a clear agreement between the data and the model predictions . The plateau reached was q ≈ 10 ants . cm−1 . s−1 . Thus , even though increasing density k generated more contacts impacting negatively the flow q through their effect on traveling time T , when k < 5 ant . cm−2 ants moved faster , which positively impacted the flow q . These two effects counterbalanced , leading to a linear increase of the flow q with density k ( phase 1 ) . For k > 8 ants . cm−2 , despite the crowdedness of the trail , ants maintained a constant flow q . The speed v ( k ) continued to decay due to the increase of contacts , but this negative effect on the flow q ( k ) was offset by the increase of k . We recovered this phenomenon from our model by estimating the limit flow q as the density k increases: ( 4 ) limk→∞qk=LΔT⋅0 . 61⋅α≈11 . 09 In other words , the flow in Figure 6B would merely increase for larger values of density k . However , experimentally the flow would have to decay eventually as the ant occupancy on the bridge cannot increase indefinitely . Given that the unoccupied portion of the bridge decayed with the density , it is remarkable that the number of contacts increased only linearly with the density according to Figure 4A ( or one could argue even sub-linearly for k > 10 ants . cm−2 ) . Interestingly , we also observed that ants restrained themselves from leaving the nest to prevent overcrowding as k never exceeded 18 ant . cm−2 even though we increased colony size and reduced bridge width ( number of ants entering the bridge for the largest colonies ± CI95: 2 . 69 ± 0 . 04 , 4 . 34 ± 0 . 03 and 5 . 05 ± 0 . 03 ants . sec−1 for the 5 , 10 and 20 mm bridges ) . Furthermore , U-turns were seldom once the ants were traveling on the bridge ( probability of turning back once on the bridge = 0 . 01 ) .
Traffic jams are ubiquitous in human society where individuals are pursuing their own personal objective ( Youn et al . , 2008 ) . In contrast , ants share a common goal: the survival of the colony , thus they are expected to act cooperatively to optimize food return . In pedestrian and car traffic ( Banks , 1999; Helbing et al . , 2005 ) , at occupancy levels above 0 . 4 the flow starts to decline , while in ants , the flow showed no sign of declining even when occupancy reached 0 . 8 . Here , by investigating traffic dynamics on a large range of densities , we demonstrated that ants seem to be immune to traffic congestion . The traffic of pedestrians or vehicles is often ruled by external constraints ( enforced rules ) , time consuming interactions ( avoidance behavior ) and negative feedbacks ( jamming ) . In contrast , in ants , traffic is governed by positive feed-backs ( trail following and reinforcement ) and interactions which are time consuming but often beneficial for the colony as they promotes information transfer ( Bouchebti et al . , 2015; Burd and Aranwela , 2003; Farji-Brener et al . , 2010; Fourcassié et al . , 2010; Gordon , 2019 ) . Nonetheless , time spent interacting with other ants has strong consequences on traffic dynamics ( Gravish et al . , 2015 ) . Short interaction time allows a smooth traffic over a wide range of densities whereas long interaction time promotes rapid slowing of traffic ( Gravish et al . , 2015 ) . In fire ants , Gravish et al . ( 2015 ) found that interaction time was relatively short ( 0 . 45 s ) and that the flow was relatively unaffected by density up to a certain critical value beyond which they observed the formation of traffic jams ( Gravish et al . , 2015 ) . Here , we found that Argentine ants have an interaction time approximately twice shorter than fire ants , which could explain why we observed a smooth and efficient mixing of opposite flows over a broader range of densities . The exact nature of the mechanisms used by Argentine ants to keep the traffic flowing in this study remains elusive , yet when density on the trail increases , ants seemed to be able to assess crowding locally , and adjusted their speed accordingly to avoid any interruption of traffic flow . Moreover , ants restrained themselves from entering a crowded path and ensured that the capacity of the bridge was never exceeded that is the maximum value of the flow allowed by the bridge width . Traffic regulation on trails ultimately allows Argentine ants to maintain a high rate of food return to the nest , an essential asset in the context of food competition occurring in natural environments . This balance between positive feedback and negative feedback appears to be generic in ants . For instance , in the nest-building context , behaviors such as ‘individual idleness and retreating’ at overcrowded digging sites supports an optimal accessibility of space facilitating excavation during nest construction in fire ants ( Aguilar et al . , 2018 ) . Similarly , carpenter ants placed under threat and forced to escape through a narrow door distribute themselves uniformly over the space available instead of rushing toward the door and trampling on others ( Parisi et al . , 2015 ) . As a result , contrary to humans ( Helbing et al . , 2000 ) , they avoid clogging and evacuate efficiently ( Parisi et al . , 2015 ) . Lastly , in garden ants , the interplay between trail following behavior and collisions allows ants to cope with bottleneck situations ( Dussutour et al . , 2006; Dussutour et al . , 2005; Dussutour et al . , 2004 ) . Overall , our results extend previous research on ant traffic organization ( Aguilar et al . , 2018; Burd et al . , 2002; Fourcassié et al . , 2010; Gravish et al . , 2015; Hönicke et al . , 2015; John et al . , 2009 ) and show that ant prevented traffic jams from occurring and behaved as a self-organized biological adaptive system ( Camazine et al . , 2003; Hemelrijk , 2005 ) . Similar phenomena of self-regulation ought to be found in other complex systems such as migrating animals ( Buhl et al . , 2006 ) , cell machinery ( Leduc et al . , 2012 ) , swarm of robots ( Aguilar et al . , 2018 ) and data traffic ( Valverde and Solé , 2004 ) and could provide inspiration in all disciplines , ranging from molecular biology to automotive engineering . Although comparing ant traffic to human traffic might be a delicate task , as traffic in humans is neatly separated into unidirectional roadways , the rise of autonomous vehicles paves the way for new strategies to optimize traffic flow .
We used the Argentine ant Linepithema humile , an invasive species that uses mass recruitment through pheromone trails to exploit abundant food sources . Argentine ants have multiple queens and form supercolonies . Established supercolonies can contain up to 108 workers ( Giraud et al . , 2002 ) . Usually , foraging in Argentine ants is associated with high traffic between the nest and the food source in order to provide food for the whole colony . These ants are monomorphic ( 2 . 5 mm long ) . The ants were collected in Toulouse ( France ) . All the statistical analysis were done using R version R 3 . 5 . 0 ( Crawley , 2012 ) . All the parameters of the fundamental diagrams were fitted using a nonlinear least squares fit procedure ( command nls; Baty et al . , 2015 ) . This procedure used Gauss-Newton algorithm to find the parameters that minimize the mean square error between the experimental data and the model prediction . The results of the estimations for the four functions are given in Supplementary file 1 . A model selection using the Akaike weights ( AW ) has been conducted to assign a conditional probability for each statistical model . Source codes are available ( Poissonnier et al . , 2019 ) . Data are available from the Dryad Digital Repository: https://doi . org/10 . 5061/dryad . 8q58jg3 ( Poissonnier et al . , 2019 ) . | Humans and ants are among the few species that engage in two-way traffic . Maintaining a smooth and efficient traffic flow while avoiding collisions is challenging for humans . Yet ants seem to be masters of traffic management . They can efficiently move back and forth between their nests and food without overtaking or passing each other , forming a steady stream of traffic . Few studies have looked at how ants maintain such a smooth flow even as the number of ants on a path increases . Now , Poissonnier , Motsch et al . have designed an experiment to investigate whether ants can maintain their steady stream of traffic when their path to food gets more crowded . This involved manipulating the density of ants using a combination of different sized colonies ( ranging from 400 to 25 , 600 Argentine ants ) and changing the width of the bridge connecting the ants to their source of food . The experiment was repeated 170 times , and data was collected on traffic flow , speed of the ants , and number of collisions . For pedestrians and car traffic , the flow of movement will slow down if occupancy levels reach over 40% . Whereas in ants , the flow of traffic showed no signs of declining even when bridge occupancy reached 80% . The experiments revealed that ants do this by adjusting their behavior to their circumstances . They speed up at intermediate densities , avoid collisions at large densities , and avoid entering overcrowded trails . Studying ant traffic management has been a source of inspiration for scientists working with large groups of interacting particles in many fields . This includes molecular biology , statistical physics , and telecommunications . It may also have relevance for managing human traffic , particularly as scientists develop autonomous vehicles that might be programmed to work together cooperatively as ants do . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology"
] | 2019 | Experimental investigation of ant traffic under crowded conditions |
To position the mitotic spindle within the cell , dynamic plus ends of astral microtubules are pulled by membrane-associated cortical force-generating machinery . However , in contrast to the chromosome-bound kinetochore structure , how the diffusion-prone cortical machinery is organized to generate large spindle-pulling forces remains poorly understood . Here , we develop a light-induced reconstitution system in human cells . We find that induced cortical targeting of NuMA , but not dynein , is sufficient for spindle pulling . This spindle-pulling activity requires dynein-dynactin recruitment by NuMA’s N-terminal long arm , dynein-based astral microtubule gliding , and NuMA’s direct microtubule-binding activities . Importantly , we demonstrate that cortical NuMA assembles specialized focal structures that cluster multiple force-generating modules to generate cooperative spindle-pulling forces . This clustering activity of NuMA is required for spindle positioning , but not for spindle-pole focusing . We propose that cortical Dynein-Dynactin-NuMA ( DDN ) clusters act as the core force-generating machinery that organizes a multi-arm ensemble reminiscent of the kinetochore .
Forces generated at dynamic plus-ends of microtubules drive directional movement of chromosomes and the mitotic spindle to achieve successful cell division ( Inoué and Salmon , 1995 ) . During animal mitosis , dynamic plus-ends of microtubules emanating from the spindle interact with two macro-molecular complexes; kinetochores and the cortical force-generating machinery . Kinetochores consist of more than 100 different proteins assembled on centromeric DNA and surround dynamic microtubule plus-ends using multiple fibril-like microtubule-binding proteins and/or ring-like couplers to harness the energy of microtubule depolymerization for chromosome segregation ( Cheeseman , 2014; Dimitrova et al . , 2016; McIntosh et al . , 2008 ) . In contrast , the cortical force-generating machinery assembles on the plasma membrane and pulls on the dynamic plus-ends of astral microtubules to define spindle position and orientation ( Galli and van den Heuvel , 2008; Gönczy , 2008; Grill and Hyman , 2005 ) . Spindle positioning determines daughter cell fate by controlling the distribution of polarized cell fate determinants and daughter cell size during both symmetric and asymmetric cell division ( di Pietro et al . , 2016; Kiyomitsu , 2015; Morin and Bellaïche , 2011; Williams and Fuchs , 2013 ) . In metaphase human cells , the cortical machinery consists of evolutionary conserved protein complexes , including the cytoplasmic dynein motor , its binding partner dynactin , and the cortically-anchored NuMA-LGN-Gαi complex ( Figure 1A ) ( Kiyomitsu and Cheeseman , 2012 ) . Prior work has conceptualized that the cortical complex is distributed along the cell cortex and individually pulls on astral microtubules using dynein-based motility and/or by controlling microtubule dynamics ( Kiyomitsu and Cheeseman , 2012; Kotak and Gönczy , 2013; Laan et al . , 2012 ) . However , compared to the focal kinetochore structure , how this diffusion-prone membrane-associated complex efficiently captures and pulls on dynamic plus-ends of astral microtubules remains poorly understood . Here , we sought to understand the mechanisms of cortical pulling-force generation by reconstituting a minimal functional unit of the cortical force-generating complex in human cells using a light-induced membrane tethering . We found that cortical targeting of NuMA is sufficient to control spindle position , and that NuMA makes multiple , distinct contributions for spindle pulling through its N-terminal dynein recruitment domain , central long coiled-coil , and C-terminal microtubule-binding domains . In addition , we demonstrate that NuMA assembles focal clusters at the mitotic cell cortex that coordinate multiple dynein-based forces with NuMA’s microtubule binding activities . We propose that the cortical Dynein-Dynactin-NuMA clusters ( hereafter referred to as the cortical DDN clusters ) act as the core spindle-pulling machinery that efficiently captures astral microtubules and generates cooperative pulling forces to position the mitotic spindle .
To understand the molecular mechanisms that underlie cortical force generation , we sought to reconstitute a minimal functional unit of the cortical force-generating machinery in human cells using a light-induced hetero-dimerization system ( iLID ) ( Guntas et al . , 2015 ) . In this system , cytoplasmic RFP-Nano fusion proteins can be targeted to a locally illuminated region of the mitotic cell cortex by interacting with membrane-bound iLID ( Figure 1A; Figure 1—figure supplement 1A–B; and Video 1 ) . Because the N-terminal fragment of NuMA is sufficient to recruit dynein-dynactin to the cell cortex ( Kotak et al . , 2012 ) , we first sought to manipulate endogenous NuMA . We established triple knock-in cell lines that stably express membrane-targeted BFP-iLID ( Mem-BFP-iLID ) , a NuMA-RFP-Nano fusion ( Figure 1A; Figure 1—figure supplement 1C–E ) , and SNAP-tagged dynein heavy chain ( DHC ) or the dynactin subunit p150 ( Figure 1—figure supplement 1F–G ) . To prevent cortical recruitment of NuMA by the endogenous LGN-Gαi complex , we depleted LGN by RNAi ( Figure 1A middle , 1B t = 0:00; Figure 1—figure supplement 1H ) . We then continuously illuminated the cortical region next to one of spindle poles ( indicated by red circles in Figures ) with a 488 nm laser to induce NuMA-RFP-Nano targeting . Light illumination induced the asymmetric cortical accumulation of NuMA-RFP-Nano within a few minutes ( Figure 1B–C ) , which subsequently recruited DHC-SNAP and p150-SNAP ( Figure 1B–C; Figure 1—figure supplement 2A–C ) . The level of light-induced cortical NuMA is about three times higher than that of endogenous NuMA in metaphase , but similar to that in anaphase ( Figure 1—figure supplement 1I–J ) . Importantly , following asymmetric NuMA-RFP-Nano targeting , the mitotic spindle was gradually displaced toward the light-illuminated region in 82 . 4% of cells ( n = 17 , Figure 1B , D–E , and Video 2 ) , whereas spindle displacement and cortical dynein recruitment was never observed by targeting RFP-Nano alone ( n = 6 , Figure 1D and Figure 1—figure supplement 2D ) . Additionally , we found that light-induced repositioning of cortical NuMA is sufficient to drive spindle rotational re-orientation ( Figure 1F and Video 3 ) , and that light-induced NuMA targeting also causes spindle displacement in 71 . 4% of Gαi ( 1 + 2 + 3 ) depleted cells ( n = 7 , Figure 1—figure supplement 2E–F ) . These results indicate that light-induced cortical recruitment of the Dynein-Dynactin-NuMA ( DDN ) complex is sufficient , and that LGN/Gαi are dispensable for generating cortical spindle-pulling forces in human cells . Cortical pulling forces are supposed to be generated by dynein-based motility on astral microtubules and/or astral microtubule depolymerization coupled with cortical anchorage ( Grill and Hyman , 2005 ) . To understand the contributions of astral microtubules to the spindle movement caused by light-induced cortical NuMA , we disrupted or stabilized astral microtubules using the microtubule-targeting drugs , nocodazole or taxol , respectively . In control cells , the metaphase spindle contains visible astral microtubules ( Figure 2A , left ) and is displaced following light-induced NuMA-RFP-Nano targeting ( Figure 2B , D–E ) . In contrast , when astral microtubules were selectively disrupted by treatment with 30 nM nocodazole ( Figure 2A , middle ) , the spindle was no longer displaced in 56% of cells ( n = 5/9 cells ) , and only partially displaced in the remaining 44% of cells ( n = 4/9 ) ( Figure 2C–E ) , despite presence of cortical dynein ( Figure 2C t = 5:30 ) . This suggests that astral microtubules are required for spindle pulling by the light-induced cortical DDN complex . Treatment with 10 μM taxol stabilized astral microtubules based on increases in both the length and number of astral microtubules 1 min after addition of taxol ( Figure 2A , right ) ( Rankin and Wordeman , 2010 ) . Importantly , even in the presence of 10 μM taxol , the spindle was gradually displaced toward the light-illuminated region ( Figure 2H-G , t = 5:00 ) . In these taxol-treated cells , the velocity of spindle movement was slower than that observed in control cells ( Figure 2F–I ) , suggesting that depolymerization of astral microtubules may also contribute to force generation , although this reduced velocity might be caused alternatively by cortical pushing by stabilized astral microtubules . In these experiments , we visualized spindle microtubules with 50 nM SiR-tubulin ( Lukinavičius et al . , 2014 ) , a fluorescent docetaxel derivative , and confirmed the presence of 10 μM taxol by the decrease of SiR-tubulin intensity ( Figure 2G t = 0:00 ) , likely due to competition for the same microtubule-binding site . These results suggest that the light-induced cortical DDN complex generates cortical pulling forces by using dynein-based motility on astral microtubules even if microtubule depolymerization is inhibited . Recently , ciliobrevin D was developed as a specific dynein inhibitor ( Firestone et al . , 2012 ) . This compound inhibits dynein’s microtubule gliding and ATPase activity , but not the association between ADP-bound dynein and microtubules in vitro . To understand the requirement of these dynein activities for force generation , we next sought to analyze spindle displacement following ciliobrevin D treatment . In HCT116 cells , we found that ciliobrevin D treatment in interphase caused mitotic phenotypes including chromosome misalignment similar to dynein degradation ( Natsume et al . , 2016 ) under 0 . 5% , but not 10% , FBS culture conditions ( Figure 3—figure supplement 1A–D ) , consistent with a previous report ( Firestone et al . , 2012 ) . We next added ciliobrevin D in metaphase-arrested cells . Although dynein activity is required to maintain spindle bipolarity , we found that spindle bipolarity was maintained for ~30 min following the treatment of ciliobrevin D , and was gradually disrupted during the subsequent 30–60 min ( Figure 3—figure supplement 1E–G ) . We next performed the optogenetic spindle-pulling assay during the initial 60 min of ciliobrevin treatment according to the Procedure depicted in Figure 3A . In control cells , light-induced targeting of NuMA displaced the spindle in 80% of cell ( n = 10 , Figure 3B and D ) . In contrast , the spindle was not displaced in 75% of ciliobrevin D-treated cells ( n = 12 , Figure 3C–D ) , whereas dynein was normally recruited to the cell cortex and the bipolar spindle structure was maintained during the assay . These results suggest that light-induced NuMA not only recruits , but also activates dynein at the cell cortex to generate cortical pulling forces . A dimerized version of the yeast dynein motor domain is sufficient to position microtubule asters in microfabricated chambers ( Laan et al . , 2012 ) . To understand the sufficiency of cortical dynein for generating spindle-pulling forces within a human cell , we next directly targeted dynein to the cell cortex ( Figure 3E ) . Similar to the NuMA-RFP-Nano fusion , endogenously tagged Nano-mCherry-DHC asymmetrically accumulated at the light-illuminated region within several minutes ( Figure 3F; Figure 3—figure supplement 1H ) , and subsequently recruited SNAP-tagged endogenous p150/dynactin to this cortical region ( Figure 3F–G; Figure 3—figure supplement 1I ) . However , endogenous NuMA-SNAP was not recruited to the light illuminated region ( Figure 3H; Figure 3—figure supplement 1I ) , and the spindle was not displaced toward dynein/dynactin-enriched cortex ( Figure 3F right , and Figure 3I ) despite the fact that substantial levels of dynein were recruited to the cell cortex ( compare Figure 3G to Figure 1C ) . These results suggest that cortical dynein targeting is not sufficient for generating cortical pulling forces in human cells , consistent with recent studies demonstrating that human dynein is auto-inhibited ( Torisawa et al . , 2014; Zhang et al . , 2017 ) and dynactin and cargo adaptors are required to activate dynein motility ( McKenney et al . , 2014; Schlager et al . , 2014; Zhang et al . , 2017 ) . Although we cannot exclude the possibility that iLID-Nano mediated cortical targeting of DHC may impair some cortical dynein functions or assemblies in human cells , cortical dynein anchoring with ePDZ-LOVp system in C . elegans is also insufficient to generate cortical pulling forces ( Fielmich et al . , 2018 ) . The above results suggest that NuMA is required to activate dynein at the cell cortex . Thus , we next sought to define the minimal functional region of NuMA as a dynein adaptor ( Figure 4A ) . Importantly , our truncation analyses revealed that the NuMA N-terminal region contains a Spindly-like motif sequence ( Figure 4B–E; Figure 4—figure supplement 1A–G ) that was recently identified as a conserved binding motif for the pointed-end complex of dynactin in dynein cargo adaptors ( Gama et al . , 2017 ) . We found that NuMA wild type ( WT ) fragment ( 1-705 ) , but not a Spindly-motif ( SpM ) mutant containing alanine mutations in the Spindly-motif ( Figure 4D ) , recruited dynein to the light-illuminated cortical region ( Figure 4F and Figure 4—figure supplement 1H ) . However , the NuMA ( 1-705 ) WT and longer NuMA ( 1–1700 ) fragments were unable to fully displace the spindle despite the presence of substantial levels of cortical dynein ( Figure 4B–C , H–I; Figure 4—figure supplement 1I–L ) , whereas ectopically expressed full length NuMA ( 1–2115 ΔNLS ) was able to displace the spindle in ~40% of cells ( Figure 4G; the NLS was deleted to eliminate dimerization with endogenous NuMA by spatially separating exogenously expressed constructs from the nuclear-localized endogenous NuMA before G2 release . In contrast , exogenously expressed NLS containing NuMA-RFP-Nano ( 1–2115 ) accumulated in the nucleus before G2 , but was unable to displace the spindle efficiently ( 11 . 1% , n = 9 ) , likely due to weak cortical anchorage by hetero-dimerization with endogenous NuMA lacking RFP-Nano ) . These results suggest that NuMA recruits dynein-dynactin via its N-terminal Spindly motif , likely to activate dynein’s motility at the mitotic cell cortex similarly to other dynein cargo adaptors ( Gama et al . , 2017; McKenney et al . , 2014; Schlager et al . , 2014 ) . However , despite this activation , additional NuMA domains are required to produce cortical spindle-pulling forces . At kinetochores , a multiplicity of microtubule-binding activities is required to generate cooperative pulling forces ( Cheeseman et al . , 2006; Schmidt et al . , 2012 ) . Because NuMA’s C-terminal region contains two microtubule-binding domains ( MTBD1 , and MTBD2 ) ( Figure 5A and Figure 5—figure supplement 1A ) ( Chang et al . , 2017; Du et al . , 2002; Gallini et al . , 2016; Haren and Merdes , 2002 ) , direct binding of NuMA to astral microtubules may generate cooperative forces in parallel with dynein-dynactin recruitment as recently proposed by Seldin et al ( Seldin et al . , 2016 ) . Consistent with this , we found that a Nano fusion with a NuMA ( 1–1895 ) fragment , which lacks both microtubule-binding domains , was unable to fully displace the spindle regardless of cortical dynein recruitment ( Figure 5B–C; Figure 5—figure supplement 1B ) . Similarly , NuMA ( 1–1985 ) , which lacks only the C-terminal microtubule-binding domain ( MTBD2 ) , was unable to displace the spindle ( Figure 5B , D; Figure 5—figure supplement 1C ) . In contrast , NuMA Δex24 , which lacks exon 24 thus disrupting MTBD1 and an NLS ( Figure 5A ) ( Gallini et al . , 2016; Seldin et al . , 2016; Silk et al . , 2009 ) , was able to recruit dynein and displace the spindle similarly to the NuMA-ΔNLS construct ( Figure 5B , E–F; Figure 4—figure supplement 1J ) . Because the corresponding mouse NuMA Δex22 mutant shows spindle orientation defects in mouse keratinocytes and the epidermis ( Seldin et al . , 2016 ) , this region may have specific roles in different cell types . Alternatively , weak defects in the NuMA Δex24 mutant may be suppressed by targeting increased levels of cortical NuMA Δex24 in this assay . These results indicate that NuMA’s microtubule binding domains , particularly MTBD2 , play critical roles for the ability of the DDN complex to generate spindle-pulling forces . The work described above defines two important molecular features for cortical force generation: dynein recruitment/activation through the Spindly-like motif and a distinct direct microtubule-binding activity by NuMA . To test whether these features are sufficient to generate cortical pulling forces , we next expressed a fusion construct , NuMA ( N + C ΔNLS ) , that contains both its dynein-recruiting N-terminal and microtubule-binding C-terminal domains , but lacks a ~1000 aa region of its central coiled-coil ( Figure 5A #12 ) . The NuMA fusion , but not the C-terminal domain ( 1700–2115 ) alone ( NuMA-C ) , recruited DHC-SNAP to the light-illuminated region ( Figure 5G–I; Figure 5—figure supplement 1D ) . However , the NuMA ( N + C ΔNLS ) fusion was unable to fully displace the spindle ( Figure 5G; Figure 4—figure supplement 1J ) . These results indicate that NuMA’s 200 nm long , central coiled-coil ( Harborth et al . , 1999 ) also functions with its N-terminal and C-terminal domains to efficiently capture and pull on astral microtubules . Our results reveal that NuMA has multiple functional modules for force generation . However , considering the sophisticated kinetochore structure that surrounds a plus-end of microtubule with multiple microtubule-binding proteins ( Cheeseman , 2014; Dimitrova et al . , 2016 ) , we next sought to define the architecture of the cortical attachment site that is required to efficiently capture and pull on dynamic plus-ends of astral microtubules . Intriguingly , we found that NuMA constructs containing its C-terminal region displayed punctate cortical signals , which tended to be even more evident in smaller constructs ( e . g . Figure 5H–I ) . These results suggest that NuMA forms oligomeric structures at the mitotic cell cortex as observed in vitro ( Harborth et al . , 1999 ) . To understand mechanisms of the NuMA’s C-terminal oligomerization/clustering at the mitotic cell cortex , we took advantages of a NuMA-C 3A fragment , which eliminates CDK phosphorylation sites ( Compton and Luo , 1995 ) allowing NuMA to localize to the metaphase cell cortex independently of LGN ( Kiyomitsu and Cheeseman , 2013 ) . Similar to the NuMA-C-RFP-Nano ( Figure 5I ) , GFP-NuMA-C 3A displayed punctate cortical signals ( Figure 6A–B #C1 ) , which was distinct from that of its cortical interacting partners - phospholipids and 4 . 1 proteins ( Kiyomitsu and Cheeseman , 2013; Kotak et al . , 2014; Mattagajasingh et al . , 2009; Zheng et al . , 2014 ) – that localize homogenously to the cell cortex ( Figure 6—figure supplement 1A–B ) . Interestingly , the punctate NuMA-C 3A patterns intercalated with cortical actin localization , and still localized following the disruption of actin polymerization ( Figure 6—figure supplement 1C ) . These results suggest that the NuMA C-terminal fragment self-assembles on the membrane independently of its cortical binding partners and actin cytoskeleton . Importantly , by analyzing different truncations and mutants , we found that a 100 aa region ( aa: 1700–1801 ) of NuMA adjacent to its 4 . 1 binding domain is required for the formation of punctate foci ( Figure 6A–B , compare #C1 to #C2 ) , and further that a highly conserved 10 amino acid region , E1768-P1777 ( Figure 6C ) , is necessary for cluster formation ( Figure 6B , see 5A-2 and 5A-3 alanine mutants; Figure 6—figure supplement 1D–F ) . Consistently , the 1700–1895 region of NuMA is required for the NuMA fragments to display punctate cortical signals ( compare Figure 4H to Figure 5C; Figure 6—figure supplement 1G ) . These results suggest an exciting possibility that NuMA assembles a specialized structure to produce large spindle-pulling forces at the cell cortex . Above we identified NuMA mutants ( 5A-2 , 5A-3 ) that are unable to form clusters at the mitotic cell cortex ( Figure 6B–C ) . To test the functional importance of the novel clustering behavior of NuMA , we next analyzed cortical force generation by full length NuMA wild-type ( WT ) compared to the 5A-3 mutant using Nano fusions . In cells expressing NuMA ( 1–2115 ΔNLS ) -RFP-Nano ( WT ) , NuMA and DHC-SNAP became gradually detectable as punctate foci ( Figures 6D , 4:30 and 11:00 ) , and the spindle was displaced towards the light-illuminated region ( Figures 6D , 13:00 ) . In contrast , when the NuMA 5A-3 mutant was targeted to the cell cortex , both NuMA 5A-3 mutant and DHC failed to form punctate foci ( Figure 6E; Figure 6—figure supplement 1H ) , similarly to GFP-NuMA-C 5A-3 ( Figure 6B ) , and the spindle was not fully displaced ( Figure 5B #14 , Figure 6E; Figure 6—figure supplement 1H ) . These results indicate that NuMA’s clustering activity correlates with the generation of cortical pulling forces . To further probe functional importance of the NuMA’s clustering activity , we next replaced endogenous NuMA with either NuMA WT or the 5A-3 mutant using the auxin-induced degron ( AID ) system ( Figure 7A ) ( Natsume et al . , 2016 ) . Consistent with the above results , endogenous NuMA fused to mAID-mClover-FLAG tag ( NuMA-mACF ) displayed punctate cortical signals that colocalized with dotted signals of SNAP-tagged dynein and LGN ( Figure 7—figure supplement 1A–C ) . When the endogenous NuMA-mACF was degraded , 80% of mitotic cells ( n = 63 ) displayed abnormal spindles with unfocused microtubules ( Figure 7B #2; Figure 7—figure supplement 1D–E ) , consistent with the NuMA KO phenotypes in human hTERT-RPE1 cells ( Hueschen et al . , 2017 ) . However , both NuMA WT and the 5A-3 mutant were able to rescue these abnormal spindle phenotypes ( Figure 7B #3 and #4 ) , suggesting that clustering of NuMA is dispensable for microtubule focusing at the spindle poles . In contrast , when endogenous NuMA was replaced with NuMA 5A-3 mutant , the metaphase spindle was tilted and randomly oriented on the x-z plane ( 26 . 8 ± 20 . 7° , n = 37 , Figure 7C; Figure 7—figure supplement 1F ) whereas the spindle in NuMA WT cells was oriented parallel to the substrate ( 10 . 7 ± 9 . 6° , n = 34 , Figure 7C ) as observed in control metaphase cells ( 11 . 5 ± 11 . 8° , n = 41 , Figure 7C ) . These results suggest that NuMA’s C-terminal clustering is required for proper spindle orientation . We note that the 5A-3 mutation site contains Y1774 ( Figure 6C ) , which is phosphorylated by ABL1 kinase and contributes to proper spindle orientation ( Matsumura et al . , 2012 ) . However , treatment with the ABL1 kinase inhibitor Imatinib caused only a mild spindle orientation phenotype ( 12 . 3 ± 14 . 7° , n = 27 , Figure 7C ) , suggesting that the spindle mis-orientation phenotype observed in the 5A-3 mutant is largely attributable to defects in NuMA clustering . Taken together , these results indicate that clustering activity of NuMA is required at the mitotic cell cortex , but not at the spindle poles , for generating cortical pulling forces . Thus , NuMA has a location-dependent structural function that clusters multiple DDN complexes to efficiently capture and pull on dynamic plus ends of astral microtubules .
Here , we applied a light-induced targeting system , iLID ( Guntas et al . , 2015 ) , for in cell reconstitution of the cortical force-generating machinery ( e . g . Figure 1A–B ) . Our work demonstrates that light-induced targeting of NuMA , but not dynein , is sufficient to control spindle position and orientation in human cells . This is consistent with recent findings that mammalian dynein requires cargo adaptors to activate its motility in vitro ( McKenney et al . , 2014; Schlager et al . , 2014; Zhang et al . , 2017 ) . In addition , our findings suggest that LGN/Gαi are dispensable for force generation , and instead act as receptors that specify the position of NuMA at the cell membrane . Consistent with this model , LGN-independent pathways that target NuMA to the cell cortex have been reported , such as Dishevelled ( Ségalen et al . , 2010 ) and phospho-lipids ( Zheng et al . , 2014 ) . Thus , we propose that the Dynein-Dynactin-NuMA ( DDN ) complex is a universal core unit that constitutes the cortical force-generating machinery , whereas LGN and other receptors specify the targeting of the DDN complex to the membrane . Our work demonstrates four distinct functions for NuMA at the mitotic cell cortex . First , NuMA recruits dynein-dynactin through its N-terminal region . We found that the conserved Spindly-like motif in NuMA is required for dynein recruitment ( Figure 4D–F ) . NuMA may directly interact with the dynactin pointed-end complex through this Spindly-like motif similarly to other dynein cargo adaptors ( Gama et al . , 2017 ) , and activate dynein motility at the mitotic cell cortex . Second , the central long coiled-coil of NuMA is required for spindle pulling ( Figure 5G ) . Purified NuMA displays a long ( ~200 nm ) rod-shaped structure that shows flexibility with a main flexible-linker region near the middle of central coiled coil ( Harborth et al . , 1999 ) . Longer flexible arms of NuMA may increase the efficiency of astral microtubule capture by the dynein-dynactin complex , similarly to fibril-like Ndc80 complexes and CENP-E motors at kinetochores ( Kim et al . , 2008; McIntosh et al . , 2008 ) . Third , NuMA contributes to cortical force generation with its own C-terminal microtubule-binding domains ( MTBDs ) ( Figure 5C ) , particularly MTBD2 ( Figure 5D ) . Because this region is also required to prevent hyper-clustering ( Figure 5—figure supplement 1C right ) , and is sufficient for cortical localization in anaphase ( Figure 6—figure supplement 1F C#3 , T . K . unpublished observation ) , this region may play multiple roles for cortical pulling-force generation . Interestingly , a NuMA C-terminal fragment containing MTBD1 ( aa: 1811–1985 , called NuMA-TIP , Figure 5—figure supplement 1A ) accumulates at microtubule tips , and remains associated with stalled and/or deploymerizing microtubules ( Seldin et al . , 2016 ) . By using its two microtubule-binding domains , NuMA may harness the energy of microtubule depolymerization for pulling on astral microtubules similar to the human Ska1 complex or yeast Dam1 ring complex at kinetochores , both of which track with depolymerizing microtubules ( Schmidt et al . , 2012; Westermann et al . , 2006 ) . Finally , we demonstrate that NuMA generates large pulling forces by clustering the DDN complexes through its C-terminal clustering domain ( Figure 6C–E ) , similar to lipid microdomains on phagosomes that achieve cooperative force generation of dynein ( Rai et al . , 2016 ) . Previous studies demonstrated that the 1700–2003 region of NuMA is required for oligomerization in vitro ( Harborth et al . , 1999 ) . We defined the 1700–1801 region of NuMA as a clustering domain ( CD ) required for clustering of NuMA-C 3A , and found that the CD containing 1700–1895 region of NuMA is sufficient for NuMA fragments to form clusters at the mitotic cell cortex when targeted as a Nano fusion ( Figure 4H and Figure 5C ) . Because this 1700–1895 fragment itself localizes to the cytoplasm , and showed no punctate signals ( Figure 6—figure supplement 1D and G #C6 ) , the clustering activity of this region may be enhanced by its recruitment and concentration at membranes , as observed for CRY2 clusters ( Che et al . , 2015 ) . Consistently , NuMA’s clustering function is required for spindle pulling at the cell cortex ( Figure 6E and Figure 7C ) , but not for microtubule focusing at spindle poles ( Figure 7B ) . Interestingly , spindle pole focusing requires both NuMA’s C-terminal microtubule binding and N-terminal dynein-dynactin binding modules , but not its central long coiled-coil ( Hueschen et al . , 2017 ) . Whereas NuMA-dynein complexes generate active forces within cells , NuMA’s multiple modules appear to be differently utilized depending on the context . Our live-cell imaging revealed that DDN clusters gradually assemble at the cell cortex and then displace the spindle ( Figures 5H and 6D ) . Based on the results obtained in this study , we propose a multiple-arm capture model of astral microtubules by the DDN clusters ( Figure 7D; Figure 7—figure supplement 1G ) . Following nuclear envelope break down , cytoplasmic NuMA and DDN complexes are recruited to the mitotic cell cortex by binding to the LGN/Gαi complex , and then assemble DDN clusters on the cell cortex via the NuMA C-terminal domain . In vitro , up to 10–12 NuMA dimers self-assemble and form ring-like structures with an average diameter of 48 ± 8 nm ( Harborth et al . , 1999 ) ( Figure 7—figure supplement 1G ) , which are similar to those of the central hub of yeast kinetochores ( 37 ± 3 nm ) ( Gonen et al . , 2012 ) , and of the Dam1 ring complex ( about 50 nm ) which encircles a single kinetochore microtubule ( Miranda et al . , 2005; Westermann et al . , 2005 ) . Given that the NuMA MTBD interacts with depolymerizing microtubules ( Seldin et al . , 2016 ) , dynein-dynactin moves along the lattice of microtubules , and astral microtubules tends to interact with the cell cortex through an end-on configuration in pre-anaphase cells ( Kozlowski et al . , 2007; Kwon et al . , 2015; Samora et al . , 2011 ) , it is tempting to speculate that the DDN cluster encircles or partially wrap the plus tip of a single astral microtubule with NuMA’s MTBDs , and holds the lateral wall of the astral microtubule with multiple dynein/dynactin-containing arms ( Figure 7D; Figure 7—figure supplement 1G ) . Future work using super-resolution imaging and in vitro reconstitution will reveal the precise architecture of the interaction between astral microtubule tips and the cortical DDN cluster . This multiple-arm capture by the DDN cluster leads to larger cooperative pulling forces by increasing the number of both dynein-dynactin containing modules and NuMA’s microtubule binding per an astral microtubule . Additionally , this clustering may contribute to force generation by increasing both the stability of the DDN complex at the membrane , and the frequency for dynein-dynactin to capture or re-bind to astral microtubules . Alternatively , astral microtubule binding of the DDN complex may also assist cluster formation on the cell cortex . To produce pulling forces at dynamic plus-ends of microtubules , the cortical force-generating machinery appears to develop multiple molecular and structural features analogous to the kinetochore ( Cheeseman , 2014; Dimitrova et al . , 2016 ) . In conclusion , our optogenetic reconstitution and AID-mediated replacement reveal that the cortical DDN cluster acts as a core spindle-pulling machinery in human cells . Analyzing the structure and regulation of the DDN cluster will provide further information to understand the basis of spindle positioning in both symmetric and asymmetric cell division , and the general principles for microtubule plus-end capture and pulling .
Plasmids for CRISPR/Cas9-mediated genome editing were constructed according to the protocol described in Natsume et al . , ( Natsume et al . , 2016 ) . To construct CRISPR/Cas9 vectors , pX330-U6-Chimeric_BB-CBh-hSpCas9 ( #42230 , Addgene , Cambridge , MA ) was used ( Ran et al . , 2013 ) . PAM and 20 bp single guide RNA sequences were selected by the optimized CRISPR design tool ( http://crispr . mit . edu ) ( Supplementary File 2 ) . To construct donor plasmids containing homology arms for NuMA ( ~500 bp homology arms ) , p150 ( ~200 bp arms ) and DHC ( N-terminal , ~500 bp arms ) , a gene synthesis service ( Genewiz , South Plainsfield , NJ ) was used . To construct the donor plasmid for DHC ( C-terminal ) , a ~2 , 000 bp sequence was amplified by PCR from genomic DNA and then cloned into the pCR2 . 1-TOPO vector . A BamHI site was introduced at the center of the 2 , 000 bp fragment to facilitate the subsequent introduction of cassettes encoding tag and selection marker genes . To express Mem-BFP-iLID from the AAVS1 locus , membrane-targeted BFP2 ( ‘Mem’ from Neuromodulin; Clontech , Mountain View , CA ) was fused to the N-terminus of iLID ( #60411 , Addgene ) with a 53-amino acid ( aa ) linker derived from pIC194 ( Kiyomitsu and Cheeseman , 2012 ) ( #44433 , Addgene ) , and the resulting fusion construct was introduced between the AfeI and HindIII sites in pMK231 ( AAVS1 CMV-MCS-Puro , #105924 , Addgene ) . Note that the Venus-iLID-caax construct ( #60411 , Addgene ) was able to recruit RFP-Nano , but not NuMA-RFP-Nano to the membrane . To construct the RFP-Nano-NeoR cassette , a tagRFPt-Nano fragment ( #60415 , Addgene ) was introduced between the SacI and MfeI sites in pMK277 ( #72793 , Addgene ) . The RFP-Nano-NeoR cassette was excised by BamHI and cloned into the BamHI site in the donor plasmid containing NuMA’s homology arms . A 24-aa linker sequence containing 4 × GGGS was introduced between the last codon of NuMA and the first codon of RFP . To construct the Nano-mCherry cassette , the Nano coding sequence was fused to the N-terminal region of mCherry from pIC194 with a 2 × GGGS linker . To express Nano-mCherry-DHC , the BSDR sequence from pIC242 ( Kiyomitsu and Cheeseman , 2012 ) ( #44432 , Addgene ) was linked to the Nano-mCherry sequence with a P2A sequence , and the resulting BSDR-P2A-Nano-mCherry cassette , which contained a BamHI site at each end , was inserted into the BamHI site of the donor plasmid for DHC ( N-terminal ) . A 47-aa linker sequence derived from pIC 194 was introduced between the last codon of mCherry and the start codon of DHC . To generate the SNAP-HygroR cassette , the mCherry coding sequence in pMK281 ( #72797 , Addgene ) was replaced with the SNAPf coding sequence ( N9186 , New England BioLabs , Ipswich , MA ) using In-Fusion® cloning ( Takara Bio , Ōtsu , Japan ) . The SNAP-HygroR cassette was excised by BamHI and cloned into the BamHI site of the donor plasmids . To make the DHC donor plasmid containing a SNAP-BSDR cassette , HygroR of the SNAP-HygroR cassette was replaced with BSDR from pIC242 using In-Fusion® cloning . To conditionally express NuMA-RFP-Nano constructs from the Rosa 26 locus , a fragment containing Tet-On 3G , the TRE3GS promoter , and a multiple cloning site ( MCS ) derived from pMK240 ( Tet-On-AAVS1-MCS-PuroR , #105925 , Addgene ) was introduced into pMK247 ( Rosa26-CMV-MCS-HygroR , #105926 , Addgene ) , which contains homology arms for the Rosa 26 locus . An RFP-Nano coding sequence was integrated between MluI and AgeI in the MCS , and NuMA fragments were subsequently inserted into the MluI site . NuMA truncation fragments and mutants were generated by PCR using NuMA cDNA ( Compton and Luo , 1995; Kiyomitsu and Cheeseman , 2012 ) as a template , and the sequences were confirmed by DNA sequencing . These NuMA fragments encode isoform 2 ( aa: 1–2101 ) , which lacks a 14-aa region ( aa: 1539–1552 ) in the longer isoform 1 . However , the human NuMA constructs presented in the present study conform to isoform 1 ( aa: 1–2115; NP_006176 ) to avoid confusion . To construct mAID-mClover-3×FLAG NeoR , a 3 × FLAG sequence with a GGGS linker was introduced at the C-terminus of mClover of pMK289 ( #72827 , Addgene ) by PCR . To conditionally express mCherry-NuMA WT or the 5A-3 construct from Rosa 26 locus , a fragment containing the TRE3GS promoter and the MCS derived from pMK240 was introduced into pMK247 . The mCherry coding sequence derived from pIC 194 was integrated between the MluI and AgeI sites in the MCS , and the NuMA fragments were subsequently inserted Between the SalI and AgeI site . HCT116 and HeLa cells were cultured as described previously ( Kiyomitsu and Cheeseman , 2012; Natsume et al . , 2016; Tungadi et al . , 2017 ) . No mycoplasma contamination was detected by MycoAlert Mycoplasma Detection Kit ( Lonza ) . Knock-in cell lines were generated according to the procedures described in Natsume et al . , ( Natsume et al . , 2016 ) with minor modifications . CRISPR/Cas9 and donor plasmids were transfected into the cell lines using Effectene ( Qiagen , Venlo , Netherlands ) . For drug selection , 1 μg/mL puromycin ( Wako Pure Chemical Industries , Osaka , Japan ) , 800 μg/mL G418 ( Roche , Basel , Switzerland ) , 200 μg/mL hygromycin B ( Wako Pure Chemical Industries ) , and 8 μg/mL blasticidin S hydrochloride ( Funakoshi Biotech , Tokyo , Japan ) were used . Selection medium was replaced with fresh selection medium 4–5 days after starting selection . After 10–14 days , colonies grown on a 10 cm culture dish were washed once with PBS , picked up with a pipette tip under a microscope ( EVOS XL , Thermo Fisher Scientific , Waltham , MA ) located on a clean bench , and subsequently transferred to a 96-well plate containing 50 μL of trypsin-EDTA . After a few minutes , these trypsinized cells were transferred to a 24-well plate containing 500 μL of the selection medium , and then further transferred to a 96-well plate ( 200 μL per well ) for the preparation of genomic DNA . The remaining cells in the 24-well plate were grown and frozen using Bambanker Direct ( Nippon Genetics , Tokyo , Japan ) . For the preparation of genomic DNA , cells in the 96-well plate were washed once with PBS and then mixed with 60 μL of DirectPCR® lysis solution ( Viagen Biotech , Los Angeles , CA ) containing 0 . 5 mg/mL proteinase K ( Wako Pure Chemical Industries ) . The 96-well plate was sealed with an aluminum plate seal and incubated first at 56°C for 5–6 hr , then at 80°C for 2–3 hr in a water bath . To confirm the genomic insertion , PCR was performed using 1–2 μL of the genomic DNA solution and Tks Gflex DNA polymerase ( Takara Bio ) . The cell lines and primers used in this study are listed in Supplementary Files 1 and 3 , respectively . Antibodies against tubulin ( DM1A , Sigma-Aldrich , 1:2 , 000 ) , NuMA ( Abcam , 1:1 , 000 ) , DHC ( Santa Cruz Biotechnology , 1:500 ) , p150 ( BD Transduction Laboratories , 1:1 , 000 ) , SNAP ( New England BioLabs , 1:1 , 000 ) , LGN ( BETHYL Laboratories , 1:2 , 000 ) , Gαi-1 ( Santa Cruz Biotechnology , 1:100 ) , OsTIR1 ( Kanemaki Laboratory , 1:1 , 000 ) , and H3S10P ( Abcam , 1:500 ) were used for western blotting . Imaging was performed using spinning-disc confocal microscopy with a 60 × 1 . 40 numerical aperture objective lens ( Plan Apo λ , Nikon , Tokyo , Japan ) . A CSU-W1 confocal unit ( Yokogawa Electric Corporation , Tokyo , Japan ) with three lasers ( 488 , 561 , and 640 nm , Coherent , Santa Clara , CA ) and an ORCA-Flash4 . 0 digital CMOS camera ( Hamamatsu Photonics , Hamamatsu City , Japan ) were attached to an ECLIPSE Ti-E inverted microscope ( Nikon ) with a perfect focus system . A stage-top incubator ( Tokai Hit , Fujinomiya , Japan ) was used to maintain the same conditions used for cell culture ( 37°C and 5% CO2 ) . For light illumination , a Mosaic-3 digital mirror device ( Andor Technology , Belfast , UK ) and a 488 nm laser ( Coherent ) were used . The microscope and attached devices were controlled using Metamorph ( Molecular Devices , Sunnyvale , CA ) . For immunofluorescence in Figure 2A , cells were fixed with PBS containing 3% paraformaldehyde and 2% sucrose for 10 min at room temperature . Fixed cells were permeabilized with 0 . 5% Triton X-100 for 5 min on ice , and pretreated with PBS containing 1% BSA for 10 min at room temperature after washing with PBS . Microtubules and DNA were visualized using 1:1000 anti-α-tubulin antibody ( DM1A , Sigma-Aldrich , St . Louis , MO ) and 1:5000 SiR-DNA ( Spirochrome ) , respectively . Images of multiple z-sections were acquired by spinning-disc confocal microscopy using 0 . 2 μm spacing and camera binning 1 . Maximally projected images from 15 z-sections were generated with Metamorph . For time-lapse imaging of living cells , cells were cultured on glass-bottomed dishes ( CELLview , #627870 , Greiner Bio-One , Kremsmünster , Austria ) and maintained in a stage-top incubator ( Tokai Hit ) to maintain the same conditions used for cell culture ( 37°C and 5% CO2 ) . Three z-section images using 0 . 5 μm spacing were acquired every 30 s with camera binning 2 . Maximally projected z-stack images were shown in figures unless otherwise specified . Microtubules and actin were stained with 50 nM SiR-tubulin and 50 nM SiR-actin ( Spirochrome ) , respectively , for >1 hr prior to image acquisition . DNA was stained either 20 nM SiR-DNA ( Spirochrome ) or 50 ng/mL Hoechst 33342 ( Sigma-Aldrich ) for >1 hr before observation . To visualize SNAP-tagged proteins , cells were incubated with 0 . 1 μM SNAP-Cell 647 SiR or TMR-STAR ( New England BioLabs ) for >2 hr , and those chemical probes were removed before observation . For drug treatment , cells were incubated with drugs at the following concentrations and duration: nocodazole , 330 nM ( high dose ) for 18–24 hr and 30 nM ( low dose ) for 1–4 hr; paclitaxel , 10 μM for 1–10 min; cytochalasin D , 1 μM for 1–10 min; MG132 , 20 μM for 1–4 hr ( Figure 4—figure supplement 1B ) ; RO-3306 , 9 μM for 20 hr; imatinib , 10 μM for 24 hr ( Matsumura et al . , 2012 ) ; doxycycline hyclate ( Dox ) , 2 μg/mL ( Figure 4—figure supplement 1B ) ; Ciliobrevin D , 75 μM . To express NuMA-RFP-Nano constructs from the Rosa 26 locus in LGN-depleted cells , cells were treated with LGN siRNA ( Kiyomitsu and Cheeseman , 2012 ) and Dox at 24 hr and 48 hr , respectively , according to the procedure described in Figure 4—figure supplement 1B . RO-3306 was added at 48 hr to cells that were then synchronized at G2 at 68 hr . The NuMA-RFP-Nano fusion protein was expressed in most cells , but its expression frequency was reduced in cells that expressed longer NuMA fragments . siRNAs targeting Gαi-1 isoforms ( Kiyomitsu and Cheeseman , 2012 ) were obtained from Dhamacon . To compare the intensities of cortically targeted NuMA-Nano fusions , images of NuMA-Nano fusions and DHC-SNAP were acquired using the same parameters ( Exposure time: NuMA , 1000 msec; DHC , 500 msec ) , except for Figure 1B ( NuMA , 1500 msec; DHC , 500 msec ) . To optimize image brightness , same linear adjustments were applied using Fiji and Photoshop . Supplemental movie files were generated using Metamorph and Fiji . To activate the auxin-inducible degradation of NuMA-mAID-mClover-3FLAG ( mACF ) , cells were treated with 2 μg/mL Dox and 500 μM indoleacetic acid ( IAA ) for 20–24 hr . Cells with undetectable mClover signals were analyzed . A small population of cells showed mClover signals even after being treated with Dox and IAA . For replacement experiments , either mCherry-NuMA WT or the 5A-3 mutant was expressed from the Rosa 26 locus following Dox treatment . This caused the cells to simultaneously express OsTIR1 from the AAVS1 locus to initiate the auxin-inducible degradation of endogenous NuMA-mACF . Except for Figure 1—figure supplement 1B , HCT116 cells expressing Mem-BFP-iLID and NuMA-Nano fusion proteins were treated with RO-3306 and MG-132 according to the procedure described in Figure 4—figure supplement 1B to increase the proportion of metaphase-arrested cells . To target Nano fusion proteins at the metaphase cell cortex , cells were illuminated using a Mosaic-3 digital mirror device ( Andor Technology ) at the indicated regions ( circles with a diameter of 1 . 95 μm for Figure 1—figure supplement 1B , and that of 2 . 82 μm for other figures ) with a 488 nm laser pulse ( 500 msec exposure , 25 mW ) . To manually control the frequency of the light pulse and the position of the illuminated region during time-lapse experiments , a custom macro was developed using Metamorph . Using this macro , indicated regions were illuminated ~10 times with the light pulse during time intervals ( 30 s ) between image acquisitions . The illuminated position was adjusted to precisely illuminate the cortical region of each cell . In response to the expression level of the Nano fusion proteins , the frequency of the light pulse was reduced to prevent the targeting of Nano fusion proteins throughout the cell cortex . To reposition NuMA-RFP-Nano at the mitotic cell cortex in Figure 1F , a cortical region adjacent to the spindle axis was illuminated . The light-illuminated region was changed once the spindle started to move but before the spindle was completely attached to the cell cortex . Spindles that rotated by approximately 90° within 15 min were counted . Cortical and cytoplasmic fluorescence intensities were determined using Fiji by calculating the mean pixel intensity along three different straight lines ( length 3 μm , width three pixels ) drawn along the cell cortex showing Nano signals or the cytoplasm near the cell cortex but without any aggregations . The background intensity was subtracted from each measurement . The distance from the pole to the cell cortex was measured using Metamorph or Fiji . Line scans for cortical fluorescence intensity were generated using Fiji by calculating the mean pixel intensity along the segmented line ( width three pixels ) drawn along the cell cortex . Kymographs were generated using Photoshop ( Adobe Systems , San Jose , CA ) . Spindle displacement was judged by the definition given in Figure 4—figure supplement 1I . In addition , cells that satisfied the following conditions were analyzed; ( 1 ) NuMA-RFP-Nano fusion proteins were asymmetrically recruited at the light-illuminated region , but not distributed to a whole cell cortex . ( 2 ) The cortical intensities of NuMA-Nano fusion proteins were higher than that of NuMA Δex24-RFP-Nano ( Figure 5F ) . ( 3 ) DHC-SNAP was detectable at the light-illuminated region except for the case of the cortical targeting of NuMA-C ( #13 ) . ( 4 ) The spindle was monitored for >10 min , and not vertically rotated . ( 5 ) The bipolar spindle was properly formed without severe membrane blebbing . To determine the significance of differences between the mean values obtained for two experimental conditions , Student’s t-tests or Mann-Whitney tests ( Prism 6; GraphPad Software , La Jolla , CA ) were used as indicated in the figure legends . | Almost every time a cell divides , it must share copies of its genetic material between two new daughter cells . A large molecular machine called the mitotic spindle makes this happen . The spindle is made of protein filaments known as microtubules that radiate out from two points at opposite ends of the cell . Some of these filaments attach to the genetic material in the center of the cell; some extend in the other direction and anchor the spindle to the cell membrane . The anchoring filaments – also known as astral microtubules – can position the mitotic spindle , which controls whether the cell splits straight down the middle ( to give two identically sized cells ) or off-center ( which gives cells of different sizes ) . The force required to move the spindle comes from complexes of proteins under the cell membrane that contain a molecular motor called dynein , its partner dynactin , and three other proteins – including one called NuMA . The astral microtubules interact with this force-generating machinery , but it was unclear how these proteins are arranged at the membrane . One way to explore interactions in a protein complex is to use a light-induced reconstitution system . This technique involves molecules that will bind together whenever a light shines on them . Fusing these molecules with different proteins means that experimenters can control exactly where , and when , those proteins interact . Okumura et al . have now used a light-induced reconstitution system to understand how the force-generating machinery positions the spindle in human cells . One of the system’s molecules was fused to a protein located at the cell membrane; the other was fused to either the dynein motor or NuMA protein . Using light to move dynein around on the membrane did not move the spindle . Yet , changing the position of NuMA , by moving the light , was enough to rotate the spindle inside the cell . Understanding how these complexes of proteins work increases our understanding of how cells divide . Using the light-induced system to move the spindle could also reveal more about the role of symmetric and asymmetric cell division in organizing tissues . In particular , being able to manipulate the position and size of daughter cells will provide insight into how cell division shapes and maintains tissues during animal development . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2018 | Dynein–Dynactin–NuMA clusters generate cortical spindle-pulling forces as a multi-arm ensemble |
Prominent theories suggest that compulsive behaviors , characteristic of obsessive-compulsive disorder and addiction , are driven by shared deficits in goal-directed control , which confers vulnerability for developing rigid habits . However , recent studies have shown that deficient goal-directed control accompanies several disorders , including those without an obvious compulsive element . Reasoning that this lack of clinical specificity might reflect broader issues with psychiatric diagnostic categories , we investigated whether a dimensional approach would better delineate the clinical manifestations of goal-directed deficits . Using large-scale online assessment of psychiatric symptoms and neurocognitive performance in two independent general-population samples , we found that deficits in goal-directed control were most strongly associated with a symptom dimension comprising compulsive behavior and intrusive thought . This association was highly specific when compared to other non-compulsive aspects of psychopathology . These data showcase a powerful new methodology and highlight the potential of a dimensional , biologically-grounded approach to psychiatry research .
Compulsivity is a theoretical clinical phenomenon that reflects the loss of control over repetitive self-deleterious behavior seen in a range of disorders , most notably obsessive-compulsive disorder ( OCD ) and addiction ( Everitt and Robbins , 2005; Gillan and Robbins , 2014 ) . But what are the underlying neural , computational , or psychological mechanisms ? Researchers have suggested that compulsivity in these disorders may be partially explained by an imbalance between two different modes of control , which are more and less flexible ( Everitt and Robbins , 2005; Graybiel and Rauch , 2000 ) . In particular , a deficit in deliberative , ‘goal-directed’ control may leave individuals vulnerable to rely excessively on forming more rigid habits . Habits are behaviors that animals and humans learn to execute automatically when presented with familiar environmental cues ( Dickinson , 1985 ) . While habits are typically very useful , allowing us to efficiently perform routine actions while expending minimal cognitive effort , they cannot adapt flexibly to new situations . To override our habits , organisms are capable of ‘goal-directed behavior’ . This refers our ability to make more considered choices , reflecting both ( i ) knowledge of the outcomes that our actions typically produce and ( ii ) our current motivation for those outcomes ( Dickinson and Balleine , 1994 ) . Consistent with the hypothesis that compulsion is linked to an imbalance between these modes of control , deficits in goal-directed learning have been observed across a range of putatively compulsive disorders such as drug addiction ( Sjoerds et al . , 2013; Voon et al . , 2015 ) , obsessive-compulsive disorder ( OCD ) ( Voon et al . , 2015; Gillan et al . , 2011; 2014a; 2014b; 2015a ) and also binge-eating disorder ( Voon et al . , 2015 ) . These deficits in goal-directed control have been linked to abnormal structure and function of the caudate and medial orbitofrontal cortex ( Voon et al . , 2015; Gillan et al . , 2015a ) , suggesting that they may be a promising target for understanding the etiology of these disorders and thus for future treatment development . Critically , the scope of the relationship between goal-directed learning deficits and psychopathology , and particularly their specificity to compulsive versus non-compulsive aspects has not been established . In fact , a similar deficit in goal-directed control was recently reported in other patient groups ( Alvares et al . , 2014; Morris et al . , 2015 ) , including those diagnosed with social anxiety disorder and schizophrenia , at least the former of which is not characterized by repetitive compulsive acts . This casts serious doubt over the hypothesis that goal-directed deficits are a neurocognitive mechanism that is partly responsible for psychiatric compulsivity . This lack of specificity is unfortunately ubiquitous in psychiatry research ( Lipszyc and Schachar , 2010; Bickel et al . , 2012 ) , a result , we suggest , of the broader issue that psychiatric diagnostic categories do not reflect the most discrete and neurobiologically informative phenomena . Of particular relevance to the present study are the high rates of co-morbidity between OCD and social anxiety disorder ( Ruscio et al . , 2010 ) , the preponderance of OCD symptoms in the schizophrenia poopulation ( Poyurovsky and Koran , 2005 ) , and more broadly that the vast majority of patients diagnosed with obsessive-compulsive disorder ( OCD ) meet the criteria for another lifetime psychiatric disorder ( Ruscio et al . , 2010 ) . Given these major overlaps , dissociating the neurocognitive bases for these respective diagnostic categories in their current form may be untenable . Indeed , the Diagnostic and Statistical Manual of Mental Disorders ( DSM ) , now in it’s fifth edition ( American Psychiatric Association , 2013 ) was developed to provide a reliable , descriptive psychiatric taxonomy , rather than an etiologically valid one . As such it is difficult to clearly discriminate the diagnostic categories it defines on the basis of genetics , neuroimaging , or indeed any of the modern tools of cognitive neuroscience . These issues have been described in detail by others ( Cuthbert and Kozak , 2013; Hyman , 2007; Robbins et al . , 2012 ) , and have been recognized by the National Institute of Mental Health ( NIMH ) , which has launched the Research Domain Criteria ( RDoC ) initiative , aiming to identify biologically plausible , trans-diagnostic markers of psychiatric disturbances ( Insel et al . , 2010 ) . Although progress towards this goal has already been made by studies examining dissociable clusters of patients within groups diagnosed with the same disorder ( Brodersen et al . , 2014; Fair et al . , 2012 ) , the identification of robust , generalizable and specific markers that contribute to psychiatric co-morbidity has been curtailed by the small sample sizes that are typical of patient studies . Accordingly , we hypothesized that a dimensional approach leveraging the efficiencies of large-scale online data collection among healthy individuals could be used to determine the precise psychiatric phenotype associated with deficits in goal-directed control , and test the specificity of this relationship with respect to other aspects of psychopathology . We hypothesized that this phenotype would broadly relate to compulsive behavior , which is seen across multiple disorders , including OCD and addiction ( Gillan et al . , 2015b ) , but were interested to reveal the scope and generality of this , e . g . with respect to impulsivity , a putatively related clinical phenotype ( Robbins et al . , 2012 ) . We also wished to study how any psychiatric correlates of goal-directed control relate to variation in age and IQ , two more general factors that have been shown to covary both with goal-directed control and with some aspects of psychopathology ( Eppinger et al . , 2013; Schad et al . , 2014; Sandstrom et al . , 1998 ) . To this end , rather than diagnosed patients , we used two large general-population samples collected online via Amazon’s Mechanical Turk ( AMT ) to test ( i ) if compulsivity as indicated by self-report OCD symptoms is associated with individual differences in goal-directed learning , ( ii ) if this association generalizes to self-report symptoms of other DSM diagnostic categories that involve compulsivity , and ( iii ) if this association is specific to compulsive versus non-compulsive psychopathology . Goal-directed control has recently been computationally formalized as arising from a form of reinforcement learning known as ‘model-based’ ( Daw et al . , 2011 ) , which can be expressed as an individual difference measure that has been shown to predict how likely individuals are to form habits ( Gillan et al . , 2015c ) . Using this well-validated task ( Figure 1 ) ( Daw et al . , 2011 ) , we found support for all three postulates . In Experiment 1 , we found that total scores on a self-report questionnaire measuring the severity of OCD symptoms were tracked by normal variation in model-based learning in the general population , but not by self-report anxiety or depression symptoms . In Experiment 2 , we replicated the association with self-report OCD symptoms and showed that it generalized to a broader set of psychiatric symptoms that similarly involve failures in exerting control over self-deleterious behaviors , specifically alcohol addiction , eating disorders and impulsivity . Once again , we found tentative evidence for specificity with respect to non-compulsive aspects of psychopathology . Next , we conducted a factor analysis , which indicated the existence of three latent symptom dimensions that cut across the nine different questionnaires assessed in this study . Crucially , the second symptom dimension identified was characterized by ‘Compulsive Behavior and Intrusive Thought’ , in which items were most consistently drawn from the questionnaires assessing symptoms of OCD , eating disorders and addiction , pertaining not just to repetitive compulsive behaviors ( as was our prediction ) , but also to associated preoccupations and cognitive distortions . This factor , which was defined independently of task performance , was a significant predictor of deficits in model-based learning . Crucially , this effect was highly specific to this factor , when directly compared to the two other factors identified in this analysis , ‘Anxious-Depression’ and ‘Social Withdrawal’ . 10 . 7554/eLife . 11305 . 003Figure 1 . Two-step reinforcement learning task used to assess goal-directed ( model-based ) learning . ( a ) Subjects chose between two fractals , which probabilistically determined whether they would transition to the orange or blue second stage state . For example , the fractal on the left had a 70% chance of leading to the blue second stage state ( ‘common’ transition ) and a 30% chance of leading to the orange state ( ‘rare’ transition ) . These transition probabilities were fixed and could be learned over time . In the second stage state , subjects chose between two fractals , each of which was associated with a distinct probability of being rewarded with a 25 cents coin . The probability of receiving a reward associated with each second stage fractal could also be learned , but ( unlike the transition structure ) these drifted slowly over time ( 0 . 25 < P <0 . 75 , panel b ) . This meant that in order to earn the most rewards possible , subjects had to track which second stage fractals were currently best as they changed over time . Reward probabilities depicted ( 34% , 68% , 72% , 67% ) refer to example trial 50 , denoted by the vertical dashed line in ( b ) . ( b ) Drifting reward probabilities determined by Gaussian Random Walks for 200 trials with grey horizontal lines indicating boundaries at 0 . 25 and 0 . 75 . ( c ) Schematic representing the performance of a purely ‘model-free’ learner , who only exhibits sensitivity to whether or not the previous trial was rewarded vs . unrewarded , and does not modify their behavior in light of the transition that preceded reward . ( d ) Schematic representing the performance of a purely ‘model-based’ learner , who is more likely to repeat an action ( i . e . ‘stay’ ) following a rewarded trial , only if the transition was common . If the transition to that rewarded state was rare , they are more likely to switch on the next trial . DOI: http://dx . doi . org/10 . 7554/eLife . 11305 . 003
In Experiment 1 , we tested the hypothesis that individual differences in total scores on a questionnaire assessing the severity of OCD symptoms are associated with normal variation in goal-directed control , rather than necessitating the categorical comparison of OCD patient vs . control groups . Participants ( N = 548 ) first completed a reinforcement-learning task that quantifies individual differences in goal-directed ( ‘model-based’ ) learning , which is operationalized as a parameter estimate from a logistic regression analysis predicting choices in the task ( see Materials and methods and refs [Daw et al . , 2011; Gillan et al . , 2015c] ) . Next , we administered a short Intelligence Quotient ( IQ ) test , followed by self-report questionnaires assessing symptoms of OCD , along with depression and trait anxiety , which we did not expect to be associated with goal-directed deficits . In line with our hypothesis , there was a significant association between scores on the OCD questionnaire and goal-directed deficits ( i . e . a negative relationship between OCD severity and model-based learning; β = −0 . 040 , Standard Error ( SE ) = 0 . 02 , p=0 . 049 ) when ( as in all analyses reported henceforth ) controlling for age , IQ and gender , which have been previously reported to covary with goal-directed behavior ( Eppinger et al . , 2013; Schad et al . , 2014; Sandstrom et al . , 1998 ) . Specifically , for each increase of 1 standard deviation ( SD ) in the total score on the OCD questionnaire , model-based learning was reduced by 14% . No such relationship was observed for self-report depression ( β = −0 . 016 , SE = 0 . 02 , p=0 . 439 ) or trait anxiety ( β = −0 . 006 , SE = 0 . 02 , p=0 . 777 ) severity ( Table 1 , Figure 2A ) . Moreover , the relationship between total scores on the OCD questionnaire and goal-directed deficits survived inclusion of the depression and trait anxiety total scores in the same model as covariates ( β = −0 . 048 , SE = 0 . 02 , p=0 . 04 ) . These data indicate that deficits in goal-directed control are a marker of normal variation in OCD symptomatology in the general population . 10 . 7554/eLife . 11305 . 004Table 1 . Self-report questionnaire total scores and model-based learning . DOI: http://dx . doi . org/10 . 7554/eLife . 11305 . 004Questionnaireβ ( SE ) z-valuep-valueExperiment 1 ( N=548 ) OCD-0 . 040 ( 0 . 02 ) -1 . 970 . 049 *Depression-0 . 016 ( 0 . 02 ) -0 . 770 . 439Trait Anxiety-0 . 006 ( 0 . 02 ) -0 . 280 . 778Experiment 2 ( N=1413 ) Eating Disorders-0 . 037 ( 0 . 01 ) -3 . 30<0 . 001 ***Impulsivity-0 . 034 ( 0 . 01 ) -3 . 010 . 007 **OCD-0 . 026 ( 0 . 01 ) -2 . 340 . 020 *Alcohol Addiction-0 . 025 ( 0 . 01 ) -2 . 180 . 029 *Schizotypy-0 . 017 ( 0 . 01 ) -1 . 480 . 14Depression-0 . 010 ( 0 . 01 ) -0 . 870 . 385Trait Anxiety-0 . 008 ( 0 . 01 ) -0 . 680 . 498Apathy-0 . 001 ( 0 . 01 ) -0 . 060 . 953Social Anxiety0 . 008 ( 0 . 01 ) 0 . 680 . 496*p<0 . 05; **p<0 . 01; ***p<0 . 001 . SE=standard error . Each row reflects the results from an independent analysis where each questionnaire total score ( z-transformed ) was entered as SymptomScorez in the following model: glmer ( Stay ~ Reward * Transition * SymptomScorez + Reward * Transition * ( IQz + Agez + Gender ) + ( Reward * Transition + 1 | Subject ) ) . Model-based statistics refer to the following interaction: SymptomScorez x Reward x Transition . For each , positive β values indicate that the symptom score is associated with greater model-based learning , while negative β values indicate that the symptom score is associated with reduced model-based learning . 10 . 7554/eLife . 11305 . 005Figure 2 . Associations between Goal-directed ( model-based ) deficits and self-reported psychopathology . The y-axes indicate the% change in model-based learning for each change of 1 standard deviation ( SD ) of clinical symptoms . Error bars denote standard error . ( a ) In Experiment 1 , total scores on a self-report questionnaire assessing OCD symptoms in a general population sample were associated with deficits in goal-directed ( model-based ) learning . Specifically , for each increase of 1 SD in OCD symptoms reported , model-based learning was 14% lower than the group mean . No effects were observed in depression or trait anxiety . ( b ) In Experiment 2 , the results from Experiment 1 were replicated: OCD symptoms were associated with deficits in goal-directed learning , while total scores on questionnaires assessing depression and trait anxiety were not . We found that the association between compulsive behavior and goal-directed deficits generalized to symptoms associated with other disorders that are similarly characterized by a loss of control over behavior , alcohol addiction , eating disorders and impulsivity . No significant effects were observed for scores on questionnaires assessing schizotypy , depression , trait anxiety , apathy or social anxiety . DOI: http://dx . doi . org/10 . 7554/eLife . 11305 . 005 In Experiment 2 , we aimed to test the reliability , generalizability and specificity of this finding in a larger cohort of task-naïve subjects ( N = 1413 , based on a power analysis given the aforementioned results ) . The procedure was identical to that in Experiment 1 , except for the addition of several more clinical questionnaires . To test for generalizability , we assessed symptoms associated with other disorders that have been hypothesized to have compulsive features . In addition to the OCD questionnaire used in Experiment 1 , these pertained to alcohol addiction , eating disorders , along with aspects of impulsivity and schizotypy ( Everitt and Robbins , 2005; Poyurovsky and Koran , 2005; Robbins et al . , 2012; Godier and Park , 2014 ) . To test for specificity , in addition to the mood symptoms assessed previously ( depression and trait anxiety ) we added self-report measures assessing social anxiety and apathy; we also predicted that non-compulsive aspects of schizotypy and impulsivity might not be associated with goal-directed control . In this independent sample , we replicated the results from Experiment 1; scores on the OCD questionnaire were significantly associated with goal-directed deficits ( β=−0 . 026 , SE=0 . 01 , p=0 . 020 ) , while controlling for age , gender and IQ ( Table 1 , Figure 2B ) . As we hypothesized , this effect generalized to phenotypically disparate manifestations of psychiatric compulsivity: total scores on self-report measures of eating disorder severity ( β=−0 . 037 , SE=0 . 01 , p<0 . 001 ) , impulsivity ( β=−0 . 034 , SE=0 . 01 , p=0 . 007 ) and alcohol addiction ( β=−0 . 025 , SE=0 . 01 , p=0 . 029 ) . Also as predicted , we found no significant associations between goal-directed deficits and total scores on the depression ( β=−0 . 01 , SE=0 . 01 , p=0 . 385 ) , apathy ( β=−0 . 001 , SE=0 . 01 , p=0 . 953 ) , trait anxiety ( β=−0 . 008 , SE=0 . 01 , p=0 . 498 ) or social anxiety ( β=0 . 008 , SE=0 . 01 , p=0 . 496 ) questionnaires . We found no significant association between self-report levels of schizotypy and goal-directed control ( β=−0 . 017 , SE=0 . 01 , p=0 . 14 ) , possibly reflective of the great deal of heterogeneity within this particular psychiatric construct . Previous studies using this task have assessed an individual’s goal-directed learning in either of two ways: predicting their choices using either a regression model ( as reported above ) or the fit of a more elaborate computational learning model , which the regression model approximates . In separate analyses using fits of the computational learning model ( see Materials and methods; Supplementary file 5A ) , all of the aforementioned results were recapitulated , with the exception that the relationship between OCD and model-based learning in Experiment 1 fell short of significance ( but was significant in Experiment 2 ) and schizotypy reached significance in Experiment 2 as a negative predictor of model-based learning . Given both the heterogeneity within , and the high correlation across these questionnaires ( e . g . , Depression and Trait Anxiety scores correlate at r=0 . 81 ) these questionnaires , assessing the statistical specificity of these effects by including their total scores in the same model is both methodologically and conceptually fraught . To address this issue , we conducted a factor analysis based on the 209 individual questionnaire items , thereby reducing the collinearity across scores on these psychiatric questionnaires . Note that this analysis was carried out on the questionnaire scores alone , without reference to the results on the reinforcement learning task . We found evidence for three dissociable factors ( ‘dimensions’ ) that cut across the nine questionnaires from which items were drawn , which we labeled ‘Anxious-Depression’ , ‘Compulsive Behavior and Intrusive Thought’ and ‘Social Withdrawal’ , based on the loadings of individual items ( Supplementary file 2A–C , Table 2 , Figure 3A ) . Although the labeling of factors is of course a subjective process , quantitatively speaking , ‘Compulsive Behavior and Intrusive Thought’ had high and consistent loadings from almost all items pertaining to eating disorders ( Mean loading=0 . 36 , SD=0 . 15 ) , OCD ( Mean loading=0 . 50 , SD=0 . 06 ) and addiction ( Mean loading=0 . 31 , SD=0 . 07 ) , which have all been couched as ‘compulsive’ disorders in the literature ( Everitt and Robbins , 2005; Gillan and Robbins , 2014; Godier and Park , 2014 ) ( Table 2 ) . In addition to picking up every self-report item that pertained to compulsive behavior from our question pool , the loadings on Factor 2 were not confined to compulsive behaviors , but equally featured items pertaining to related patterns of thought , i . e . obsessions , preoccupations , or intrusive thoughts . We cannot speak to causality here , but this suggests that repetitive behavior and repetitive , irrational patterns of thought are not orthogonal symptom dimensions , but perhaps share a common neurobiological root . Items from the impulsivity scale ( of which the total score was a significant predictor of goal-directed deficits ) did not load as strongly or consistently on this factor ( M=0 . 15 , SD=0 . 15; significantly less than the former three questionnaires , Eating Disorders vs . Impulsivity: t ( 52 ) =5 . 178; OCD vs . Impulsivity: t ( 41 ) =11 . 379; Alcohol Addiction vs . Impulsivity: t ( 33 ) =4 . 342 , all p<0 . 001 ) ( Table 2 ) . 10 . 7554/eLife . 11305 . 006Table 2 . Means and standard deviations ( in parentheses ) of loadings onto Factor 1 ‘Anxious-Depression’ , Factor 2 ‘Compulsive Behavior and Intrusive Thought’ and Factor 3 ‘Social Withdrawal’ factors for each questionnaire . DOI: http://dx . doi . org/10 . 7554/eLife . 11305 . 006Anxious-depressionCompulsive behavior and intrusive thoughtSocial withdrawal ( Factor 1 ) ( Factor 2 ) ( Factor 3 ) Alcohol addiction0 . 15 ( 0 . 05 ) 0 . 31 ( 0 . 07 ) -0 . 23 ( 0 . 06 ) Apathy0 . 44 ( 0 . 16 ) -0 . 05 ( 0 . 13 ) 0 . 04 ( 0 . 13 ) Depression0 . 38 ( 0 . 23 ) 0 . 14 ( 0 . 14 ) 0 . 04 ( 0 . 06 ) Eating disorders-0 . 05 ( 0 . 10 ) 0 . 36 ( 0 . 15 ) 0 . 06 ( 0 . 06 ) Impulsivity0 . 24 ( 0 . 22 ) 0 . 15 ( 0 . 15 ) -0 . 11 ( 0 . 11 ) OCD-0 . 05 ( 0 . 14 ) 0 . 50 ( 0 . 06 ) 0 . 09 ( 0 . 07 ) Schizotypy0 . 16 ( 0 . 11 ) 0 . 18 ( 0 . 13 ) 0 . 08 ( 0 . 14 ) Social anxiety0 . 04 ( 0 . 05 ) 0 . 08 ( 0 . 09 ) 0 . 57 ( 0 . 14 ) Trait anxiety0 . 52 ( 0 . 17 ) 0 . 15 ( 0 . 16 ) 0 . 13 ( 0 . 08 ) Scores greater than 0 . 25 are emboldened to highlight the dominant constructs . 10 . 7554/eLife . 11305 . 007Figure 3 . Trans-diagnostic factors . ( a ) Factor analysis on the correlation matrix of 209 questionnaire items suggested that 3-factor solution best explained these data . Factors were ‘Anxious-Depression’ , ‘Compulsive Behavior and Intrusive Thought’ and ‘Social Withdrawal’ . Item loadings for each factor are presented on the top , left and bottom sides of the correlation matrix , color-codes indicate the questionnaire from which each item was drawn . ( b ) These factors were entered into mixed-effects models , revealing that only the Factor 2 ‘Compulsive Behavior and Intrusive Thought’ was associated with goal-directed deficits , the effect size ( 17% reduction in model-based learning for every 1 SD increase in ‘Compulsive Behavior and Intrusive Thought’ ) was larger than for any individual questionnaire , and pairwise contrasts revealed that these deficits were specific to this factor , compared to Factor 1 ‘Anxious-Depression’ and Factor 3 ‘Social Withdrawal’ . The y-axes indicate the% change in model-based learning for each change of 1 standard deviation ( SD ) of clinical symptomatology . Error bars denote standard error . DOI: http://dx . doi . org/10 . 7554/eLife . 11305 . 007 We next tested for an association between subjects’ scores on these three factors and their , separately measured , goal-directed performance . When tested alone , ‘Compulsive Behavior and Intrusive Thought’ was significantly associated with deficits in goal-directed learning ( β=−0 . 046 , SE=0 . 01 , p<0 . 001 ) , and this effect size was greater than that of any of the questionnaires used in this study , corresponding to a 17% reduction in model-based learning for an increase of 1 SD in ‘Compulsive Behavior and Intrusive Thought’ ( Figure 3B , Table 3 ) . There were no significant effects of Factor 1 ( β=−0 . 001 , ( 0 . 01 ) , p=0 . 92 ) or Factor 3 ( β=0 . 013 , SE=0 . 01 , p=0 . 24 ) on model-based learning . Finally , we directly compared the associations between goal-directed deficits and these factors by including them in the same model and conducting planned contrasts . We found that deficits in goal-directed control were highly specific to the ‘Compulsive Behavior and Intrusive Thought’ ( vs . ‘Anxious-Depression’ , β=−0 . 062 , SE=0 . 02 , p=0 . 001; vs . ‘Social Withdrawal’ , β=−0 . 089 , SE=0 . 02 , p<0 . 001 ) . Moreover , when included in the same model with the other factors , ‘Social Withdrawal’ ( onto which addiction and aspects of impulsivity load negatively ) emerged as a significant positive predictor of goal-directed control over action ( β=0 . 031 , SE=0 . 01 , p=0 . 014 ) . To test the extent to which the relationship between goal-directed deficits and ‘Compulsive Behavior and Intrusive Thought’ is truly continuous , we carried out a supplementary analysis in which this factor was entered as a quadratic term in our model , thereby testing for a nonlinear effect . We found no evidence for nonlinearity ( beta=−0 . 0016 , p=0 . 822 ) , and the linear effect remained significant when included in this model ( beta=−0 . 045 , p=0 . 001 ) . Similarly , we repeated our analyses in subsets of our population comprising either ‘putative patients’ ( defined as those who scored in the top 25% on a given self-report measure ) or subjects in the normal range ( bottom 75% ) and the results were broadly consistent across sub-samples ( Materials and methods , Supplementary file 3 ) . 10 . 7554/eLife . 11305 . 008Table 3 . Trans-diagnostic factors and model-based learning . DOI: http://dx . doi . org/10 . 7554/eLife . 11305 . 008Constructβ ( SE ) z-valuep-valueIndependent Models‘Anxious-Depression’ ( Factor 1 ) -0 . 001 ( 0 . 01 ) 0 . 100 . 920‘Compulsive Behavior and Intrusive Thought’ ( Factor 2 ) -0 . 046 ( 0 . 01 ) -4 . 06<0 . 001 ***‘Social Withdrawal’ ( Factor 3 ) 0 . 013 ( 0 . 01 ) 1 . 180 . 238Covariate Model‘Anxious-Depression’ ( Factor 1 ) 0 . 003 ( 0 . 01 ) 0 . 280 . 781‘Compulsive Behavior and Intrusive Thought’ ( Factor 2 ) -0 . 058 ( 0 . 01 ) -4 . 71<0 . 001 ***‘Social Withdrawal’ ( Factor 3 ) 0 . 031 ( 0 . 01 ) 2 . 450 . 014**p<0 . 05; **p<0 . 01; ***p<0 . 001 . SE=standard error . Top panel shows results from Independent Models . Bottom panel shows results from Covariate Model , where trans-diagnostic factors were entered together into the same model: glmer ( Stay ~ Reward * Transition * ( Factor1z + Factor2z + Factor3z + IQz + Agez + Gender ) + ( Reward * Transition + 1 | Subject ) ) . Statistics refer to the interaction between scores on each factor and Reward x Transition , i . e . the extent to which that score is associated with changes in model-based learning . Positive β values indicate that the symptom score is associated with greater model-based learning , while negative β values indicate that the symptom score is associated with reduced model-based learning . Finally and in complement to the unsupervised factor analysis used to define ‘Compulsive Behavior and Intrusive Thought’ , we carried out a fully supervised analysis ( regression with elastic net regularization ) to identify directly from the individual questionnaire items those most predictive of goal-directed learning , as assessed using the regression model . Supporting our previous conclusions , those items that predicted model-based deficits in the negative direction substantially overlapped with items with above-threshold loadings on ‘Compulsive Behavior and Intrusive Thought’ ( 75% overlap; Supplementary file 4 ) . One noteworthy pattern arises among the exceptions . The supervised analysis also identified several additional items from the impulsivity questionnaire , which had not loaded on ‘Compulsive Behavior and Intrusive Thought’ , but did predict goal-directed learning . Those were items that tracked subjects’ motivation to engage with the experimental paradigm , e . g . “I ( do not ) like to think about complex problems” . Other , more compulsivity-relevant items from the impulsivity scale , involving compulsive shopping and general loss of control over action , were identified in both analyses . The former items are likely of little clinical relevance , but can explain the strong association between impulsivity total scores and goal-directed deficits , despite the fact that impulsivity did not load strongly onto ‘Compulsive Behavior and Intrusive Thought’ . In addition to tracking one well-delineated aspect of psychopathology , we found that task performance was significantly related to other measures collected in this study . First , although individual variation in ‘model-free’ performance on the learning task did not track any of the scores from our psychiatric questionnaires ( Supplementary file 1C ) , in Experiment 1 , model-free performance did relate significantly to age ( Supplementary file 1B , Reward*Age interaction ) . ‘Model-based’ learning was also related to age and IQ . In particular , higher IQ was associated with increases in goal-directed , ‘model-based’ learning . In contrast to the effect of age on ‘model-free’ learning , older people were significantly less ‘model-based’ compared to their younger counterparts . All of these results were replicated in Experiment 2 . Additionally , the larger sample size in Experiment 2 allowed us to detect small but significant associations between gender and model-free and model-based learning . Males were significantly less model-free and more model-based relative to females tested in this study . Importantly , all of these effects are controlled for ( by including age , IQ , and gender as additional covariates ) in the analyses relating learning to psychiatric symptoms .
Here , we tested the utility of a dimensional approach to investigating the neurocognitive basis of compulsivity using two large-scale general population samples . Evidence from multiple complimentary analyses supported the conclusion that ‘Compulsive Behavior and Intrusive Thought’ is a symptom dimension associated with deficits in goal-directed control that links features of multiple psychiatric disorders; most notably symptoms of OCD , addiction , and eating disorders . Interestingly , this dimension goes beyond the uncontrolled behaviors that have been previously associated with compulsivity , to include obsessions , preoccupations and intrusive thoughts . That self-report scores of OCD and addiction symptoms were associated with these deficits is consistent with previous research in patient populations ( Sjoerds et al . , 2013; Voon et al . , 2015; Gillan et al . , 2011; 2014a; 2014b; 2015a ) , and extends these results for the first time to a general population sample . Likewise , binge-eating disorder has also been previously associated with reduced goal-directed control in one patient study and an animal model ( Voon et al . , 2015; Furlong et al . , 2014 ) . Critically , the results of the present study extend this finding to self-report symptoms of other subtypes of eating disorders , suggesting that Compulsive Behavior and Intrusive Thought ( and associated deficits in goal-directed control ) are a key component of more aspects of eating disorders than previously documented . An entirely consistent exception was that items relating to exerting control over food intake ( e . g . “I display self-control around food” ) did not load strongly on the ‘Compulsive Behavior and Intrusive Thought’ factor . A previous study reported an association between social anxiety disorder and deficits in goal-directed control ( Alvares et al . , 2014 ) . Using self-report social anxiety symptom scores in our general population sample , we did not replicate this finding , and in fact observed a trend towards enhanced goal-directed control associated with social anxiety symptoms . Specifically , in most analyses social anxiety symptoms ( both total scores and the ‘Social Withdrawal’ factor ) was unrelated to task performance . We did however observe a significant positive association between the ‘Social Withdrawal’ factor and goal-directed control in one analysis , while controlling for the other factors in the same analysis . This result should be interpreted with caution , given that the association was not sufficiently robust to predict goal-directed control alone , but this serves to illustrate that ‘Social Withdrawal’ trended towards predicting better goal-directed control , not worse . Two explanations for the discrepant findings between the present study and the prior investigation with diagnosed social anxiety disorder patients are the differences in sample size between our respective studies and that the co-morbidities reported for the social anxiety disorder population of the study by Alvares and colleagues ( 2014 ) could not be controlled for and may have driven the reported association . This underscores the importance of a dimensional approach to psychiatric phenotyping . Schizophrenia has also been previously associated with deficits in goal-directed control ( Morris et al . , 2015 ) , a finding that was partially supported by the present study ( to the limited extent that ‘schizotypy’ , measured here , has implications for schizophrenia as a clinical condition ) . Consistent with the heterogeneous nature of schizophrenia , where two diagnosed patients can have entirely non-overlapping symptoms ( American Psychological Association , 2013 ) , we did not find a significant association between the total score on the schizotypy questionnaire and deficits in goal-directed control ( although this was significant in a second analysis based on a full computational model ) . However , using our trans-diagnostic approach , we found that in particular ‘unusual experiences’ characteristic of schizotypy loaded onto the ‘Compulsive Behavior and Intrusive Thought’ factor , which in turn was a strong predictor of goal-directed deficits . This finding converges with studies highlighting that delusions are more closely linked to executive deficits than the negative symptoms of schizophrenia ( Lysaker et al . , 1998; 2003 ) . In terms of clinical phenomenology , schizophrenia and OCD share a common pattern of abnormal beliefs and as DSM-5 and others have noted , the distinction between a delusion in schizophrenia and a strongly held belief in OCD is often blurred ( Poyurovsky and Koran , 2005; American Psychological Association , 2013 ) . These data suggest that ‘Compulsive Behavior and Intrusive Thought’ , which comprises automatic behaviors as well as associated repetitive thoughts , may be common to both schizophrenia and OCD and explained by deficits in goal-directed control . Earlier work investigating deficits in goal-directed learning in compulsive patient populations did not employ a positive clinical control ( Voon et al . , 2015 ) , therefore until now the possibility that goal-directed deficits were non-specific , i . e . evident in all psychiatric populations , remained untested . For instance , prior studies have found a consistent association between stress and goal-directed learning deficits ( Otto et al . , 2013; Schwabe and Wolf , 2009 ) , which might in principle mediate non-specific effects due to the considerable burdens of mental illness . Here , we tested this possibility rigorously in two independent samples . We found no association between ‘Anxious-Depression’ and deficits in goal-directed control , and moreover the specificity of goal-directed deficits to ‘Compulsive Behavior and Intrusive Thought’ was confirmed through direct statistical comparisons . Prior work has shown that the model-based learning deficits predict the presence of habits using a devaluation probe ( Gillan et al . , 2015c ) , providing a tentative mechanism through which the goal-directed deficits observed in the present study might cause the development of compulsive behaviors . Indeed , this converges with prior work showing that when OCD patients are performing habits , they show dysfunctional hyperactivity in the caudate ( Gillan et al . , 2015a ) , a region associated with goal-directed control over behavior ( Dolan and Dayan , 2013 ) . An outstanding question , however , is the extent to which excessive stimulus-response habit learning also contributes to Compulsive Behavior and Intrusive Thought . The model-free component of the task we employed in the present study did not relate significantly to psychiatric symptomatology , as indeed we had hypothesized because it also does not appear to be sensitive to slow habitual learning ( indeed , unlike the model-based component of the task , it does not predict devaluation [Gillan et al . , 2015c] ) . Future work is needed to develop a computational marker of individual differences in stimulus-response habit formation , so that this possibility can directly be tested . Another interesting question that emerges from these data is how deficits in goal-directed control might result in both cognitive distortions ( which take the form of obsessions in OCD , such as a fear of germs ) and compulsive behavior ( e . g . repetitive hand-washing ) , which our factor analysis suggested are inextricably linked . One possibility was raised by a recent study , which demonstrated that just like low-level stimulus-response behaviors , more abstract goal selection can also be rendered habitual ( Cushman and Morris , 2015 ) . If these habitual cognitive actions can be conceived as a sort of ‘habit of thought , ’ this might indicate a common mechanism for both compulsive behavior and the related repetitive patterns of thought ( i . e . ‘habits of thought’ ) . An alternative possibility posits that obsessive thoughts may develop as a result of compulsive behavior ( Gillan and Robbins , 2014 ) . Evidence for this idea comes from a study where OCD patients were found to engage in post-hoc rationalization in order to explain a series of habitual responses ( Gillan et al . , 2014b ) . The notion is that in OCD , experiencing a recurrent urge to wash one’s hands might cause a patient to infer that they are concerned about hygiene . Future , longitudinal work will be needed to dissect the temporal dynamics of these symptom features to test these hypotheses , which are not mutually exclusive . Researchers have suggested that ‘Impulsivity’ and ‘Compulsivity’ are partially overlapping neurocognitive features relevant for many psychiatric disorders ( Robbins et al . , 2012 ) . The present study offers some insights in this regard . While the total score of the impulsivity scale was a strong predictor of goal-directed deficits , it did not load significantly onto the ‘Compulsive Behavior and Intrusive Thought’ factor , suggesting it has an independent association with goal-directed deficits . The supervised analysis identified the items from the impulsivity scale that best predicted goal-directed deficits . In terms of the overlap between the impulsivity questionnaire items and Factor 2 , the two above-threshold predictors of model-based deficits were “I spend or charge more than I earn” and “I do things without thinking” , each of which is qualitatively characteristic of compulsive , habitual behavior . Importantly , the three items that did not overlap with ‘Compulsive Behavior and Intrusive Thought’ , but still predicted model-based learning , tracked subjects’ general interest in engaging with the task ( e . g . “I do not like puzzles” , “I do not like to think about complex problems” ) . We suggest that these items may not be of particular clinical importance , but simply serve as a marker of how likely individuals are to engage with the task material . In summary , while a small subset of the impulsivity items contributed to ‘Compulsive Behavior and Intrusive Thought’ , impulsivity as assessed by our scale was mostly distinct . Of course , impulsivity as a construct itself involves a broad range of potentially distinct behaviors , such as impatient inter-temporal choice preferences and premature responding ( Dalley et al . , 2011 ) . Further work will be need to assess how such behaviors relate to the features measured here; notably , our large-scale online methodology is well suited for examining such questions . As has been shown for other tests that broadly fall within the category of executive function ( Arffa , 2007 ) , model-based learning was also associated with IQ and age ( and gender in experiment 2 only ) . Although these effects were controlled for in all analyses and therefore do not bias the interpretation of our results , they highlight the fact that the coupling between model-based learning and ‘Compulsive Behavior and Intrusive Thought’ is far from perfect . One particularly interesting observation is that as people get older , they show greater deficits in model-based learning ( Supplementary file 1B ) , but fewer psychiatric symptoms on all nine questionnaires collected in the present study ( Supplementary file 1A ) , in line with prior work with diagnosed patients ( Kessler et al . , 2005 ) . This incongruence suggests that there may be multiple dissociable processes responsible for model-based learning . Future studies are needed to dissect this somewhat complex construct into its constituent parts ( as has been already attempted for other executive tasks [Miyake et al . , 2000] ) , with a view to identifying the simpler component that is specific to the compulsive phenotype . Relatedly , future work might test if working memory might conceivably contribute to this association observed in the present study ( Otto et al . , 2013 ) . Also , the strength of the association between a clinical phenotype and an underlying mechanism is fundamentally limited by the accuracy with which we can assess that phenotype . Aside from issues of relatively low reliability of self-report clinical symptoms ( e . g . self-report OCD , r=0 . 71 [Hajcak et al . , 2004] ) , we are also limited by the questions we ask . For example , in the present study we did not account for pathological gambling or trichotillomania , which are similarly defined clinically by a loss of control over repetitive behavior ( Potenza , 2008; Chamberlain et al . , 2007 ) and therefore may contribute noise to our signal . It is clear that iterative improvements to both self-report assessment and behavioral testing are needed to increase effect sizes and further refine the neurobiological characterization of Compulsive Behavior and Intrusive Thought suggested by these data . Although we have labeled the three factors that emerged from our unsupervised analysis based on theoretical considerations , we acknowledge that this is an inherently subjective process and that some may rightfully disagree with our choice of terminology . An important distinction to be made here is that although this labeling process was subjective , the way in which these clusters were identified was not . We first identified a heretofore-unrecognized collection of trans-diagnostic psychiatric symptoms based on their inter-correlations and then validated this clustering by demonstrating an association with neurocognitive performance in an independent task . ‘Compulsive Behavior and Intrusive Thought’ is not intended to be a fixed or final definition – rather it is hoped that future work can ( i ) use the clusters defined in this study to find closer links between biological markers and clinical and ( ii ) improve and augment these clusters through further data-driven evaluations . More broadly , we hope that this methodology can be employed in many other areas of psychiatry where the considerable issues of heterogeneity within and homogeneity across the existing diagnostic categories is curtailing efforts to delineate the precise neurobiological basis of psychiatric problems . In the present study , we did not screen for psychiatric disorders , favoring the acquisition of a large sample within which we could leverage normal variation in psychopathology . Although our results converge with prior work using this neurocognitive marker in compulsive disorders ( Voon et al . , 2015 ) , future studies will be needed to test if these dimensional results map onto clinically diagnosed patients . For example , based on the results of the present study , we hypothesize that the co-morbidity between OCD and addiction might be largely explained by a common deficit in goal-directed control . Conversely , the co-morbidity between OCD and anxiety disorders might be explained by an orthogonal ( equally important ) symptom dimension . This kind of exciting work should be coupled with studies aiming to use such trans-diagnostic markers to predict treatment response on an individual basis within the existing diagnostic categories . Altogether , these data suggest that ‘Compulsive Behavior and Intrusive Thought’ together constitute a dimensional psychiatric phenotype that can be tracked in the general population and is linked to deficits in goal-directed control over action , which has a clear neurobiological foundation ( Dolan and Dayan , 2013 ) . These data highlight the utility of a computational approach to psychiatry ( Montague et al . , 2012 ) and specifically our novel approach of leveraging large datasets , online testing , and normal variation in psychopathology to isolate the neurocognitive basis of psychiatric dimensions that may be relevant for multiple disorders . More broadly , the results of this study constitute progress toward realizing the promise of the RDoC initiative , suggesting that dimensional markers of psychiatric disturbances may map more closely to underlying biological states than do the overlapping and heterogeneous definitions of DSM disorders .
Data were collected online using Amazon’s Mechanical Turk ( AMT ) . Participants were paid a base rate ( Experiment 1: $2 , Experiment 2: $2 . 50 ) in addition to a bonus based on their earnings during the reinforcement-learning task ( In each experiment , M=$0 . 54 , SD=0 . 04 ) . Subjects were based in the USA ( i . e . had a US billing address with an associated US credit card , debit card or bank account ) , 95% of their previous tasks were approved and were 18 years or older . Participants in Experiment 1 ( N=548 ) were 357 females ( 65% ) and 191 males with ages ranging from 18 to 72 ( M=35 , SD=11 ) . Using the effect size of the relationship between OCD symptoms and model-based learning observed in Experiment 1 , we estimated that to achieve 80–90% power on a two-tailed test with a significance level of p<0 . 05 , the sample size needed in Experiment 2 between 1223–1637 subjects . Experiment 2 , participants ( N=1413 ) were 823 females ( 58% ) and 590 males with ages ranging from 18 to 76 ( M=33 , SD=11 ) . The research team did not know participants’ identities; participants provided their consent online by clicking ‘I Agree’ after reading the study information and consent language in accordance with procedures approved by the New York University Committee on Activates Involving Human Subjects . In line with suggestions made in the literature with respect to studies conducted using Amazon’s Mechanical Turk ( AMT ) , several a priori exclusion criteria were applied to ensure data quality ( Crump et al . , 2013 ) . Prior to completing the RL task subjects completed a practice phase , which consisted of written instructions , passively viewing 20 trials demonstrating the probabilistic nature of the associations between second stage fractals and subsequent 25c rewards , and actively participating in 20 trials demonstrating the probabilistic transition structure of the task ( i . e . selecting a top-stage box on each trial and observing the transition to second-stage states ) . After this practice phase , participants were required to correctly answer a 3-item basic comprehension test regarding the rules of the reinforcement-learning task ( Gillan et al . , 2015c ) . If subjects failed to answer the questions correctly , they were sent back to the beginning and required to repeat the instructional section prior to re-taking the comprehension test . Participants were permitted to repeat this cycle as many times as was necessary for them to pass this test and continue to the main experiment . The RL instructions and associated comprehension test were always administered first , followed by the RL task , then the IQ test and finally the self-report psychiatric assessments . Within the self-report section , the order of the questionnaires was fully randomized . Exclusions based on task performance/engagement were applied sequentially , in the order listed below . Reinforcement-Learning Task Exclusion Criteria: Subjects were excluded if they missed more than 10% of trials ( Exp1: n=11; Exp2: n=62 ) , responded on the same key on more than 95% of trials on which they registered a response ( Exp1: n=46; Exp2: n=85 ) or had implausibly fast reaction times , i . e . ± 2 standard deviations from the mean ( Exp1: n=9; Exp2: n=18 ) . Clinical Questionnaires Exclusion Criterion: In an effort to identify participants who were not reading the questions prior to selecting their responses , we included one catch item: “If you are paying attention to these questions , please select 'A little' as your answer” . Very few subjects failed to select the appropriate response to this catch question; those that did were excluded ( Exp1: n=0; Exp2: n=6 ) . IQ Test Exclusion Criterion: Participants who did not answer correctly to any of the IQ questions were excluded from further analysis ( Exp1: n=32; Exp2: n=87 ) . The adaptive character of the test meant that participants responding incorrectly received increasingly easy items; consistently failing to respond correctly indicates that given participants might have been inattentive or dishonest . In total , 98/646 ( 15% ) subjects who submitted data were excluded in Experiment 1 and 258/1671 ( 15% ) were excluded in Experiment 2 . We tested post hoc if subjects excluded on the basis of RL task performance were typical in terms of psychiatric self-report and other assessments . In study 1 , we found that those subjects who were excluded had lower symptoms of OCD ( t ( 604 ) =2 . 477 , p=0 . 014 ) , trait anxiety ( t ( 604 ) =2 . 225 , p=0 . 027 ) , and a trend towards lower levels of depression ( t ( 604 ) =1 . 799 , p=0 . 073 ) . These differences were not observed in Study 2 , where all questionnaire total scores were not significantly different across groups ( p>0 . 05 ) . For both Experiment 1 and 2 , results presented in this paper are not changed by the inclusion of these subjects in the analyses . To assess individual differences in goal-directed learning , we used a reinforcement-learning task ( Daw et al . , 2011 ) that distinguishes goal-directed ( 'model-based' ) learning from basic temporal difference ( 'model-free' ) learning . Model-based learning , like ‘goal-directed learning’ , reflects the extent to which individuals integrate contingency information with estimations of outcome value to make choices , and predicts whether or not individuals can exert control over their habits in a devaluation test ( Gillan et al . , 2015c; Friedel et al . , 2014 ) . While model-free learning has been suggested to capture slow incremental learning characteristic of habit-formation itself , empirical studies using sequential decision tasks have not detected this relationship ( Gillan et al . , 2015c; Friedel et al . , 2014 ) , and this converges with the empirical observation that deficits in model-based ( but not model-free ) learning have been observed in compulsive disorders ( Voon et al . , 2015 ) . The design of the task is presented in Figure 1 . On each trial , subjects were presented with a choice between two fractals ( 2 . 5 s choice window ) . Each fractal usually ( i . e . ‘common’ transitions: 70% , Figure 1A , white arrow ) led to a particular second state ( orange or blue ) displaying another two fractal options . Selecting one of the fractals in the second stage resulted in participants being probabilistically rewarded with a picture of a 25¢ coin . There was a unique probability of receiving a reward associated with each second stage fractal , and these drifted slowly and independently over time ( never being less than 0 . 25 or greater than 0 . 75 ) . Responses were indicated using the left ( ‘E’ ) and right ( ‘I’ ) keys . Critically , on 30% of ‘rare’ trials ( Figure 1A , grey arrow ) , choices uncharacteristically led to the alternative second state . A purely ‘model-free’ learner makes choices based solely on whether or not they were rewarded the last time they performed this action , regardless of whether the transition was rare or common ( Figure 1C ) . A ‘model-based’ learner , in contrast , makes decisions based not only on the history of reward , but also the transition structure of the task , i . e . the environmental contingency ( Figure 1D ) . For example , if a choice was followed by a rare transition to a second state , and that second state was rewarded , a model-based learner would be more likely to choose the alternate action on the next trial , because this is more likely to return them to that rewarding second state . A model-free learner , on the other hand , would be more likely to repeat that same action again , making no adjustment based on the transition type . We used a logistic regression based on this logic to identify from their switching patterns the extent to which each participant exhibited goal-directed ( model-based , vs . model-free ) choices ( Daw et al . , 2011 ) . Intelligence Quotient ( IQ ) was approximated using a Computerized Adaptive ( CAT ) based on a bank of n=26 items similar to those used in Raven's Standard Progressive Matrices ( SPM: [Raven , 2000] ) . The item bank was built using two parameter logistic Item Response Theory model ( 2pl: [Baker , 1992] ) . Item parameters were estimated using an online piloting sample of 760 participants ( not included in the present study ) that took both the test used in this study and original SPM . Items retained in the item bank were characterized by parameters ( item-fit and discrimination ) comparable or better than original SPM items . The length of the CAT test was 5 items ( plus one non-diagnostic starting items ) . The items , including the starting item , were selected using Maximum Fisher Information criterion ( va der Linden et al . ) with a randomesque parameter of n=3 ( Kingsbury and Zara , 1989 ) . The scores were estimated using a Bayes Modal estimator ( Birnbaum , 1969 ) . Estimates based on the piloting sample showed that the score based on a 5-item CAT correlates relatively highly ( r=0 . 77 ) with a score of a full SPM test . In both Experiments 1 & 2 , subjects completed self-report questionnaires assessing obsessive-compulsive disorder ( OCD ) using the Obsessive-Compulsive Inventory – Revised ( OCI-R ) ( Foa et al . , 2002 ) , depression , using the Self-Rating Depression Scale ( SDS ) ( ZUNG , 1965 ) and trait anxiety was assessed using the trait portion of the State-Trait Anxiety Inventory ( STAI ) ( Spielberger et al . , 1983 ) . In Experiment 2 , subjects were additionally assessed for alcohol addiction using the Alcohol Use Disorder Identification Test ( AUDIT ) ( Saunders et al . , 1993 ) , apathy using the Apathy Evaluation Scale ( AES ) ( Marin et al . , 1991 ) , eating disorders using the Eating Attitudes Test ( EAT-26 ) ( Garner et al . , 1982 ) , impulsivity using the Barratt Impulsivity Scale ( BIS-10 ) ( Patton et al . , 1995 ) , schizotypy scores using the Short Scales for Measuring Schizotypy ( Mason et al . , 2005 ) and social anxiety using the Liebowitz Social Anxiety Scale ( LSAS ) ( Liebowitz , 1987 ) . Means of these total scores are presented in Supplementary file 1A , along with their relationship to age , gender and IQ . In Experiment 2 , subjects also completed some additional self-report assessments that were unrelated to the present study and will be published elsewhere . These self-report assessments were fully randomized within the psychiatric assessment component of the procedure . Logistic regression analyses were conducted using mixed-effects models implemented with the lme4 package in the R programming language , version 3 . 1 . 1 ( http://cran . us . r-project . org ) . The model tested if subjects’ choice behavior ( coded as switch: 0; stay: 1 , relative to the previous choice ) was influenced by Reward ( coded as rewarded: 1; unrewarded: -1 ) , Transition ( coded as common: 1 , rare: -1 ) , and their interaction , on the preceding trial . A main effect of reward indicates that there is a significant contribution of model-free learning to choice behavior . An interaction between Reward and Transition indicates that there is a significant contribution of model-based learning to choice behavior . Within-subject factors ( the intercept , main effects of reward and transition , and their interaction ) were taken as random effects , i . e . allowed to vary across subjects . First , we tested our basic logistic regression model , which included age , gender and IQ as fixed effects covariates . We used Bound Optimization by Quadratic Approximation ( bobyqa ) with 1e5 functional evaluations . The model was specified in the syntax of R as follows: Stay ~ Reward * Transition * ( IQz + Agez + Gender ) + ( Reward * Transition + 1 | Subject ) In each Experiment , we found a significant main effect of Reward ( 'model-free' ) and a significant Reward x Transition interaction ( 'model-based' ) ( Figure 4 , Supplementary file 1B ) . There was also an unhypothesized significant main effect of Transition and an interaction between Transition and IQ , such that subjects were more likely to stay following a common transition and individuals higher in IQ showed this pattern more strongly . This seemingly anomalous effect is likely a side effect of additional structure in the choices that the regression model fails to capture . In particular , in the full computational model , choices are impacted by incremental learning that accrues over trials , such that a choice on some trial is affected by rewards on multiple preceding trials . Although the regression model considers only the most recent trial’s rewards , some aspects of additional learning might be correlated with the transition term , producing small bias that can be detected given the large sample size of the current study ( Skatova et al . , 2013 ) . For instance , the full model tends to encounter a negative reward prediction error immediately following a rare transition , which is driven by learning about second-stage state values driven by rewards received on previous trials . Such structure is more interpretably subsumed within the model-based and model-free learning terms in the fits of the fuller computational model , where , notably , the key results were all recapitulated ( see below ) . 10 . 7554/eLife . 11305 . 009Figure 4 . Behavioral data from experiments 1 ( N=548 ) and 2 ( N=1413 ) . Error bars denote standard error . Data illustrate that consistent with previous studies ( Daw et al . , 2011 ) , participants use a mixture of model-based and model-free learning to guide choice . Associated statistics are presented in Supplementary file 1B . *p<0 . 05 **p<0 . 01 ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11305 . 009 To test the hypothesis that a symptom severity of a given clinical construct ( 'SymptomScore' ) was associated with model-based learning deficits , we included the total score for each questionnaire ( z-scored ) as a between-subjects predictor and tested for interactions with all other factors in the model . We included age , gender and IQ ( all z-scored ) as fixed effects predictors interacted with Reward , Transition and Reward x Transition , to control for potentially confounding relationships between model-based learning and these covariates of no-interest . We hypothesized that there would be a significant three-way interaction between Reward , Transition and SymptomScore , only if those symptoms pertained to compulsive patterns of behavior . Specifically , we expected that greater severity of self-reported compulsive symptoms ( i . e . OCD , addiction , eating disorders and aspects of impulsivity ) would be predictive of reductions in model-based control over action . In the syntax of the lme4 package , the specification for the regression was the same as above with the addition of the SymptomScorez , as follows: Stay ~ Reward * Transition * ( SymptomScorez + IQz + Agez + Gender ) + ( Reward * Transition + 1 | Subject ) In Experiment 1 , three models were tested in which ‘SymptomScorez’ refers to the z-scored OCD , Trait Anxiety and Depression total scores in each respective model . Additionally , in Experiment 1 , we also tested a model where self-report symptoms of OCD , trait anxiety and depression were included in the same model , to illustrate that the association with OCD symptoms survived the exclusion of shared variance . This was specified as follows: Stay ~ Reward * Transition * ( OCDz + TraitAnxietyz + Depressionz + IQz + Agez + Gender ) + ( Reward * Transition + 1 | Subject ) In Experiment 2 , due to the high correlations across the different clinical scales , including all of the questionnaires in the same model would not produce an interpretable result - such that meaningful shared variance would be lost . Therefore , the associations between model-based learning and each questionnaire were assessed using separate models for each questionnaire ( SymptomScorez , as specified above ) . As expected based on prior literature in this area ( Voon et al . , 2015 ) , there was no relationship between clinical symptomatology and model-free learning in either Experiment ( Supplementary file 1C ) . Note that we tested this model without gender ( as gender was not itself significant in the model ) , and the results do not change - the effect of OCD symptoms on model-based learning remains significant ( β=−0 . 041 , SE=0 . 02 , p=0 . 043 ) . We nonetheless include gender in the presented models for Experiment 1 for consistency with Experiment 2 , where gender effects were observed . In order to ( i ) reduce the collinearity between the total scores for each of the 9 questionnaires employed and ( ii ) investigate the possibility that a more parsimonious latent trans-diagnostic structure could explain item-level responses in this dataset , we employed factor analysis using Maximum Likelihood Estimation ( MLE ) . Factor analysis was conducted using the factanal ( ) function from the Psych package in R , with an oblique rotation ( oblimin ) . Two hundred and nine individual questionnaire items were entered as measured variables into the factor analysis . As responses on the schizotypy scale were binary at the item-level , a heterogeneous correlation matrix was computed using the hector function in polycor package in R . This allowed for Pearson correlations between numeric variables , polyserial correlations between numeric and binary items and polychoric correlations between binary variables . Factor selection was based on Cattell’s criterion ( Cattell , 1966 ) ; wherein a sharp transition from horizontal to vertical ( ‘elbow” ) indicates that there is little benefit to retaining additional factors . The scree-plot was analyzed using an objective implementation of this criterion , the Cattell-Nelson-Gorsuch ( CNG ) test , which computes the slopes of all possible sets of three adjacent eigenvalues and determines the point at which there is the greatest differences in slope ( nFactors package in R ) ( Gorsuch and Nelson , 1981 ) . The CNG test indicated the existence of a 3-factor latent structure ( Figure 5 ) , which comprises factors that we labeled ‘Anxious-Depression’ , ‘Compulsive Behavior and Intrusive Thought’ and ‘Social Withdrawal’ based on the strongest individual item loadings ( Supplementary files 2A , 2B & 2C , respectively ) . Although Cattell’s criterion is perhaps the most widely utilized rule-of-thumb for factor selection , we acknowledge that there are many alternatives and indeed another objective method , ‘Parallel Analysis’ ( Drasgow and Lissak , 1983 ) , suggests an 8-factor solution to our data . This model was not only less parsimonious than the 3-factor solution , but in addition , a post hoc analysis revealed that it was also quantitatively inferior at predicting task performance when these 8 factors were entered as predictors in a mixed effects model ( as per our main task analyses ) . Specifically , both Akaike Information Criterion ( AIC ) and Bayesian Information Criterion ( BIC ) were lower for the mixed effects model with covariates derived from the 3-factor solution relative to the 8-factor solution , indicating that this model was the best at predicting behavior . 10 . 7554/eLife . 11305 . 010Figure 5 . Scree plot of eigenvalues . The outer frame shows the eigenvalues for every possible factor solution , N=209 . Inset is data for the first 20 potential factor solutions only . An empirically defined elbow , where Eigenvalues begin to level out , was identified at factor 4 using the nFactors package in R , provideing evidence for a 3-factor solution ( Cattell , 1966 ) , indicated in orange . *p<0 . 05 **p<0 . 01 ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11305 . 010 As outlined in the results section , factors were labeled based on items that loaded most strongly and consistently . For the ‘Anxious-Depression’ factor , the highest average loadings came from the Trait Anxiety questionnaire ( M=0 . 52 , SD=0 . 17 ) , followed by Apathy ( M=0 . 44 , SD=0 . 16 ) and Depression ( M=0 . 38 , SD=0 . 23 ) ( Table 2 ) . No other questionnaires reached the 0 . 25 average loading threshold we apply throughout this manuscript , but impulsivity came very close ( M= 0 . 24 , SD=0 . 22 ) . Those impulsivity items that loaded most consistently reflected a tendency to not plan for the future and reduced ability to concentrate . Factor 3 was labeled ‘Social Withdrawal’ . This factor was dominated by items from the Social Anxiety questionnaire ( M=0 . 57 , SD=0 . 14 ) , and interestingly did not have a significant contribution from trait anxiety ( M=0 . 13 , SD=0 . 17 ) . We chose the term ‘withdrawal’ primarily to distinguish this factor from the original social anxiety disorder questionnaire . Interestingly this factor had borderline negative contributions from the alcohol addiction scale , which were low but consistent ( M=−0 . 23 , SD=0 . 06 ) . Overall , this factor describes a phenotype that fears and avoids social situations , but interestingly also thinks excessively about future events and appears risk averse . A mixed effects logistic regression analysis was conducted to test the extent to which ‘Anxious-Depression’ , ‘Compulsive Behavior and Intrusive Thought’ , and ‘Social Withdrawal’ factors predicted deficits in goal-directed control over action . Specifically , these three factors were entered as z-scored fixed effect predictors in the basic model described above ( i . e . interacted with Reward , Transition and Reward*Transition ) , while controlling for age , gender and IQ: Stay ~ Reward * Transition * ( Factorz + IQz + Agez + Gender ) + ( Reward * Transition + 1 | Subject ) The extent to which a factor is related to deficits in goal-directed control is indicated by the presence of a significant Reward*Transition*Factorz interaction ( in the negative direction ) . Unlike the analysis of the original questionnaire total scores , in addition to testing the predictors separately in independent models , here we also tested a model where all three clinical predictors were included in the same model , which allowed us to statistically compare their effect sizes and thereby make claims about the specificity of our effects to compulsive ( versus non-compulsive ) aspects of psychopathology . Table 3 shows the effects for model-based learning . There were no effects on model-free reinforcement learning . To test the extent to which these results reflect a continuous relationship between model-based learning and Factor 2 ( ‘Compulsive Behavior and Intrusive Thought’ ) , constructed subsets of our total sample comprising either ‘putative patients’ ( defined as those who scored in the top 25% on a given self-report measure ) or subjects in the normal range ( bottom 75% ) . We then repeated our analyses in these sub-samples . The slopes of the regression lines were consistent across all analyses , such that the relationship between model-based deficits and Factor 2 were observed both in individuals reporting the most severe symptoms and those in the normal range ( see Supplementary file 3 ) . In all 9 analyses with individuals in the ‘normal range’ , the relationship between Factor 2 and model-based deficits were significant at p<0 . 05 . In 5/9 analyses with ‘probably patients’ who were in the top 25% of symptom severity , the relationship between Factor 2 and model-based deficits were significant at p<0 . 05 . This analysis had just ¼ of the total sample and was therefore severely underpowered . But nonetheless , the direction and slope of the effect were consistent across the board , providing evidence to suggest that these relationships will likely generalize to patient populations . In addition to the factor analysis , we also carried out a fully supervised analysis to identify the individual items that explained the most independent variance in goal-directed learning using linear regression with elastic net regularization . Elastic Net ( Zou and Hastie , 2005 ) regularization imposes a hybrid of both L1- and L2-norm penalties ( i . e . , penalties on the absolute ( L1 norm ) and squared values of the β weights ( L2 norm ) ) . This allows relevant but correlated coefficients to coexist in a sparse model fit , by doing automatic variable selection and continuous shrinkage simultaneously , and selects or rejects groups of correlated variables . Least absolute shrinkage and selection operator LASSO , ( Tibshirani , 1996 ) and ridge regression ( Hoerl and Kennard , 1970 ) are special cases of the Elastic Net . The dependent measure in this analysis was each subject’s model-based score ( i . e individual subject’s coefficients for reward x transition , corrected for age , IQ and gender , from the analysis in Experiment 2 , Supplementary file 1B ) . All predictor data were first feature scaled ( z-score transformed ) . We implemented ten-fold cross-validation with nested cross-validation for tuning and validating the model . Briefly , to implement cross-validation , the data were randomly split into 10 groups . A model was then generated based on 9 training groups , and then applied to the remaining independent testing group . Each group served as the testing group once , resulting in 10 different models , and predictions for every subject based on independent data . Nested cross-validation involved subdividing the 9 training groups ( i . e . , 90% of the sample ) into a further 10 groups ( ‘inner’ folds ) . Within these 10 inner folds , 9 were utilized for training a model over a range of 50 alpha ( 0 . 01–1 ) and 50 lambda ( 0 . 0001–1 ) values , where alpha is the weight of lasso versus ridge optimization and lambda is the regularization coefficient . This generated a resulting model fit on the inner fold test set for each possible combination of alpha and lambda . The mean fit over all 10 inner folds for each combination of alpha and lambda was then calculated and then used to determine the optimal parameters for the outer fold . We conducted 100 iterations of regularization with tenfold validation and retained items that were significant predictors of model-based learning in >=95% of final models . The overall model was significant , with the median cross-validated p=0 . 00003 , median cross-validated r=0 . 11 . Twenty-eight features met these criteria and are listed in Supplementary file 4 . The logistic regression analyses presented are a simplified method for analyzing the data , but as this approach only considers events taking place on the trial immediately preceding choice , it does not fully capture the influence of slow , incremental learning that takes place over many trials . These analyses have been shown to produce very similar results , particularly when estimating model-based learning ( Gillan et al . , 2015c; Otto et al . , 2013 ) ( indeed they are correlated at 0 . 87 here ) . Nonetheless , to complement these analyses , we verified that the relationship between model-based learning and compulsive behavior holds in the full computational instantiation of model-based and model-free reinforcement learning . For this analysis , choices were modeled as arising due to the weighted combination of model-free and model-based reinforcement learning . The model is equivalent to that used by Otto et al ( Otto et al . , 2013 ) , which is itself a simplified variant of the one used by Daw et al ( Daw et al . , 2011 ) . At each trial t , a participant makes a stage-1 choice c1 , t , occasioning a transition to a stage-2 state st where she makes another choice c2 , t and receives reward rt . At stage 2 , subjects are assumed to learn a value function over states and choices , Qtstage2 ( s , c ) , whose value for the chosen action is updated in light of the reward received at each trial according to a delta rule , Qt+1stage2 ( st , c2 , t ) = ( 1−α ) Qtstage2 ( st , c2 , t ) +rt . Here , α is free learning rate parameter , and ( in this and all analogous update equations throughout ) we have omitted a factor of α from the last term of the update , which is equivalent to rescaling the rewards and Qs by 1/α and the corresponding weighting parameters β by α . ( Otto et al . , 2013 ) The probability that a subject will make a particular stage-2 choice is modeled as governed by these choices according to a logistic softmax , with free inverse temperature parameter βstage2: Pc2 , t=c α exp βstage2Qtstage2 ( st , c ) , normalized over both options c . Stage-1 choices are modeled as determined by the weighted combination of both model-free and model-based value predictions about the ultimate , stage-2 value of each stage-1 choice . Model-based values QMB are given by the learned values of the corresponding stage-2 state , maximized over the two actions: QtMB ( c ) =maxc2 ( Qtstage2 ( s , c2 ) ) , where s is the stage-2 state predominantly produced by stage-1 choice c . Model-free values are learned by two learning rules , each of which updates according to a delta rule with a different estimate of the second-stage-value: TD ( 0 ) , where Qt+1MF0 ( c1 , t ) = ( 1−α ) QtMF0+Qtstage2 ( st , c2 , t ) , and TD ( 1 ) , where Qt+1MF1 ( c1 , t ) = ( 1−α ) QtMF1+rt . Stage-1 choice probabilities are then given by a logistic softmax , with contributions from each value estimate , each weighted by its own free inverse temperature parameter: P ( c1 , t=c ) ∝exp ( βMBQtMB ( c ) +βMF0QtMF0 ( c ) +βMF1QtMF1 ( c ) +βstickI ( c=c1 , t−1 ) ) . Here , I ( c=c1 , t−1 ) is a binary indicator for the choice that repeats the one made on the previous trial; the corresponding weight βstick measures the tendency to alternate or perseverate regardless of feedback . At the conclusion of each trial , the value estimates Q ( of all three sorts ) for all unchosen actions and unvisited states are decayed multiplicatively by ( 1−α ) . Altogether , the model has six free parameters: five weights β and a learning rate α . These represent a minor change of variables with respect to the equations in Otto et al . ( 2013 ) : In particular , by separating the TD ( 0 ) and TD ( 1 ) stages of the model-free update into separate Q values , we split Otto et al . ’s aggregate model-free weight βMF version into two variables , thereby also replacing their eligibility trace parameter λ which encodes the balance between the two updates and eliminating the ( 0 , 1 ) boundaries associated with that variable . Following estimation , we reconstruct the aggregate model-free weighting as βMF=βMF0+αβMF1 , where the factor of α accounts for the difference in scaling between the two weights arising from the omission of α from the update equations . For each participant , we estimated the free parameters of the model by maximizing the likelihood of her sequence of choices , jointly with group-level distributions over the entire population using an Expectation Maximization procedure ( Huys et al . , 2011 ) implemented in the Julia language ( Bezanson et al . , 2012 ) . We extracted the per-subject model-based and model-free weightings βMB and βMF as indices of the strength of each sort of learning for further analysis of individual differences . Specifically , we used subject-level estimates of model-based and model-free learning from the computational model as dependent variables in regression analyses where clinical characteristics ( i . e . questionnaire total scores and factors from factor analysis ) were independent variables . The results of the full reinforcement-learning model mirrored that of the logistic regression analysis in almost every respect . The two differences were that when estimated using the computational model , the relationship between self-report OCD symptoms and goal-directed learning in Experiment 1 fell short of reaching significance at p<0 . 05 ( Supplementary file 5A ) . The size of this effect was similar in Experiment 2 , but with the benefit of an increased sample size was highly significant , indicating this was an issue of statistical power . Secondly , while Schizotypy did not reach significance as a predictor of model-based deficits using the regression model , it was a significant predictor model-based learning when computationally estimated . There were no relationships between self-report psychopathology and model-free learning defined using the computational model . A side-by-side comparison of the predictive power of model-based learning defined using the computational model versus one-trial back regression analysis is presented in Supplementary file 5B . | When an individual resists the temptation to stay out late in order to get a good night’s sleep , he or she is exercising what is known as “goal-directed control” . This kind of control allows individuals to regulate their behaviour in a deliberate manner . It is thought that a reduction in goal-directed control may be linked to compulsiveness or compulsivity , a psychological trait that involves excessive repetition of thoughts or actions . Furthermore , evidence shows that goal-directed control is reduced in people with compulsive disorders , such as obsessive-compulsive disorder ( or OCD ) and drug addiction . However , failures of goal-directed control have also been reported in other mental health conditions that are not linked to compulsivity , such as social anxiety disorder . The fact that reduced goal-directed control is found across various mental health conditions highlights a core issue in modern psychiatric research and treatment . Mental health conditions are typically defined and diagnosed by their clinical symptoms , not by their underlying psychological traits or biological abnormalities . This makes it difficult to determine the cause of a specific disorder , as its symptoms are often rooted in the same psychological and biological traits seen in other mental health conditions . To start to tackle this issue , Gillan et al . used a strategy that allowed them to look at compulsivity as a “trans-diagnostic dimension”; that is , as something that exists on a spectrum and is not specific to one disorder but involved in numerous different mental health conditions . Nearly 2 , 000 people completed an online task that assessed goal-directed control , and filled in questionnaires that measured symptoms of various mental health conditions . Gillan et al . showed that , as expected , people with reduced goal-directed control were generally more compulsive , and that this relationship could be seen in the context of both OCD and other compulsive disorders such as addiction . Further , by leveraging the efficiency of online data collection to collect such a large sample , Gillan et al . were also able to examine how much different symptoms co-occurred in people . This enabled them to use a statistical technique to pick out three trans-diagnostic dimensions – compulsive behaviour and intrusive thought , anxious-depression and social withdrawal – and found that only the compulsive factor was associated with reduced goal-directed control . In fact , reduced goal-directed control was found to be more closely related to compulsivity than the symptoms of traditional mental health disorders including OCD . These findings show that research into the causes of mental health conditions and perhaps ultimately diagnosis and treatment – all of which have traditionally approached specific disorders in isolation – would benefit greatly from a trans-diagnostic approach . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Characterizing a psychiatric symptom dimension related to deficits in goal-directed control |
Collection of high-throughput data has become prevalent in biology . Large datasets allow the use of statistical constructs such as binning and linear regression to quantify relationships between variables and hypothesize underlying biological mechanisms based on it . We discuss several such examples in relation to single-cell data and cellular growth . In particular , we show instances where what appears to be ordinary use of these statistical methods leads to incorrect conclusions such as growth being non-exponential as opposed to exponential and vice versa . We propose that the data analysis and its interpretation should be done in the context of a generative model , if possible . In this way , the statistical methods can be validated either analytically or against synthetic data generated via the use of the model , leading to a consistent method for inferring biological mechanisms from data . On applying the validated methods of data analysis to infer cellular growth on our experimental data , we find the growth of length in E . coli to be non-exponential . Our analysis shows that in the later stages of the cell cycle the growth rate is faster than exponential .
The last decade has seen a tremendous increase in the availability of high-quality large datasets in biology , in particular in the context of single-cell level measurements . Such data are complementary to ‘bulk’ measurements made over a population of cells . They have led to new biological paradigms and motivated the development of quantitative models ( Osella et al . , 2017; Facchetti et al . , 2017; Ho et al . , 2018; Soifer et al . , 2016; Jun et al . , 2018; Amir and Balaban , 2018; Kohram et al . , 2021 ) . Nevertheless , they have also led to new challenges in data analysis , and here we will point out some of the pitfalls that exist in handling such data . In particular , we will show that the commonly used procedure of binning data and linear regression may hint at specific functional relations between the two variables plotted that are inconsistent with the true functional relations . As we shall show , this may come about due to the ‘hidden’ noise sources that affect the binning procedure and the phenomenon of ‘inspection bias’ where certain bins have biased contributions . One of our main take home messages is the significance of having an underlying model ( or models ) to guide/test/validate data analysis methods . The underlying model is referred to as a generative model in the sense that it leads to similar data to that observed in the experiments . The importance of a so-called generative model has been beautifully advocated in the context of astrophysical data analysis ( Hogg et al . , 2010 ) , yet biology brings in a plethora of exciting differences: while in physics noise from measurement instruments often dominates , in the biological examples we will dwell on here it is the intrinsic biological noise that can obscure the mathematical relation between variables when not handled properly . In the following , we will illustrate this rather philosophical introduction on a concrete and fundamental example , albeit e pluribus unum . We will focus on the analysis of the Escherichia coli growth curves obtained via high throughput optical microscopy . Nevertheless we anticipate the conceptual points made here – and demonstrated on a particular example of interest – will translate to other types of measurements , which make use of microscopy but also beyond . Binning corresponds to grouping data based on the value of the x-axis variable , and finding the mean of the fluctuating y-axis variable for this group . By removing the fluctuations of the y-variable , the binning process often aims to expose the ‘true’ functional relation between the two variables which can be used to infer the underlying biological mechanism . While binning may provide a smooth non-linear relation between variables , linear regression is used to find a linear relationship between the variables . In addition to binning , we use the ordinary least squares regression where the slope and the intercept of the best linear fit line are obtained by minimizing the squared sum of the difference between the dependent variable raw data and the predicted value . Here , the best fit/the best linear fit is obtained using the raw data and not the binned data . Similar to binning , the assumption underlying linear regression is that our knowledge of x-axis variable is precise while the noise is in the y-axis variable . It is important to discuss the sources of fluctuations in the y-axis variable before we proceed . In biology , fluctuations in the variables arise inevitably from the intrinsic variability within a cell population . Cells growing in the same medium and environment have different characteristics ( e . g . growth rate ) due to the stochastic nature of biochemical reactions in the cell ( Kiviet et al . , 2014 ) . For example , the division event is controlled by stochastic reactions , whose variability leads to cell dividing at a size smaller or larger than the mean . In this paper , when modeling the data , we will consider the intrinsic noise as the only source of variability and assume that the measurement error is much smaller than the intrinsic variation in the population . One example of the use of binning and linear regression is shown in Figure 1A where size at division ( Ld ) vs size at birth ( Lb ) is plotted using experimental data obtained by Tanouchi et al . for E . coli growing at 25 °C ( Tanouchi et al . , 2017 ) . In Figure 1A , the functional relation between length at division and length at birth for E . coli is observed to be linear and close to Ld=Lb+ΔL ( see the Experimental data section for details ) . The relation obtained allows us to hypothesize a coarse-grained biological model known as the adder model as shown in Figure 1B in which the length at division is set by addition of length ΔL from birth ( Soifer et al . , 2016; Harris and Theriot , 2016; Si et al . , 2019; Amir , 2014; Campos et al . , 2014; Taheri-Araghi et al . , 2015; Eun et al . , 2018 ) . This previously discussed example demonstrates and reiterates the use of statistical analysis on single-cell data to understand the underlying cell regulation mechanisms . Using statistical methods such as binning and linear regression , other phenomenological models apart from adder have also been proposed in E . coli where the division length ( Ld ) is not directly ‘set’ by that at birth ( Ho and Amir , 2015; Micali et al . , 2018; Witz et al . , 2019 ) . The phenomenological models , in turn , can be related to mechanistic ( molecular-level ) models of cell size and cell cycle regulation ( Barber et al . , 2017 ) . Recent work has shed light on the subtleties involved in interpreting the linear regression results for the Ld vs Lb plot where seemingly adder behavior in length can be obtained from a sizer model ( division occurring on reaching a critical size ) due to the interplay of multiple sources of variability ( Facchetti et al . , 2019 ) . This issue is similar in spirit to those we highlight here . The volume growth of single bacterial cells has been typically assumed to be exponential ( Godin et al . , 2010; Wang et al . , 2010; Campos et al . , 2014; Cermak et al . , 2016; Soifer et al . , 2016; Iyer-Biswas et al . , 2014 ) . Assuming ribosomes to be the limiting component in translation , growth is predicted to be exponential and growth rate depends on the active ribosome content in the cell ( Scott et al . , 2010; Lin and Amir , 2018; Metzl-Raz et al . , 2017 ) . Under the assumption of exponential growth , the size at birth ( Lb ) , the size at division ( Ld ) , and the generation time ( Td ) are related to each other by , ( 1 ) ln ( LdLb ) =λTd , where λ is the growth rate . Understanding the mode of growth is important for example , due to its potential effects on cell size homeostasis . Exponentially growing cells cannot employ a mechanism where they control division by timing a constant duration from birth but such a mechanism is possible in case of linear growth ( Amir , 2014; Kafri et al . , 2016; Ho et al . , 2018 ) . Linear regression performed on ln ( LdLb ) vs ⟨λ⟩Td plot , where ⟨λ⟩ is the mean growth rate , was used to infer the mode of growth in the archaeon H . salinarum ( Eun et al . , 2018 ) , and in the bacteria M . smegmatis ( Logsdon et al . , 2017 ) and C . glutamicum ( Messelink et al . , 2020 ) , for example . If the best linear fit follows the y = x trend , the resulting functional relation might point to growth being exponential . A corollary to this is the rejection of exponential growth when the slope and intercept of the best linear fit deviate from one and zero , respectively ( Messelink et al . , 2020 ) . Thus , binning and linear regression applied on single-cell data appear to provide information about the underlying biology , in this case , the mode of cellular growth . We will test the validity of such inference by analyzing synthetic data generated using generative models . We find that linear regression performed on the plot ln ( LdLb ) vs ⟨λ⟩Td , surprisingly , does not provide information about the mode of growth . Nonetheless , we show that other methods of statistical analysis such as binning growth rate vs age plots are adequate in addressing the problem . Using these validated methods on experimental data , we find that E . coli grows non-exponentially . In later stages of the cell cycle , the growth rate is higher than that in early stages .
To illustrate the pitfalls associated with binning , we use data from recent experiments on E . coli where the length at birth , the length at division and the generation time were obtained for multiple cells ( see Experimental methods and [Tiruvadi-Krishnan et al . , 2021] ) . Phase-contrast microscopy was used to obtain cell length at equal intervals of time . Note that we consider length to reflect cell size in this paper rather than other cell geometry characteristics such as surface area and volume . The length growth rate that we elucidate in the paper can be different from the cell volume growth rate as shown in Appendix 1 assuming a simple cell morphology and exponential growth . Using the same cell morphology , we also find the length growth rate to be identical to cell surface growth rate . To investigate if the cell growth was exponential , we plotted ln ( LdLb ) vs ⟨λ⟩Td for cells growing in M9 alanine minimal medium at 28 °C ( ⟨Td⟩ = 214 min ) . The linear regression of these data yields a slope of 0 . 3 and an intercept of 0 . 4 as shown in Figure 2A . The binned data and the best linear fit deviate significantly from the y = x line ( see Supplementary file 1 ) . Additionally , the binned data follows a non-linear trend and flattens out at longer generation times . We also found similar deviations in the binned data and best linear fit in glycerol medium ( ⟨Td⟩ = 164 min ) shown in Figure 2—figure supplement 1A , and glucose-cas medium ( ⟨Td⟩ = 65 min ) shown in Figure 2—figure supplement 1B . Qualitatively similar results have been recently obtained for another bacterium , C . glutamicum , in Messelink et al . , 2020 . These results might point to growth being non-exponential . Next , we will approach the same problem but with a generative model . We will first show that the ln ( LdLb ) vs ⟨λ⟩Td binned plot could not distinguish exponential growth from non-exponential growth . For that purpose , we use a previously studied model ( Eun et al . , 2018 ) which considers growth to be exponential with the growth rate distributed normally and independently between cell cycles with mean growth rate ⟨λ⟩ and standard deviation CVλ⟨λ⟩ . CVλ is thus the coefficient of variation ( CV ) of the growth rate and is assumed to be small . To maintain a narrow distribution of cell size , cells must employ regulatory mechanisms . In our model , we assume that , barring the noise due to stochastic biochemical reactions , cells attempt to divide at a particular size Ld given size at birth Lb . Keeping the model as generic as possible , we can write Ld as a function of Lb , f ( Lb ) which can be thought of as a coarse-grained model for the regulatory mechanism . Amir , 2014 provides a framework to capture the regulatory mechanisms by choosing f ( Lb ) = 2 Lb1-αL0α . L0 is the typical size at birth and α , which can take values between 0 and 2 , reflects the strength of regulation strategy . α = 0 corresponds to the timer model where division occurs on average after a constant time from birth , and α = 1 is the sizer model where a cell divides upon reaching a critical size . α = 1/2 can be shown to be equivalent to the adder model where division is controlled by addition of constant size from birth ( Amir , 2014 ) . In addition to the deterministic function ( f ) specifying division , the size at division is affected by noise ( ζ⟨λ⟩ ) in division timing . We assume it has a Gaussian distribution with mean zero and standard deviation σn⟨λ⟩ and that it is independent of the growth rate . Thus , the generation time ( Td ) can be mathematically written as Td=1λln ( f ( Lb ) Lb ) +ζ⟨λ⟩ and is influenced by growth rate noise and division timing noise . Note that replacing the time additive division timing noise with a size additive division timing noise will not affect the results qualitatively ( see ‘Model’ and ‘Exponential growth’ sections for details and Supplementary file 1 for variable definitions ) . For perfectly symmetrically dividing cells whose sizes are narrowly distributed , we find the trend in the binned data for ln ( LdLb ) vs ⟨λ⟩Td plot to be ( see section ‘Predicting the results of statistical constructs applied on ln ( LdLb ) vs ⟨λ⟩Td and ⟨λ⟩Td vs ln ( LdLb ) ’ ) , ( 2 ) y=x ( 1+1−xln ( 2 ) 1+22−ασn2CVλ2ln2 ( 2 ) ) . Fixing CVλ = σn = 0 . 15 , we show using simulations in Figure 2C the non-linear trend in the binned data even though we assumed exponential growth . Similarly , on performing linear regression on the raw data of ln ( LdLb ) vs ⟨λ⟩Td plot , we find that the slope of the best linear fit is not equal to one and the intercept is non-zero ( see Equation 27 and 28 and Figure 2C ) . Equation 2 shows that the trend in the binned data depends on the ratio of growth rate noise and division timing noise . The slope is equal to one and intercept is zero only if the noise in growth rate is negligible as compared to the division timing noise . In experiments that is rarely the case , hence , the binned data trend and the best linear fit deviate from the y = x line even though growth might be exponential . Thus , we cannot rule out exponential growth in the E . coli experiments despite the binned data trend being non-linear and the best-fit line deviating from the y = x line . Why does a non-linear relationship in the binned data for the plot ln ( LdLb ) vs ⟨λ⟩Td arise even for exponential growth ? According to the model , Ld is determined by a deterministic strategy , f ( Lb ) and a time/size additive division timing noise . The noise component which affects Ld and subsequently the quantity ln ( LdLb ) is thus the noise in division timing and not the growth rate . The generation time ( Td ) plotted on the x-axis is influenced by the noise in division timing as well as the noise in growth rate . Binning assumes that for a fixed value of the x-axis variable , the noise from other sources affects only the y-axis variable ( the binned variable ) . Similarly for linear regression , the underlying assumption is that the independent variable on x-axis is precisely known while the dependent variable on the y-axis is influenced by the independent variable and from external factors other than the independent variable . In this case , only ⟨λ⟩Td plotted on x-axis is influenced by growth rate noise while both ⟨λ⟩Td and ln ( LdLb ) are influenced by noise in division time . This does not fit the assumption for binning and linear regression and hence , the best linear fit for ln ( LdLb ) vs ⟨λ⟩Td plot might deviate from the y = x line even in the case of exponential growth . Another way of explaining the deviation from the linear y = x trend is by inspection bias , which arises when certain data is over-represented ( Stein and Dattero , 2018 ) . Cells which have a longer generation time than the mean will most likely have a slower growth rate . Thus , in Figure 2A and C , at larger values of ⟨λ⟩Td or Td , the bin averages are biased by slower growing cells , thus making ln ( LdLb ) or λTd to be lower than expected . This provides an explanation for the flattening of the trend . It follows from the previous discussion that if one bins data by ln ( LdLb ) then the assumption for binning is met . Both of the variables ⟨λ⟩Td and ln ( LdLb ) are influenced by the noise in division time but ⟨λ⟩Td plotted on the y-axis is also influenced by the growth rate noise . Thus , the y-axis variable , ⟨λ⟩Td is determined by the x-axis variable , ln ( LdLb ) , and an external source of noise , in this case , the growth rate noise . Thus , based on our model , we expect the trend in binned data and linear regression performed on the interchanged axes to follow the y = x trend for exponentially growing cells ( see section ‘Predicting the results of statistical constructs applied on ln ( LdLb ) vs ⟨λ⟩Td and ⟨λ⟩Td vs ln ( LdLb ) ’ ) . Indeed , on interchanging the axis and plotting ⟨λ⟩Td vs ln ( LdLb ) for synthetic data , we find that the trend in the binned data and the best linear fit closely follows the y = x line ( Figure 2D ) . We also find that the best linear fit follows the y = x line in the case of alanine ( Figure 2B ) , glycerol ( Figure 2—figure supplement 1A ) and glucose-cas ( Figure 2—figure supplement 1B ) . A change from non-linear behavior to that of linear on interchanging the axes is also observed in a related problem where growth rate ( λ ) and inverse generation time ( 1Td ) are considered ( Figure 2—figure supplement 2 and Section ‘Interchanging axes in growth rate vs inverse generation time plot might lead to different interpretations’ ) . Thus far , we showed for a range of models where birth controls division that the binned data trend for ln ( LdLb ) as function of ⟨λ⟩Td is non-linear and dependent on the noise ratio σnCVλ in the case of exponential growth . On interchanging the axes the binned data trend agrees with the y = x line independent of the growth rate and division time noise . However , we will show next that this agreement with the y = x trend cannot be used as a ‘smoking gun’ for inferring exponential growth from the data . To investigate this further , let us consider linear growth , which has also been suggested to be followed by E . coli cells ( Mitchison , 2005; Abner et al . , 2014 ) . The underlying equation for linear growth is , ( 3 ) Ld-Lb=λ′Td , where λ′ is the the elongation speed that is , dLdt . For cells growing linearly , the best linear fit for the plot ⟨λ⟩Td vs ln ( LdLb ) is expected to deviate from the y = x line . As before , we fix ⟨λ⟩ to be the mean of 1Tdln ( LdLb ) , agnostic of the linear mode of growth . Surprisingly , we found that for the class of models where birth controls division by a strategy f ( Lb ) and cells grow linearly , the best linear fit for ⟨λ⟩Td vs ln ( LdLb ) agrees closely with the y = x trend . On carrying out analytical calculations based on this model , we obtain the slope and the intercept of the ⟨λ⟩Td vs ln ( LdLb ) plot to be 32ln ( 2 ) ≈ 1 . 04 and –0 . 03 respectively , which is very close to that for exponential growth ( see section ‘Differentiating linear from exponential growth’ ) . This is shown for simulations of linear growth with cells following an adder model in Figure 3A . Given no information about the underlying model , Figure 3A could be interpreted as cells undergoing exponential growth contrary to the assumption of linear growth in simulations . Thus , when handling experimental data , cells undergoing either exponential or linear growth might seem to agree closely with the y = x trend . Deforet et al . , 2015 used the linear binned data trend in case of ⟨λ⟩Td vs ln ( LdLb ) plot to infer exponential growth but as we showed in this section , the linear trend does not rule out linear growth . This again reiterates our message of having a generative model to guide the data analysis methods such as binning and linear regression . For completeness , we also test the utility of ln ( LdLb ) vs ⟨Td⟩λ and its interchanged axes plots to elucidate the mode of growth ( Appendix 2 ) . We find that binning and linear regression applied on these plots can not differentiate between exponential and linear growth . To conclude the discussion of linear growth , we note that the natural plot for this growth regime is ⟨λlin⟩Td vs ld-lb and the plot obtained on interchanging the axes ( see the Linear growth section and Figure 3—figure supplement 1A and B ) . Here lb , ld and λlin are defined to be quantities Lb , Ld and λ′ , respectively , normalized by the mean length at birth . For cells growing exponentially , the best linear fit for the ⟨λlin⟩Td vs ld-lb plot is expected to deviate from the y = x line . This is indeed what is observed in Figure 3—figure supplement 1 where simulations of exponentially growing cells following the adder model are presented ( see ‘Differentiating linear from exponential growth’ for extended discussion ) . In all the cases above , the problem at hand deals with distilling the biologically relevant functional relation between two variables . However , the data is assumed to be subjected to fluctuations of various sources , and it is important to ensure that the statistical construct we are using ( e . g . binning ) is robust to these . How can we know a priori whether the statistical method is appropriate and a ‘smoking gun’ for the functional relation we are conjecturing ? The examples shown above suggest that performing statistical tests on synthetic data obtained using a generative model is a convenient and powerful approach . Note that in cases such as the ones studied here where analytical calculations may be performed , one may not even need to perform any numerical simulations to test the validity of the methods . In the last section , we showed that the plots ln ( LdLb ) vs ⟨λ⟩Td and ⟨λ⟩Td vs ln ( LdLb ) are not decisive in identifying the mode of growth . Recent works on B . subtilis ( Nordholt et al . , 2020 ) and fission yeast ( Knapp et al . , 2019 ) have used differential methods of quantifying growth namely growth rate ( = 1LdLdt ) vs age plots and elongation speed ( =dLdt ) vs age plots to probe the mode of growth within a cell cycle . Here , L denotes the size of the cell after time t from birth in the cell cycle and age denotes the ratio of time t to Td within a cell cycle ( hence it ranges from 0 to 1 by construction within a cell cycle ) . In this section , using various models of cell growth and cell cycle , we test the growth rate vs age method . Note that the growth rate vs age and the elongation speed vs age plots are not dimensionless unlike the previous plots . Using the growth rate vs age and elongation speed vs age plots , we aim to quantify the growth rate changes within a cell cycle . For cells assumed to be growing exponentially , growth rate is constant throughout the cell cycle . On averaging over multiple cell cycles , the trend of binned data is expected to be a horizontal line with value equal to mean growth rate which is indeed what we find in the numerical simulations of the adder and the adder per origin model ( Ho and Amir , 2015 ) , as shown in Figure 3B . The binned data trend in each of the models matches the theoretical predictions of growth rate ( shown as dotted lines ) . In contrast , for linearly growing cells , the elongation speed is expected to remain constant . We show this constancy using numerical simulations of linearly growing cells following the adder model ( Figure 3—figure supplement 3A ) . In accordance with this result , the growth rate is expected to decrease with cell age as λ∝11+age . This is verified in Figure 3B by again using the numerical simulations of linear growth with cells following the adder model . The binned data trend for linear growth ( green squares ) matches the theoretical predictions of λ∝11+age ( green dotted line ) . Thus , the two growth modes ( exponential and linear ) could be differentiated using the growth rate vs age plot ( for details see ‘Growth rate vs age and elongation speed vs age plots’ section ) . However , the growth rate vs age plots can be used to infer the mode of growth beyond the two discussed above . We show this by using simulations of cells following the adder model and undergoing faster than exponential or super-exponential growth ( see the Simulations section for details ) . In such a case , the growth rate is expected to increase . This increase in growth rate is shown in Figure 3B using simulations . The binned data trend ( red triangles ) again matches the growth rate mode used in the simulations ( red dotted line ) . Thus , the growth rate vs age plots are a consistent method to distinguish linear from exponential and super-exponential growths . Using the validated growth rate vs age plots , we obtained the growth rate trend for experimental data on E . coli for the three growth conditions studied in this paper ( Figure 4A-C ) . We found an increase in growth rate in all growth conditions during the course of the cell cycle . One may wonder whether such an increase may be explained by the E . coli morphology alone , due to the presence of hemispherical poles . For exponentially growing cell volume and considering a geometry of E . coli with spherical caps at the poles , the percentage increase in the growth rate of length over a cell cycle is around 3 % which is significantly smaller than that observed in our experimental data . Considering cell size trajectories ( cell size , L at time , t data ) where cell lengths were tracked beyond the cell division event ( by considering cell size in both daughter cells ) , we also found that the growth rate decreases close to division ( age ≈ 1 ) and returns to a value nearly equal to that observed at the beginning of cell cycle ( age ≈ 0 ) as shown in Figure 4—figure supplement 1 ( see ‘Growth rate vs age and elongation speed vs age plots’ section for extended discussion ) . The above question of mode of growth within a cell cycle can also be analyzed in relation to a specific event . Several studies have pointed to a change in growth rate at the onset of constriction ( Reshes et al . , 2008; Banerjee et al . , 2017 ) . This change in growth rate can be probed using growth rate vs time plots where time is taken relative to the onset of constriction as shown in Figure 4—figure supplement 2 . These plots show a decrease in growth rates at the two extremes of the plot . These decreases are due to inspection bias , where the growth rate trend is affected by the biased contribution of cells with a higher than average generation time or equivalently slower growth rate ( see ‘Growth rate vs time from specific event plots are affected by inspection bias’ section for extended discussion ) . Inspection bias is also observed when timing is considered relative to other cell events such as cell birth ( see ‘Growth rate vs time from specific event plots are affected by inspection bias’ section and Figure 3—figure supplement 2 ) . It might not always be possible to obtain growth rate trajectories as a function of time/cell age . Godin et al . instead obtained the instantaneous biomass growth speed ( dMdt ) as a function of its buoyant mass ( M ) ( Godin et al . , 2010 ) . On applying linear regression for instantaneous mass growth speed vs mass , we expect the slope of the best linear fit obtained to provide the average growth rate ( ⟨1MdMdt⟩ ) under the assumption of exponential growth while for linear growth the intercept provides the average growth speed . Using this method , biomass was suggested to be growing exponentially . This method can be applied to study the length growth rate within the cell cycle by plotting elongation speed as a function of length ( Cadart et al . , 2019 ) . We find that the binned data trend and the best linear fit of this plot follow the expected trend for linear and exponential growth as shown in Figure 3—figure supplement 3B and Figure 3—figure supplement 3D , respectively , for a cell cycle model where division is controlled via an adder mechanism from birth . However , the trend obtained appears to be model-dependent as shown in Figure 3—figure supplement 3F where the underlying cell cycle model used in the simulations is the adder per origin model . For this model , the binned data trend is found to be non-linear with the growth rate speeding up at large sizes , despite the synthetic data being generated for perfectly exponential growth . This non-linear trend can lead to growth rate being misinterpreted as non-exponential within the cell cycle ( see ‘Results of elongation speed vs size plots are model-dependent’ section for details ) . Thus , an analysis using the elongation speed vs size plot must be accompanied with an underlying cell cycle model . In summary , we found that the growth rate vs age plot was a consistent method to determine the changes in growth rate within a cell cycle . Unlike the growth rate vs age plots , the inference from the growth rate vs size plots was found to be model-dependent . Using the growth rate vs age plots , we show that the length growth of E . coli can be faster than exponential .
Statistical methods such as binning and linear regression are useful for interpreting data and generating hypotheses for biological models . However , we show in this paper that predicting the relationships between experimentally measured quantities based on these methods might lead to misinterpretations . Constructing a generic model and verifying the statistical analysis on the synthetic data generated by this model provides a more rigorous way to mitigate these risks . In the paper , we provide examples in which ln ( LdLb ) vs ⟨λ⟩Td and ⟨λ⟩Td vs ln ( LdLb ) plots fail as a method to infer the mode of growth . The binned data trend and the best linear fit for the ln ( LdLb ) vs ⟨λ⟩Td plot was found to be dependent upon the noise parameters in the class of models where birth controlled division ( Equation 2 ) . We also show that ⟨λ⟩Td vs ln ( LdLb ) plot could not differentiate between exponential and linear modes of growth ( Figures 2D and 3A ) . Thus , we conclude that the best linear fit for the above plots might not be a suitable method to infer the mode of growth but they are just one of the many correlations which the correct cell cycle model should be able to predict . We found growth rate vs age and elongation speed vs age plots to be consistent methods to probe growth within a cell cycle . The method was validated using simulations of various cell cycle models ( such as the adder , and adder per origin model , where in the latter , control over division is coupled to DNA replication ) and the binned growth rate trend agreed closely with the underlying mode of growth for the wide range of models considered ( Figure 3B ) . In the case of growth rate vs time plots , it was important to take into consideration the effects of inspection bias . We used cell cycle models to show the time regimes where inspection bias could be observed ( Figure 3—figure supplement 2 ) . In the regime with negligible inspection bias , we could reconcile the growth rate trend obtained using growth rate vs age ( Figure 4A-C ) and growth rate vs time plots ( Figure 4—figure supplement 2 ) . The authors in Messelink et al . , 2020 circumvent inspection bias in the elongation speed vs time from birth plots by focusing their analysis on the time period from cell birth to the generation time of the fastest dividing cell . The authors of Panlilio et al . , 2021 , while investigating the division behavior in the cells undergoing nutrient shift within their cell cycle , use both models and experimental data from steady-state conditions to identify inspection bias . These serve as good examples of using models to aid data analysis . Statistics obtained from linear regression such as in Figure 1A help narrow down the landscape of cell cycle models , but many have potential pitfalls lurking which might lead to misinterpretations ( Figures 2C and 3A ) . There are additional issues beyond those concerning linear regression and binning discussed here . For example , Willis and Huang , 2017 discusses Simpson’s paradox ( Simpson , 1951 ) where distinct cellular sub-populations might lead to erroneous interpretation of cell cycle mechanisms . Examples of such distinct sub-populations are found in asymmetrically dividing bacteria such as M . smegmatis ( Aldridge et al . , 2012; Logsdon et al . , 2017 ) . Another source of misinterpretation could arise from presence of measurement errors . Throughout this work , we deal with intrinsic noise and neglect measurement error . However , when measurement noise affects both x-axis and y-axis variables , the slope of the best linear fit is biased towards zero . This can lead to potentially related variables being misinterpreted as uncorrelated . Measurement errors can , however , be handled based on a model . Using a model which includes measurement error as a source of noise , we can guide the binning analysis . Using this methodology , we verified that typical measurement errors ( ≈0 . 02Lb ) Messelink et al . , 2020; Kaiser et al . , 2018 have negligible effects on the growth rate trends obtained from the experimental data used in our work . Single cell size in E . coli has been reported to grow exponentially ( Campos et al . , 2014; Wang et al . , 2010; Cermak et al . , 2016; Soifer et al . , 2016; Iyer-Biswas et al . , 2014; Godin et al . , 2010 ) , linearly ( Mitchison , 2005 ) , bilinearly ( Kubitschek , 1981 ) or trilinearly ( Reshes et al . , 2008 ) . These are inconsistent with our observations in Figure 4A-C where we find that growth can be super-exponential . The non-monotonic behavior in the fastest-growth condition is reminiscent of the results reported in Nordholt et al . , 2020 for B . subtilis . The authors of Nordholt et al . , 2020 attribute the increase in growth rate to a multitude of cell cycle processes such as initiation of DNA replication , divisome assembly , septum formation . In the two slower growth conditions ( Figure 4A-B ) , we find that the growth rate increase starts before the time when the septal cell wall synthesis starts i . e . , the constriction event . However , in the fastest growth condition ( Figure 4C ) , the timing of growth rate increase seems to coincide with the onset of constriction which is in agreement with previous findings ( Reshes et al . , 2008; Banerjee et al . , 2017 ) . It is important to distinguish between length growth and biomass growth . Oldewurtel et al . , 2021 measures biomass and cell volume and finds the mass-density variations within the cell-cycle to be small . In this paper , since we observe the length growth to be non-exponential ( Figure 4 ) , it remains to be seen whether biomass growth also follows a similar non-exponential behavior or if it is exponential as previously suggested ( Godin et al . , 2010; Oldewurtel et al . , 2021 ) . In conclusion , the paper draws the attention of the readers to the careful use of statistical methods such as linear regression and binning . Although shown in relation to cell growth , this approach to data analysis seems ubiquitous . The general framework of carrying out data analysis is presented in Figure 5 . It proposes the construction of a generative model based on the experimental data collected . Of course , we do not always know whether the model used is an adequate description of the system . What is the fate of the methodology described here in such cases ? First , we should be reminded of Box’s famous quote ‘all models are wrong , some are useful’ . The goal of a model is not to provide as accurate a description of a system as possible , but rather to capture the essence of the phenomena we are interested in and stimulate further ideas and understanding . In our context , the goal of the model is to provide a rigorous framework in which data analysis tools can be critically tested . If verified within the model , it is by no means proof of the success of the model and the method itself , and further comparisons with the data may falsify it leading to the usual ( and productive ) cycle of model rejection and improvement via comparison with experiments . However , if the best model we have at hand shows that the data analysis method is non-informative , as we have shown here on several methods used to identify the mode of growth , then clearly we should revise the analysis as it provides us with a non-consistent framework , where our modeling is at odds with our data analysis . Furthermore , testing the methods on a simplified model is still advantageous compared with the option of using the methods without any validation . To mitigate the risk of using irrelevant models , in some cases it may be desirable to test the analysis methods on as broad a class of models as possible as we have done in the paper , for example by our use of a general value of α to describe the size-control strategy within our models . Thus , guided by the model , the data analysis methods can be ultimately applied to experimental data and underlying functional relationships can be inferred . Reiterating the message of the authors in Hogg et al . , 2010 , the data analysis using this framework aims to justify the methods being used , thus , reducing arbitrariness and promoting consensus among the scientists working in the field .
Strain engineering: STK13 strain ( ΔftsN::frt-Ypet-FtsN , ΔdnaN::frt-mCherry-dnaN ) is derivative of E . coli K12 BW27783 ( CGSC#: 12119 ) constructed by λ-Red engineering ( Datsenko and Wanner , 2000 ) and by P1 transduction ( Thom , 2007 ) . For chromosomal replacement of ftsN with fluorescence derivative , we used primers carrying 40nt tails with identical sequence to the ftsN chromosomal locus and a plasmid carrying a copy of ypet preceded by a kanamycin resistance cassette flanked by frt sites ( frt-kanR-frt-Ypet-linker ) as PCR template ( a kind gift from R . Reyes-Lamothe McGill University , Canada; Reyes-Lamothe et al . , 2010 ) . The resulting PCR product was transformed by electroporation into a strain carrying the λ-Red-expressing plasmid pKD46 . Colonies were selected by kanamycin resistance , verified by fluorescence microscopy and by PCR using primers annealing to regions flanking ftsN gene . After removal of kanamycin resistance by expressing the Flp recombinase from plasmid pCP20 ( Cherepanov and Wackernagel , 1995 ) , we transferred the mCherry-dnaN gene fusion ( BN1682 strain; a kind gift from Nynke Dekker from TUDelft , The Netherlands , Moolman et al . , 2014 ) into the strain by P1 transduction . To minimize the effect of the insertion on the expression levels of the gene we removed the kanamycin cassette using Flp recombinase expressing plasmid pCP20 . Cells growth , preparation , and culturing E . coli in mother machine microfluidic devices: All cells were grown and imaged in M9 minimal medium ( Teknova ) supplemented with 2 mM magnesium sulfate ( Sigma ) and corresponding carbon sources at 28 °C . Three different carbon sources were used: 0 . 5 % glucose supplemented by 0 . 2 % casamino acids ( Cas ) ( Sigma ) , 0 . 3 % glycerol ( Fisher ) , and 0 . 3 % alanine ( Fisher ) supplemented with 1 x trace elements ( Teknova ) . For microscopy , we used mother machine microfluidic devices made of PDMS ( polydimethylsiloxane ) . These were fabricated following to previously described procedure ( Yang et al . , 2018 ) . To grow and image cells in microfluidic device , we pipetted 2–3 µl of resuspended concentrated overnight culture of OD600∼ 0 . 1 into main flow channel of the device and let cells to populate the dead-end channels . Once these channels were sufficiently populated ( about 1 hr ) , tubing was connected to the device , and the flow of fresh M9 medium with BSA ( 0 . 75 µg/ml ) was started . The flow was maintained at 5 µl/min during the entire experiment by an NE-1000 Syringe Pump ( New Era Pump Systems , NY ) . To ensure steady-state growth , the cells were left to grow in channels for at least 14 hr before imaging started . Microscopy: A Nikon Ti-E inverted epifluorescence microscope ( Nikon Instruments , Japan ) with a 100 X ( NA = 1 . 45 ) oil immersion phase contrast objective ( Nikon Instruments , Japan ) , was used for imaging the bacteria . Images were captured on an iXon DU897 EMCCD camera ( Andor Technology , Ireland ) and recorded using NIS-Elements software ( Nikon Instruments , Japan ) . Fluorophores were excited by a 200 W Hg lamp through an ND8 neutral density filter . A Chroma 41 , 004 filtercube was used for capturing mCherry images , and a Chroma 41 , 001 ( Chroma Technology Corp . , VT ) for Ypet images . A motorized stage and a perfect focus system were utilized throughout time-lapse imaging . Images in all growth conditions were obtained at 4 min frame rate . Image analysis: Image analysis was carried out using Matlab ( MathWorks , MA ) scripts based on Matlab Image Analysis Toolbox , Optimization Toolbox , and DipImage Toolbox ( https://www . diplib . org/ ) . Cell lengths were determined based on segmented phase contrast images . Dissociation of Ypet-FtsN label from cell middle was used to determine the exact timing of cell divisions . Further experimental details can also be found in Tiruvadi-Krishnan et al . , 2021 . Consider a model of cell cycle characterized by two events: cell birth and division . In our model , we assume that , barring the noise , cells tend to divide at a particular size vd given size at birth vb , via some regulatory mechanism . Hence , we can write vd as a function of vb , f ( vb ) . Amir , 2014 provides a framework to capture the regulatory mechanisms by choosing f ( vb ) = 2 vb1-αv0α . v0 is the typical size at birth and α captures the strength of regulation strategy . α = 0 corresponds to the timer model where division occurs after a constant time from birth , and α = 1 is the sizer where a cell divides on reaching a critical size . α = 1/2 can be shown to be equivalent to an adder where division is controlled by addition of constant size from birth ( Amir , 2014 ) . From here on , we would be using the length of the cell ( Lb , Ld , etc . ) as a proxy for size ( vb , vd , etc . ) . To reiterate , the length growth is not the same as cell volume growth as shown in Appendix 1 . All of the variable definitions are summarized in Supplementary file 1 . We also define lb=Lb⟨Lb⟩ and ld=Ld⟨Lb⟩ . Using this , we can write the division strategy f ( lb ) to be ld = f ( lb ) = 2 lb1-α . The total division size obtained will be a combination of f ( lb ) and noise in the division timing , the source of which could be the stochasticity in biochemical reactions controlling division . We will assume that division is perfectly symmetric i . e . , size at birth in the ( n+1 ) th generation ( lbn+1 ) is half of size at division in the nth generation ( ldn ) . Using the size additive division timing noise ( ζs ( 0 , σbd ) ) and f ( lb ) specified above , we obtain , ( 4 ) xn+1= ( 1-α ) xn+ln ( 1+ζs ( 0 , σbd ) 2 ( 1+xn ) 1-α ) , where xn = ln ( lbn ) . Size at birth ( Lb ) is narrowly distributed , hence lb≈1 and we can write x = ln ( lb ) = ln ( 1+δ ) where δ is a small number . We obtain x≪1 and , ( 5 ) x≈δ=lb-1 . The size additive noise , ζs ( 0 , σbd ) is assumed to be small and has a normal distribution with mean 0 and standard deviation σbd . Note that σbd is a dimensionless quantity . Since ζs ( 0 , σbd ) is assumed to be small and xn≪1 , we can Taylor expand the last term of Equation 4 to first order , ( 6 ) xn+1≈ ( 1-α ) xn+ζs ( 0 , σbd ) 2 . Equation 6 shows a recursive relation for cell size and it is agnostic of the mode of growth . We will show later for exponential growth that replacing the size additive noise with time additive noise does not change the structure of Equation 6 . Next , we will try to obtain the generation time ( Td ) in the case of exponentially growing cells . For exponential growth , the time at division Td is given by , ( 7 ) Td=1λln ( LdLb ) . For simplicity , we will assume a constant growth rate ( λ ) within the cell-cycle . Growth rate is fixed at the start of the cell-cycle and is given by λ = ⟨λ⟩ + ⟨λ⟩ξ ( 0 , CVλ ) , where ⟨λ⟩ is the mean growth rate and ξ ( 0 , CVλ ) is assumed to be small with a normal distribution that has mean 0 and standard deviation CVλ . CVλ denotes the coefficient of variation ( CV ) of the growth rate . This captures the variability in growth rate within cells arising from the stochastic nature of biochemical reactions occurring within the cell . In this section , we will focus on finding the equation of the best linear fit for relevant plots in the case of linear growth . The time at division for linear growth is given by , ( 38 ) Td=Ld-Lbλ′ . Note that λ′ has units of [length/time] and is defined as the elongation speed . This is different from the exponential growth rate which has units [1/time] . Here , we will work with the normalized length at birth ( lb ) and division ( ld ) , ( 39 ) Td=ld-lbλlin . Consider the normalized elongation speed to be λlin = ⟨λlin⟩+⟨λlin⟩ξlin ( 0 , CVλ , lin ) , where ⟨λlin⟩ is the mean normalized elongation speed for a lineage of cells and ξlin ( 0 , CVλ , lin ) is normally distributed with mean 0 and standard deviation CVλ , lin . Thus , the CV of elongation speed is CVλ , lin . The regulation strategy which the cell undertakes is equivalent to that in previous sections and is given by g ( lb ) = 2+2 ( 1-α ) ( lb-1 ) . Note that we can obtain g ( lb ) by Taylor expanding f ( lb ) around lb = 1 . Using the regulation strategy g ( lb ) and adding a size additive noise ζs ( 0 , σbd ) which is independent of lb , we find , ( 40 ) Td=2+2 ( 1−α ) ( lbn−1 ) +ζs ( 0 , σbd ) −lbn⟨λlin⟩ ( 1+ξlin ( 0 , CVλ , lin ) ) . Note that we chose size additive division timing noise ( ζs ( 0 , σbd ) ) for convenience in this section . However , it can be shown as done previously that the model is robust to the noise in division timing being size additive or time additive . Assuming that the noise terms ξlin ( 0 , CVλ , lin ) and ζs ( 0 , σbd ) are small , we obtain to first order , ( 41 ) Td≈ ( 1-2α ) ( lb-1 ) +1+ζs ( 0 , σbd ) -ξlin ( 0 , CVλ , lin ) ⟨λlin⟩ . The terms lb , ζs ( 0 , σbd ) and ξlin ( 0 , CVλ , lin ) are independent of each other . The standard deviation of Td ( σt ) can be calculated to be , ( 42 ) σt2= ( 1-2α ) 2σb2+σbd2+CVλ , lin2⟨λlin⟩2 . Assuming perfectly symmetric division and using ldn=g ( lbn ) +ζs ( 0 , σbd ) , we find the recursive relation for lbn to be , ( 43 ) ldn-lbn=2lbn+1-lbn= ( 1-2α ) lbn+2α+ζs ( 0 , σbd ) . Note that Equation 43 is the same as Equation 6 under the approximation xn=lbn-1 . At steady state , the standard deviation of lb is denoted by σb and using Equation 43 its value is obtained to be , ( 44 ) σb2=σbd24α ( 2-α ) . Similarly , the standard deviation of ld-lb , or equivalently λlinTd , denoted by σl , lin , is calculated to be , ( 45 ) σl , lin2=4α+14α ( 2-α ) σbd2 . For linear growth , a natural plot is ld-lb vs ⟨λlin⟩Td ( reminiscent of the ln ( LdLb ) vs ⟨λ⟩Td plot for exponential growth ) . To calculate the slope of the best linear fit , we have to calculate the correlation coefficient ρlin given by , ( 46 ) ρlin=⟨ ( ld-lb-⟨ld-lb⟩ ) ( ⟨λlin⟩Td-⟨⟨λlin⟩Td⟩ ) ⟩⟨λlin⟩σl , linσt . Again using the independence of terms lb , ζs ( 0 , σbd ) and ξlin ( 0 , CVλ , lin ) from each other , we get , ( 47 ) ρlin= ( 1-2α ) 2σb2+σbd2⟨λlin⟩σl , linσt=σl , lin⟨λlin⟩σt . The slope of best linear fit for the plot ld-lb vs ⟨λlin⟩Td is given by , ( 48 ) mtl , lin=ρlinσl , lin⟨λlin⟩σt=11+CVλ , lin24α ( 2−α ) σbd2 ( 4α+1 ) . The intercept ctl , lin is found to be , ( 49 ) ctl , lin=⟨ld−lb⟩−mtl , lin⟨⟨λlin⟩Td⟩=1−11+CVλ , lin24α ( 2−α ) σbd2 ( 4α+1 ) . On flipping the axis , the slope ( mlt , lin ) for the plot ⟨λlin⟩Td vs ld-lb is obtained to be , ( 50 ) mlt , lin=ρlin⟨λlin⟩σtσl , lin=1 . The intercept clt , lin is found to be , ( 51 ) clt , lin=⟨⟨λlin⟩Td⟩-mlt , lin⟨ld-lb⟩=0 . The best linear fit for the ⟨λlin⟩Td vs ld-lb plot follows the trend y = x . Simulations of the adder model for linearly growing cells were carried out . The deviation of the best linear fit for the ld-lb vs ⟨λlin⟩Td plot from the y = x line is shown in Figure 3—figure supplement 1A , while in Figure 3—figure supplement 1B , the best linear fit for the plot ⟨λlin⟩Td vs ld-lb is shown to agree with the y = x line . In this section , we explore the equation for the best linear fit of ⟨λlin⟩Td vs ld-lb plot in the case of exponential growth and ⟨λ⟩Td vs ln ( LdLb ) plot for linear growth . Intuitively , we expect the best linear fit in both cases to deviate from the y = x line . In this section , we will calculate the best linear fit explicitly . Surprisingly , we will find that , in the case of linear growth , the best linear fit for the ⟨λ⟩Td vs ln ( LdLb ) plot follows the y = x line closely . Let us begin with exponential growth with growth rate , λ = ⟨λ⟩ + ⟨λ⟩ξ ( 0 , CVλ ) as defined previously . Again , ξ ( 0 , CVλ ) has a normal distribution with mean 0 and standard deviation CVλ , it being the CV of the growth rate . The time at division is given by Equation 7 . The average growth rate ⟨λ⟩ = ⟨ln ( 2 ) Td⟩≈ln ( 2 ) ⟨Td⟩ . For exponential growth , we will plot ⟨λlin⟩Td vs ld-lb . As previously defined , ⟨λlin⟩ is the mean normalized elongation speed and ⟨λlin⟩=⟨1Td⟩≈1⟨Td⟩ . ⟨λlin⟩ is related to ⟨λ⟩ by , ( 52 ) ⟨λlin⟩=⟨λ⟩ln ( 2 ) . ld-lbcan be calculated by using the regulation strategy f ( lb ) introduced in the Model section and a normally distributed size additive noise ζs ( 0 , σbd ) . Note that we have chosen the noise in division timing to be size additive . However , the model is robust to the choice of type of noise as we showed in the Exponential growth section . Using Equations 5 and 6 we obtain , ( 53 ) ldn-lbn≈1+ ( 1-2α ) xn+ζs ( 0 , σbd ) . Using Equation 11 , ⟨λlin⟩Td is obtained to be , ( 54 ) ⟨λlin⟩Td=1-αxln ( 2 ) -ξ ( 0 , CVλ ) +ζs ( 0 , σbd ) 2ln ( 2 ) . To calculate the expression for mlt , lin , the slope of the best linear fit for ⟨λlin⟩Td vs ld-lb plot , we first calculate ρlin given by Equation 46 . The expression for σl , lin ( standard deviation of ld-lb ) and σt ( standard deviation of Td ) are found to be , ( 55 ) σl , lin2= ( 1-2α ) 2σx2+σbd2 , ( 56 ) σt2=1⟨λlin⟩2 ( ( ασxln ( 2 ) ) 2+CVλ2+ ( σbd2ln ( 2 ) ) 2 ) . σxis related to σn via Equation 17 . In Exponential growth section , we also showed that σn = σbd2 . Using these , we can write , ( 57 ) σx2=σbd24α ( 2-α ) . Now using the expressions for σt , σl , lin and the fact that x , ξ ( 0 , CVλ ) and ζs ( 0 , σbd ) are independent of each other , we get , ( 58 ) ρlin= ( 2α−1 ) ασx2ln ( 2 ) +σbd22ln ( 2 ) ⟨λlin⟩σl , linσt . For the plot ⟨λlin⟩Td vs ld-lb , the slope mlt , lin is given by , ( 59 ) mlt , lin=ρlinσt⟨λlin⟩σl , lin= ( 2α−1 ) ασx2ln ( 2 ) +σbd22ln ( 2 ) σl , lin2 . Inserting Equation 55 into Equation 59 and substituting σx2 given by Equation 57 , we obtain , ( 60 ) mlt , lin=1ln ( 2 ) 3α4α+1 . The intercept clt , lin is found to be , ( 61 ) clt , lin=⟨⟨λlin⟩Td⟩-mlt , lin⟨ld-lb⟩=1-1ln ( 2 ) 3α4α+1 . For the adder model ( α = 12 ) , we get the value of slope mlin , lt = 12ln ( 2 ) ≈0 . 7213 and intercept clin , lt = 1-12ln ( 2 ) ≈0 . 279 . This is different from the best linear fit obtained for same regulatory mechanism controlling division in linearly growing cells where we found that the best linear fit follows the y = x line . Intuitively , we expect the best linear fit of ⟨λlin⟩Td vs ld-lb plot to deviate from y = x line in the case of exponential growth . We showed analytically that for a class of models where birth controls division , it is indeed the case . This is also shown using simulations of the adder model in Figure 3—figure supplement 1C . In the ‘ Obtaining the best linear fit’ section , we found the best linear fit for ⟨λ⟩Td vs ln ( LdLb ) plot to follow the y = x line for exponentially growing cells where division is regulated by birth event via regulation strategy f ( lb ) . Next , we calculate the equation for the best linear fit of ⟨λ⟩Td vs ln ( LdLb ) plot given that growth is linear . The model for division control will be same as that in the Linear growth section that is , the regulation strategy for division is given by g ( lb ) = 2+2 ( 1-α ) ( lb-1 ) which is also equivalent to f ( lb ) . The linearly growing cells grow with elongation speed λlin = ⟨λlin⟩ ( 1+ξlin ( 0 , CVλ , lin ) ) . As discussed before , ξlin ( 0 , CVλ , lin ) has a normal distribution with mean 0 and standard deviation CVλ , lin , it being the CV of the elongation speed . Using Equations 5 and 6 , we get , ( 62 ) ln ( LdLb ) =ln ( 2 ) -αxn+ζs ( 0 , σbd ) 2 . Using Equations 5 and 52 , we obtain from Equation 41 , ( 63 ) ⟨λ⟩Td=ln ( 2 ) + ( 1-2α ) ln ( 2 ) x+ln ( 2 ) ζs ( 0 , σbd ) -ln ( 2 ) ξlin ( 0 , CVλ , lin ) . Since x , ξlin ( 0 , CVλ , lin ) and ζs ( 0 , σbd ) are uncorrelated , the standard deviation of ln ( LdLb ) and Td denoted by σl and σt respectively are calculated to be , ( 64 ) σl2=α2σx2+σbd24 , ( 65 ) σt2=ln2 ( 2 ) ⟨λ⟩2 ( ( 1-2α ) 2σx2+σbd2+CVλ , lin2 ) . We calculate the correlation coefficient for the pair ( ln ( LdLb ) , ⟨λ⟩Td ) . Since the correlation coefficient is unaffected by multiplying one of the variables with a positive constant , we can calculate the correlation coefficient for the pair ( ln ( LdLb ) , Td ) or ρexp as given by Equation 18 . Using the independence of terms x , ξlin ( 0 , CVλ , lin ) and ζs ( 0 , σbd ) , ( 66 ) ρexp=ln ( 2 ) ( σx2 ( 2α−1 ) α+σbd22 ) ⟨λ⟩σlσt . For the plot ⟨λ⟩Td vs ln ( LdLb ) , the slope mlt of the best linear fit is given by , ( 67 ) mlt=ρexpσt⟨λ⟩σl=ln ( 2 ) ( σx2 ( 2α−1 ) α+σbd22 ) σl2 . Inserting Equation 64 into Equation 67 and using Equation 57 , we get , ( 68 ) mlt=32ln ( 2 ) ≈1 . 0397 . Similarly the intercept ( clt ) for the plot ⟨λ⟩Td vs ln ( LdLb ) is found to be , ( 69 ) clt=⟨⟨λ⟩Td⟩-mlt⟨ln ( LdLb ) ⟩=ln ( 2 ) ( 1-32ln ( 2 ) ) ≈-0 . 0275 . This is very close to y = x trend obtained for the same regulatory mechanism controlling division in exponentially growing cells ( Figure 3A ) . In the previous sections , we found that binning and linear regression on the plot ln ( LdLb ) vs ⟨λ⟩Td , and the plot obtained by interchanging the axes , were inadequate to identify the mode of growth . In this section , we try to validate the growth rate vs age plot as a method to elucidate the mode of growth . In addition to cell size at birth and division and the generation time , cell size trajectories ( cell size , L vs time from birth , t ) were obtained for multiple cell cycles . In our case , the cell size trajectories were collected either via simulations ( in Figure 3B ) or from experiments ( for Figure 4A-C ) at intervals of 4 min . Note that if the measurements were to be carried out at equal length intervals instead of time , the results discussed in the paper would still remain unchanged . For each trajectory , growth rate at time t or age tTd is calculated as 1L ( t ) L ( t+Δt ) -L ( t ) Δt where Δt is the time between consecutive measurements . To obtain elongation speed vs age plots , the formula before needs to be replaced with L ( t+Δt ) -L ( t ) Δt . The growth rate is interpolated to contain 200 points at equal intervals of time for each cell trajectory . The growth rate trends appear to be robust with regards to a different number of interpolated points ( from 100 to 500 points ) . To obtain the growth rate trend as a function of cell age , we use the method previously applied in Nordholt et al . , 2020 . In this method , growth rate is binned based on age for each individual trajectory ( 50 bins ) and the average growth rate is obtained in each of the bins . The binned data trend for growth rate vs age is then found by taking the average of the growth rate in each bin over all trajectories . Binning the growth rate for each trajectory ensures that each trajectory has an equal contribution to the final growth rate trend so as to avoid inspection bias . This step is especially important when data collected at equal intervals of time is analyzed . In such a case , cells with larger generation times have a greater number of measurements than cells with smaller generation times . Obtaining the growth rate trend without binning growth rate for each trajectory would have biased the binned data trend for the growth rate vs age plot to a smaller value because of over-representation by slower-growing cells ( or equivalently cells with longer generation time ) . This bias toward lower growth rate values in the growth rate vs age plots is an instance of inspection bias . In Figure 4A-C , we find the growth rate obtained from E . coli experiments to change within the cell cycle . In the two slower growth media ( Figure 4A and B ) , the growth rate is found to increase with cell age while for the fastest growth media ( Figure 4C ) the growth rate follows a non-monotonic behaviour similar to that observed in Nordholt et al . , 2020 for B . subtilis . Abrupt changes in growth rate are reported at constriction in Reshes et al . , 2008; Banerjee et al . , 2017 . We find that the growth rate changes start before constriction in the two slower growth conditions considered . One possibility is that this increase is due to preseptal cell wall synthesis ( Pazos et al . , 2018 ) . Preseptal cell wall synthesis does not require activity of PBP3 ( FtsI ) but instead relies on bifunctional glycosyltransferases PBP1A and PBP1B that link to FtsZ via ZipA . One hypothesis that can be tested in future works is that at the onset of constriction , activity from PBP1A and PBP1B starts to gradually shift to the PBP3/FtsW complex and therefore no abrupt change in growth rate is observed . In the fastest growth condition ( glucose-cas medium ) , we find that the increase in growth rate approximately coincides with onset of constriction , in agreement with the previous findings ( Reshes et al . , 2008; Banerjee et al . , 2017 ) . In Figure 4A-C , the growth rate trends are not obtained for age close to one . This is because growth rate at age = 1 is given by 1L ( Td ) L ( Td+Δt ) -L ( Td ) Δt and this requires knowing the cell lengths beyond the division event ( L ( Td+Δt ) ) . To estimate growth rates at age close to one , we approximate L ( Td+Δt ) to be the sum of cell sizes of the two daughter cells . In order to minimize inspection bias , we considered only those cell size trajectories which had L ( t ) data for 12 min after division ( corresponding to an age of approximately 1 . 1 ) . However , the growth rate trends in all three growth media were robust with regards to a different time for which L ( t ) was considered ( 4 min to 20 min after division ) . We use the binning procedure discussed before in this section . To validate this method , we applied it on synthetic data obtained from the simulations of exponentially growing cells following the adder and the adder per origin model . Cells were assumed to divide in a perfectly symmetric manner and both of the daughter cells were assumed to grow with the same growth rate , independent of the growth rate in the mother cell . The growth rate trends for the two models considered ( adder and adder per origin ) are expected to be constant even for cell age >1 . We found that the growth rate trends were indeed approximately constant as shown in Figure 4—figure supplement 1D . We also considered linear growth with division controlled via an adder model . The daughter cells were assumed to grow with the same elongation speed , independent of the elongation speed in the mother cell . In this case , we expect the elongation speed trend to be constant for cell age >1 . This is indeed what we observed as shown in the inset of Figure 4—figure supplement 1D . We used this method on E . coli experimental data and found that the growth rate trends obtained for the three growth conditions ( Figure 4—figure supplement 1A–1C ) were consistent with that shown in Figure 4A-C in the relevant age ranges . For cell age close to one , we found that the growth rate decreased to a value close to the growth rate near cell birth ( age ≈ 0 ) for all three growth conditions considered . In summary , we find that the growth rate vs age plots are a consistent method to probe the mode of cell growth within a cell cycle . To probe the growth rate trend in relation to a specific cell cycle event , for example cell birth , growth rate vs time from birth plots are obtained for simulations of exponentially growing cells following the adder model . In the growth rate vs time from birth plot , the rate is found to stay constant and then decrease at longer times ( Figure 3—figure supplement 2C ) even though cells are exponentially growing . Because of inspection bias ( or survivor bias ) , at later times , only the cells with larger generation times ( or slower growth rates ) ‘survive’ . The average generation time of the cells averaged upon in each bin of Figure 3—figure supplement 2C is shown in Figure 3—figure supplement 2D . The decrease in growth rate in Figure 3—figure supplement 2C occurs around the same time when an increase in generation time is observed in Figure 3—figure supplement 2D . Thus , the trend in growth rate is biased toward lower values at longer times . The problem might be circumvented by restricting the time on the x-axis to the smallest generation time of all the cell cycles considered ( Messelink et al . , 2020 ) . To check for growth rate changes at constriction , we used plots of growth rate vs time from constriction ( t-Tn ) . Growth rate trends obtained from E . coli experimental data show a decrease at the edges of the plots ( Figure 4—figure supplement 2A and C , and 2E ) . These deviate from the trends obtained using the growth rate vs age plots ( Figure 4A-C ) . To investigate this discrepancy , we use a model which takes into account the constriction and the division event . Currently , it is unknown how constriction is related to division . For the purpose of methods validation , we use a model where cells grow exponentially , constriction occurs after a constant size addition from birth , and division occurs after a constant size addition from constriction . Note that other models where constriction occurs after a constant size addition from birth while division occurs after a constant time from constriction , as well as a mixed timer-adder model proposed in Banerjee et al . , 2017 , lead to similar results . We expect the growth rate trend to be constant for exponentially growing cells . However , we find using numerical simulations that it decreases at the plot edges both before and after the constriction event ( Figure 3—figure supplement 2A ) . This decrease can be attributed to inspection bias . The average growth rate in time bins at the extremes are biased by cells with smaller growth rates . This is shown in Figure 3—figure supplement 2B where the average generation time for the cells contributing in each of the bins of Figure 3—figure supplement 2A is plotted . The time at which the growth rate decreases on both sides of the constriction event is close to the time at which the average generation time increases . For example , in alanine medium , the generation time for each of the bins is plotted in Figure 4—figure supplement 2B . The average generation time for the cells contributing to each of the bins is almost constant for the timings between –80 min and 20 min . Thus , for this time range the changes in growth rate are not because of inspection bias but are a real biological effect . The behavior of growth rate within this time range in Figure 4—figure supplement 2A is in agreement with the trend in growth rate vs age plot of Figure 4A . On accounting for inspection bias , the growth rate vs age plots agree with the growth rate vs time from constriction plots in other growth media as well ( Figure 4—figure supplement 2C , Figure 4—figure supplement 2E ) . Thus , growth rate vs time plots are also a consistent method to probe growth rate modulation in the time range when avoiding the regimes prone to inspection bias . Cells assumed to undergo exponential growth have elongation speed proportional to their size . In the case of exponential growth , the binned data trend of the plot elongation speed vs size is expected to be linear with the slope of the best linear fit providing the value of growth rate and intercept being zero . In this section , we use the simulations to test if binning and linear regression on the elongation speed vs size plots are suitable methods to differentiate exponential growth from linear growth ( Cadart et al . , 2019 ) . To test the method , we generate cell size trajectories using simulations of the adder model with a size additive division timing noise and assuming exponential growth . Elongation speed at size L ( t ) is calculated for each trajectory as L ( t+Δt ) -L ( t ) Δt where Δt is the time between consecutive measurements ( = 4 min in our case ) . Each trajectory is binned into 10 equally sized bins based on their cell sizes and the average elongation speed is obtained for each bin . The final trend of elongation speed as a function of size is then obtained by binning ( based on size ) the pooled average elongation speed data of all the cell cycles . We find that the binned data trend is linear with the slope of the best linear fit close to the average growth rate considered in the simulations ( Figure 3—figure supplement 3D ) . This is in agreement with our expectations for exponential growth . In order to check if this method could differentiate between exponential growth and linear growth , we used simulations of the adder model undergoing linear growth to generate cell size trajectories for multiple cell cycles . For linear growth , elongation speed is expected to be constant , independent of its cell size . The binned data trend for the elongation speed vs size plot is also obtained to be constant for the simulations of linearly growing cells ( Figure 3—figure supplement 3B ) . The intercept of the best linear fit obtained is close to the average elongation speed considered in the simulations . The binned data trend for linear and exponential growth are clearly different as shown in Figure 3—figure supplement 3B and Figure 3—figure supplement 3D , respectively , and this result holds for a broad class of models where the division event is controlled by birth and the growth rate ( for exponential growth ) /elongation speed ( for linear growth ) is distributed normally and independently between cell-cycles . Next , we consider the adder per origin cell cycle model for exponentially growing cells ( Ho and Amir , 2015 ) . In this model space , the cell initiates DNA replication by adding a constant size per origin from the previous initiation size . The division occurs on average after a constant time from initiation . For exponentially growing cells , the binned data trend is still expected to be linear as before . Instead , we find using simulations that the trend is non-linear and it might be misinterpreted as non-exponential growth ( Figure 3—figure supplement 3F ) . Thus , the results of binning and linear regression for the plot elongation speed vs size is model-dependent . So far , our discussion was focused on the question of mode of single-cell growth . A related problem regards the relation between growth rate ( λ ) and the inverse generation time ( 1Td ) . On a population level , the two are clearly proportional to each other . However , single-cell studies based on binning showed an intriguing non-linear dependence between the two , with the two variables becoming uncorrelated in the faster-growth media ( Kennard et al . , 2016; Iyer-Biswas et al . , 2014 ) . Within the same medium , the binned data curve for the plot λ vs 1Td flattened out for faster dividing cells . The trend in the binned data was different from the trend of y = ln ( 2 ) x line as observed for the population means . A priori one might speculate that the flattening in faster dividing cells could be because the faster dividing cells might have less time to adapt their division rate to transient fluctuations in the environment . Kennard et al . , 2016 insightfully also plotted 1Td vs λ and found a collapse of the binned data for all growth conditions onto the y = 1ln ( 2 ) x line . These results are reminiscent of what we previously showed for the relation of ln ( LdLb ) and ⟨λ⟩Td . In the following , we will elucidate why this occurs in this case using an underlying model and predicting the trend based on it . We use simulations of the adder model undergoing exponential growth . The parameters for size added in a cell cycle and mean growth rates are extracted from the experimental data . CV of growth rate is assumed lower in faster growth media as observed by Kennard et al . Using this model , we could obtain the same pattern of flattening at faster growth conditions that is observed in the experiments ( Figure 2—figure supplement 2A ) . The population mean for λ and 1Td follows the expected y = ln ( 2 ) x equation ( shown as black dashed line ) as was the case in experiments . Intuitively , such a departure from the expected y = ln ( 2 ) x line for the single-cell data can again be explained by determining the effect of noise on variables plotted on both axes . As previously stated Td is affected by both growth rate noise and noise in division timing while growth rate fluctuates independently of other sources of noise . This does not agree with the assumption for binning as noise in division timing affects the x-axis variable rather than the y-axis variable . In such a case , the trend in the binned data might not follow the expected y = ln ( 2 ) x line . However , on interchanging the axes , we would expect the assumptions of binning to be met and the trend to follow the y = 1ln ( 2 ) x line ( Figure 2—figure supplement 2B ) . | All cells – from bacteria to humans – tightly control their size as they grow and divide . Cells can also change the speed at which they grow , and the pattern of how fast a cell grows with time is called ‘mode of growth’ . Mode of growth can be ‘linear’ , when cells increase their size at a constant rate , or ‘exponential’ , when cells increase their size at a rate proportional to their current size . A cell’s mode of growth influences its inner workings , so identifying how a cell grows can reveal information about how a cell will behave . Scientists can measure the size of cells as they age and identify their mode of growth using single cell imaging techniques . Unfortunately , the statistical methods available to analyze the large amounts of data generated in these experiments can lead to incorrect conclusions . Specifically , Kar et al . found that scientists had been using specific types of plots to analyze growth data that were prone to these errors , and may lead to misinterpreting exponential growth as linear and vice versa . This discrepancy can be resolved by ensuring that the plots used to determine the mode of growth are adequate for this analysis . But how can the adequacy of a plot be tested ? One way to do this is to generate synthetic data from a known model , which can have a specific and known mode of growth , and using this data to test the different plots . Kar et al . developed such a ‘generative model’ to produce synthetic data similar to the experimental data , and used these data to determine which plots are best suited to determine growth mode . Once they had validated the best statistical methods for studying mode of growth , Kar et al . applied these methods to growth data from the bacterium Escherichia coli . This showed that these cells have a form of growth called ‘super-exponential growth’ . These findings identify a strategy to validate statistical methods used to analyze cell growth data . Furthermore , this strategy – the use of generative models to produce synthetic data to test the accuracy of statistical methods – could be used in other areas of biology to validate statistical approaches . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"physics",
"of",
"living",
"systems",
"microbiology",
"and",
"infectious",
"disease"
] | 2021 | Distinguishing different modes of growth using single-cell data |
Biophysical properties of neurons become increasingly diverse over development , but mechanisms underlying and constraining this diversity are not fully understood . Here we investigate electrophysiological characteristics of Xenopus tadpole midbrain neurons across development and during homeostatic plasticity induced by patterned visual stimulation . We show that in development tectal neuron properties not only change on average , but also become increasingly diverse . After sensory stimulation , both electrophysiological diversity and functional differentiation of cells are reduced . At the same time , the amount of cross-correlations between cell properties increase after patterned stimulation as a result of homeostatic plasticity . We show that tectal neurons with similar spiking profiles often have strikingly different electrophysiological properties , and demonstrate that changes in intrinsic excitability during development and in response to sensory stimulation are mediated by different underlying mechanisms . Overall , this analysis and the accompanying dataset provide a unique framework for further studies of network maturation in Xenopus tadpoles .
Electrophysiological properties of neurons become increasingly diverse over development in ways that are critical for proper nervous system function and maturation ( Turrigiano and Nelson , 2004; Marder and Goaillard , 2006 ) . Perturbation of these processes can have broad and devastating consequences leading to neurodevelopmental disorders such as mental retardation , autism , and schizophrenia ( Rice and Barone , 2000; Belmonte et al . , 2004; Pratt and Khakhalin , 2013 ) . It remains unclear , however , to what degree this diversity in electrophysiological tuning reflects intrinsic developmental differentiation , and how much it reflects the particular activation history of a given neuron , as well as the constraints that shape how well neurons adapt to changes in their input patterns . The adaptability of electrophysiological properties is central for allowing developing neural circuits to maintain functional stability , while simultaneously providing flexibility for accommodating developmental changes . One mechanism that contributes to this balance is homeostatic plasticity , whereby neurons adjust their synaptic and intrinsic properties based on the activity of the circuit in which they are embedded ( Daoudal and Debanne , 2003; Desai , 2003; Turrigiano and Nelson , 2004; Ibata et al . , 2008; Turrigiano , 2008; Marder , 2011 ) . Homeostatic plasticity allows developing circuits to function stably by maximizing their dynamic range as new inputs become incorporated ( Bucher et al . , 2005; Marder and Goaillard , 2006; Pratt and Aizenman , 2007 ) . This is particularly relevant to developing animals: their nervous system must be functional and able to interact with its environment even as nascent circuitry is still developing . One place where this adaptability in synaptic and intrinsic properties is particularly salient , is in the optic tectum of Xenopus laevis tadpoles—a midbrain area that processes inputs from visual , auditory , and mechanosensory systems ( Cline , 1991; Ewert , 1997; Cline , 2001; Ruthazer and Cline , 2004; Ruthazer and Aizenman , 2010 ) . Sensory inputs to the tectum are strengthened over development , resulting in increasingly robust synaptic responses , yet this strengthening is accompanied with decreases in intrinsic excitability that may function to maintain a stable dynamic range in this circuit ( Pratt and Aizenman , 2007 ) . As a consequence , visually guided behaviors , such as collision avoidance , improve and become more tuned to specific stimuli ( Dong et al . , 2009 ) . Changes in sensory environment can also elicit homeostatic plasticity in tectal cells , resulting in adjustment of both synaptic and intrinsic properties ( Aizenman et al . , 2003; Deeg and Aizenman , 2011 ) . Since homeostatic plasticity coordinates changes of different cellular properties , over time it is expected to constrain these properties , limiting ways in which they can co-vary within the population of cells ( O'Leary et al . , 2013 ) : for example , strong excitatory synaptic drive results in lower intrinsic excitability . Coordinated changes in different physiological properties may contribute to diversification of cell tuning that happens as networks mature , creating and shaping differences in cell phenotypes both between cell types as they emerge ( Ewert , 1974; Frost and Sun , 2004; Kang and Li , 2010; Nakagawa and Hongjian , 2010; Liu et al . , 2011 ) , and within each cell type in a functional network ( Tripathy et al . , 2013; Elstrott et al . , 2014 ) . These considerations suggest that multivariate distributions of different physiological properties sampled across many cells in a network may contain unique information both about current tuning of this network , and the mechanisms behind this tuning that may act through local recalibration of properties in individual cells ( O'Leary et al . , 2013 ) . Yet relatively few studies have attempted this kind of analysis on a large scale so far . Here we perform a large-scale electrophysiological census of retinorecipient neurons in the developing Xenopus laevis tectum to better understand the electrophysiological variability of tectal neurons in development , and in response to a need for homeostatic change . Using a comprehensive suite of tests we describe relationships between 33 electrophysiological variables , and show that both the variability and the predictability of multivariate cell tuning increases over development , and undergo changes in response to sensory stimulation . By analyzing groups of neurons that produce similar spike trains , we also show that similar spiking behaviors may be supported by different combinations of underlying electrophysiological properties .
We recorded from 155 deep-layer , retinorecipient tectal cells across developmental stages 43 to 49 ( Nieuwkoop and Faber , 1994 ) from 42 animals , measuring from 9 to 33 different electrophysiological variables in each cell ( median of 26 variables per cell; see Figure 1 and Supplementary file 1 for a graphical description and a concise table of variables , and the Materials and methods for a detailed description of each variable ) . Of 155 cells , 35 cells contributed to all 33 variables , 62 contributed to 30 variables , 124 to 20 variables , and 154 to 10 variables . Across different variables , the least covered variable had measurements from 64 cells , while the most covered one had measurements from 154 cells ( median of 134 cells per variable ) ; in total , 18% of all possible observations were missing . The dataset containing analysis parameters is available online as Supplementary file 2 . The entire dataset including raw electrophysiology files has also been made available and can be accessed with the following doi:10 . 5061/dryad . 18kk6 . 10 . 7554/eLife . 11351 . 003Figure 1 . A review of cell properties characterized in this study ( See methods for a detailed description of every measurement ) . ( A ) Response to voltage clamp steps after passive current subtraction; full currents on the left , a zoom-in look at early active currents on the right . Red vertical lines show how voltage-gated sodium ( INa ) , stable potassium ( IKS ) , and transient potassium currents ( IKT ) were measured . ( B ) IV curve for INa . ( C ) IV curve for IKS . ( D ) IV curve for IKT . ( E ) Spiking responses to current step injections of different amplitudes; first response to produce a spike is shown in blue; response generating maximal number of spikes is shown in black . ( F ) An expanded look at the first spike produced by the cell in response to step current injections . ( G ) Number of spikes as a function of step current injection amplitude . ( H ) A trace of membrane potential recorded from the cell in response to cosine current injections of varying frequency . ( I ) Spike-raster of 10 consecutive responses to cosine injections shown in H . ( J ) Average number of spikes in response to a single cosine injection as a function of cosine period . ( K ) Average number of spikes per cosine wave in response to injections of shortest period ( 10 ms; leftmost group in panels H and I . ( L ) Sample trace of excitatory synaptic currents recorded in voltage clamp mode in response to optic chiasm stimulation with a 30 ms inter-stimulus interval . ( M ) Total postsynaptic charge as a function of inter-stimulus interval . ( N ) Average postsynaptic current showing time-windows that were used to build the "Monosynapticity ratio" . Panels ( A–D ) originate from one cell; panels ( E–M ) were recorded in another cell; both from stage 49 animals . DOI: http://dx . doi . org/10 . 7554/eLife . 11351 . 003 With 33 different cell parameters , we had 528 potential pairwise correlations to assess . Ninety of these correlations were significant after FDR adjustment ( α = 0 . 05 , corresponding to Pearson correlation p-value threshold of about 8e−3; Figure 2A ) ; all correlations with r > 0 . 5 ( n=11 ) were also significant for Spearman correlation after same FDR adjustment . We found that there were at least three distinct clusters of tightly correlated variables representing different neural characteristics: tectal cell spikiness ( a tendency to produce more spikes ) ; measurements of spike shape and dynamics that strongly anti-correlated with spikiness; and intrinsic neuronal properties , such as membrane capacitance and amplitudes of intrinsic ionic currents . Some of the more salient correlations are shown in Figure 2B . 10 . 7554/eLife . 11351 . 004Figure 2 . Correlations between cell properties . ( A ) Diagram illustrating significant correlations and their strengths ( r-values ) for all pairs of cell properties measured . Positive correlations are shown in red; negative correlations in blue; the width and darkness of each line are proportional to respective r-value . ( B ) Selected significant correlations between cell variables . ( C ) Variables sorted by the total variance each of them can explain in the full dataset ( amounts are not additive and do not add up to 100% ) . Colors indicate the biological nature of each variable ( red for spiking , green for spike shape , blue for spike timing , orange for ionic currents , yellow for activation potentials , gray for passive properties , and purple for synaptic properties ) . The white dashed line shows the expected variance explained for an ideally uncorrelated variable . DOI: http://dx . doi . org/10 . 7554/eLife . 11351 . 004 The presence of correlations in our dataset suggested that not all variables were independent , which is expected for sets of variables that describe different aspects of common underlying cellular qualities , such as spikiness or synaptic effectiveness ( McGarry et al . , 2010 ) . To make sure that none of the variables were redundant ( brought no new information to the set ) , or too noisy ( having no interactions with other variables in the set ) , we ran the so-called Principal Variable Analysis , quantifying the total amount of variance in the dataset explained by each variable ( Mccabe , 1984 ) . We found that the most informative , and thus least independent , variables were related to the number of spikes the cell was able to generate ( N spikes to cosine and step injections explained 15% and 14% of total variance , respectively ) , or different aspects of spike shape ( 9–13% of total variance; Figure 2C , top of the list ) . Conversely , some variables did not serve as good linear predictors for other properties in the set ( bottom of Figure 2C ) , suggesting that they were more independent — the lowest variable explained 4% of total variance , which was still more than the 3% expected for a fully uncorrelated variable . In summary , it can be seen from Figure 2C that no single variable was 'too good' in explaining overall variance in the dataset , but also that none of them fell at or below the predicted noise level; there was no clear division of variables into distinct groups , but rather a smooth decline across the list . Finally , variables of different biological nature were diversely distributed across the parts of the plot corresponding to higher and lower explanatory power . From this we concluded that our dataset was well balanced for exploratory analysis , offering a healthy mix of independent and interacting variables ( Guyon and Elisseeff , 2003 ) . Knowing that the maximal number of spikes in response to current step injections can alone explain 14% of total variance in the dataset made us wonder which protocol of those we used was the most informative in the sense of best capturing the electrophysiological identity of each tectal cell . To answer this question , we ran the Principal Variance Analysis on a combination of variables coming from different protocols . We found that 9 variables from the step current injection protocol together explained 42% of total variance in the set; 7 variables quantifying IV-curves explained 34%; 6 variables from cosine current injections explained 33%; 5 synaptic variables and 4 passive membrane properties both accounted for 21%; and 2 variables describing miniature postsynaptic currents ( mEPSCs ) predicted 11% of total variance ( shares of explained variance don't have to add up to 100% ) . Based on this analysis , we conclude that the step current injection protocol ( see Figure 1E–G ) is thus one of the most informative , and should be recommended for fast profiling of cell types in the future . We next asked which electrophysiological characteristics changed with development as tadpoles matured from stage 43 , a time when axons from the eye make first functional connections in the brain ( Holt and Harris , 1983 ) , to stage 49 , when tectal networks become sufficiently refined to support spatially coordinated visual behaviors and multisensory integration ( Deeg et al . , 2009; Dong et al . , 2009; Xu et al . , 2011; Khakhalin et al . , 2014 ) . To compensate for the unevenness of data availability across cell properties and developmental stages , we combined data from stages 43–44 , 45–46 , and 48–49 , as these groupings have proven useful in previous studies ( Pratt et al . , 2008; Dong et al . , 2009; Sharma and Cline , 2010 ) . We did not group stage 47 with others however , as there were indications that this transient stage in development may be unique in terms of tadpole behavior , network excitability , and average tectal cell properties ( Pratt and Aizenman , 2007; Bell et al . , 2011 ) . We found that 12 cell properties changed significantly across these developmental periods ( PANOVA < 0 . 05 ) , with 5 values decreasing overall , and 6 variables increasing ( Figure 3; also see Supplementary file 3 for summary table ) . 10 . 7554/eLife . 11351 . 005Figure 3 . Changes in cell properties with age . All cell properties that significantly changed with development are shown here as mean values ( central line ) and standard deviations ( whiskers and shading ) . Transitions between points are shown as shape-preserving piecewise cubic interpolations . DOI: http://dx . doi . org/10 . 7554/eLife . 11351 . 005 While average values changed in both directions during development , the effect on variability of cellular properties was much more consistent: 14 variables out of 33 increased their variability from stages 45–46 to 48–49 ( PV < 0 . 05 , corresponding to increases in standard deviation of 40% and higher; see Supplementary file 3 ) . Among others , spiking inactivation ( as measured by 'Wave decay’ ) , maximal amplitudes of sodium and slow potassium ionic currents , the frequency of minis , and the total synaptic charge all experienced an almost twofold increase in variability . Only two properties out of 33 became less variable over development: synaptic resonance and synaptic resonance width . These data suggest that by stages 48–49 , neurons became more electrophysiologically diverse than they were at stages 45–46 . To visualize and explore the patterns behind co-dependence and co-variance of variables in our dataset , and to better measure common features that underlie these correlations , we used principal component analysis ( PCA ) . As not every variable was measured in every cell , we used an iterative Bayesian version of PCA known as 'PCA with missing values' ( Ilin and Raiko , 2010 ) , followed by a promax oblique rotation . We extensively verified the validity of our PCA analysis , comparing it to standard PCA on restricted and imputed data , PCA on rank-transformed data , as well as two most common non-linear dimensionality reduction approaches: Isomap and Local Linear Embedding ( see Materials and methods ) . We concluded that our PCA analysis was the most appropriate analysis for for this data set , and performed better than local non-linear approaches , with the first two principal components explaining 15% and 8% of total variance respectively ( this total of 23% of variance explained would have corresponded to ~35% of variance if we had every type of observation in every cell; see Materials and methods for details ) . A loading plot ( Figure 4A ) shows contributions of individual variables from the dataset to rotated PCA components . Points on the plot are colored according to the biological nature of each variable ( Figure 4 , see legend ) ; variables shown on the right contributed positively to the first component ( C1 ) , while those on the left contributed negatively to this component; variables in the upper part of the cloud contributed positively to the second component ( C2 ) , while those at the bottom contributed negatively to it . Consistent with high predictive value of individual variables related to spiking , we found that C1 describes the overall 'Spikiness' of each cell . Cells with large values of C1 generated many sharp and narrow spikes in response to current injections ( Figure 4A , green points describing spike shape , left; red points describing number of spikes , right ) . These cells spiked strongly in response to prolonged current injections ( Figure 4A , red point for 'Spiking resonance width , ' upper right quadrant ) ; had low accommodation , both in spike amplitude ( Figure 4A , green , left bottom corner ) and inter-spike interval ( Figure 4A , blue , left top corner ) , and had high trial-to-trial spike-timing precision ( low jitter; Figure 4A , blue point , left top corner ) . Properties of cells with different C1 and C2 values can also be illustrated in a modified score-plot , in which traces of membrane potential responses to a 100 pA step current injection are arranged within the C1-C2 principal component space ( Figure 4B ) . Note the difference in spiking responses between cells on right ( high C1 ) and left ( low C1 ) sides of Figure 4B . Cells with large C1 also tended to be involved in polysynaptic networks ( Figure 4A , purple point for 'Monosynapticity coefficient , ' lower left quadrant ) and did not exhibit short-term facilitation of synaptic inputs during repeated stimulation ( Figure 4A , purple point for 'Synaptic PPF , ' lower left quadrant ) . Cells with low values of C1 exhibited opposite traits: they produced few broad , squat , quickly accommodating spikes ( Figure 4B , left ) , were not recruited in recurrent networks , and tended to have strong synaptic facilitation that could potentially indicate high plasticity of synaptic inputs ( Kleschevnikov et al . , 1997 ) . 10 . 7554/eLife . 11351 . 006Figure 4 . Principal Component Analysis ( PCA ) . ( A ) Loading-plot , presenting contribution of individual cell properties to the first two PCA components ( see detailed description in the text ) . Points are colored with regards to how they describe the spikiness of the cell ( red ) , shape of spikes ( green ) , their temporal properties ( blue ) , ionic currents ( orange ) , passive electrical properties ( gray ) , or synaptic properties of the cell ( purple ) . ( B ) Modified score-plot showing how individual cells score on first two PCA components , with responses of respective cells to step current injections used instead of standard plot markers . Responses on the right are spikier that those on the left , while responses on the bottom have a greater passive component than responses on the top . DOI: http://dx . doi . org/10 . 7554/eLife . 11351 . 006 The second component ( C2 ) can be loosely dubbed 'Current density': cells with high values of C2 had large intrinsic ionic currents ( voltage-gated sodium INa and slow potassium IKS currents ) , high membrane capacitance ( Cm ) and low membrane resistance ( Rm ) , consistent with a larger membrane surface , and received strong synaptic inputs , in terms of both frequency and amplitude of mEPSCs . These cells produced frequent and sharp spikes ( Figure 4A , blue point for 'Spike ISI' and green points for 'Spike rise-time' and 'Spike width , ' lower part of the plot ) , but also tended to have higher values of spike-timing jitter and inter-spike interval accommodation . Conversely , cells with low values of C2 behaved as smaller cells electrophysiologically ( low Cm , high Rm ) , and had weak intrinsic and spontaneous synaptic currents . As principal neurons in the optic tectum have relatively uniform geometrical cell body sizes ( Lazar , 1973 ) , these differences in electrophysiological properties may indicate different levels of electrical coupling between the cell bodies , where the recording was performed , and both dendritic and axonal compartments , where the currents are generated . After classifying all cells according to their position in the PCA space ( Figure 5A ) , we were able to visualize developmental maturation of tectal cells as movements of point clouds within the C1-C2 plane , and as changes of these clouds' shapes ( Figure 5B ) . It can be seen from Figure 5B that as cells matured , their representations in the PCA space migrated from the left to the right side of the plot . Indeed , the Spikiness ( C1 ) changed with age ( PANOVA = 1e−5 ) , first increasing from stages 43–44 , through 45–46 , and up to stage 47 ( PMW < 4e−3 at each transition , N = 11 , 64 and 24 respectively ) , and then decreasing again at stages 48–49 ( PMW = 0 . 02 , N = 24 , 56; Figure 5B ) . The value of C2 , or 'Current density' , did not change much over development ( PANOVA = 0 . 08 ) except for a slight decrease between stages 47 and 48–49 ( PMW = 0 . 02 , N = 24 , 56 ) . 10 . 7554/eLife . 11351 . 007Figure 5 . Evolution of PCA component scores with development . ( A ) Score-plot of PCA scores for all cells from the main dataset , with cells colored by the developmental stage of the animal: from reds for stages 43–44 , through yellows for stages 45–46 , to blues for stages 48–49 . Note that most red cells are located on the left , while most light blue cells ( stage 47 ) are located on the right . ( B ) Isolated sub-clouds of points from panel ( A ) , shown on the same axes as panel ( A ) , and illustrating the progression of different stages through the score-plot ( N points = 11 , 64 , 24 , and 56 ) . Estimated density kernels for sub-clouds are shown as colored backgrounds . Stages 43–47 illustrate that the cloud moved to the right , while at stages 48–49 it moved back to the center , and expanded at all directions ( see quantification in the text ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11351 . 007 The clouds of points shown in Figure 5B also differ in size and structure , with the stages 45–46 cloud being more compact and simple , and the stages 48–49 cloud being more sprawled with an uneven distribution of points . To quantify cloud size , we compared medians of pairwise Euclidian distances between points in the PCA space and found that they were larger at stage 48–49 than at any other developmental group ( PMW < 5e−7 for all comparisons ) . Likewise , the value of 'Current density' ( C2 ) was more variable at stages 48–49 than at stages 45–46 ( PV = 0 . 004 , N = 64 , 56 ) , and the average pairwise difference between cells in the original 33-dimentional space was 22% larger at stages 48–49 than at stages 45–46 ( PTT < 1e−11; see Materials and methods for details ) . This suggests that from stages 42 to 47 , neurons changed their properties consistently , gradually getting more spiky , but then scaling their average spikiness back at stages 48–49 , while simultaneously expanding to the whole range of possible C1-C2 values , increasing the diversity of electrophysiological tuning across individual tectal cells . We then quantified the presence of internal structure within the original 33-dimentional datasets , separately for younger and older cells , by performing multiple imputation on the data , and then running two different types of unsupervised feature analyses that identify structure in multivariate distributions: the hierarchical cluster analysis that looks for subclouds of points within a larger cloud , and local PCA ( as opposed to previously described global PCA that was run on the full set of data ) to quantify linear interdependencies between cell properties within each age group . For cluster analysis , we used the agglomerative nesting coefficient AGNES ( Struyf et al . , 1996 ) and found ( Figure 6A ) that the degree of clustering was 36% larger at stages 48–49 compared to stages 45–46 ( 0 . 67 ± 0 . 04 and 0 . 49 ± 0 . 03 respectively , PTT < 1e−11 ) , reflecting a substantial increase in within-group heterogeneity . For local PCA analyses , we looked at the amount of total variance explained by the first 2 components within each stage group as a measure of internal structure and linear interactions between different variables . We found ( Figure 6B ) that despite an increase in total variance and heterogeneity in older cells , locally run PCA explained a slightly higher degree of variance in these cells ( 0 . 35 ± 0 . 04 ) than in younger cells at stages 45–46 ( 0 . 31 ± 0 . 02; PTT < 1e−11 ) suggesting that in more mature networks , cell properties are more coordinated with each other . 10 . 7554/eLife . 11351 . 008Figure 6 . Internal structure of cell property distributions . ( A ) Agglomerative clustering coefficients for properties of naïve cells at stages 45–46 , 48–49 , and cells after sensory stimulation ( stage 49 ) . Higher values correspond to higher levels of clustering ( grouping ) ; lower values correspond to more Gaussian-like unimodal distributions . ( B ) The amount of within-group variance explained by the first two components of PCA for the same groups of data . Higher values correspond to higher correlations between different electrophysiological variables in the set . Both plots show means ± standard deviations of results obtained in the original 33-dimensional space after multiple imputation with subsampling . DOI: http://dx . doi . org/10 . 7554/eLife . 11351 . 008 While our analysis of neuronal maturation demonstrated that several cellular properties changed over development , both in terms of their average values and their variability , we were not able to tell whether these phenomena represented genetically determined developmental programs , or if they reflected experience-dependent adaptations of electrophysiological properties resulting from the cumulative sensory experience of each cell ( Dong et al . , 2009; Munz et al . , 2014 ) . It is known from previous studies that tadpoles exposed for several hours to strongly patterned visual stimulation show synaptic ( Aizenman et al . , 2002; Aizenman and Cline , 2007 ) , intrinsic ( Aizenman et al . , 2003; Dong et al . , 2009 ) and morphological ( Sin et al . , 2002 ) changes in individual tectal cells , as well as in tectal network activity ( Pratt et al . , 2008 ) , and animal behavior ( Dong et al . , 2009 ) . We provided four hours of strongly patterned visual stimulation to an experimental group of stage 49 tadpoles and then measured and analyzed same 33 variables from 65 cells ( across 19 animals ) . Of 33 electrophysiological properties , 8 changed significantly between naïve and visually stimulated cells ( see Supplementary file 3 ) . As previously described ( Aizenman et al . , 2003 ) , visually stimulated cells spiked more in response to step current injections , producing on average 7 . 1 ± 4 . 7 spikes per injection , compared to only 4 . 6 ± 3 . 3 spikes in naïve cells ( PMW = 5e−3 , N = 51 , 60; Figure 7A ) . Interestingly , while for naïve cells the number of spikes produced in response to cosine injections strongly correlated with the number of spikes in response to step injections ( r = 0 . 82 , Pcorr = 2e−27 , N = 108 ) , visually stimulated cells did not change their spiking response to cosine injections ( 0 . 8 ± 0 . 5 for naïve , 0 . 9 ± 0 . 5 for stimulated cells; p = 0 . 3 , N = 38 , 60 ) , indicating that changes in spikiness in development , and after visual experience , may be implemented by different biophysical mechanisms . Other variables that increased after stimulation included the absolute value of the holding current , and spike threshold potential ( Figure 7B ) . Three properties decreased after visual stimulation: membrane capacitance ( Figure 7C ) , speed of response build-up during cosine stimulation ( "Wave build-up" ) , and jitter . 10 . 7554/eLife . 11351 . 009Figure 7 . Visual stimulation changes some cell properties . ( A–C ) Visual stimulation increased the spikiness of stage 49 cells in response to step current injections ( A ) , increased spike threshold potential ( B ) , and decreased membrane capacitance of tectal cells ( C ) . ( D ) A survey of cell properties that significantly changed either during development ( red filled markers ) , in response to visual stimulation ( blue hollow circles ) , or to neither variable ( small black markers ) . Properties that changed both in development and after stimulation are shown as red markers with blue circles surrounding them . The position of each marker on the plot is defined by the share of variability explained by developmental stage or visual stimulation , presented as η2 effect size value , and taken with a sign that reflects the direction of the change . Properties that did not change significantly are labeled by their number ( see 'Materials and methods' or Supplementary file 3 for the full list ) . ( E ) Projection of cells from visually stimulated s49 animals ( black ) into PCA space defined by the analysis of naïve dataset , with naïve cells from s48-49 animals shown in blue . Shading shows estimated density kernels for respective groups . DOI: http://dx . doi . org/10 . 7554/eLife . 11351 . 009 Figure 7D shows η2 effect sizes and marks statistical significance for changes during development and in response to sensory stimulation ( Ferguson , 2009 ) . It can be seen that the overlap between normal changes in development and changes after visual stimulation is very small: only 3 electrophysiological properties changed significantly both in development and after stimulation — of these , one ( Spike threshold ) changed in opposite directions , and two others changed in the same direction . These results suggest that while visually stimulated cells from stage 49 animals seem to spike similarly to naïve stage 47 animals , the biophysical mechanisms that underlie high cellular excitability in these groups are not likely to be shared . Curiously , the only electrophysiological property that was significantly affected by both developmental stage and sensory history of the animal , and that could conceivably contribute to differences in cell excitability , was membrane capacitance — a variable that is usually interpreted as an estimation of cell size , and thus does not make an obvious first candidate as a underlying mechanism for short-term intrinsic cellular plasticity . Significant changes in membrane capacitance ( Figure 7D , low left corner ) and spike threshold potential ( Figure 7D , low right corner ) may indicate a change in electrical coupling between cell soma , from which recordings were performed , and the spike-generating axon hillocks , similar to what has been described in other preparations ( Grubb and Burrone , 2010; Kuba et al . , 2010 ) . To better compare changes after visual stimulation to those occurring over development , we projected electrophysiological data from sensory stimulated cells into the original PCA space defined by data from naïve tadpoles , and compared the resulting point cloud to points from naïve stage 48–49 animals ( Figure 7E ) . Compared to the naïve group ( Figure 7E , blue ) the 'Spikiness' ( C1 ) of neurons from visually stimulated animals ( Figure 7E , black ) was higher ( PMW = 0 . 02 , N = 56 and 60 for naïve and visually stimulated groups ) , while average 'Current density' ( C2 value ) of stimulated cells did not change ( PMW = 0 . 06 ) . The variances of C1 and C2 did not change significantly ( PV > 0 . 05 ) , however the size of the cloud , as measured by median pairwise Euclidean distance between points in PCA space , decreased in stimulated cells ( PMW < 1e−8 ) . Likewise , the average pairwise difference between cells in the original 33-dimensional space decreased by 11% after visual stimulation ( 29 ± 1 for stimulated , compared to 33 ± 1 for naïve stage 48–49 cells; PTT < 1e−15 ) . This change in cloud size is almost certainly because 8 out of 33 electrophysiological properties became significantly less variable after visual stimulation: namely , maximal Na+ and slow K+ voltage-gated currents , transient K+ current activation potential , spiking threshold , jitter , frequency and amplitude of miniature EPSCs , and synaptic resonance inter-stimulus interval ( PV < 0 . 05 for each of these comparisons ) . Only two properties were more variable in visually stimulated cells: mean number of spikes in response to step injections and membrane resistance Rm ( PV < 0 . 01 for both ) . We then analyzed the internal structure of naïve and stimulated datasets from stage 48–49 animals in the same way as it was done for different developmental stages . We found that the amount of internal heterogeneity , as quantified by the agglomerative nesting coefficient , was reduced in stimulated neurons ( 0 . 62 ± 0 . 02 ) compared to naïve cells ( 0 . 67 ± 0 . 04; PTT < 1e−15; Figure 6A ) , while the share of total variance explained by first two components of local PCA was higher in stimulated cells ( 0 . 41 ± 0 . 02 ) than in naïve cells ( 0 . 35 ± 0 . 04; PTT < 1e−15; Figure 6B ) . Together these findings suggest that visual stimulation of tectal neurons increased their propensity to spike and lowered overall cell-to-cell variability and within-group heterogeneity , making cells more alike . Visual stimulation also made different cell properties more interdependent and predictable , consistent with the consequences of a strong homeostatic constraint experienced by these neurons . Our data suggest that spiking properties of tectal neurons can be modulated in several different ways during development and in response to sensory stimulation . We used our dataset to try to infer biological processes that could underlie adjustment of spike output . To deduce these processes , we looked for basic intrinsic properties that could serve as good predictors of cell spiking output across all experimental groups ( 220 cells ) . We ran a set of competitive sequential linear regression models ( Calcagno and de Mazancourt , 2010 ) to test whether we could predict the number of spikes produced by cells in response to current step injections based on their low-level electrophysiological properties , namely: membrane resistance ( Rm ) , membrane capacitance ( Cm ) , and both ionic current activation potentials and maximal amplitudes for sodium ( INa ) , stable potassium ( IKS ) and transient potassium ( IKT ) voltage-gated currents ( 8 variables total ) . We found that across all cells , the pair of properties that explained most of spike-output variance was a pair of activation potentials for sodium and stable potassium voltage-gated currents ( INa and IKS; 18% of variance without interaction , 19% with interaction ) . After activation potentials were considered , the next most informative variable was membrane resistance ( Rm , 7% of variance; 13% with interactions ) followed by voltage-gated sodium channel amplitude ( 2% of explained variance , 4% with interactions ) , and membrane capacitance ( 5% without , 15% with interactions ) . Together these 5 variables explained 33% of variation in the number of spikes if taken without interactions , and as much as 51% if multiple-order interactions were included . Unfortunately , linear models did not help to differentiate between variables underlying tuning of spike output during development and after sensory stimulation , as none of the effects disappeared ( based on PF < 0 . 05 criterion ) when both developmental stage and experimental manipulation were factored in . At the same time , the relative abundance of significant interactions ( 7 out of 26 investigated in the model ) and the high share of variance in spiking output explained by these interactions ( 18% , compared to main linear effects of 33% ) indicated that simple electrophysiological properties included in our analysis do not regulate cell spiking independently , but are likely to be constrained ( O'Leary et al . , 2013 ) . For example , the cell property that predicted most of spiking output variance , voltage-gated sodium current activation potential , interacted significantly ( PF < 0 . 05 ) with slow potassium current activation potential , membrane resistance and capacitance , as well as several second and third-order combinations of these variables . In practical terms it means that to keep the spike output of a tectal cell constant , a change in any of these variables should be accompanied by a balancing correction of sodium current activation potential , and vice versa . To further investigate this point , we analyzed distributions of low-level electrophysiological properties in a subset of cells that had similar spiking output in response to current injections ( Figure 8 ) . 10 . 7554/eLife . 11351 . 010Figure 8 . Low-level cell properties are a bad predictor for spiking output . ( A ) Multidimensional scaling of differences between cell spiking outputs onto a 2D plane . Cells that produced similar trains of spikes in response to step current injections , both in terms of the total number of spikes , input-output curve , and spike latency , are located nearby . ( B ) Spike-raster for several subsets of 6 cells each shown in panel A . Spiking outputs of cells are very different between the groups , but are closely matched within each group . ( C ) Groups of cells from panels A and B , projected into PCA space that describes the full variability of cell properties . Clusters of cells are still visible , but they are no longer compact , and groups are partially overlapping . ( D–F ) The same groups of cells are shown on correlation plots for meaningful ( both biologically and statistically; see text ) pairs of cell properties: threshold potentials for voltage-gated sodium and stable potassium currents ( D ) , membrane resistance and capacitance ( E ) , and ionic currents amplitudes ( F ) . The clusters are strongly overlapping , suggesting that cells in which a small subset of properties match can be tuned to produce strikingly different spiking outputs . Threshold data in panel D is renormalized to avoid overplotting ( see 'Materials and methods' ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11351 . 010 To objectively identify cells with similar spike outputs , we combined each cell’s responses to consecutive step injections of increasing current amplitudes into one trace and extracted spike-timing data from this trace . We then applied a commonly-used standard cost-based metric sensitive to both number of spikes and their timing to quantify similarity between spike-trains of different cells ( Victor and Purpura , 1996 , 1997 ) , and used multidimensional scaling to represent a matrix of pairwise cell-to-cell distances in a 2D plot ( Figure 8A ) . We used this analysis to select groups of cells ( 6 in each group ) that generated similar spike trains in response to current step injections ( Figure 8B ) by pulling 5 nearest neighbors to arbitrarily selected reference cells . Although the spike-trains produced by cells within each group were very similar to each other , we found that these cells did not form compact clusters when projected on the C1-C2 PCA space , but resembled sparse constellations that were partially overlapping between groups ( Figure 8C ) . These results suggest that although spiking output was similar between these cells , they differed drastically in other electrophysiological aspects . Finally , we combined these two approaches and labeled groups of similarly-spiking cells in correlograms of electrophysiological values that were found to be best linear predictors for cell spiking output , as described above . These variables included activation potentials for sodium and slow potassium voltage-gated ionic currents ( Figure 8D ) ; passive properties , such as membrane resistance and capacitance ( Figure 8E ) , and ionic current amplitudes ( Figure 8F ) . Although spiking responses of cells ( Figure 8B ) were similar to each other within each group , and strikingly different between the groups , corresponding markers formed neither clusters nor layered structures indicative of low-dimensional constraints that could link different properties together ( Figure 8D , E , F ) . This suggests that in our system , cells with similar spiking phenotypes may have very diverse underlying electrophysiological properties , and conversely , cells that are strikingly different in their spiking output can have very similar low-level physiological properties ( Figure 8 , black and red points respectively ) .
In this study we systematically assessed cell-to-cell electrophysiological variability of primary neurons in the optic tectum of Xenopus tadpoles across several developmental periods and in response to sensory stimulation . Our results indicate that during development cells in the deep layer of the tectum become more diverse — although at the stages we studied they do not split into distinct non-overlapping cell types that are reported in the tecta of other species and at later stages of development in frogs ( Lazar , 1973; Ewert , 1974; Grüsser and Grüsser-Cornehls , 1976; Frost and Sun , 2004; Kang and Li , 2010; Nakagawa and Hongjian , 2010; Liu et al . , 2011 ) . We also found that several key electrophysiological properties of tectal cells change over development . We confirmed previously described changes in the average intrinsic excitability of tectal cells with age ( Pratt and Aizenman , 2007 ) , and showed that at these stages most physiological differences between cells are linked to their overall spikiness ( based on the results of Principal Variable Analysis , Principal Component Analysis , and the comparison of statistical efficiency of different protocols ) . More importantly , we report an increased diversification of cell phenotypes at later developmental stages , and a shrinkage of this diversity in response to strong sensory stimulation . The cell-to-cell variability remained relatively low at stages 43–47 , and different electrophysiological parameters were more random with respect to each other , both in terms of clustering and linear interdependencies between different variables . By stages 48–49 cell variability in the tectum increased , and some internal structure in the PCA cloud began to emerge , with patterns of cell properties agglomerating into clusters , which although poorly resolved at the PCA plot , were noticeable through the quantitative clustering analysis . Complementing previously described receptive field refinement ( Dong et al . , 2009 ) , and temporal decorrelation of spiking activity ( Xu et al . , 2011 ) , this tuning and differentiation of cell properties likely reflects maturation of tectal networks . An increase in cell tuning variability is reminiscent of reports from other experimental models , including mammalian sensory cortex ( Jadhav et al . , 2009; Yassin et al . , 2010 ) , where the non-uniformity of neuronal recruitment thresholds was shown to be a common feature of developed , functional networks ( Elstrott et al . , 2014 ) . This emerging structure and differentiation of cell properties was , however , decreased by strongly patterned visual stimulation , which reduced cell-to-cell variability , making neurons more similar to each other electrophysiologically . At the same time , the amount of variance explained by linear correlations between different variables increased after visual stimulation . This suggests that sensory stimulation , and associated homeostatic plasticity ( Aizenman et al . , 2003; Dong et al . , 2009 ) left a predictable trace in the mutual arrangement of different physiological properties within each cell ( Turrigiano et al . , 1994; Dong et al . , 2009; Munz et al . , 2014 ) . These predictable traces and correlations were then picked up by local factor analysis , making our results similar to reports from the stomatogastric ganglion model ( O'Leary et al . , 2013 ) . We also show that the shift in neuronal excitability induced by visual stimulation was supported by different underlying electrophysiological properties than were the similar changes in excitability observed during development . Among the practical consequences of this study , we point to developmental stage 47 as a likely candidate for the critical tuning period for tectal network maturation . We describe a previously undocumented sharp , transient increase in excitability in tectal cells during stage 47 , providing an explanation for the previously unarticulated practice of aggregating developmental data over stages 45–46 and 48–49 , but avoiding pools of stage 47 neurons with other stages ( Pratt et al . , 2008; Deeg et al . , 2009; Dong et al . , 2009; Sharma and Cline , 2010; Xu et al . , 2011; Khakhalin and Aizenman , 2012; Spawn and Aizenman , 2012 ) . This transient development stage , which lasts for only about 12–18 hr , and is traditionally defined solely on the basis of embryonic morphology ( Nieuwkoop and Faber , 1994 ) , was accompanied by rapid changes in cell tuning variability , and a powerful ( almost two-fold ) increase in cell excitability . This intriguing developmental pattern could be explored in the future as a model for a critical period in development . Finally , our analysis of neurons with similar spiking outputs demonstrated that strikingly different combinations of underlying low-level electrophysiological properties can lead to similar spiking phenotypes , and conversely , that the predictability of spiking phenotype from any small set of cell properties is low . This reinforces the notion that the conflicting biological goals of developmental flexibility and stability in response to perturbations rely on the redundancy of parameters underlying dynamic behavior of these systems ( Marder and Taylor , 2011; Marder et al . , 2014 ) . The consequence of this redundancy is that multiple parameter configurations can produce phenomenologically identical patterns of network activation ( Goaillard et al . , 2009; Caplan et al . , 2014 ) . Altogether , our results provide a promising framework for studying mechanisms of network maturation and calibration in Xenopus tadpoles , as well as a unique dataset that will be helpful to inform computational modeling of the optic tectum . In future studies we plan to combine electrophysiological identification of single cells with the transcriptional mapping of relevant genes ( Nelson et al . , 2006; Schulz et al . , 2006 ) to further advance our understanding of the molecular biology underlying development and plasticity in dynamic systems .
Wildtype Xenopus laevis adults were bred overnight through natural mating in the Brown University animal care facility . Females were primed with 800U human chorionic gonadotropic ( hCG ) ; males were primed with 300U hCG ( [1000 U/mL]; Sigma-Aldrich; St . Louis , MO ) . Embryos were collected the following day; cleaned by removal of unhealthy/unfertilized oocytes , and kept in a variant of 10% Steinberg’s Solution ( also known as ½x MR ) [in mM: 5 . 8 NaCl , 0 . 067 KCl , 0 . 034 Ca ( NO3 ) 2 · 4H2O , 0 . 083 MgSO4 · 7H2O , 5 HEPES; pH 7 . 4] in incubators at 18–21°C under a 12:12 light:dark cycle . Developmental stages are determined according to Nieuwkoop and Faber ( Nieuwkoop and Faber , 1994 ) . Under our rearing conditions , tadpoles reach stages 44–46 at 9–12 days post-fertilization ( dpf ) , and 48/9 at 18–20 dpf . Animals between stages 43 and 49 were used in experiments . Tadpoles used to characterize development of tectal electrophysiological properties were taken directly from the 18–21°C incubators , while those stage 49 tadpoles that were used to assess homeostatic changes in the tectum were first placed in a custom black acrylic box with four rows of four green LEDs flashing in sequence at 1 Hz for 4 hr . Tadpole brains were prepared as described in ( Aizenman et al . , 2003 ) . All experiments were performed between ZT 3–9 ( 10:00–16:00 EST ) , where ZT 0 is lights-on for a diurnal animal . In brief , tadpoles were anesthetized with 0 . 02% ( w/v ) tricaine methanosulfonate ( MS-222 ) in 10% Steinberg’s solution and brains were then dissected out in HEPES-buffered extracellular media ( containing in mM: 115 NaCl , 4 KCl , 3 CaCl2 , 3 MgCl2 , 5 HEPES , 10 glucose , 10 µM glycine; pH 7 . 2 at 255 mOsm/Kg ) . To access the soma layer of the tectum , brains were filleted along the dorsal midline and extracted for pinning to a submerged block of Sylgard 184 Silicone Elastomer ( Dow Corning; Midland , MI ) in a custom recording chamber at room temperature ( 23°C ) . Using a large-bore glass electrode , the ventricular membrane was suctioned to reveal the tectal cell body layer . Tectal cells were visualized using a Nikon ( Tokyo , Japan ) FN1 light microscope with a 60x water-immersion objective . While a visually heterogeneous population of tectal neurons were selected , care was taken to only patch those principal tectal neurons that looked healthy ( clear , no granulation ) and to avoid particularly large cells ( size and shape ) that might be mesencephalic trigeminal neurons ( Pratt and Aizenman , 2009 ) . To ensure valid comparisons across stages of development , we restricted our recordings to the middle third of the tectum , thus reducing developmental variability along the rostro-caudal axis ( Wu et al . , 1996; Khakhalin and Aizenman , 2012; Hamodi and Pratt , 2014 ) . All cells were recorded within 3 hr of dissection . Drugs and chemicals were obtained from Sigma ( Sigma-Aldrich; St . Louis , MO ) . Glass electrodes were pulled on a Sutter P97 or P1000 puller ( Sutter Instruments; Novato , CA ) from either Corning 7056 thin wall capillary glass tubing ( G75165T-4 , Warner Instruments; Hamden , CT ) or Sutter thick wall capillary glass tubing ( B150-86-10 ) to a tip resistance of 8–12 MΩ . The electrodes were then filled with a K-gluconate-based intracellular saline ( containing in mM: 100 K-gluconate , 5 NaCl , 8 KCl , 1 . 5 MgCl2 , 20 HEPES , 10 EGTA , 2 ATP , 0 . 3 GTP; pH 7 . 2 at 255 mOsm/Kg ) through a 200 nm syringe filter ( #171 , Nalgene-Thermo; Waltham , MA ) . Filled electrodes were placed in an Axon headstage containing a AgCl wire and controlled by a motorized micromanipulator ( MX7600R and MC1000e-R , Siskiyou; Grants Pass , OR ) . Electrode was lowered into the chamber with slight positive pressure in the pipette until the target cell was contacted , at which time negative pressure was applied to form a high resistance seal ( >1GΩ ) , and then again to break through the neuronal membrane . Whole cell patch clamp electrophysiological signals were measured with an Axon Instruments MultiClamp 700B amplifier , filtered with a 5 kHz band-pass filter , and digitized at 10 kHz by an Axon Instruments DigiData 1440A , and acquired with pCLAMP 10 software ( Molecular Devices; Sunnyvale , CA ) . Upon initial whole cell patch , a seal test and membrane test measured the following parameters: cell membrane capacitance ( Cm ) , cell membrane resistance ( Rm ) , access resistance ( Ra ) , and holding current ( I hold ) necessary to keep the membrane at -65 mV . Each clamped neuron was subjected to a series of voltage clamp and current clamp protocols to assess its characteristics . First , cells were held at -65 mV in voltage clamp to ensure that Na+ channels are fully de-inactivated . Then cells were stepped for 150ms to membrane potentials from -65 mV to 115 mV in 20 mV increments to get the data for IV curves . Cells were then switched to current clamp and subjected to ten square pulses of current from 0 pA to 180 pA in 20 pA increments . We next explored potential resonances in tectal neurons ( Hutcheon and Yarom , 2000 ) by probing their ability to fire in response to cosine current injections of varying frequencies . Each sweep contained five separate 200 ms bouts of cosine injections with frequencies of 100 , 50 , 30 , 25 , and 20 Hz , and peak amplitude of 135 pA . To determine health of the neurons , they were switched back to voltage clamp and subjected to the IV-curve voltage step protocol once again . After that , the spontaneous activity was continuously recorded for one minute at −45 mV holding potential . Continuing to hold at -45mV , cells were then subjected to synaptic stimulation protocols , in which the optic chiasm ( OCh ) was stimulated with a bipolar stimulating electrode ( FHC , Bowdoin , ME ) to activate retinal ganglion cell axons . Five stimuli of 150 µA to 800 µA and duration of 180 µs were provided at varying frequencies , with inter-stimulus intervals of 10 , 20 , 30 , 40 , 50 , 100 , 150 , 200 , 250 , and 300 ms; the protocol was repeated 5 times . Finally , the IV-curve voltage step protocol was repeated a third time to ensure the health of the cell and stability of the data . The microscope was then switched to the 10x objective , and the cell location was recorded ( Khakhalin and Aizenman , 2012 ) . The cell was suctioned away from the tissue and the process was repeated for up to 6 neurons . Some cells did not survive the entire series of protocols , but as long as the nearest IV-curve recording from these cells was stable , they were included in the dataset . No potentials reported in this paper were corrected for the expected junction potential of +12 mV . Here we introduce and enumerate variables that were included in the analysis , and are presented further . In total , up to 33 different variables were measured for each cell , with 4 variables coming from a standard seal test; 6 variables from the IV-curve protocol; 10 variables from the step current injection protocol; 6 from the protocol of cosine-shaped current injections of different frequencies; 5 from the synaptic stimulation protocol , and 2 from the recording of spontaneous postsynaptic potentials . Data analysis was performed in MATLAB ( Mathworks , MA ) and R Studio . As many variables analyzed in this paper were not normally distributed , and to stay consistent , we preferred Mann-Whitney two-sample tests for comparing data between groups; p-values of this test were reported as PMW . At the same time , to make referencing and subsequent meta-analysis possible , we always report means and standard deviations in the text . Other statistical tests used in this study include Pearson correlation ( Pcorr ) , one-way ANOVA ( PANOVA ) , multiple linear regression test ( PF ) , Welch generalization of Student’s t-test ( PTT ) , and F-test for equality of variance ( PV ) . When exploring potential correlations between variables we used False Discovery Rate ( FDR ) procedure to adjust for multiple testing , keeping the false discovery rate at 0 . 05 ( Benjamini and Hochberg , 1995 ) . The FDR procedure was performed separately for each massive set of comparisons , as indicated in the text . In post-hoc pairwise comparisons of data groups after ANOVA tests we used a more conservative Bonferroni correction , but reported uncorrected p-values . To illustrate correlation levels between variables ( Figure 2A ) we used a custom MATLAB script inspired by the circular diagram from the 'Circos' visualization package ( Krzywinski et al . , 2009 ) . As not every pair of variables was available in every cell , different correlation tests were run on different numbers of points ( from 38 to 154; median of 108 ) . We verified our Pearson correlation-based analysis by calculating Spearman correlation coefficients; after FDR 112 Spearman correlations were significant ( as opposed to 90 for Pearson ) , and all Pearson correlations with r>0 . 5 ( n=11 ) , including all shown in Figure 2B , remained significant for Spearman calculation . We ran the Principal Variables Analysis in R — package 'subselect' ( Mccabe , 1984; Cadima and Jolliffe , 2001; Cadima et al . , 2004 ) — using the correlation matrix of original non-imputed data , which was computed on all available pairwise observations for each pair of variables . We then compared and presented squared RM coefficients ( Cadima and Jolliffe , 2001 ) , to make the results of this analysis comparable to that of factor analysis below . As not all measurements were available for every cell in the dataset , for factor analysis we used a generalization of a standard Principal Component Analysis ( PCA ) procedure , called the Variational Bayesian PCA , or the PCA with missing values ( PCA-MV; ( Ilin and Raiko , 2010 ) ; MATLAB m-files are available at the web-site of Bayesian Group , Aalto University , Finland ) . To verify that the PCA-MV procedure is applicable to our data , we selected a subset of 52 cells and 27 variables to form a full matrix free of missing values . A standard PCA was then run on this reduced data set , and the results of it were compared to the results of PCA-MV run on the full data set . The scree plot of a standard PCA on restricted data suggested that two first components ( explaining 20% and 16% of variation respectively ) could be interpreted meaningfully; the remaining components representing individual variability of the data , or 'noise’; see ( Shabalin and Nobel , 2013 ) for references . We therefore only present the first two PCA-MV components in this paper ( explaining 15% and 8% of total variance respectively ) . Component scores found by standard PCA and PCA-MV on the same subset of cells ( N = 52 ) were highly correlated ( r = 0 . 90 and 0 . 85 for components 1 and 2 respectively; p < 3e−15 ) , suggesting that the PCA-MV is indeed applicable to our data set . As 18% of all possible measurements in our dateset were missing , the total share of explained variance ( 23% ) was almost certainly underestimated , and cannot be compared to variance explained by standard PCA . This statement is obvious if you consider that the estimation of total variance in a set with randomly missing values is unbiased , yet explained variance is biased , as missing values cannot contribute to the calculation: while predictions for them are available , the values themselves are not present . As described above , a standard PCA run on a subset of data with full representation of all cells and variables explained 36% of total variance . Similarly , when we performed multiple imputation of missing values using R package 'Mi' ( Su et al . , 2011 ) ( see below for details ) , a standard PCA procedure explained on average 35 ± 5% of total variance . Some of the variables we assessed were distributed non-normally , and we attempted to run PCA-MV on renormalized rank-transformed variables , and compared the results of this analysis to that of PCA-MV run on raw variables . Rank-based normalization improved the amount of variance explained by the first component ( from 15% on raw data to 21% on rank-transformed data ) , but did not improve explanatory value of higher components . Upon visual comparison of score-plots and loading-plots , we concluded that the relative arrangement of individual cells ( score-plot ) , as well as contributing variables within the 2D plane of first two components ( loading-plot ) , did not change enough to justify the use of rank-transformation . All analysis reported in the paper was therefore performed on raw variables . While linear approaches to factor analysis , such as PCA or Multidimensional Scaling , are usually considered to be safe and preferable methods when noisy and weakly correlated data are concerned ( Nowak et al . , 2003; Sobie , 2009; McGarry et al . , 2010 ) , we compared the performance of PCA to the two most popular non-linear 2D ordination approaches: Isomap and Local Linear Embedding . To quantify the quality of 2D ordination we looked at how well the 2D map preserved pairwise differences between points in the original 33D space , using the squared correlation coefficient R2 between 2D and 33D distances as an output measure ( Pedhazur , 1982 ) . Based on this metric , PCA preserved 51 ± 2% of variance in pairwise differences ( 5 alternative Bayesian imputations of missing data using R package 'Mi’ ) . The quality of Isomap projection improved as the projection became less and less local , from 17 ± 4% for isomap based on 3 closest neighbors for each point , to 36 ± 7% based on 20 closest neighbors; still it was substantially lower than for PCA . The Local Linear Embedding approach ( R package 'lle’ , based on ( Kouropteva et al . , 2005 ) also produced better results as more neighbors were considered , with the local best solution achieved at 19 neighbors explaining 27 ± 9% of variance in pairwise differences ( as opposed to 51 ± 2% for PCA ) . Based on these results we concluded that for our data linear factor analysis approach is not only adequate , but also the most appropriate . In all cases testing was performed on centered and normalized data . We also attempted restricting the number of variables included in PCA by pre-screening them based on their Principal Variables rank and leaving only variables that explained high amounts of total variance in the dataset . At 17 variables ( one half of the original set , corresponding to total explained variance threshold of 6% ) PCA-MV explained 40% of total variance in the set ( as opposed to 23% for full data PCA-MV ) , and some of the effects we describe in the paper became more prominent ( for example , F-value for changes in PCA cloud size across developmental stages increased from 75 for the full set to 118 for restricted set ) . However we decided not to present PCA of restricted data in the paper , as thinning out of the multivariate dataset is generally not recommended for exploratory analysis when there is no objective post-hoc test to justify the use of one restricted model over another ( Guyon and Elisseeff , 2003 ) . We therefore only report it here as another validation of the method . To simplify interpretation of loading- and score-plots we performed 'promax' oblique rotation of first two PCA-MV components using a standard 'rotatefactors' routine from MATLAB statistics toolbox . This approach maximizes varimax criterion using orthogonal rotation , and further simplifies the projection by applying Procrustes oblique rotation , using orthogonal rotation from the first step as a target . To compare cloud sizes in the PCA space , we calculated all possible pairwise 2D Euclidian distances between the cells and compared their medians . To illustrate positions , shapes and spreads of clusters at the PCA scores plot in Figure 5 and Figure 7 we used the Kernel Density Estimation procedure ( Botev et al . , 2010 ) . To compare cluster sizes in the original 33-dimensional space , we centered and normalized the data , performed 50 alternative Bayesian imputations of missing values using R package 'Mi' ( Su et al . , 2011 ) , and for each imputation subsampled 10 different sets of 50 points to compensate for small differences in original dataset sizes ( n = 64 , 56 and 60 for naïve stage 44–45 , naïve stage 48–49 and visually stimulated stage 48–49 respectively ) . For each sampled subset of points we computed city-block ( "Manhattan" ) pairwise distances between all cells in the subset , and calculated the median of these values . Finally , we used a t-test to compare sets of medians between data groups ( resulting in n = 500 values for each group ) . To quantify the amount of internal heterogeneity within data sets , we ran agglomerative nesting analysis ( AGNES from package 'cluster’ ) in R , and used the agglomerative nesting coefficient as a measure of heterogeneity ( Struyf et al . , 1996 ) . In practice , we used the same imputation procedure as described above ( 50 imputations , each contributing to 10 subsets of 50 points each ) , applied the AGNES clustering procedure to these data , saved agglomerative clustering coefficients , and finally compared them across data groups . Similarly , for 'local within-group PCA' we used same imputation / subsetting procedure , ran PCA on each set , and calculated the share of variance explained by first two components . To statistically link the number of spikes produced in response to step injections to basic electrophysiological properties of each cell we built a family of general linear models using non-marginal sequential GLM statistics in R package 'glmulti' ( Calcagno and de Mazancourt , 2010 ) . | Brains consist of many cells called neurons: billions of them in a human brain , and hundreds of thousands in the brain of a small fish or a frog tadpole . Many of these neurons are very much alike , and work together to process information in the brain . Yet while they are similar , they are not exactly identical . One of the reasons for these differences seems to be to allow each neuron to contribute something unique to the overall working of the brain . By looking at how individual neurons within a specific type differ from each other , it is possible to understand more about how they work together . Ciarleglio , Khakhalin et al . have now compared the properties of the neurons in a part of the brain of a developing frog tadpole that processes sensory information . This showed that these neurons appear relatively similar to each other in young tadpoles . However , as the tadpoles grow and their brains become more elaborate the neurons become increasingly diverse , and their properties become more unique and nuanced . One possible explanation is that this diversity reflects new types of neurons being formed; another , that the differences between the neurons reflect how these cells have adapted to different patterns of sensory input they may have experienced . To distinguish between these two possibilities , Ciarleglio , Khakhalin et al . provided a group of older tadpoles with strobe-like visual stimulation and observed that this caused the neurons to become more similar once again . This suggests that neurons can change their response properties to adapt to the type of sensory input they receive , which would allow the animal to better process different types of sensory information . The data collected through these experiments could now be used to build computational models of this part of the tadpole brain . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Multivariate analysis of electrophysiological diversity of Xenopus visual neurons during development and plasticity |
Muscle development and regeneration require delicate cell cycle regulation of embryonic myoblasts and adult muscle satellite cells ( MuSCs ) . Through analysis of the Polo-like kinase ( Plk ) family cell-cycle regulators in mice , we show that Plk1’s expression closely mirrors myoblast dynamics during embryonic and postnatal myogenesis . Cell-specific deletion of Plk1 in embryonic myoblasts leads to depletion of myoblasts , developmental failure and prenatal lethality . Postnatal deletion of Plk1 in MuSCs does not perturb their quiescence but depletes activated MuSCs as they enter the cell cycle , leading to regenerative failure . The Plk1-null MuSCs are arrested at the M-phase , accumulate DNA damage , and apoptose . Mechanistically , Plk1 deletion upregulates p53 , and inhibition of p53 promotes survival of the Plk1-null myoblasts . Pharmacological inhibition of Plk1 similarly inhibits proliferation but promotes differentiation of myoblasts in vitro , and blocks muscle regeneration in vivo . These results reveal for the first time an indispensable role of Plk1 in developmental and regenerative myogenesis .
Adult skeletal muscle has an efficient regenerative capacity in response to muscle injury or physiological stimuli ( i . e . intense exercise training ) . Muscle regeneration relies on a population of muscle resident stem cells , known as muscle satellite cells ( MuSCs ) . These cells are located in a unique niche between the basal lamina and the plasma membrane of myofibers ( Mauro , 1961 ) and remain in a quiescent state until activated by regenerative signals . Upon activation , MuSCs proliferate to generate a pool of myoblasts that eventually return to the quiescent state to replenish the MuSCs pool or differentiate to repair muscle injuries . Thus , cell cycle regulation of MuSCs is critical for precise control of the number of myoblasts that is needed for muscle regeneration . Reduced regenerative capacity was observed when a muscle was exposed to an inhibitor of mitotic division , colchicine ( Pietsch , 1961 ) or irradiation ( Quinlan et al . , 1995 ) . Knockout of Cdkn1b , Cdkn1c or Rb ( negative cell cycle regulators ) results in aberrant satellite cell activation and proliferation ( Chakkalakal et al . , 2014; Hosoyama et al . , 2011; Mademtzoglou et al . , 2018 ) . Positive cell cycle regulators , including cyclin A , B , D , E , F and G , are upregulated in activated MuSCs to regulate cell cycle progression ( Cenciarelli et al . , 1999; De Luca et al . , 2013; Fukada et al . , 2007 ) . Deregulation of cell cycle regulators p16 and p21 , and Notch signaling in quiescent MuSCs in old mice leads to proliferative senescence , accumulation of DNA damage , mitotic catastrophe and high frequency of cell death ( Li et al . , 2015; Liu et al . , 2018 ) . The polo-like kinases ( PLKs ) are a conserved subfamily of Ser/Thr protein kinases that play pivotal roles in cell cycle regulation . The PLK family contains five members ( PLK1-5 ) in mammals , all except for PLK5 contain an amino-terminal Ser/Thr kinase domain ( Archambault and Glover , 2009; de Cárcer et al . , 2011; Liu , 2015 ) . Among the PLK kinases , PLK1 is the most conserved and best known for its role in mitosis via phosphorylation of different substrates ( Barr et al . , 2004 ) . PLK1 also participates in modulating DNA replication and DNA damage checkpoints ( Takaki et al . , 2008 ) . Overexpression of PLK1 is observed in several human tumors , including prostate and ovarian cancers , and muscle cell-derived rhabdomyosarcoma ( Hugle et al . , 2015; van Vugt and Medema , 2005 ) . Inhibition of PLK1 by small interfering RNA or pharmacological inhibitors exerts antitumor effect in vitro and in vivo , providing strong preclinical and clinical support for the use of PLK1 inhibitors in cancer therapy ( Degenhardt and Lampkin , 2010; Lens et al . , 2010 ) . Outside the cancer field , the role of PLK1 in normal mitotic cells especially stem cells are poorly understood . As Plk1 gene deletion leads to embryonic lethality in mice , zebrafish , Drosophila and yeast ( Jeong et al . , 2010; Lu et al . , 2008; Ohkura et al . , 1995; Sunkel and Glover , 1988 ) , conditional deletion of Plk1 is necessary to understand its tissue or cell-type-specific functions . In this study , we used myogenic cell-specific targeted mutation to show that Plk1 is absolutely required for mitosis and survival of myogenic cells during muscle development and regeneration in mice .
To establish the relevance of Polo-like kinases in myogenesis , we surveyed the expression of Plk1–Plk4 ( Plk5 was not surveyed as it does not have a kinase domain ) at various time points during CTX-induced muscle regeneration . Activation and proliferation of MuSCs peaks at 3 days post injury ( DPI ) , and the overall architecture of the muscle is restored by 10 DPI ( CHARGÉ and Rudnicki , 2004 ) . The mRNA levels of Plk1 , Plk3 , and Plk4 were all transiently up-regulated after muscle injury , reaching peak expression levels at 3 DPI and returning to the preinjury levels at 10 DPI , but Plk1 exhibited the most prominent fold change ( increased by >13 fold ) at 3 DPI ( Figure 1A ) . The expression pattern of Plk1 corresponded to those of myogenic transcriptional factors Pax7 and MyoG at both mRNA ( Figure 1B ) and protein ( Figure 1C ) levels . We also surveyed the mRNA levels of Plk1-4 during differentiation of primary myoblasts isolated from limb muscles . Compared with day 0 ( proliferating myoblast ) , Plk1 , Plk3 , and Plk4 levels were all down-regulated during myogenic differentiation ( Figure 1D ) . Among these , Plk1 and Plk4 exhibited the most robust down-regulation ( Figure 1D ) . The expression pattern of Plk1 was inversely correlated to the expression of myogenic differentiation makers Myog and eMyhc , which were robustly upregulated during differentiation ( Figure 1E ) . Consistently , Plk1 levels progressively declined from embryonic day 17 . 5 ( E17 . 5 ) to postnatal day 90 ( P90 ) during limb muscle differentiation and maturation in vivo ( Figure 1F ) . Since Plk1 is the most dynamically regulated Plks during myogenesis , we focused on Plk1 for the rest of the current study . Non-myogenic cells such as fibroadipogenic progenitors , endothelial cells , and infiltrating inflammatory cells are also present in the muscle during regeneration . To assure that the above-observed Plk1 dynamics are specific to MuSCs , we used an established antibody to detect Plk1 protein expression in MuSCs that were co-labeled with Pax7 , an established marker of MuSCs ( Seale et al . , 2000 ) . In the freshly isolated EDL myofibers ( Day 0 ) carrying quiescent MuSCs , Plk1 immunofluorescence signal was undetectable in any Pax7+ MuSCs ( Figure 2A ) . After culture , Plk1 signal appeared in activated MuSCs that co-express MyoD ( Day 1 ) , and then was abundantly expressed in clusters of MuSCs progenies or myoblasts ( Days 2–3 ) , with positive signal only detected in MyoD+ cells ( Figure 2A , n > 200 cells ) , which marks activated MuSCs ( Olguin and Olwin , 2004; Zammit et al . , 2004 ) . We also analyzed mRNA levels of Plk1 in FACS-purified MuSCs isolated from non-injured and injured muscle tissues , representing quiescent and activated MuSCs , respectively . Plk1 mRNA level was 15-fold higher in activated than in quiescent MuSCs ( Figure 2B ) . Additionally , bFGF , a growth factor known to promote the proliferation of MuSCs , increased the level of Plk1 in cultured myoblasts ( Figure 2C ) . Collectively , these lines of evidence demonstrate that Plk1 expression is dynamically regulated in MuSCs during their quiescence , activation , proliferation and differentiation . To assess the role of Plk1 in muscle development in vivo , we crossed the MyodCre with Plk1f/f mice to generate the myoblast-specific Plk1 knockout ( MyodCre::Plk1f/f , or Plk1MKO ) mice ( Figure 3A ) . In this model , Plk1 should be deleted in all muscle progenitors during development as MyodCre marks all embryonic myogenic cells ( Kanisicak et al . , 2009 ) . After several intercrosses of heterozygous mice , no live Plk1MKO pups were obtained , indicating that myoblast-specific knockout of Plk1 resulted in prenatal lethality . Therefore , embryos were harvested in utero at E16 . 5 , several days before birth . From 69 embryos analyzed , we obtained 13 motionless embryos that were subsequently confirmed by PCR to be Plk1MKO ( Supplementary file 1 ) . Plk1MKO embryos were shorter in body length and transparent in appearance ( Figure 3B ) . Histological sections through limbs revealed an absence of central-nucleated myofibers that were otherwise stained by eosin and embryonic myosin heavy chain ( eMyHC ) in the Plk1MKO embryos ( Figure 3C–D ) . In contrast , eosin and eMyHC staining reveals the development of various groups of muscles in the WT control embryos ( Figure 3C–D ) . To determine the effect of Plk1MKO on myogenic progenitors , we labeled limb muscle sections with Pax7 antibody . Whereas numerous Pax7+ myoblasts were detected in the WT control limb muscles , the Plk1MKO limb muscles were completely depleted of Pax7+ myoblasts ( Figure 3E ) , suggesting that Plk1 is necessary for the generation or maintenance of embryonic myoblasts in the limb muscles . In addition , no Ki67+ cells were Pax7+ in the Plk1MKO limb muscles ( Figure 3E ) , suggesting that proliferative failure may be another driver of the depletion of myoblasts . These results indicate that loss of Plk1 in myogenic progenitors blocks their proliferation and survival , leading to failure in skeletal muscle development and embryonic lethality . The prenatal lethality of the Plk1MKO mice precludes the exploration Plk1 function in postnatal MuSCs and muscle regeneration . To circumvent embryonic lethality , we established tamoxifen ( TMX ) -inducible Pax7CreER::Plk1f/f ( Plk1PKO ) mice to specifically delete Plk1 in Pax7-expressing MuSCs upon TMX injection in adult mice ( Figure 4—figure supplement 1A ) . WT control and Plk1PKO mice were intraperitoneally ( IP ) injected with TMX for five days followed by 14 days of chasing ( Figure 4—figure supplement 1A ) . Knockout efficiency was shown by the lack Plk1 signal in MuSCs isolated from Plk1PKO myofibers but not in WT MuSCs after 72 hr of culture ( Figure 4—figure supplement 1B ) . In the absence of injury , no morphological differences were observed between Plk1PKO and WT muscles based on H and E staining of tibialis anterior ( TA ) muscle cross-sections at Day 14 after TMX-induced deletion of Plk1 ( Figure 4—figure supplement 1C–D ) . Immunofluorescence staining of Pax7 showed that the number of MuSCs ( Pax7+/DAPI+ ) in Plk1PKO mice was ~70% of that in WT mice , and no MyoG+ differentiated cells were observed in either WT or KO muscles ( Figure 4—figure supplement 1E–G ) . These results confirm the efficiency of the Plk1PKO conditional KO mouse model and demonstrate that the loss of Plk1 do not lead to obvious morphological changes in uninjured muscles . We then examined muscle regeneration of the Plk1PKO mice to determine the effect of Plk1 KO on MuSCs . TA muscles from both Plk1PKO and WT mice were degenerated by CTX after TMX-induced Plk1 deletion ( Figure 4A ) . At 7 DPI , the masses of WT TA muscles were recovered to 70% of preinjury levels , whereas the TA muscle masses of Plk1PKO mice were only recovered by ~40% ( Figure 4B–C ) . Histologically , the WT TA muscles were uniformly replaced by newly regenerated myofibers characterized by central-nucleated myofibers ( Figure 4D ) . In contrast , TA muscles of Plk1PKO mice were devoid of newly regenerated myofibers ( Figure 4D ) . Pax7+ cells were diminished in Plk1PKO mice , together with an absence of any Dystrophin-expressing myofibers ( Figure 4E–F ) . Furthermore , very few MyoG+ cells and no eMyHC+ myofibers were observed in Plk1PKO mice ( Figure 4—figure supplement 2A–C ) . Compared with uniformly regenerated muscles in WT mice , Plk1PKO mice had no Pax7+KI67+ cells , but a compensatory increase in the number of Pax7–KI67+ cells ( Figure 4—figure supplement 2D–E ) , indicating depletion of MuSCs and increased proliferation of non-myogenic cells . To further distinguish if Plk1 KO leads to regenerative delay or deficiency , we examined TA muscles at 21 DPI ( Figure 4—figure supplement 2F ) , at which time the CTX-injured muscles were completely repaired , and the muscle weights were restored to the non-injured levels in the WT mice ( Figure 4—figure supplement 2G ) . However , the TA muscle mass of the CTX-injured muscles was only 25% of the mass of the non-injured muscles in the Plk1PKO mice ( Figure 4—figure supplement 2G ) . Newly regenerated myofibers ( indicated by central-nucleation ) were uniformly packed in the WT TA muscles ( Figure 4—figure supplement 2H ) . In contrast , TA muscles of Plk1PKO mice were devoid of newly regenerated myofibers , and were infiltrated by Plin1 +adipocytes ( Figure 4—figure supplement 2H–I ) and F4/80+ macrophages ( Figure 4—figure supplement 2I ) . These results revealed that Plk1 KO leads to ablation of MuSCs and severe regenerative failure . To further dissect if Plk1 function in MuSCs and muscle regeneration depends on its kinase activity , we treated regenerating muscles with BI2536 , a selective inhibitor of Plk1 activity ( Figure 4—figure supplement 3A ) . BI2536 treatment ( injection of 50 μl into the TA muscle at a final dosage of 0 . 4 μg/g body weight ) significantly inhibited regeneration of CTX-injured muscles ( Figure 4—figure supplement 3B ) . H and E staining of vehicle control and BI2536-treated muscle cross-sections revealed that BI2536-treated muscles had increased interstitial space filled with inflammatory infiltration and very few newly regenerated myofibers ( Figure 4—figure supplement 3C ) . Consistent to Plk1 KO results , BI2536 treatment did not have any effects on morphology of uninjured muscles ( Figure 4—figure supplement 3D ) , suggesting Plk1 mainly functions to regulate proliferating MuSCs during regeneration . BI2536 treatment reduced the number of Pax7+ cells by~60% in regenerating muscles , but not in uninjured muscle sections ( Figure 4—figure supplement 3E–F ) . Similarly , BI2536 treatment did not affect the abundance of MuSCs in uninjured EDL myofibers , but significantly reduced the number of MuSCs on CTX-injured myofibers ( Figure 4—figure supplement 3G–H ) . These data together suggest that genetic deletion or pharmacological inhibition of Plk1 similarly inhibits the proliferation of MuSCs and prevents muscle regeneration . To investigate how Plk1 regulate cell cycle progression of MuSCs , we synchronized WT myoblasts by double thymidine block and released them to allow synchronous cell cycle progression . Plk1 expression was significantly increased as cells enter G2 and peaked at metaphase , indicated by phosphohistone H3 ( pHH3 ) labeling ( Figure 5A ) . The cell-cycle-dependent expression pattern suggests a key role of Plk1 in regulating metaphase progression of MuSCs . To further determine cell cycle defects of Plk1 KO MuSCs , we analyzed proliferation of MuSCs attached on cultured EDL myofibers . After 3 days of culture , the WT MuSCs formed clusters of 6–8 myoblasts on the myofibers . However , MuSCs from Plk1PKO mice failed to complete the cytokinesis and were arrested at mitosis stage , evident from the dispersed distribution of nuclear Pax7 and MyoD signals surrounding metaphase and telophase chromosomes ( Figure 5B ) . The numbers of myoblasts were also significantly decreased on the Plk1PKO relative to WT myofibers ( Figure 5—figure supplement 1A ) . We also isolated primary myoblast from Plk1PKO and used 4-hydroxy-tamoxifen ( 4-OHT ) to induce acute Plk1 deletion in proliferating myoblasts . The high efficiency of 4-OHT-induced Plk1 deletion was verified by WB and DNA recombination analysis ( Figure 5D ) . Compared to the control ( Methanol ) treatment in which the number of myoblasts increased exponentially , the numbers of 4-OHT-treated myoblast only slightly increased in the first 12 hr , followed by a gradual decrease ( Figure 5E ) , suggestive of cell death . The numbers of myoblasts continually decreased when the myoblasts were cultured for more than 2 days ( Figure 5—figure supplement 1B ) . Flow cytometry analyses revealed a significant increase of tetraploid cells 24 after 4-OHT induced KO of Plk1 ( Figure 5F ) . DAPI staining also demonstrated that over 60% of Plk1-null myoblast underwent a cell cycle arrest at M-phase with condensation of the chromatin and the disappearance of the nucleolus ( Figure 5G ) . The Plk1 inhibitor BI2536 was also used to treat cultured myofibers carrying MuSCs and primary myoblasts in vitro . When EDL myofibers were cultured for 3 days in the absence or presence of 100 nM BI2536 , BI2536 treatment not only reduced the total number of myoblast clusters per myofiber by ~70% , but also reduced the number of cells per clusters , with no cluster containing more than four cells ( Figure 5—figure supplement 1C–D ) . In contrast , 40% of the vehicle-treated myofibers contained clusters of more than four cells per cluster ( Figure 5—figure supplement 1D ) . Consistently , BI2536 also suppressed the proliferation of primary myoblasts dose-dependently over various timepoints ( Figure 5—figure supplement 2A–B ) . Nuclear staining with DAPI indicated that more than 40% of the BI2536-treated myoblasts were arrested at an undivided stage , resulting in tetraploid-like cells ( Figure 5—figure supplement 2C–D ) , indicating an unsuccessful division of myoblasts by Plk1 inhibition . Together , these data reveal a key role of Plk1 in cytokinesis and indicate that loss or inhibition of Plk1 leads to proliferative defect and cell cycle arrest at M-phase . We next questioned what’s the fate of the Plk1-null myoblasts that were arrested at M-phase . To test this , we first checked the cell viability by antibody staining with cleaved Caspase-3 ( C-Cas3 ) , a marker for apoptosis ( Igney and Krammer , 2002 ) . As a positive control , 100 mM H2O2 was used to treat MuSCs on EDL myofibers cultured for 72 hr , which resulted in robust C-Cas3 signal ( Figure 6—figure supplement 1A ) . Next , Plk1PKO EDL myofibers were cultured in vehicle control ( Methanol ) or 4-OHT to induce Plk1 deletion . None of the methanol-treated myoblasts on myofibers were C-Cas3 positive , but all the 4-OHT treated MyoD+ myoblasts were C-Cas3 positive at 48 and 72 hr after the treatment ( Figure 6A ) . Morphologically , the 4-OHT-treated myoblasts were arrested in the cell cycle before the first division ( Figure 6A ) . All activated myoblasts were C-Cas3 negative at 20 hr in both treatment group , indicative of a cell-cycle-dependent Cas3 activation . In situ terminal deoxynucleotidyl transferase dUTP nick-end labeling ( TUNEL ) also showed that MuSCs on EDL myofibers were all TUNEL+ without finishing the first division after 42 hr culture ( Figure 6B ) . Myoblasts cultured from Plk1PKO mice were also 100% C-Cas3+ or TUNEL+ when induced with 4-OHT following 2 days of culture , but control groups only had ~5% C-Cas3+ or TUNEL+ cells ( Figure 6C–F ) . In addition , there were no presence of MyoG+ differentiated myoblasts on EDL myofibers from Plk1PKO mice after 72 hr of culture ( Figure 6—figure supplement 1B ) . Apoptosis of Plk1 null myoblast was also confirmed by Annexin V-FITC Apoptosis Staining , which showed a dramatically increment of FITC/PI double positive cells 36 hr after 4-OHT induction when compared to the control group ( Figure 6—figure supplement 1C–D ) . These results suggest that Plk1 deletion induced loss of MuSCs due to apoptosis but not differentiation . As Plk1 KO activates Cas3 , which is necessary for myogenic differentiation ( Larsen et al . , 2010 ) , we also examined the effect of Plk1 inhibition on differentiation of myoblasts . As deletion of Plk1 in proliferating myoblasts leads to cell death and precludes analysis of cell differentiation , we used BI2536 to inhibit Plk1 activity in myoblasts . Interestingly , myosin heavy chain ( marked by MF20 ) , markers of myogenic differentiation , was observed in 15% of myoblasts treated with BI2536 , but rarely observed in control myoblasts cultured in growth media without BI2536 ( Figure 6—figure supplement 2A–B ) . This suggests that Plk1 inhibition promotes premature differentiation of myoblasts . When BI2536 was added to myoblasts at the onset of serum withdrawal-induced myogenic differentiation , the fusion index at 48 hr was increased by 30% compared to vehicle treatment controls ( Figure 6—figure supplement 2C–D ) . The protein and mRNA levels of MF20 were increased by Plk1 inhibition under both growth and differentiation conditions ( Figure 6—figure supplement 2E–F ) . These results demonstrate that pharmacological inhibition of Plk1 in primary myoblasts promote their differentiation . To understand how Plk1 deletion and inhibition leads to the apoptosis of myoblasts , we first examined DNA damage response ( DDR ) , revealed by the presence of phosphorylated histone 2A family member X ( γH2AX ) ( Paull et al . , 2000 ) . We found that Plk1 null MuSCs ( labeled by Cav1 ) that were arrested at undivided stage had very strong γH2AX signal ( Figure 7A ) . This is in sharp contrast to control cells that contained very few small puncta of γH2AX signal ( Figure 7A ) . Similarly , ~80% of cultured Plk1 KO primary myoblasts were strongly γH2AX+ at 2 days after 4-OHT induction , whereas only ~23% control WT myoblast exhibited weak γH2AX signal ( Figure 7B–C ) . We also analyzed DNA fragmentation of BI2536 treated myoblasts using single-cell gel electrophoresis assay , revealing that 70% of myoblasts accumulated DNA damage after BI2536 treatment ( Figure 7D ) . Among these , 55% of myoblasts were scored as moderate damage ( Classes 1–3 ) and 15% were scored as maximal damage ( Class 4 ) ( Figure 7E ) . In contrast , 95% control myoblasts were fragmentation-free and only 5% control myoblasts displayed moderate DNA damage ( Figure 7E ) . We also synchronized cell cycle with double thymidine block and release ( Figure 7—figure supplement 1A ) . The Plk1 KO ( 4-OHT treated ) myoblasts exhibited more abundant γH2AX signal than control myoblasts at 8 hr after the release , but no differences in TUNEL signal were observed between the two groups ( Figure 7—figure supplement 1B ) , suggesting DNA damage precedes apoptosis . At 12 hr after release , TUNEL+ myoblasts were significantly increased in 4-OHT treated groups ( ~80% ) and all the 4-OHT-treated myoblast were TUNEL positive at 16 hr and 24 hr after release . Consistently , we also found that γH2AX signal was increased by approximately fourfold in BI2536-treated myoblasts compared to control myoblasts ( Figure 7—figure supplement 1C–D ) . These results are in agreement with the pivotal role of Plk1 in DNA damage checkpoint regulation ( Takaki et al . , 2008 ) , and demonstrate that deletion of Plk1 in MuSCs accumulates DNA damage response and leads to apoptosis . Finally , we explored the molecular mechanism mediating the defective DDR and apoptosis of Plk1-null MuSCs . The tumor suppressor p53 , known to play a pivotal role in DDR and cell survival , was reported to be upregulated by Plk1 inhibition in cancer cells ( Liu and Erikson , 2003 ) . WB analysis showed that p53 protein level was massively elevated in Plk1 KO myoblasts compared to control myoblasts ( Figure 7F ) . In contrast , Plk1 KO did not affect the level of Phospho-ATR ( Figure 7F ) , which has been reported to play a role in G2 checkpoint ( Cimprich and Cortez , 2008 ) . To examine if p53 upregulation is responsible for cell apoptosis , we used Pifithrin-α ( 2 nM , Sigma ) to inhibit p53 in control and Plk1 KO myoblasts . We found that 4-OHT-induced deletion of Plk1 led to more than 80% of apoptotic cells ( C-Caspase three positive ) within 36 hr , and inhibition of p53 significantly reduced apoptosis of Plk1-null myoblasts ( Figure 7G , H ) . However , inhibition of ATR failed to rescue the apoptosis of Plk1-null myoblasts ( Figure 7H ) , suggesting that apoptosis occurred prior to the G2-checkpoint . Interestingly , inhibition of p53 or ATR/ATM alone in WT myoblasts increased their apoptosis ( Figure 7H ) , consistent with the notion that p53 and ATR promotes cell survival in the absence of DNA damage ( Reinhardt et al . , 2007; Sakaguchi et al . , 1998 ) . Altogether , these results point to the upregulation of p53 as a key driver of cell cycle arrest and apoptosis in the Plk1-null MuSCs .
The expansion of MuSCs plays a crucial role for muscle development and repair following injury and a variety of signal pathways are involved in the regulation of myoblast proliferation and expansion ( Conboy and Rando , 2002; Jones et al . , 2001; Jones et al . , 2005; Perdiguero et al . , 2007 ) . Although Plk1 has been shown to play an indispensable role in cell cycle and mitotic progression of non-myogenic cell types especially cancer cells ( Barr et al . , 2004 ) , its function in MuSCs and myogenesis has not been investigated . Here , we identify an irreplaceable role of Plk1 in controlling MuSC proliferation . Specifically , we found that Plk1-deficient MuSCs lose their proliferative capacity , accumulate marks of DNA damage response and undergo apoptosis without completing mitosis , leading to failure of muscle development and defective muscle regeneration in response to injury . PLK1 is highly expressed in various human tumors and proliferating cells during embryonic development ( Lu et al . , 2008; van Vugt and Medema , 2005 ) . Our findings extend this feature to the adult stem cells and demonstrate that Plk1 activity synchronizes with the proliferative ability of MuSCs . Previous studies have reported that constitutive deletion of Plk1 leads to embryonic lethality at morula stage ( E3 . 5 ) due to mitotic aberrancies ( Wachowicz et al . , 2016 ) . Similar mitotic defects were also observed in our myogenic progenitor-specific Plk1 KO mice , which died prenatally due to failure in muscle development . Compared to our model , Pax7−/− mice exhibit progressive loss of the MuSC lineage with reduced muscle size , myonuclei number and myofiber diameters , leading to poor viability and early death within the first 3 weeks of life ( Seale et al . , 2000 ) . Myf5 and MyoD double knockout mice , but not the Myf5 or MyoD single knockout mice , show complete lack of skeletal muscle formation ( Rudnicki et al . , 1992; Rudnicki et al . , 1993 ) . Myogenin-null mice die after birth from severe and global muscle deficiency ( Hasty et al . , 1993 ) . Our myogenic progenitor specific Plk1 KO mice die earlier than these KO mice lacking one of the key myogenic transcription factors , demonstrating an indispensable role of Plk1 in embryonic muscle development . Activation of MuSCs is a crucial step for muscle regeneration . Evidence shows that a lack of MuSCs contributes to defective regeneration and muscle weakness ( Relaix and Zammit , 2012 ) . Indeed , previous studies have demonstrated that blockage of cell proliferation by colchicine treatment ( Pietsch , 1961 ) or irradiation ( Quinlan et al . , 1995 ) drastically reduces muscle regenerative capacity . In addition , MuSCs function is also controlled by various signals in the local microenvironment , called stem cell niche ( Kuang et al . , 2008 ) . This niche consists of various cells and extracellular factors that function to regulate MuSCs during muscle regeneration ( Kuang et al . , 2008 ) . Our study identifies Plk1 as an intrinsic regulator of MuSCs , but what extrinsic factors regulate the dynamic expression of Plk1 remains to be elucidated . It is reported that ~ 50% reduction of MuSCs induced by Pten deletion could sufficiently maintain muscle regeneration , and muscle regeneration fails only when the SC population drops to 10% or less ( Yue et al . , 2017 ) . In our model , MuSCs-specific Plk1 deletion only leads to ~30% reduction in MuSCs , yet the remainder 70% MuSCs fail to regenerate the injured muscle after CTX injury due to cell cycle arrest . This leads to infiltration of fibro-adipogenic progenitors ( FAPs ) , macrophages and adipocytes that occupy the muscle after CTX-induced regeneration . These results highlight the absolute requirement of Plk1 in cell cycle progression of Pax7-expressing MuSCs during regenerative myogenesis . PLKs are key regulators of many cell-cycle-related events , including chromosome segregation , centrosome maturation , bipolar spindle formation , regulation of anaphase-promoting complex , and execution of cytokinesis ( Barr et al . , 2004 ) . Conditional knockout of Plk1 leads to defective polyploidization and cell death in megakaryocytes ( Trakala et al . , 2015 ) . Plk1 inhibition also prevented pancreatic β cell proliferation and G2/M cell-cycle phase progression in a dose-dependent manner ( Shirakawa et al . , 2017 ) . In addition to proliferative defects , Plk1-null MuSCs also undergo apoptosis . This observation is consistent with previous reports that Plk1 depletion induces apoptosis in cancer cells ( Liu and Erikson , 2003 ) . During normal mitosis processes , if DNA damages are detected a signaling cascade will be initiated to enforce cell cycle arrest ( checkpoint activation ) , followed by DNA repair process . If DNA repair fails or if excessive DNA lesions are accumulated , an apoptosis process will be triggered . Phosphorylated H2AX could form nuclear foci within 1 min at the sites of DNA double-strand breaks and thus represents a sensitive marker of DNA damages ( Paull et al . , 2000 ) . Here , we reported that after Plk1 deletion , most MuSCs exhibit robust γH2AX signal , suggesting that a failure of DNA repair may underpin the observed cell death . Previous work has also shown that PLK1 plays a crucial role during recovery from G2 DNA damage checkpoint through targeting multiple factors such as ATR/Chk1 , ATM/Chk2 and p53 pathways ( Bassermann et al . , 2008; Li et al . , 2017 ) . We show that pharmacological inhibition of p53 or ATM/ATR significantly increases the apoptosis of wildtype myoblasts , potentially due to inhibition of p53 and ATM-dependent DNA damage response . The Plk1-null cells also underwent apoptosis , manifested by markers of activated Caspase-3 , DNA damage response ( γH2AX ) , DNA fragmentation and TUNNEL labeling . Previous studies have shown that the DNA damage response and activation of Caspase-3 are key steps in myogenic differentiation ( Burgon and Megeney , 2018; Fernando et al . , 2002; Fortini et al . , 2012; Larsen et al . , 2010 ) . In agreement with this concept , we also found that pharmacological inhibitions of Plk1 activates Caspase-3 and promotes myogenic differentiation . However , the Plk1-null myoblasts fail to express the myogenic differentiation marker myogenin despite activation of Caspase-3 ( Figure 4—figure supplement 1F and Figure 6—figure supplement 1B ) , due to the cell cycle blockage and apoptosis prior to terminal differentiation . These results demonstrate that cell cycle-dependent dynamic regulation Plk1 is key for MuSC proliferation and differentiation . PLK1 knockout in tumor cells induces DNA damage and causes p53 activation ( Li et al . , 2017; Liu and Erikson , 2003 ) , but activated p53 could also transactivate proapoptotic genes , leading to cell death ( Liu and Erikson , 2003 ) . This explains why p53 inhibition partially rescues apoptosis of Plk1-null myoblasts that express high levels of p53 . Our results demonstrate that a proper level of p53 is critical for MuSC homeostasis . Indeed , it has been reported that regeneration-induced loss of quiescence in p53-deficient MuSCs results in tumor formation ( Preussner et al . , 2018 ) . Similarly , lower levels of p53 is observed in aged MuSCs that exhibit a high frequency of cell death in the transient expansion phase of muscle regeneration ( Liu et al . , 2018 ) . PLK1 is highly expressed in several cancer types , and thus represents a druggable target in cancer therapeutics . BI2536 , one of the most effective Plk1 inhibitors , induces apoptosis of rhabdomyosarcoma cells when synergistically used with microtubule-destabilizing drugs , but a low dose of BI2536 ( 7 nM ) has no effect on C2C12 myoblasts ( Hugle et al . , 2015 ) . Consistently , we report that low dose of BI2536 ( 1–10 nM ) have no effect on the proliferation of MuSCs-derived primary myoblasts . However , BI2526 treatment profoundly inhibits MuSCs function during muscle regeneration , cautioning the potential side effect of PLK1 inhibition in skeletal muscle homeostasis and cancer cachexia . Impaired regenerative ability has been well-established in aged muscle owing in part to decreased number and functionality of MuSCs ( Jang et al . , 2011 ) , including defective dividing capability ( Liu et al . , 2018 ) and quiescent maintenance failure ( Chakkalakal et al . , 2012 ) . Similar aging features were observed in our Plk1 null or BI2536 treated MuSCs , suggesting that increasing Plk1 activity may preserve functions of MuSCs in aged muscle .
MyodCre ( #014140 ) , Pax7CreER ( #012476 ) mice were obtained from Jackson Laboratory and housed in the animal facility with free access to water and standard rodent chow . Plk1f/f mice was a gift from Dr . Cárcer from Spanish National Cancer Research Centre ( CNIO ) , Madrid , Spain ( Wachowicz et al . , 2016 ) . Mice were genotyped by PCR of ear DNA using primers listed in Supplementary file 2 . The genotypes of experimental KO and associated control animals are as follows: Plk1PKO ( Pax7CreER::Plk1f/f ) and wild type ( Plk1f/f ) , Plk1MKO ( MyodCre::Plk1f/f ) and wild type ( Plk1f/f ) . Mouse maintenance and experimental use were performed according to protocols approved by the Purdue Animal Care and Use Committee . Muscle regeneration was induced by injections of cardiotoxin ( CTX , sigma ) into the tibialis anterior ( TA ) muscles of 8–12 week-old male mice . Mice were anesthetized using a ketamine-xylazine cocktail , and 50 μl saline or 50 μl of 10 μM CTX was injected into TA muscles in the absence or presence of 10 μg BI2536 . Muscles were then harvested at different days post-injection to assess the completion of regeneration , repair and gene expression . Tamoxifen ( Calbiochem ) was prepared in corn oil at a concentration of 10 mg ml−1 , and experimental and control mice were injected intraperitoneally at 2 mg per day per 20 g body weight for 5 days to induce Cre-mediated deletion . Primary myoblasts were isolated from hind limb skeletal muscle of 6-week-old male mice as previously described ( Yue et al . , 2017 ) . Muscle tissues were minced and digested in type II collagenase and Dispase B mixture ( Roche ) . Digested cells were harvested and cultured in growth media , F-10 Ham’s medium ( Thermo Fisher Scientific ) supplemented with 20% fetal bovine serum ( FBS , Atlanta ) , 4 ng/ml basic fibroblast growth factor ( Thermo Fisher Scientific ) and 1% penicillin-streptomycin ( Thermo Fisher Scientific ) on collagen-coated dishes . Primary myoblasts were isolated and purified after 2–3 times of pre-plate . For in vitro genetic deletion , 4-OHT ( 0 . 4 μM , Calbiochem ) was added in culture medium for 1 day to induce Cre-mediated deletion . P53 ( Pifithrin-α , Sigma ) and ATM/ATR ( CGK733 , Sigma ) inhibitor were used according to the manufactory’s protocol and treated the myoblast together with 4-OHT or Methanol . Muscle differentiation was induced using 80% confluence of isolated primary myoblasts in Dulbecco's Modified Eagle Medium ( DMEM , Sigma , ) supplemented with 2% horse serum ( Sigma ) . Differentiated cells were kept in differentiation media for further analysis . The amount of DNA present in the cell was detected by Flow cytometry using propidium iodide ( PI ) staining . For each single test , Pax7CreER::Plk1f/f myoblasts from three individual were harvest from a 10 cm dish after treated with 4-OHT or Methanol for 1 day . Then the myoblasts were washed in PBS for twice and fixed in pre-cooled 70% ethanol for overnight . After another two times of wash with PBS , myoblasts were centrifuged , and the pellet was resuspended with PBS containing 10 μg/ml ribonuclease . Suspended myoblasts were incubated at 37°C for 1 hr and the 10 μl ( 2 mg/ml ) PI was added and incubated at 4°C for more than 1 hr before analysis using a BD flow cytometer . Apoptosis of myoblasts was detected by Flow cytometry using Alexa Fluor 488 annexin V and PI ( Invitrogen , V13241 ) double staining . For each single test , Pax7CreER::Plk1f/f myoblasts from two individual were digested using trypsin from two 10 cm dishes after treated with 4-OHT or Methanol for 36 hr . Then the myoblasts were washed with pre-cooled PBS and centrifuged . The pellet was resuspended with 100 μl of 1X annexin-binding buffer , then 5 µL Alexa Fluor 488 annexin V and 1 µL 100 µg/mL PI working solution was added to each cell suspension . The mixtures were incubated at room temperature for 15 min then 400 μl 1X annexin-binding buffer was added before analysis using a BD flow cytometer . Single fibers were isolated from extensor digitorum longus ( EDL ) muscles of adult mice as previously described ( Pasut et al . , 2013 ) . Briefly , intact EDL muscles were digested in 0 . 2% type I collagenase ( Sigma ) in DMEM and incubated for approximately 1 hr at 37°C . Fibers were then liberated from the muscle bulk using graded glass pipettes . Suspended fibers were cultured in 60 mm horse-serum-coated plates in DMEM supplemented with 10% FBS , 4 ng/ml basic fibroblast growth factor ( Promega ) , and 1% penicillin-streptomycin for 3 days . Freshly isolated fibers and cultured fibers were then fixed in 4% paraformaldehyde ( PFA ) for subsequent immunofluorescent analysis . MuSCs from at least 20 fibers were stained and count for statistical analysis . Fresh TA muscles and hind limb were embedded in optimal cutting temperature compound ( OCT , Tissue-Tek ) and frozen in isopentane that was chilled on dry ice . Frozen muscles were then cut into 10 μm-thick cross-sections by using a Leica CM1850 cryostat ( Leica ) . For hematoxylin and eosin staining , the slides were first stained in hematoxylin for 15 min , rinsed in running tap water and then stained in eosin for 1 min . Slides were dehydrated in graded ethanol and Xylene , and then covered using Permount . Stained images were captured with a Nikon D90 digital camera installed on a Leica DM6000 ( Leica ) inverted microscope . Immunofluorescence was performed on cross-sections , myofiber explants , primary myoblasts and differentiated myotubes . Briefly , samples were fixed in 4% PFA ( paraformaldehyde ) for 5 min and then permeabilized and blocked in PBS containing 5% goat serum , 2% bovine serum albumin ( BSA ) , 0 . 2% Triton X-100 , and 0 . 1% sodium azide for 1 hr . Samples were subsequently incubated with primary antibodies ( Key resources table ) overnight at 4°C . After washing with PBS , the samples were incubated with secondary antibodies and DAPI for 1 hr at room temperature . Fluorescent images were captured with a CoolSnap HQ charge coupled-device camera ( Photometrics ) by using a Leica DM6000 microscope ( Leica ) . 10 separated images were taken and count in each experimental groups . For TUNEL and Pax7 staining , slides were fixed in 4% PFA for 10 min and then subjected to the TUNEL reaction using the CF488A TUNEL Assay Apoptosis Detection Kit ( Biotium ) according to the manufacturer’s instructions . For negative control , samples were added TUNEL reaction buffer without TdT Enzyme . Samples treated with H2O2 ( 100 mM ) for 30 min before TUNEL staining was set up as positive control . Counterstaining of Pax7 was then performed as regular immunofluorescence staining procedure . Total RNA from muscle tissue and myoblast was extracted by using Trizol reagent ( Thermo Fisher Scientific ) . The first-strand cDNA was generated with random primer with MMLV Reverse Transcriptase ( Thermo Fisher Scientific ) . Real-time PCR reactions were performed with a SYBR green PCR kit ( Roche ) in the Roche LightCycler 480 System ( Roche ) . Primers for the genes of interest were all derived from the primer bank ( Harvard Medical School ) and were listed in Supplementary file 2 . Gene expression was determined with the 2-ΔΔCt relative quantification method and normalized to 18 s expression . Protein was extracted from homogenized muscle tissue or muscle cells ( For Plk1-null myoblasts , each sample represent proteins extract from one 10 cm culture dish ) with RIPA buffer that contained a protease inhibitor cocktail ( Sigma ) and phosphatase inhibitors NaF and Na3VO4 . Protein concentration was measured using the BCA protein quantification kit ( Pierce ) . Equal amounts of each protein sample were loaded for electrophoresis ( Bio-Rad ) . Proteins were then transferred to a PVDF membrane ( Biorad ) and incubated with primary antibodies , followed by anti-rabbit or anti-mouse immunoglobulin G-horseradish peroxidase secondary antibody ( Cell Signaling Technology ) . Signals were detected using fluorescence or chemiluminescence Western blot detection reagent ( Santa Cruz Biotechnology ) on a FluorChem E system ( Protein Simple ) . Antibodies used for western blot analysis were listed in Key resources table . Single cell gel electrophoresis assay was performed in primary myoblast as previously described ( Collins , 2004 ) . Briefly , 10 , 000 cells were collected in 10 μl PBS and mixed with 75 μl of 0 . 5% low-melting point agarose . The cell-agarose mixture was then placed on a chilled and fully frosted slide with a 1% normal-melting point agarose coating layer . Subsequently , the slide was submerged in lysis solution ( 2 . 5 M NaCl , 100 mM Na2EDTA , 10 mMTris , pH 10 , 1% Sodium Sarcosinate with 1% triton X-100% and 10% DMSO being 1 hr before use ) overnight at 4°C . Electrophoreses were carried out in alkaline electrophoresis buffer ( 1 mM Na2EDTA and 300 mM NaOH , pH >13 ) at 24 v , 300 milliamperes for 30 min . The slide was then neutralized and stained with 1 μg/ml DAPI for 15 min . Images were captured with CoolSnap HQ charge coupled-device camera ( Photometrics ) by using a Leica DM6000 microscope ( Leica ) . 100 nuclei were measured by their tail DNA content for each treatment , and scored five classes ( 0–4 ) according to previous protocol ( Collins , 2004 ) . Class 0 represent 0% DNA is in tail and class 4 means 100% DNA is in the tail , respectively . Classes 1–3 are in between 0 and 100% and with an increment from 1 to 3 . All analyses were conducted with Student's t-test ( two-tail ) . All experimental data were presented as mean ± SEM . Comparisons with p values < 0 . 05 or <0 . 01 were considered statistically significant . | Muscles have their own population of stem cells , called muscle satellite cells . These cells are essential for muscle growth and repair . In healthy adult muscles , they spend most of their time inactive , but when there is an injury , they reawaken and start dividing . Some of the new cells return to an inactive stem cell state to await the next injury . The rest mature into new muscle cells or join with damaged muscle fibres to help them repair . The cell cycle is the series of events that a cell goes through from its birth until it divides . In muscle satellite cells , progression through the cell cycle is tightly controlled to ensure they divide and grow the correct amount . One of the proteins responsible for controlling the cell cycle is Polo-Like Kinase 1 ( PLK1 ) , but studying this protein is difficult . A common way to investigate a protein's effect is to delete the gene that makes it and observe the consequences . However , PLK1 is so essential to life that yeast , flies , zebrafish and mice all die when the gene is missing . Jia et al . deleted the gene that makes PLK1 only in mouse muscle satellite cells to find out the role this protein plays in controlling the cell cycle in stem cells . Deleting the gene that codes for PLK1 before the mice were born was lethal . The embryos failed to develop mature muscle fibres , and they died . But deleting the gene after the mice were born had a different effect . The muscles developed normally , but they were unable to heal when injured . The same healing problem also happened when healthy mice received a drug that blocked the function of PLK1 protein . A closer look at the muscle satellite cells revealed the source of the problem . Without PLK1 , the cells got stuck part way through their cell cycle , just before they were due to divide . They tried to become muscle cells , but they did not make it . Instead , the muscle satellite cells started to act as though their DNA had been damaged , and then they self-destructed . Muscle satellite cells become less able to divide as we get older . They can also malfunction in some types of degenerative muscle diseases . Understanding how muscle satellite cells control their cell cycle could help us to find out what causes them to go wrong . Further work to understand PLK1 also has potential implications for cancer treatment . PLK1 blockers have been used to stop cancer cells from dividing , but Jia et al . ’s findings show that this kind of drug may also hamper the ability of muscle to repair damage . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"developmental",
"biology"
] | 2019 | A requirement of Polo-like kinase 1 in murine embryonic myogenesis and adult muscle regeneration |
Mitochondrial dysfunction and subsequent metabolic deregulation is observed in neurodegenerative diseases and aging . Mutations in the presenilin ( PSEN ) encoding genes ( PSEN1 and PSEN2 ) cause most cases of familial Alzheimer’s disease ( AD ) ; however , the underlying mechanism of pathogenesis remains unclear . Here , we show that mutations in the C . elegans gene encoding a PSEN homolog , sel-12 result in mitochondrial metabolic defects that promote neurodegeneration as a result of oxidative stress . In sel-12 mutants , elevated endoplasmic reticulum ( ER ) -mitochondrial Ca2+ signaling leads to an increase in mitochondrial Ca2+ content which stimulates mitochondrial respiration resulting in an increase in mitochondrial superoxide production . By reducing ER Ca2+ release , mitochondrial Ca2+ uptake or mitochondrial superoxides in sel-12 mutants , we demonstrate rescue of the mitochondrial metabolic defects and prevent neurodegeneration . These data suggest that mutations in PSEN alter mitochondrial metabolic function via ER to mitochondrial Ca2+ signaling and provide insight for alternative targets for treating neurodegenerative diseases .
In all metazoans , mitochondria are essential organelles and mitochondrial dysfunction is frequently observed in neurodegenerative diseases and aging . While mitochondria have a well-established role in ATP generation , they also have a vital role in Ca2+ homeostasis . Several studies have shown that release of endoplasmic reticulum ( ER ) Ca2+ results in the elevation of mitochondrial Ca2+ levels which in turn stimulates metabolic activity of the mitochondria ( Das and Harris , 1990; Glancy and Balaban , 2012; Hansford and Zorov , 1998; McCormack and Denton , 1993; Mildaziene et al . , 1995; Wernette et al . , 1981 ) . Thus , insults or deregulated signaling between the ER and mitochondria can cause mitochondrial dysfunction and affect cellular fitness . Alzheimer’s disease ( AD ) is the leading cause of dementia in the elderly and accounts for between 60–80% of all cases of dementia . Familial Alzheimer’s disease ( FAD ) is a subset of AD where there is a genetic predisposition to the disease as a result of mutations predominantly in the presenilin genes , PSEN1 and PSEN2 ( ALZFORUM , 2016 ) . PSENs are ~50 kDa multipass transmembrane proteins that primarily localize to the ER ( Bezprozvanny and Mattson , 2008 ) and are enriched in the ER membrane associated with mitochondria ( Area-Gomez et al . , 2009 ) . Notably , PSENs are the critical aspartyl protease component of the gamma-secretase complex , a multi-subunit protease that resides in cellular membranes . Gamma-secretase is involved in the cleavage of amyloid precursor protein ( APP ) into beta amyloid ( Abeta ) peptides , and in the cleavage and activation of several other transmembrane proteins , including Notch ( Beel and Sanders , 2008 ) . Dominating the study of AD pathogenesis is the amyloid hypothesis which ascertains that altered PSEN function leads to an increase in the toxic Abeta42 peptide , and the eventual accumulation of amyloid plaques in the brain leads to neurodegeneration and dementia observed in AD patients ( Hardy , 2006 ) . However , it is noteworthy that while amyloid plaques serve as a histopathological hallmark of AD , there is no correlation between plaque load and the severity of dementia , and postmortem analyses of some AD patients with exaggerated cognitive decline have shown a lack of plaque formation ( Giannakopoulos et al . , 2003; Guillozet et al . , 2003; Terry et al . , 1991 ) . Also , using neuroimaging techniques , extensive plaque formation has been observed in people with no cognitive impairment ( Nordberg , 2008; Villemagne et al . , 2008 ) . Concomitantly , there is a large body of work that implicates exacerbated ER Ca2+ release as the causative agent in AD pathogenesis ( Bandara et al . , 2013; Chan et al . , 2000; Cheung et al . , 2008; Green et al . , 2008; Leissring et al . , 1999; Stutzmann et al . , 2004; Tu et al . , 2006 ) . Moreover , there is evidence of enhanced ER-mitochondria crosstalk in cells from FAD patients with mutations in PSEN1 , PSEN2 , and APP , as well as patients with sporadic AD ( Area-Gomez et al . , 2012 ) and enhanced ER to mitochondrial Ca2+ transfer is observed in cells expressing FAD-mutant PSEN2 ( Zampese et al . , 2011 ) . Additionally , reactive oxygen-mediated oxidative damage has been observed in brains of AD patients ( Lovell and Markesbery , 2007; Lovell et al . , 2011 ) and an increase in hydrogen peroxide levels is observed in AD mice models even prior to the appearance of plaques ( Manczak et al . , 2006 ) . However , despite decades of research , a clear connection between these seemingly disparate observations does not exist that explains the pathogenesis of AD . In C . elegans , disruption of the PSEN ortholog , SEL-12 , leads to ER Ca2+ dysregulation that causes mitochondrial disorganization and reduced organismal health ( Sarasija and Norman , 2015 ) . Here , we examine whether these mitochondrial defects observed in sel-12 mutants lead to mitochondrial metabolic defects that can result in neuronal dysfunction . From these analyses , we find that sel-12 mutations result in increased mitochondrial Ca2+ concentration , accelerated oxidative phosphorylation ( OXPHOS ) , elevated reactive oxygen species ( ROS ) generation and oxidative stress mediated neurodegeneration . Remarkably , we demonstrate that reducing ER to mitochondrial Ca2+ transfer in sel-12 mutants can normalize mitochondrial Ca2+ levels and function , reduce the levels of ROS and suppress neurodegeneration . Moreover , we demonstrate similar mitochondrial dysfunction in fibroblasts isolated from FAD patients . Lastly , we show that treating sel-12 mutants with a mitochondria-targeted antioxidant suppresses the neurodegenerative phenotypes observed in sel-12 mutants . Since C . elegans do not encode an Abeta peptide ( Daigle and Li , 1993; McColl et al . , 2012 ) , our data indicates that neurodegeneration occurs in presenilin mutants by faulty ER to mitochondria Ca2+ transfer leading to mitochondrial metabolic dysfunction , completely independent of Abeta signaling , and consummates in the rise of mitochondrial ROS generation to detrimental levels . Therefore , our study provides a comprehensive delineation of AD pathogenesis in an intact animal model and its importance is underscored by the fact that we provide evidence that Abeta signaling does not appear necessary for neurodegeneration .
Previously , we observed mitochondrial disorganization in the body wall muscle of sel-12 animals , which could be rescued by reducing Ca2+ release from the ER or mitochondrial Ca2+ uptake ( Sarasija and Norman , 2015 ) . This led us to hypothesize that mutations in sel-12 result in increased mitochondrial Ca2+ levels . To test this hypothesis , we expressed in the body wall muscle , a mitochondrial matrix targeted GCaMP6 ( a genetically encoded Ca2+ indicator ) along with a red fluorescent protein , mCherry , as an expression control ( Figure 1—figure supplement 1A ) . We normalized the fluorescence intensity of the GCaMP6 to that of mCherry to determine the relative mitochondrial matrix Ca2+ concentration in day one adult wild type animals and two sel-12 mutants , sel-12 ( ar131 ) and sel-12 ( ty11 ) . sel-12 ( ar131 ) mutants carry a missense mutation in the sel-12 gene , which is a conserved change seen in FAD patients ( ALZFORUM , 2016; Levitan and Greenwald , 1995 ) and sel-12 ( ty11 ) mutants contain a premature stop codon in the sel-12 gene , which is a predicted null mutation ( Cinar et al . , 2001 ) . Strikingly , we observe that the relative fluorescence intensity of GCaMP6 to mCherry is 2 . 4 and 3 . 4 fold higher in sel-12 ( ar131 ) and sel-12 ( ty11 ) animals , respectively , compared to wild type animals ( Figure 1A ) . Therefore , consistent with our hypothesis , sel-12 mutants have higher levels of mitochondrial matrix Ca2+ . Since sel-12 , like PSEN1 , is widely expressed , we examined mitochondrial Ca2+ levels in a subset of neurons using the same ratiometric GCaMP6/mCherry construct . Similar to our observation of body wall muscle mitochondrial Ca2+ levels , we detected higher fluorescence levels of GCaMP6 in the ALM and PLM light touch mechanosensory neurons of sel-12 mutants consistent with elevated mitochondrial Ca2+ levels in the nervous system ( Figure 1—figure supplement 1B ) . Although mitochondria can act as a Ca2+ sink , they also respond dynamically to increases in Ca2+ levels . Indeed , an increase in mitochondrial matrix Ca2+ levels results in the coordinated up regulation of the tricarboxylic acid ( TCA ) cycle and OXPHOS , increasing ATP output and ROS levels ( Das and Harris , 1990; Glancy and Balaban , 2012; Hansford and Zorov , 1998; McCormack and Denton , 1993; Mildaziene et al . , 1995; Wernette et al . , 1981 ) . Thus , to investigate whether the elevated Ca2+ levels observed in sel-12 mutant mitochondria influence mitochondrial activity , we first analyzed ATP levels as a general read out of mitochondrial function . Remarkably , we observed that ATP levels are higher in sel-12 ( ar131 ) , sel-12 ( ty11 ) and sel-12 ( ok2078 ) ( an additional null allele ) animals when compared to age matched day one wild type animals ( Figure 1B ) . Next , to determine whether this elevation in ATP levels was a result of amplified OXPHOS , we examined mitochondrial respiration by measuring the oxygen consumption rate ( OCR ) in sel-12 mutant and wild type animals using a Seahorse XFp Analyzer . While wild type animals had a basal OCR of 3 . 9 pmol/min/worm , sel-12 ( ar131 ) , sel-12 ( ty11 ) and sel-12 ( ok2078 ) animals respired at a rate of 5 . 8 , 6 . 4 and 5 . 5 pmol/min/worm , respectively ( Figure 1C ) . Similarly , when challenged with carbonyl cyanide 4- ( trifluoromethoxy ) phenylhydrazone ( FCCP ) , an uncoupler of mitochondrial OXPHOS , we detected an increased maximal OCR of 11 . 7 , 10 . 3 and 10 pmol/min/worm in sel-12 ( ar131 ) , sel-12 ( ty11 ) and sel-12 ( ok2078 ) animals respectively when compared to the maximal OCR of 7 . 2 pmol/min/worm in wild type animals ( Figure 1D ) . These data indicate that sel-12 mutants have elevated OXPHOS activity , which results in higher OCR and ATP levels . Consistent with elevated mitochondrial activity , we have previously reported high ROS levels in sel-12 mutants ( Sarasija and Norman , 2015 ) . Interestingly , as sel-12 mutants age , mitochondrial respiration dramatically decreases compared to wild type animals suggesting a reduction of mitochondrial function in old animals ( Figure 1—figure supplement 1C , D ) . The elevated levels of ATP , ROS and mitochondrial Ca2+ as well as the high OCR observed in young adult sel-12 mutants could arise due to increased mitochondrial content in sel-12 mutants . Thus , to investigate this possibility , we first used qRT-PCR to measure relative mitochondrial DNA ( mitoDNA ) copy number versus nuclear DNA in day one adult wild type and sel-12 mutants . We found that the relative mitoDNA copy number in sel-12 mutants is similar to wild type animals ( Figure 1—figure supplement 1E ) . Additionally , we measured fluorescence intensity of day one adult animals expressing a TOM20::GFP integrated transgene that labels the outer mitochondrial membrane in the body wall muscle . We found that fluorescence intensity between wild type and sel-12 mutants are indistinguishable ( Figure 1—figure supplement 1F ) . These data suggest that the changes in ATP and mitochondrial Ca2+ levels and OCR are not caused by an increase in mitochondrial biogenesis or mitochondrial accumulation . A recent study on astrocytes differentiated from induced pluripotent stem cells ( iPSCs ) derived from AD patients with PSEN1 mutations demonstrated high OCR and ROS levels in these cells , which could be rescued by CRISPR/Cas9 correction of the mutation ( Oksanen et al . , 2017 ) . These data along with our data suggest a conserved role of PSEN1 and SEL-12 in regulating mitochondrial activity . To investigate this further , we analyzed mitochondrial activity in skin fibroblast cells isolated from FAD patients with PSEN1 mutations . First , we examined ATP levels to obtain an approximation of mitochondrial activity in control and FAD fibroblasts . As we observed in sel-12 mutants , we found that ATP levels are significantly higher in FAD fibroblasts compared to control fibroblasts ( Figure 1E ) . Next , we investigated mitochondrial respiratory rate by measuring oxygen consumption in FAD and control fibroblasts . Again , consistent with the sel-12 mutants and similar to FAD astrocytes ( Oksanen et al . , 2017 ) , we found that the basal and maximal OCRs are significantly higher in FAD fibroblasts compared to control fibroblasts ( Figure 1F , G ) . Lastly , since we previously found high levels of ROS in sel-12 mutants ( Sarasija and Norman , 2015 ) and ROS is produced by mitochondrial respiratory activity , we examined ROS levels in FAD and control fibroblasts . Like sel-12 mutants and FAD astrocytes ( Oksanen et al . , 2017 ) , we found elevated ROS in the FAD fibroblasts compared to control fibroblasts ( Figure 1H ) . These observations are consistent with PSEN1 having a role in mitochondrial activity in human fibroblasts similar to the role SEL-12 has in C . elegans . In addition to body wall muscle mitochondria , we previously observed mitochondrial disorganization in a subset of interneurons in sel-12 mutants ( Sarasija and Norman , 2015 ) and therefore , sought to determine whether mutations in sel-12 lead to mitochondrial disorganization in other neuronal cell types and , if so , investigate whether this leads to a physiological defect in neuronal function in vivo . We focused on touch receptor neurons , a class of mechanosensory neurons that respond to light touch . Light touch to the body of C . elegans is sensed by six mechanosensory neurons ( ALML/R , PLML/R , AVM , and PVM ) whose structural and functional neurodegeneration has been well-characterized ( Pan et al . , 2011; Tank et al . , 2011; Toth et al . , 2012 ) . Using transgenic animals that express mCherry targeted to the outer mitochondrial membrane in mechanosensory neurons ( Hsu et al . , 2014 ) , we compared the mitochondrial structure in wild type animals , sel-12 ( ar131 ) and sel-12 ( ty11 ) at day one of adulthood . Unlike wild type animals , which maintain their mitochondria mostly in a connected network ( Figure 2A ) ( Hsu et al . , 2014 ) , we found that 53 . 3 and 55% of sel-12 ( ar131 ) and sel-12 ( ty11 ) mutants , respectively , have discontinuous morphology of ALM neuronal mitochondria compared to the 16 . 7% of wild type animals with discontinuous mitochondria morphology ( Figure 2A ) . This data suggests that SEL-12 function is required to maintain mitochondrial structure and is consistent with our previous observations of mitochondrial disorganization in the body wall muscle and interneurons of sel-12 mutants ( Sarasija and Norman , 2015 ) . Given the extent of structural disorganization observed in the mechanosensory neuronal mitochondria of sel-12 mutants , we decided to investigate whether sel-12 mutations lead to a behavioral defect , and examined light touch response in these animals . A freely crawling animal when touched by an eyebrow hair just posterior of the pharynx ( anterior touch ) responds normally with a reversal in their motion and when touched slightly anterior of the anus ( posterior touch ) , will move forward; a response that dampens as the animal ages into mid-late adulthood ( Pan et al . , 2011; Tank et al . , 2011; Toth et al . , 2012 ) . We examined the response to anterior and posterior light touch in sel-12 ( ar131 ) , sel-12 ( ty11 ) , and sel-12 ( ok2078 ) at L4 larval stage ( last larval stage before adulthood ) , and days 1 and 3 of adulthood . From these analyses , we found that sel-12 mutants respond similarly to wild type animals in their L4 larval stage but show a significant reduction in the response to light touch at day 1 of adulthood with only 45 . 1 , 39 and 39 . 2% positive response in sel-12 ( ar131 ) , sel-12 ( ty11 ) and sel-12 ( ok2078 ) mutants , respectively , compared to the 83 . 1% positive response in wild type animals ( Figure 2B ) . Moreover , this response to light touch progressively worsens when this behavior is examined in day three adults ( Figure 2B ) . In order to determine whether the mechanosensation defect in adult animals is due to the loss of SEL-12 function in the nervous system , we investigated mechanosensation in sel-12 ( ty11 ) null mutants expressing wild type SEL-12 under a pan-neuronal or muscle-specific promoter ( Sarasija and Norman , 2015 ) . Pan-neuronal expression of wild type SEL-12 rescues the mechanosensory defects in sel-12 ( ty11 ) mutants while the expression of wild type SEL-12 under a muscle-specific promoter fails to rescue ( Figure 2C ) . Since we have found mitochondrial disorganization in the mechanosensory neurons ( Figure 2A ) and mechanosensory defects ( Figure 2B ) in adult sel-12 mutants , we next examined the morphology of these neurons for any signs of neurodegeneration . In healthy animals , ALM neurons comprise a small circular soma , a long anterior process that extends into the nerve ring and an occasional short posterior process with no defined function ( Figure 3A top , middle ) . Previously , it has been shown that as adult wild type animals age they show ectopic branches sprout from the ALM neuronal soma , which is considered to be a sign of age associated neurodegeneration ( Pan et al . , 2011; Tank et al . , 2011; Toth et al . , 2012 ) . While ALM in L4 larval stage sel-12 mutants is indistinguishable from wild type animals , strikingly sel-12 mutants show ectopic branches emanating from the soma as adults ( Figure 3A top , Figure 3B ) . Day one sel-12 ( ty11 ) adults show on average two sprouts/ALM soma compared to 0 . 56 sprouts/ALM soma in day one wild type adults ( Figure 3A , B ) . Furthermore , as they age further , the sel-12 ( ty11 ) animals develop upwards of 3 sprouts/ALM soma ( while age matched wild type animals only have one sprout/ALM soma ) ( Figure 3A , B ) , paralleling the loss in mechanosensation ( Figure 2B ) . Also , supporting a neuronal tissue specific role of SEL-12 , we found that pan-neuronal expression of wild type SEL-12 decreases the sprouting in adult sel-12 ( ty11 ) animals from 2 . 8 sprouts/ALM soma to 1 . 2 sprouts/ALM soma while muscle specific expression of SEL-12 does not ( Figure 3—figure supplement 1A ) . Along with the structural aberrations observed in the ALM soma , the axonal processes of mechanosensory neurons are known to undergo age-associated neurodegeneration ( Pan et al . , 2011; Tank et al . , 2011; Toth et al . , 2012 ) . Strikingly , we observe morphological abnormalities in the axonal processes of the mechanosensory neurons in adult sel-12 ( ty11 ) animals at a higher frequency than in age-matched wild type animals . The ALM and PLM axonal processes of sel-12 ( ty11 ) animals at day 1 of adulthood develop a wavy appearance at a frequency of ~7 per 10 ALM/PLM processes while age-matched wild type animals show only ~2 per 10 ALM/PLM processes ( Figure 3A bottom , Figure 3C ) . The appearance of irregular lesions that appear to give way to breaks in the neuronal processes is also noted . Day one adult sel-12 ( ty11 ) animals have irregular lesions ( Figure 3A bottom , Figure 3D ) and breaks in their neuronal processes ( Figure 3A bottom , Figure 3E ) at a rate of ~6 and ~2 per 10 ALM/PLM processes , respectively compared to wild type animals at ~1 and ~0 per 10 ALM/PLM processes , respectively . sel-12 mutants appear normal at the L4 larval stage but as they enter adulthood , their propensity of developing abnormal processes , lesions and breaks also increases ( Figure 3C–E ) similar to our observation with ectopic sprouting ( Figure 3B ) . Taken together , the neuronal structural aberrations and decline in mechanosensory perception in adult sel-12 mutants appear to be structural and functional indicators of neurodegeneration in these animals rather than symptoms of developmental defects . Presenilin is the catalytic component of the gamma-secretase complex that is responsible for proteolytic cleavage and activation of Notch signaling . Due to the pervasive nature of Notch signaling , we investigated whether the mechanosensory neuronal defects observed in sel-12 mutants are the result of Notch signaling loss . C . elegans possess two genes encoding Notch orthologs , lin-12 and glp-1 . We assayed mechanosensation in both lin-12 and glp-1 mutants to determine if loss of Notch signaling was sufficient to cause mechanosensory defects akin to sel-12 mutants . Unlike sel-12 mutants , both lin-12 and glp-1 mutants displayed responses to light touch at 86 . 4 and 78 . 5% as day one adults similar to the 83 . 1% positive response in day one aged wild type animals ( Figure 4A ) . Moreover , the light touch response of lin-12 and glp-1 mutants was indistinguishable from wild type animals even at day 2 and day 3 of adulthood ( Figure 4A ) , suggesting that Notch signaling is not required for mechanosensation . Gamma-secretase activity is also involved in the cleavage of several other transmembrane proteins , which includes , but is not limited to , APP ( Beel and Sanders , 2008 ) . Thus , we next investigated whether mechanosensation is mediated in a gamma-secretase dependent but Notch-independent manner . We pharmacologically inhibited gamma-secretase activity using Compound E which specifically blocks SEL-12 but not HOP-1 ( another C . elegans presenilin homolog ) mediated gamma-secretase activity ( Francis et al . , 2002; Sarasija and Norman , 2015 ) . In order to obtain complete lack of gamma-secretase activity , we treated hop-1 ( ar179 ) null animals with Compound E and examined their response to light touch . The gamma-secretase inhibited animals responded at 87 . 1 , 79 . 8 and 78 . 3% to light touch at day 1 , 2 and 3 of adulthood respectively , which is similar to wild type animals at 83 . 1 , 69 . 3 and 67 . 3% ( Figure 4B ) . Furthermore , analysis of ALM mechanosensory neuron morphology in animals treated with Compound E did not show ectopic sprouts as is observed in sel-12 mutants ( Figure 4C ) . While compound E treatment clearly shows loss of gamma-secretase function regarding Notch signaling ( e . g . glp-1 like sterility ) , it is unclear whether the drug is targeting the nervous system . To address this caveat , we generated a proteolytic dead version of SEL-12 . The gamma-secretase activity of presenilin is dependent on the presence of two intact aspartate residues , D257 and D385 in PSEN1 . Disruption of one aspartate residue leaves PS catalytically nonfunctional ( Kim et al . , 2005; Wolfe et al . , 1999 ) . Thus , to disrupt aspartyl proteolytic activity of SEL-12 , we generated a D226A mutation ( equivalent to D257A in PSEN1 ) in our sel-12 genomic rescue construct and tested whether this mutant construct could rescue the light touch defect observed in sel-12 mutants . Consistent with gamma-secretase inactivation , the SEL-12 D226A mutant does not rescue the loss of Notch signaling required for vulval development leaving the sel-12 animals with their characteristic egg-laying defect . However , sel-12 ( ty11 ) animals expressing the gamma-secretase dead SEL-12 , similar to the animals expressing wild type SEL-12 rescue construct show improved response to touch at 85 and 82 . 8% , respectively , comparable to the response of wild type animals at 79 . 8% ( Figure 4D ) . Similar rescue of mechanosensation was observed in sel-12 ( ar131 ) expressing either wild type or gamma-secretase dead SEL-12 ( Figure 3—figure supplement 1B ) . In addition , sel-12 animals expressing the gamma-secretase dead SEL-12 rescued the neurodegenerative phenotypes , including waves , lesions , breaks , and ectopic sprouting , comparable to the animals expressing wild type SEL-12 ( Figure 4—figure supplement 1 ) . These data indicate that gamma-secretase activity does not have a role in maintenance of mechanosensory neuronal integrity or function . The major histopathological hallmark of Alzheimer’s disease is the presence of amyloid plaques; aggregates of Abeta peptides formed from the consecutive cleavage of APP by beta and gamma secretases . Since C . elegans do not form Abeta peptides due to their lack of beta-secretase and APP ( Daigle and Li , 1993; McColl et al . , 2012 ) , we used a transgenic line , which overexpresses human Abeta1-42 peptide in the nervous system ( pan-neuronal expression ) to investigate whether Abeta overexpression causes similar phenotypes as sel-12 mutants ( Wu et al . , 2006 ) . Previous analyses of these Abeta1-42 overexpressing animals have demonstrated neuronal dysfunction ( Ahmad and Ebert , 2017; Bravo et al . , 2018; Dosanjh et al . , 2010; Wu et al . , 2006 ) . Thus , it is clear that Abeta1-42 overexpression is toxic and causes cellular defects . Here , we find that pan-neuronal Abeta1-42 expressing day one adult animals ( Figure 5—figure supplement 1 ) show a mechanosensory deficit similar to that of sel-12 mutants ( Figure 5A ) ; however , they do not show any neuronal structural aberrations ( Figure 5B , C ) as we observed in sel-12 mutants . Furthermore , both mitochondrial structure and mitochondrial calcium levels in the Abeta1-42 overexpressing animals appear wild type ( Figure 5D , E ) . These data suggest that while the ectopic overexpression of Abeta1-42 likely perturbs neuronal function , unlike sel-12 mutants , it appears to be doing so without affecting axon morphology and mitochondrial calcium homeostasis . Since we previously demonstrated that reducing ER Ca2+ release in sel-12 mutants markedly improved the mitochondrial disorganization that occurs in body wall muscle ( Sarasija and Norman , 2015 ) and given the presence of elevated mitochondrial Ca2+ in sel-12 mutants ( Figure 1A , Figure 1—figure supplement 1B ) , we hypothesized that exacerbated ER Ca2+ signaling in the nervous system is promoting the mechanosensory neuronal defects observed in these animals by causing heightened OXPHOS mediated ROS production . To test this hypothesis , we introduced a calreticulin null mutation , crt-1 ( jh101 ) , which reduces ER Ca2+ release ( Michalak et al . , 1999; Xu et al . , 2001 ) , into the sel-12 mutant background , and determined whether the crt-1 null mutation could alter mitochondrial Ca2+ load using a mitochondrial targeted GCaMP6/mCherry Ca2+ sensor . From these analyses , we found that crt-1 null animals have a significantly reduced GCaMP6 fluorescence compared to wild type animals and the introduction of the crt-1 null mutation into sel-12 ( ty11 ) animals normalized their GCaMP6 fluorescence to wild type levels ( Figure 6A ) . Similarly , we found that introduction of an unc-68 null mutation ( unc-68 encodes the only ryanodine receptor in the C . elegans genome ) into the sel-12 mutant background also normalized the mitochondrial GCaMP6 fluorescence to wild type levels ( Figure 6—figure supplement 1A ) . Next , we explored the impact this normalization of mitochondrial Ca2+ level has on mitochondrial disorganization , OCR and ROS levels in sel-12 mutants . Similar to our studies of mitochondrial structure in the body wall muscle ( Sarasija and Norman , 2015 ) , we observe a dramatic improvement of mitochondrial organization in crt-1; sel-12 double mutants compared to sel-12 mutants ( Figure 6B , C ) . Next , to ascertain whether exacerbated ER Ca2+ release is responsible for the elevated OCR observed in sel-12 mutants , we determined the OCR in crt-1; sel-12 and unc-68; sel-12 mutant animals . We found that crt-1; sel-12 and unc-68; sel-12 double mutants have basal OCR and maximal OCR rates comparable to wild type and significantly lower than sel-12 ( Figure 6D , E , Figure 6—figure supplement 1B , C ) . These data are consistent with the notion that higher Ca2+ release from the ER leads to elevated OXPHOS in sel-12 mutants . Additionally , this elevated OXPHOS activity can explain the increased levels of ROS we previously observed in sel-12 mutants ( Sarasija and Norman , 2015 ) . Since we found that the OCR was decreased in sel-12 animals that had reduced ER Ca2+ release , indicating decreased OXPHOS activity , we investigated whether crt-1; sel-12 mutant animals would have a reduced ROS burden . To test this , we quantified ROS formation and discovered that crt-1; sel-12 animals have ROS levels comparable to wild type animals and lower than sel-12 mutants animals ( Figure 6F ) . These data suggest that mutations in sel-12 lead to exacerbated ER Ca2+ release , which causes mitochondrial morphology changes and increased metabolic activity resulting in elevated ROS . In order to determine whether the improvement of mitochondrial organization and function of sel-12 mutants in the presence of the crt-1 or unc-68 null mutation could rescue the sel-12 neurodegenerative phenotypes , we assessed mechanosensory neuron morphology in crt-1; sel-12 and unc-68; sel-12 double mutants and behavioral response to light touch in crt-1; sel-12 double mutants . Strikingly , the ectopic sprouting , wave-like processes , lesions and breaks of the ALM mechanosensory neurons observed in sel-12 mutants are strongly suppressed in crt-1; sel-12 and unc-68; sel-12 animals ( Figure 6G , Figure 6—figure supplement 1D–E , Figure 6—figure supplement 2A–C ) . Consistent with the absence of neurodegenerative morphology of the mechanosensory neurons , the crt-1; sel-12 animals also showed improved response to light touch with a 77 . 7% response compared to 48 . 4% response of sel-12 animals ( Figure 6H ) . This suggests that loss of SEL-12 function results in deregulated ER Ca2+ release which promotes neuronal mitochondrial disorganization , enhanced OXPHOS and high ROS production which leads to neurodegeneration in sel-12 mutants . Ca2+ uptake into the mitochondrial matrix is mediated by the highly conserved mitochondrial Ca2+ uniporter ( Baughman et al . , 2011; Csordás et al . , 2013; De Stefani et al . , 2011 ) . Prior work from our lab has shown that the introduction of a null mutation in the gene encoding the mitochondrial Ca2+ uniporter , mcu-1 , which reduces mitochondrial Ca2+ uptake ( Xu and Chisholm , 2014 ) , into sel-12 ( ty11 ) animals improves the body wall muscle mitochondria structure similar to the crt-1 mutation in sel-12 ( ty11 ) animals ( Sarasija and Norman , 2015 ) . Therefore , we investigated the role mitochondrial Ca2+ uptake has on the neurodegenerative phenotypes observed in sel-12 ( ty11 ) animals by generating mcu-1; sel-12 ( ty11 ) double mutants . In order to determine the effect the absence of the mitochondrial Ca2+ uniporter has on mitochondrial Ca2+ levels , we measured mitochondrial Ca2+ levels and found that mcu-1 null animals have reduced mitochondrial Ca2+ levels compared to wild type animals ( Figure 7A ) . Also , akin to our analyses of the crt-1; sel-12 and unc-68; sel-12 double mutants , mcu-1; sel-12 ( ty11 ) animals , show a reduction of mitochondrial Ca2+ load ( Figure 7A ) . Furthermore , as we previously observed in the body wall muscle ( Sarasija and Norman , 2015 ) , we found that the mcu-1 mutation improved the organization of neuronal mitochondria in sel-12 mutants ( Figure 7B ) . Consistent with the improvement in mitochondrial structure and Ca2+ load , mcu-1; sel-12 ( ty11 ) animals also show normal ROS and OCR levels similar to what was observed in sel-12 mutants with reduced ER Ca2+ release ( Figure 7C–E ) . Along with these observations , phenotypes of structural neurodegeneration such as ectopic sprouting , wave-like processes , lesions and breaks are also decreased in the mcu-1; sel-12 ( ty11 ) animals compared with sel-12 ( ty11 ) mutants ( Figure 7F , Figure 7—figure supplement 1A–C ) . Moreover , the mcu-1; sel-12 ( ty11 ) animals show 78 . 8% response to light touch compared to 48 . 4% response of sel-12 ( ty11 ) animals ( Figure 7G ) . Taken together these data suggest that reducing mitochondrial Ca2+ uptake normalizes mitochondrial metabolic activity and suppresses the neurodegenerative phenotype observed in sel-12 ( ty11 ) animals . Next , to determine whether reducing mitochondrial Ca2+ uptake could alleviate the mitochondrial metabolic dysregulation observed in FAD fibroblasts ( Figure 1E–H ) , we treated control fibroblasts and fibroblasts isolated from FAD patients with Ruthenium360 ( Ru360 ) . Ru360 is a cell-permeable mitochondrial Ca2+ uniporter inhibitor . Consistent with the improvement in mitochondrial function observed in the mcu-1; sel-12 ( ty11 ) animals , treatment of FAD fibroblasts with Ru360 resulted in a decrease of both ATP and ROS levels in these cells ( Figure 7H , I ) . Previously , we found that reducing mitochondrial fission in sel-12 mutants could improve mitochondrial morphology ( Sarasija and Norman , 2015 ) . This was accomplished by knocking down the gene encoding DRP-1 by RNA interference ( RNAi ) . DRP-1 is a dynamin-1-like protein that regulates mitochondrial fission ( Labrousse et al . , 1999 ) . Thus , we sought to determine whether inhibiting mitochondrial fission could normalize mitochondrial activity of the sel-12 mutants and prevent neurodegeneration . We knocked down the expression of drp-1 in sel-12 mutants by drp-1 ( RNAi ) and analyzed mitochondrial structure , mitochondrial calcium levels , OXPHOS status , neuronal structure , and response to soft touch in these animals . While drp-1 ( RNAi ) rescues the mitochondrial structure in the sel-12 ( ty11 ) animals ( Figure 8—figure supplement 1A ) , it is unable to restore mitochondrial calcium levels , OCR , and neurodegeneration ( Figure 8—figure supplement 1B–F ) . This data suggests that defects in mitochondrial morphology may be a symptom but not the cause of the neurodegeneration observed in sel-12 mutants . Since we have observed high mitochondrial activity in young adult sel-12 mutants and that the reduction of ER Ca2+ release or mitochondrial Ca2+ uptake can reduce mitochondrial activity and improve the neurodegeneration observed in sel-12 mutants , we next sought to determine whether directly reducing mitochondrial activity could suppress the sel-12 neurodegenerative phenotype . To reduce mitochondrial activity , we used doxycycline to inhibit mitochondrial function ( Moullan et al . , 2015 ) . sel-12 ( ty11 ) animals grown in the presence of doxycycline showed inhibition of mitochondrial function as evidenced by the suppression of the high mitochondrial respiration rate observed in sel-12 mutants ( Figure 8A , B ) . Interestingly , doxycycline treatment also rescued the mechanosensory defects in sel-12 ( ty11 ) animals , both in touch response and axon morphology ( Figure 8C–E ) . Moreover , to test whether elevated mitochondrial activity causes the elevated ATP and ROS levels and OCR observed in FAD fibroblasts , we treated control and FAD fibroblasts with doxycycline and measured these outputs . Consistent with FAD cells showing high mitochondrial activity , doxycycline treatment reduced ATP and ROS levels as well as OCR in FAD fibroblasts ( Figure 8F–I ) . These data taken together with our observations of increased mitochondrial Ca2+ levels in sel-12 mutants and the ability of reducing Ca2+ transfer from the ER to the mitochondrial in suppressing the neurodegeneration associated with sel-12 mutants indicates that exacerbated ER Ca2+ transfer to the mitochondria boosts mitochondrial activity causing neurodegeneration in the sel-12 mutants . Thus far , our analyses have found that mutations in sel-12 result in deregulation of ER Ca2+ release and mitochondrial Ca2+ uptake , which leads to increased OXPHOS , and increased global ATP as well as ROS levels and results in neurodegeneration . To gain further insight into the cause of neurodegeneration in sel-12 mutants , we sought to determine the mechanism underlying the neurodegenerative phenotypes observed in sel-12 mutants . First , to investigate whether there is an elevation in mitochondrial ROS in the mechanosensory neurons undergoing neurodegeneration , we used a mitochondrial matrix targeted roGFP , a reduction-oxidation sensitive green fluorescent protein ( Cannon and Remington , 2008; Melentijevic et al . , 2017 ) and discovered that sel-12 ( ty11 ) animals have elevated mitochondrial ROS compared to wild type animals ( Figure 9A ) . Since OXPHOS leads to the production of superoxides , high levels of which can be toxic to cells and we observed elevated OXPHOS and mitochondrial ROS in sel-12 mutants , we hypothesized that mitochondrial superoxide production could be causing the neurodegeneration observed in sel-12 mutants . To investigate the impact of mitochondrial superoxide production on the mechanosensory defects observed in sel-12 mutants , we exposed sel-12 ( ty11 ) animals to a mitochondria-targeted superoxide-scavenger called MitoTEMPO or triphenylphosphonium ( TPP ) , which lacks the superoxide-scavenging moiety and acts as a control . Consistent with MitoTEMPO reducing mitochondrial ROS levels without altering mitochondrial Ca2+ levels , we found that sel-12 ( ty11 ) animals treated with 500 µM MitoTEMPO showed significant reduction in oxidized neuronal mitochondrial roGFP fluorescence levels ( Figure 9—figure supplement 1A ) while mitochondrial Ca2+ levels remained similar to control sel-12 ( ty11 ) animals ( Figure 9—figure supplement 1B ) . Next , we examined neuronal structure and behavioral response to light touch . sel-12 ( ty11 ) animals raised on MitoTEMPO show a decrease in the number of ectopic neuronal sprouts from two sprouts/ALM in control sel-12 ( ty11 ) animals to one sprouts/ALM ( Figure 9B ) while TPP treated animals showed no such improvement ( Figure 9B ) . Also , MitoTEMPO treatment resulted in a normal behavioral response to light touch at 83 . 1% similar to wild type animals at 82 . 6% and much improved compared to the control and TPP treated sel-12 ( ty11 ) animals ( Figure 9C ) . However , MitoTEMPO treatment does not rescue the mitochondrial morphology defects seen in sel-12 ( ty11 ) animals ( Figure 9—figure supplement 1C ) , suggesting that oxidative stress is not directly causing mitochondrial structural disorganization . Together our data suggest that deregulated Ca2+ transfer from the ER to the mitochondria increases OXPHOS resulting in superoxide generation triggering oxidative stress mediated neurodegeneration .
Mutations in the genes encoding PSEN1 and PSEN2 are the most frequent cause of FAD . Despite the identification of PSEN involvement in AD over 20 years ago , the mechanism causing neurodegeneration in FAD patients has remained elusive . In this study , we provide evidence that mutations in the gene encoding a C . elegans PSEN homolog , sel-12 result in ER to mitochondria Ca2+ signaling defects , elevated mitochondrial Ca2+ levels , mitochondrial structural disorganization , dysfunction of mitochondrial metabolism and subsequent oxidative stress mediated neurodegeneration . We also demonstrate that the structural and functional neurodegeneration observed in sel-12 mutants are independent of the gamma-secretase activity of SEL-12 and are a result of mitochondria generated superoxide that arise due to increased ER to mitochondria Ca2+ signaling resulting in amplified OXPHOS . Moreover , we show that skin fibroblasts isolated from FAD patients with mutations in PSEN1 show a similar mitochondrial phenotype as observed in C . elegans sel-12 mutants , demonstrating a conserved role of PSEN in mitochondrial regulation . While dysregulation in ER Ca2+ signaling ( Chan et al . , 2000; Cheung et al . , 2008; Green et al . , 2008; Leissring et al . , 1999; Stutzmann et al . , 2004 ) , mitochondrial dysfunction ( Area-Gomez et al . , 2012; Kipanyula et al . , 2012; Zampese et al . , 2011 ) , and deleterious effects of ROS have been observed separately in various in-vivo and in-vitro models of AD , here we provide evidence that link these observations and demonstrate the central role played by mitochondrial metabolism in AD pathogenesis using an intact animal model of AD where neurodegeneration arises due to altered ER to mitochondrial Ca2+ signaling , in the absence of Abeta peptides . In order to maintain neuronal function and health , optimal mitochondrial activity is critical to provide energy and concomitantly manage ROS production . Furthermore , mitochondria have a critical role in buffering Ca2+ and Ca2+ in turn modulates mitochondrial activity . Indeed , increased mitochondrial Ca2+ aids in the allosteric activation of several TCA cycle enzymes , including pyruvate dehydrogenase α-ketoglutarate dehydrogenase , and isocitrate dehydrogenase ( McCormack and Denton , 1993 ) , stimulates the ATP synthase ( complex V ) ( Das and Harris , 1990 ) , α-glycerophosphate dehydrogenase ( McCormack and Denton , 1993; Wernette et al . , 1981 ) and the adenine nucleotide translocase ( ANT ) ( McCormack and Denton , 1993; Mildaziene et al . , 1995 ) . Thus , the increase in mitochondrial Ca2+ concentration results in the coordinated upregulation of the TCA cycle and OXPHOS machinery , allowing for increased oxygen consumption , and ATP and ROS production . Here , we discover that mutations in sel-12 lead to deregulated ER and mitochondrial Ca2+ signaling resulting in increased mitochondrial Ca2+ levels , which promotes mitochondrial metabolic activity . Additionally , we provide evidence that this elevated metabolic activity elevates ROS levels in sel-12 mutants and the deleterious effects of ROS results in the emergence of structural and functional phenotypes of neurodegeneration . Upon reducing ER Ca2+ release or mitochondrial Ca2+ uptake , the mitochondria of sel-12 mutants are no longer disorganized and show normal OCR , and the high levels of ROS and neurodegenerative phenotypes observed in sel-12 mutants are suppressed . These data indicate that SEL-12 activity is required for normal ER to mitochondrial Ca2+ signaling , proper mitochondrial function and mitochondrial superoxide maintenance . Consistent with PSENs having a role in ER Ca2+ regulation , several studies in a variety of PSEN FAD models as well as tissue samples isolated from AD patients have shown that loss of PSEN function leads to perturbed ER Ca2+ signaling ( Chan et al . , 2000; Green et al . , 2008; Ito et al . , 1994; Leissring et al . , 1999; Smith et al . , 2005; Tu et al . , 2006 ) . Furthermore , PSENs are predominantly ER membrane proteins that are enriched in the mitochondria-associated ER membrane ( Area-Gomez et al . , 2009 ) . The mitochondria-associated ER membrane is a highly active subcompartment of the ER that is physically connected to the mitochondria and is critical for several metabolic functions , such as Ca2+ homeostasis , cholesterol metabolism and the synthesis and transfer of phospholipids between the ER and mitochondria . Importantly , elevated ER to mitochondrial contact and crosstalk has been observed in PSEN1 and PSEN2 knockout cells , in cells expressing FAD mutant PSEN2 , and in skin fibroblasts from FAD and sporadic AD patients ( Area-Gomez et al . , 2012; Filadi et al . , 2016; Kipanyula et al . , 2012; Zampese et al . , 2011 ) . Although the functional importance of these observations has not been resolved , our analyses of skin fibroblasts isolated from FAD patients , indicates that there is an up regulation of OXPHOS , ATP production and ROS generation that can be reduced by inhibiting mitochondrial Ca2+ uptake , similar to what we observe in sel-12 mutants . We also demonstrate that in sel-12 mutants neurodegeneration is caused by elevated ER to mitochondrial transfer of Ca2+ . Taken together , these data present an inclusive model that indicates that PSEN function on the ER membrane is required for normal ER Ca2+ transfer to the mitochondria and in the absence of optimal PSEN function , excessive Ca2+ is transferred to the mitochondria leading to mitochondrial dysfunction and an elevation of mitochondrial generated superoxide radicals that promote neurodegeneration ( Figure 9D ) . However , recent analyses of PSEN1/2 knockout mouse embryonic fibroblasts ( MEFs ) , unlike our studies , have shown reduced mitochondrial activity suggesting PSEN function is vital for mitochondrial function ( Contino et al . , 2017; Pera et al . , 2017 ) . Significantly , one of these studies demonstrated that the reduction of mitochondrial function in PSEN1/2 KO MEFs could be rescued by PSEN2 and not PSEN1 , implicating a role of PSEN2 in mitochondrial function in MEFs ( Contino et al . , 2017 ) . While the reduced mitochondrial activity might appear to be at odds with our study , a recent analysis of astrocytes derived from iPSCs isolated from skin cells from patients with PSEN1 mutations also showed elevated OCR and ROS production , which could be rescued by CRISPR/Cas9 repair of the PSEN1 mutation ( Oksanen et al . , 2017 ) . Thus , illustrating an important role of PSEN1 in the regulation of mitochondrial activity in astrocytes similar to SEL-12 in C . elegans . Also of note , we did observe in day eight vs . day one adult sel-12 mutants a drastic reduction of mitochondrial activity in comparison to aged matched wild type animals indicating that in aged animals mitochondria function becomes impaired in sel-12 mutants , consistent with the decreased mitochondrial function as a result of PSEN mutations ( Figure 1—figure supplement 1C , D ) . The impairment of mitochondrial function with age in sel-12 ( ty11 ) animals could be a result of increased mitochondrial activity resulting in ROS-mediated oxidative damage at earlier stages of adulthood . The best-studied function of PSEN is its role as the proteolytic component of the gamma-secretase complex and its cleavage of APP and Notch . Gamma-secretase mediated cleavage is required for the activation of Notch signaling and emphasizes the critical role of PSEN in mediating Notch function ( De Strooper et al . , 1997 ) . Also , it is the gamma-secretase mediated cleavage of APP that leads to the generation of Abeta peptides of various lengths , including the toxic Abeta42 ( Bezprozvanny and Mattson , 2008; Hardy , 2006 ) . Here , we provide genetic and pharmacological evidence that gamma-secretase activity of SEL-12 is not required for neurodegenerative phenotypes observed in sel-12 mutants . While we are not able to rescue Notch signaling defects using a sel-12 protease dead genetic construct , we are able to rescue the mechanosensory behavioral defects observed in sel-12 null mutants ( Figure 4D , Figure 4—figure supplement 1A-D ) . Furthermore , using a gamma-secretase inhibitor , which abolished Notch signaling , we do not recapitulate the neurodegenerative phenotypes observed in sel-12 mutants ( Figure 4B , C ) . Consistently , our analyses of Notch mutants also did not reveal mechanosensory defects ( Figure 4A ) . These results demonstrate that the role of SEL-12 in the regulation of mitochondrial function and nervous system fitness does not require gamma-secretase activity or Notch signaling . In the last two decades , research into AD therapeutics has relied heavily on the amyloid hypothesis , which holds the toxic Abeta peptides and their aggregation to form plaques responsible for AD pathogenesis . However , the exact role Abeta peptides have in AD is not clear , which is further emphasized by the failure of phase III trials of anti-Abeta therapies like the gamma-secretase inhibitor , semagacestat ( Doody et al . , 2013 ) and solanezumab , a monoclonal antibody targeting amyloid plaques ( Le Couteur et al . , 2016 ) . This suggests the need to shift focus from Abeta peptides to more proximal causes of AD . Therefore , C . elegans provides a unique system for investigating the mechanisms underlying AD . Besides the usual strengths ( e . g . rapid growth cycle , genetic manipulation , cell biological approaches , simple nervous system ) , C . elegans encodes a single APP-like protein , apl-1 . Unlike APP , APL-1 , like its APL-1 mammalian homologs ( APLP1 and ALPL2 ) , lack the Abeta peptide sequence . Furthermore , APL-1 lacks the beta-secretase recognition sequences and the C . elegans genome does not encode a beta-secretase; therefore , it is unlikely that C . elegans produce Abeta peptides ( Daigle and Li , 1993; McColl et al . , 2012 ) and indeed , no Abeta peptides have been detected in C . elegans ( McColl et al . , 2012 ) . Thus , C . elegans provides a model system that can explore the mechanism underlying AD without the confounding role of Abeta . Excitingly , utilizing this experimental system , we find that in the absence of Abeta , mutations in PSEN result in deregulated ER to mitochondria Ca2+ signaling that causes elevated mitochondrial Ca2+ which hyperactivates mitochondrial respiration leading to the accumulation of superoxides and the subsequent reduction of neuronal cell function and health . In spite of our findings indicating that the role of SEL-12 in neurodegeneration is independent of Abeta accumulation , due to the fact that worms do not express Abeta , some toxic effects of Abeta accumulation have been demonstrated in C . elegans . Overexpression of human Abeta1-42 results in chemosensory defects , locomotor defects ( Ahmad and Ebert , 2017; Dosanjh et al . , 2010; Fong et al . , 2016; Wu et al . , 2006 ) and touch defects ( Figure 5A ) . Moreover , similar to the drastic reduction of OXPHOS we observe in older adult sel-12 mutants compared to age matched wild type animals ( Figure 1—figure supplement 1C , D ) , recent studies have shown that overexpression of human Abeta1-42 in the nervous system or body wall muscle results in the reduction of OXPHOS ( Fong et al . , 2016; Sorrentino et al . , 2017 ) . However , there are some important distinctions between overexpression of human Abeta1-42 and mutations in sel-12 . Although overexpression of human Abeta1-42 in the nervous system disrupts touch response similar to sel-12 mutants ( Figure 5A ) , it fails to result in hallmark sel-12 mutation defects such as axonal morphology phenotypes , disorganized mitochondria , and excessive calcium loading ( Figure 5B–E ) . Therefore , the mechanism of Abeta1-42 overexpression appears distinct from presenilin mutations . These important findings suggest that an initial insult in FAD , since C . elegans does not produce Abeta peptides , is elevated ROS production from the mitochondria due to an increase in ER Ca2+ transfer . Thus , Abeta accumulation may be a secondary , albeit critical , component to the pathogenesis of the disease .
C . elegans are grown on NGM plates between 16–25°C . For all experiments , animals are synchronized by bleaching plates containing gravid animals and the progeny are allowed to hatch in M9 buffer . The synchronized L1 larvae are then grown on NGM plates seeded with OP50 at 20°C , unless otherwise noted , until the necessary stage of growth for each experiment . glp-1 ( e2141 ) and their wild type counterparts were raised for experiments at 25°C from L1 stage . For experiments that required sterilization , age synchronized L4 animals are moved to 0 . 5 mg/ml 5-fluorour-aci1-2'-deoxyribose ( FUDR ) containing agar plates seeded with OP50 . The following strains were used in this study: Wild type ( N2 ) , sel-12 ( ar131 , ty11 and ok2078 ) X , hop-1 ( ar179 ) I , crt-1 ( jh101 ) V , mcu-1 ( tm6026 ) IV , glp-1 ( e2141 ) III , lin-12 ( ok2215 ) III , unc-68 ( r1162 ) V , CL2355 smg-1 ( cc546 ) dvIs50 [unc-119p::Abeta1-42] I , zdIs5 [mec-4p::GFP] I , uIs69 [unc-119p::sid-1+myo-2p::mCherry] V , jsIs609 [mec-4p::MLS::GFP] , zcIs14 [myo-3p::tomm-20::GFP] , twnEx8 ( mec-7p::tomm-20::mCherry , myo-2p::gfp ) , zhsEx17 [Pmec-4mitoLS::ROGFP] , takEx214 ( myo3p::sel-12::SL2::mCherry; myo2p::mCherry ) , takEx222 ( rab3p::sel-12::SL2::mCherry; ttx3p::GFP ) . Cell culture media and reagents were purchased from Invitrogen ( Waltham , MA ) and Corning ( Cellgro , Manassas , VA ) . FAD ( AG07872 , AG07768 , AG06848 , AG08170 ) and normal control ( AG08379 , AG07871 , AG08701 , AG08509 ) human skin fibroblasts cell lines were obtained from NIA Aging Cell Culture Repository ( Coriell , Camden , NJ ) . Cell lines were authenticated by AmpFLSTR Indentifiler Plus PCR Amplification Kit ( ThermoFisher Scientific ) and tested for absence of mycoplasma by MycoSEQ Mycoplasm Detection System ( Life Technologies ) by Coriell Cell Repositioires . All cell lines were grown at 37°C under humidified air containing 5% CO2 . Cells were grown in DMEM . DMEM medium ( 4 . 5 mg/L glucose , 110 mg/L sodium pyruvate ) was supplemented with penicillin ( 100 U/ml ) , streptomycin ( 100 μg/ml ) , and 15% fetal bovine serum . Cell culture protocol followed in accordance with those provided by the cell supplier . The sel-12 genomic fragment was subcloned from F35H12 cosmid ( Sanger Center ) into pBluescript KS+ using SacI and XhoI . The D226A mutation was generated by site directed mutagenesis using the following primers: 5’ GCTGTTTGTTATCTCGGTTTGGGcTCTGGTTGCCGTGCTCACACC 3’ and 5’ GGTGTGAGCACGGCAACCAGAgCCCAAACCGAGATAACAAACAGC 3’ . The final product was confirmed by DNA sequencing . The myo-3p::mito-GCaMP6f::SL2::mCherry construct was made by PCR amplification of myo-3p::mito-GCaMP6f from pKN#18 , which contains myo-3p::mito-GCaMP6f and was gene synthesized by Genscript , and SL2::mCherry::unc-54 3’ UTR from pKN#7 , which contains SL2::mCherry::unc-54 3’ UTR and was gene synthesized by Genscript . These products were combined together using Gibson cloning . The mitochondrial matrix targeting sequence was obtained from the N-terminal cytochrome C oxidase subunit VII ( Akerboom et al . , 2013 ) . The final product was confirmed by DNA sequencing . To generate transgenic animals , Qiagen Midi prepared DNA constructs were injected into N2 animals following standard procedures ( Mello and Fire , 1995 ) . Transgenic animals were selected using ttx-3p::GFP as a marker and appropriate subcellular localization was confirmed by fluorescence microscopy . The basal mitochondrial Ca2+ concentration in the body wall muscles and mechanosensory neurons was measured using takEx347 and takEx415 animals , respectively expressing myo-3p::mito-GCaMP6f::SL2::mCherry and mec-7p::mito-GCaMP6f::SL2::mCherry . The fluorescence intensity of both the GCaMP6f ( genetically encoded Ca2+ indicator ) and mCherry ( used here as a expression control ) were measured using a 10X objective lens on a Zeiss Axio Observer microscope and Andor Clara CCD camera , as previously described ( Sarasija and Norman , 2015 ) . The GCaMP6 fluorescence intensity in each animal was normalized to mCherry intensity . Animals at day 1 , 2 and 3 of adulthood were evaluated for the ability to respond to light touch using an eyebrow hair glued to the end of a Pasteur pipette tip ( Chalfie and Sulston , 1981 ) . Animals were scored based on their responsiveness to a total of 10 touches; five to the anterior ( between the head and vulva ) and five touches to the posterior ( between the tail and vulva ) . If the animal moves forwards in response to a light touch in the posterior or backward after an anterior touch , respectively , then the animal receives a score of 1 for a maximum score of 10 ( 100% response ) or a minimum score of 0 ( 0% response ) . zdIs5[Pmec-4::GFP] and was introduced into various genotypes and was utilized to determine ALM and PLM neuronal morphologies ( Wu et al . , 2007 ) . Age synchronized animals are immobilized on 3% agarose pads using 0 . 1 µm diameter polystyrene microspheres ( Polysciences ) and their ALM and PLM neuronal structure is imaged under the 63X oil objective on a Zeiss Axio Observer microscope equipped with a Andor Clara CCD camera . Images were compiled using Metamorph software and scored positive for ectopic neurite sprouting when a visible GFP-labeled branch is seen stemming from the ALM , presence of wave-like bending , presence of triangular beaded lesions and breaks in the ALM or PLM axons . Mitochondrial structure is analyzed using the jsIs609 [mec-4p::MLS::GFP] , which targets GFP to the mitochondria in the mechanosensory neurons ( Fatouros et al . , 2012 ) or twnEx8 ( mec-7p::tomm-20::mCherry ) , which targets mCherry to the outer mitochondrial membrane protein via a fusion with the outer mitochondrial membrane protein , TOMM-20 ( Hsu et al . , 2014 ) . Mitochondrial structures were imaged following the protocol used for neuronal imaging ( Sarasija and Norman , 2018a ) . The mitochondria were scored as continuous if they existed as an intact circle , or discontinuous when there were breaks to this circle , under blind conditions . The surface of unseeded NGM plates are coated with 100 μl of 100 μM Compound E ( Calbiochem ) , a gamma-secretase inhibitor and seeded with OP50 the next day . Synchronized L1 staged animals are grown on these plates until day one adults , which are then used for analysis ( Francis et al . , 2002 ) . Gamma-secretase inhibition is confirmed by the glp-1-like sterility observed in the hop-1 ( ar179 ) mutants that are grown on the Compound E containing plates . Animals are moved to NGM plates containing 500 μM ( 2- ( 2 , 2 , 6 , 6-tetramethylpiperidin-1-oxyl-4-ylamino ) −2-oxoethyl ) triphenylphosphonium chloride ( mitoTEMPO ) ( Sigma ) or 500 μM triphenylphosphonium chloride ( TPP ) ( Sigma ) as L1 larvae . These animals are then used for touch assay , neuronal sprouting and mitochondrial structure analysis as Day one adults . NGM plates containing 10 µg/ml of doxycycline ( Sigma ) were seeded with the E . coli strain HT115 . Gravid adult animals from each strain were bleached and their progeny that were moved onto these plates as eggs were used for experiments as day one adults . Fibroblasts were treated with either 1 μg/ml doxycycline ( Sigma ) or 10 μM Ru360 ( Calbiochem ) for 48 hr . To ensure efficient knockdown in the nervous system we used strains expressing sid-1 pan-neuronally ( Calixto et al . , 2010 ) . These animals are then age synchronized to larval stage one and then allowed to grow on plates seeded with RNAi bacteria ( HT115 ) carrying either the empty vector or drp-1 RNAi ( Source BioScience ) ( Sarasija and Norman , 2015 ) . These animals were then used for experiments as young adults . ROS measurements were carried out as previously described ( Sarasija and Norman , 2018b ) . In Brief , animals grown on 100 mm NGM plates are washed three times with M9 and gravity separated to remove OP50 . Then , the animals are washed once in PBS and resuspended in 100 μl of PBS followed by freeze thawing and 10 s sonication . After spinning down the sample at 20 , 000 g for 10 min at 4°C , the protein concentration is determined using a BCA protein assay and equal amounts of proteins extract are used for each strain and the samples are incubated with H2DCF-DA assay ( Life Technologies ) at 37°C for 24 hr . and a Flex Station 3 Reader ( Molecular Devices ) is used to measure light intensity . Each sample is measured in duplicates or triplicates and the experiments were repeated three times ( three biological replicates ) . FAD ( AG07872 , AG07768 , AG06848 , AG08170 ) and normal control ( AG08379 , AG07871 , AG08701 , AG08509 ) human skin fibroblasts were seeded at a density of 1 . 0 × 105 cells/well in a 96 well black-walled plate with clear bottoms ( Costar , Corning , NY ) overnight . Prior to the measurement , media were removed and cells were washed with PBS . Cells were incubated in dark at 37°C with 10 μM H2DCF-DA solution in PBS per well plates was added for 30 mins . After the incubation period , solution of H2DCF-DA was removed and cells were suspended in PBS . The fluorescence was measured by Flexstation three microplate reader ( ex/em = 485/535 nm ) . Each sample is measured in triplicates and the experiments were repeated three times ( three biological replicates ) . Animals grown on 100 mm NGM plates are washed three times with M9 and gravity separated to remove OP50 . Then , the animals are washed once in TE buffer ( 100 mM Tris-Cl [pH7 . 5] , 4 mM EDTA ) and resuspended in 100 μl of TE buffer followed by freeze thawing , 10 s sonication and boiling for 10 min . After spinning down the sample at 20 , 000 g for 10 min at 4°C , the protein concentration is determined using a BCA protein assay and ATP concentration is determined as per manufacturer’s instruction using the ENLITEN ATP Assay System ( Promega , Madison , WI ) . Each sample is measured in quadruplicates and the experiments were repeated three times ( three biological replicates ) . ATP release in FAD ( AG07872 , AG07768 , AG06848 , AG08170 ) and normal control ( AG08379 , AG07871 , AG08701 , AG08509 ) human skin fibroblasts was determined using the Enliten ATP assay system ( Promega ) as described by the manufacturer . The luminescence was measured by integration over a 3 s time interval using the luminometer Flexstation three microplate reader , and normalized to protein content determined by BCA . Each sample is measured in triplicates and the experiments were repeated at least twice ( two biological replicates ) . Oxygen consumption rate ( OCR , indicative of mitochondrial OXPHOS ) in day one sterilized animals was measured using the Seahorse XFp Extracellular Flux Analyzer ( Agilent Seahorse Technologies ) adapting the protocol used to measure OCR in C . elegans using a Seahorse XF96 Extracellular Flux Analyzer ( Koopman et al . , 2016 ) . Each assay run compared the OCR between two genotypes , in three wells each and five measurements were made for each well for basal OCR and maximal OCR upon addition of FCCP . Assays were repeated on three separate days and three biological replicates were analyzed . Mitochondrial OXPHOS in human fibroblasts ( AG08379 and AG06848 ) was analyzed using a Seahorse XFp Extracellular Flux Analyzer ( Agilent Seahorse Technologies ) by measuring the OCR in real time . For OCR analysis , cells were seeded in 8-well plates designed for XFp at 25 , 000 cells per well in complete growth media . On the next day , the cells were switched to unbuffered media ( supplemented with 2 . 5 mM glucose , 1 mM pyruvate , and 1 mM glutamine ) and further incubated in a CO2 -free incubator for 1 hr prior to measurement . During measurement , oligomycin ( 1 µM ) , FCCP ( 1 µM ) , antimycin A and rotenone ( 0 . 5 µM ) were added . At the end of each assay , protein quantification was performed for normalization . All experiments were performed in triplicate and repeated three times . Age synchronized young adult animals carrying zhs17 [Pmec-4mitoLS::ROGFP] with or without treatment ( MitoTEMPO or TPP ) were immobilized on 5% agarose pad using 2 . 4 mg/ml solution of levamisole . They were then imaged on an Olympus Fluoview 1200 Laser Scanning Confocal Microscope with a 60X oil immersion objective lens . Samples were first excited with a 10% power 405 nm laser and then a 10% power 488 nm laser with sequential scanning method and a 0 . 5 um step size and GFP emission was detected . Images were quantified using ImageJ and the ratio of 405 nm channel was divided by the 488 nm channel . Animals were washed off with M9 and gravity separated to remove OP50 . This is followed by three washes with dH20 and animals were resuspended in a small volume of RIPA buffer with protease inhibitor cocktail ( Roche Diagnostic ) . Animals were flash frozen at −80C , sonicated and spun down to obtain animal pellet and lysate . The lysates were used to determine protein concentration . The worm pellets were boiled in SDS loading buffer containing 2-beta mercaptoethanol and 60 μg of protein ( determined from the protein concentration of the lysate ) was loaded in each well of a 12% acrylamide SDS-Page gel . The proteins were transferred to 0 . 2 μm nitrocellulose membrane and stained with ponceau to ensure equal loading . The nitrocellulose blots were then boiled in PBS , washed in TBS-T and blocked with 5% milk in TBS-T . The blots were incubated overnight in primary antibody against Abeta ( 6E10 , Biolegend; 1:1000 ) and then with secondary antibody ( anti-mouse IgG ( whole molecule ) -peroxidase , Sigma-Aldrich , 1:20 , 000 ) , after washes with TBS-T . The proteins were visualized using Clarity TM Western ECL Substrate ( BIO-RAD ) and X-ray film . GraphPad Prism software is used for statistical analysis . Student’s t test is used only for comparing two samples and one-way ANOVA with Tukey post test has been used when making multiple comparisons . Chi-square test was used to compare the difference in mitochondrial structural phenotypes between strains . | Alzheimer's disease is the most common type of dementia . A hallmark of this condition is progressive loss of memory , accompanied by a buildup of hard clumps of protein between the brain cells . These protein clumps , known as amyloid plaques , are a key focus of research into Alzheimer's disease . They are likely to be toxic to brain cells , but their role in the development and progression of the disease is not yet known . Though the cause of Alzheimer's disease remains unclear , an inherited form of the disease may hold some clues . Mutations in genes for proteins called presenilins cause an earlier onset form of Alzheimer's disease , in which symptoms can develop in people who are in their 40s or 50s . The presenilin proteins appear in a cell structure called the endoplasmic reticulum , which plays many roles in the normal activities of a cell . Among other things , this structure stores and releases calcium ions , and cells use these ions to send and process many signals . The cell's energy-producing powerhouses , the mitochondria , use calcium to boost their metabolic activity . This allows them to make more energy for the cell , but in the process they also make damaging byproducts . These byproducts include oxygen-containing chemicals , known as reactive oxygen species ( ROS ) , which react strongly with other molecules . While low levels of ROS are a normal part of cell activity , if the levels get too high , these chemicals can attack and damage structures within the cell . Untangling the effects of amyloid plaques and presenilins on brain cells in humans is challenging . But , a nematode worm called Caenorhabditis elegans does not form plaques , making it possible to look at presenilins on their own . Previous work in these worms has shown that presenilin mutations affect the endoplasmic reticulum and change the appearance of mitochondria . Here , Sarasija et al . extend this work to find out more about the effects presenilin mutations have on living cells . Presenilin mutations in young adult worms increased the amount of calcium released by the endoplasmic reticulum . This increased the activity of the mitochondria and caused ROS levels to rise to damaging levels . This caused stress inside the cells , and the worms started to show early signs damage to their nervous systems . Mutations that decreased the movement of calcium from the endoplasmic reticulum to the mitochondria helped to prevent the damage . Treating the mitochondria with antioxidants to mop up the extra ROS also protected the cells . This kind of damage to brain cells did not depend on amyloid plaques . Whilst the plaques are likely to be toxic , these new findings highlights the role that other chemical and biological processes might play in Alzheimer's disease . Further work to reveal the underlying cause of Alzheimer's disease may lead to new therapies to treat this condition in the future . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2018 | Presenilin mutations deregulate mitochondrial Ca2+ homeostasis and metabolic activity causing neurodegeneration in Caenorhabditis elegans |
The spliceosome must identify the correct splice sites ( SS ) and branchsite ( BS ) used during splicing . E complex is the earliest spliceosome precursor in which the 5' SS and BS are defined . Definition occurs by U1 small nuclear ribonucleoprotein ( snRNP ) binding the 5' SS and recognition of the BS by the E complex protein ( ECP ) branchpoint bridging protein ( BBP ) . We have used single molecule fluorescence to study Saccharomyces cerevisiae U1 and BBP interactions with RNAs . E complex is dynamic and permits frequent redefinition of the 5' SS and BS . BBP influences U1 binding at the 5' SS by promoting long-lived complex formation . ECPs facilitate U1 association with RNAs with weak 5' SS and prevent U1 accumulation on RNAs containing hyperstabilized 5' SS . The data reveal a mechanism for how U1 binds the 5' SS and suggest that E complex harnesses this mechanism to stimulate recruitment and retention of U1 on introns .
Intron removal by the spliceosome is an essential step in precursor messenger RNA ( pre-mRNA ) processing during eukaryotic gene expression . The spliceosome is composed of a number of subunits including the five small nuclear ribonucleoproteins ( the U1 , U2 , U4 , U5 , and U6 snRNPs ) , each containing a U-rich small nuclear RNA ( snRNA ) and several snRNP-specific proteins ( Wahl et al . , 2009 ) . Spliceosomes assemble stepwise on pre-mRNAs via a partially ordered pathway to correctly identify the boundaries of the intron and carry out the splicing reaction ( Hoskins et al . , 2011; Shcherbakova et al . , 2013; Wahl et al . , 2009 ) . Assembly often begins with ATP-independent binding of the U1 snRNP to the 5' splice site ( 5' SS ) to form the spliceosome E complex ( Michaud and Reed , 1991; Mount et al . , 1983; Séraphin et al . , 1988; Zhuang and Weiner , 1986 ) . Along with U1 , E complex also contains a number of other splicing factors ( E complex proteins , ECPs ) associated with other locations on the transcript ( Figure 1A ) . The nuclear cap binding complex ( CBC ) is bound to the 5' cap of the pre-mRNA ( Colot et al . , 1996 ) while additional ECPs associate with the intron branchsite ( BS; BBP/Mud2 in the yeast or SF1/U2AF1 and 2 in humans ) ( Abovich and Rosbash , 1997; Abovich et al . , 1994; Bennett et al . , 1992; Berglund et al . , 1997; Zamore et al . , 1992; Zhang and Rosbash , 1999 ) . E complex is the earliest spliceosome assembly intermediate conserved between yeast and humans that also marks the approximate locations for both 5' SS cleavage and the branchpoint adenosine responsible for carrying out the cleavage reaction . Ample evidence has shown that E complexes are competent intermediates for subsequent spliceosome assembly ( Seraphin and Rosbash , 1989; Séraphin et al . , 1988; Michaud and Reed , 1991 ) . In yeast , the E complex has been referred to as the commitment complex ( CC ) since at least a subset of the complexes being formed are resistant to addition of competitor RNA and can be chased into functional spliceosomes ( Seraphin and Rosbash , 1989; Séraphin et al . , 1988 ) . However , both native gel analysis of CC formation and recent single molecule analysis of U1 binding dynamics showed that many U1/pre-mRNA interactions are short-lived ( Ruby , 1997; Hoskins et al . , 2011 ) . Since these experiments did not assay ECPs , it is unknown if these transient U1 binding events originated from the lack of simultaneous binding of both U1 and ECPs or if E complex itself is dynamic . Within E complex , U1 recognizes the 5' SS by base pairing between the 5' end of the U1 snRNA and the transcript ( Mount et al . , 1983; Siliciano and Guthrie , 1988; Zhuang and Weiner , 1986 ) . Despite the high conservation of this region of the U1 snRNA , 5' SS can vary greatly in sequence . In yeast , most pre-mRNAs contain strong 5' SS that match the consensus , 5'-GUAUGU-3' ( Grate and Ares , 2002 ) . However , there are several non-consensus 5ʹ SS that are also found in yeast introns and are predicted to make fewer base pairing interactions with U1 ( Grate and Ares , 2002; Qin et al . , 2016 ) . These introns may be less efficiently spliced and often function as ‘weak’ 5' SS signals for U1 . In humans , 5' SS are often weak and >9000 sequence variants have been reported ( Roca and Krainer , 2009 ) . How U1 uses a single snRNA to facilitate spliceosome assembly at many different 5' SS of varying strengths is not yet clear . Genetic evidence from yeast indicates that ECPs may aid U1 in 5' SS recognition . ECPs are essential for the splicing of particular transcripts ( Hossain et al . , 2009; Qin et al . , 2016 ) or in the presence of certain U1 protein or snRNA mutations ( Schwer and Shuman , 2014; Chang et al . , 2010; Liao et al . , 1993; Schwer et al . , 2013 ) . The genetic data indicate an apparent redundancy of ECPs in their ability to compensate for U1 mutations and loss of multiple interactions is often necessary to impair splicing ( Chang et al . , 2012; Schwer and Shuman , 2014; Schwer et al . , 2013 ) . Many genetic interactions between U1 and ECPs converge on both the 5' end of the snRNA and the Yhc1 protein ( U1-C in humans ) ( Schwer and Shuman , 2014; Schwer et al . , 2013 ) . Crystal structures of the human U1 snRNP have revealed that U1-C abuts a portion of the snRNA/5' SS duplex , and biochemical data show that U1-C can modulate the stability of duplexes formed with weak 5ʹ SS ( Kondo et al . , 2015; Pomeranz Krummel et al . , 2009 ) . Moreover , mutations in yeast Yhc1 can either bypass the requirement for the Prp28 ATPase in U1/U6 exchange during activation or exacerbate the phenotypes of defective Prp28 mutants ( Schwer and Shuman , 2014; Chen et al . , 2001 ) . Recently , Prp28 has also been shown to play an ATP-independent role in E complex formation ( Price et al . , 2014 ) . Together these data imply that the Yhc1/U1-C protein plays a critical role in U1’s interactions with 5ʹ SS and ECPs may modulate this function . Despite these genetic and biochemical data , key mechanistic questions of how U1 interacts with RNA remain unanswered . For example , it is not clear how snRNP proteins facilitate base pairing between the snRNA and intron or how non-U1 splicing factors strengthen or weaken U1 binding to regulate 5' SS selection . To understand the steps involved in 5' SS recognition , it is essential to characterize the interaction kinetics between U1 and pre-mRNAs and how these dynamics are influenced by both cis- and trans-acting factors . We have used colocalization single molecule spectroscopy ( CoSMoS ) to observe U1 interactions with RNAs containing or lacking a consensus 5' SS as well as to monitor E complex formation between fluorescently-labeled U1 , BBP , and pre-mRNAs in yeast whole cell extract ( WCE ) . Our data reveal how dynamic E complex formation tunes U1 interactions at the 5' SS and define a pathway for efficient recruitment and retention of U1 on intron-containing transcripts .
To determine how U1 binding kinetics are influenced by splice site strength , we carried out CoSMoS experiments to measure the dwell times of individual , fluorescent U1 molecules on surface-immobilized RNA substrates in WCE ( Figure 1B ) . In these experiments , endogenous U1 snRNPs were fluorescently labeled on two protein components harboring C-terminal SNAPf tags ( Hoskins et al . , 2011 ) . U1 binding was observed by colocalization of fluorescence from U1 molecules with surface immobilized RNAs ( Hoskins et al . , 2011 ) . We used the well-characterized , model pre-mRNA substrate RP51A ( Pikielny and Rosbash , 1986 ) in these experiments and varied its consensus 5' SS sequence by reducing or increasing the number of potential base pairs with the U1 snRNA . The resulting set of RNAs ( Figure 1C and Supplementary file 1 ) were 5' capped and identical in sequence with the exception of the 5ʹ SS which was either non-functional ( RNA 1 ) , weak ( RNA 2 ) , the yeast consensus sequence ( RNA 3 ) , or hyperstabilized to be completely complementary to the 5' end of the U1 snRNA ( RNA 4 ) . It has previously been shown that U1 associates with immobilized RNAs containing a 5' SS and that removal of a functional 5' SS results in decreased surface accumulation ( Hoskins et al . , 2011 ) . However , the lifetimes of U1 interactions on these different RNAs were not compared with one another . We carried out this kinetic analysis under conditions in which the WCE was depleted of ATP by addition of glucose , which permits U1 binding and E complex formation but not subsequent steps in spliceosome assembly ( Seraphin and Rosbash , 1989 ) . We observed frequent , transient interactions between U1 and RNAs either containing or lacking a functional 5' SS ( Figure 1D , Figure 1—figure supplement 1 ) . We compared these interactions using rastergrams that simultaneously depict U1 binding events occurring on multiple RNAs . In agreement with previous results , it was apparent that U1 interacted only briefly with RNAs that lacked a 5' SS ( Figure 1E ) but stably associated with RNAs containing a consensus 5' SS ( Figure 1F ) . On both RNAs the distribution of observed dwell times was best fit using a double exponential function containing short ( τ1 with amplitude A1 ) and long ( τ2 with amplitude A2 ) kinetic parameters ( Figure 1G–J and Supplementary file 2 ) . As expected from the rastergrams , removal of the functional 5' SS resulted in a large decrease in both τ2 and its amplitude , A2 , and an accompanying increase in A1 ( Figure 1I , J and Supplementary file 2 ) . However , we were surprised to observe that the overall number of binding events showed only a modest decrease on RNAs lacking a 5' SS ( Figure 1D ) , and little change was observed in the short kinetic parameter ( τ1 ) ( Figure 1H ) . Background binding resulting from fleeting association of fluorescent U1 molecules with the slide surface was rarely seen and could not account for the number of events that were observed ( Figure 1D ) . U1 frequently colocalizes with immobilized RNAs even in the absence of a functional 5' SS . We wondered if use of a weak 5' SS would result in U1 binding kinetics more similar to those observed with the consensus sequence or those seen when the RNA lacked a 5' SS altogether . We tested this using RNA 2 which contains the 5' SS normally found in the yeast SUS1 pre-mRNA ( Figure 1C ) . This 5' SS can form fewer potential base pairs with U1 relative to the consensus ( 4 vs . 6 base pairs , respectively ) but permits splicing in vitro when placed into RP51A ( Figure 1—figure supplement 2 ) . The results were similar to those observed in the absence of a functional 5' SS with short-lived interactions occurring frequently ( Figure 1D ) and predominating the distribution . Long-lived events were rare , resulting in an A2 value nearly identical to that observed in the absence of a 5' SS and much lower than observed with the consensus ( Figure 1J ) . Ablation of the 5' end of the U1 snRNA with RNase H also resulted in nearly complete loss of the long-lived binding events and frequent short-lived interactions ( Figure 1—figure supplement 3 ) . These data suggest that both binding lifetime as well as the relative abundances of short- and long-lived binding events are influenced by 5' SS strength and the ability to pair with the snRNA . If eliminating or weakening the 5' SS reduces the longest-lived U1 events , we predicted that strengthening the 5' SS should do the opposite . We analyzed U1 interactions with RNAs containing a hyperstabilized 5' SS , capable of making up to 10 potential base pairs with the snRNA ( Figure 1C , RNA 4 ) . In this case , we frequently observed both short- and long-lived interactions ( Figure 1—figure supplement 1D ) . Unexpectedly , the parameters obtained from a fit of the dwell time distribution revealed only a slight increase in τ2 while its amplitude ( A2 ) and τ1 remained close to what were observed with the consensus 5' SS ( Figure 1H–J ) . Long-lived U1 binding events seem to be much more sensitive to a loss in base pairing than an increase , while short-lived events are not solely dependent on the ability to base pair with a strong 5ʹ SS . Do the short-lived U1 binding events represent specific or non-specific interactions between the snRNP and RNA ? It has previously been proposed that ECPs facilitate U1 recognition of splice sites , particularly those with limited complementarity to the snRNA ( Qiu et al . , 2012; Hossain et al . , 2009 ) , as well as splicing-independent occupancy of U1 on transcripts ( Görnemann et al . , 2005; Patel et al . , 2007; Spiluttini et al . , 2010 ) . It is possible that momentary U1 binding results from interactions between U1 and ECPs bound to the RNA . We tested the role of ECPs in U1 recruitment by mitigating the influence of both the CBC and BBP . Previous work has shown that the impact of the CBC on splicing in vitro can be inhibited by addition of cap analog dinucleotide ( CA ) , which acts as a competitive inhibitor of capped RNAs for binding the CBC ( Edery and Sonenberg , 1985; Konarska et al . , 1984; Lin et al . , 1985 ) . Therefore , we carried out single molecule experiments either in the absence or presence of CA in order to remove the influence of the CBC ( Figure 2A ) . We hypothesized that the influence of BBP could be alleviated by mutation of the BS sequence in the transcript to prevent stable BBP association . We tested this using a CoSMoS assay for BBP binding to immobilized RNAs ( Figure 2—figure supplement 1 ) . On WT RP51A ( RNA 3 ) , we observed both short- and long-lived binding of SNAPf-tagged BBP ( Figure 2—figure supplement 1D and Supplementary file 3 ) . Mutation of the BS eliminated the long-lived binding events but short-lived binding remained unchanged ( Figure 2—figure supplement 1F; RNA 10 ) . These short-lived events could be eliminated by simultaneous mutation of both the BS and a nearby ΨBS ( Séraphin and Rosbash , 1991; Figure 2—figure supplement 1C; RNA 7 ) . This suggests that a 5' SS and interactions with U1 alone cannot result in detectable BBP binding , in contrast with a previous model obtained by cross-linking in vivo that described U1-facilitated BBP recruitment ( Görnemann et al . , 2005 ) . It is possible that irreversible cross-linking may have captured interactions too transient for us to observe or interactions occurring between BBP and RNA sequences other than the BS ( i . e . , ΨBS ) . Regardless , since BBP appears to associate stably with the BS and briefly with the ΨBS , we were able to mitigate its influence on U1 by mutation of both of these sequences . In the presence of CA , colocalization of U1 with the tethered RNAs became strongly dependent on the 5' SS when those RNAs also lacked the BS/ΨBS sequences . We no longer observed colocalization of U1 with RNAs that either lacked a functional 5' SS or contained the weak 5' SS ( Figure 2B , C and Figure 2—figure supplement 2; RNAs 5 and 6 ) . In addition , U1 binding was near background levels after ablation of the 5' end of the snRNA and ECP mitigation , even when the RNA contained a consensus 5' SS ( Figure 1—figure supplement 3C ) . Any remaining U1/RNA binding events occurred too quickly for us to detect in these experiments , likely with lifetimes <~0 . 5 s . These data suggest that the 5' cap and BS sequence are responsible for recruitment of U1 on RNAs that lack 5' SS ( Figure 1D ) or when the 5' SS pairing region of the snRNA has been removed . In contrast , frequent binding was still observed on RNAs containing strong 5' SS and with the intact snRNA ( Figure 2B , RNAs 7 and 8 , and Figure 1—figure supplement 3 ) . Thus , short-lived binding events are not non-specific - they depend on features of the RNA and likely involve interactions between U1 and ECPs bound to the transcript . When a 5' SS is present , short-lived binding events originate from both interactions between U1 and the 5' SS and interactions between U1 and ECPs . In the absence of a 5' SS or intact snRNA , U1 is recruited to RNAs primarily through interactions with ECPs . While a strong 5' SS is sufficient for U1 recruitment , binding of U1 was significantly weakened when ECPs were mitigated ( cf . Figure 2D–F vs . Figure 1H–J and Supplementary file 2 ) . On RNAs containing a consensus 5' SS , the long-lived parameter τ2 was reduced ~3 fold and these long-lived events were much less frequent . In fact , U1 dwell times on RNAs containing a 5' SS but without ECPs were similar to those observed on RNAs lacking a 5' SS but with ECPs . This indicates that U1/ECP interactions play an important role in tuning U1 binding even at strong 5' SS and that the longest-lived U1 complexes depend on both the 5' SS and ECPs . After ECPs were mitigated , the differences between RNAs containing a consensus or hyperstabilized 5' SS became apparent . Hyperstabilization did not change τ1 ( Figure 2D ) . However , both τ2 and A2 increased approximately 2-fold upon hyperstabilization relative to the consensus , a greater change than was observed when the ECPs were not mitigated ( cf . Figure 1I and J , RNA 3 and 4 with Figure 2E and F , RNA 7 and 8 ) . These data show that increased pairing potential between U1 and the 5ʹ SS predominantly impacts the kinetics of the long-lived binding events and ECPs mask this effect . We next tested if the cap and BS/ΨBS independently influence U1 binding dynamics . Unexpectedly , while the lifetimes of the longest-lived U1 binding events ( τ2 ) could be modulated by either the cap or the BS/ΨBS , this was not true of their relative abundance ( A2 ) . In the presence of CA , we observed an increase in both τ2 and A2 for U1 binding events on RNAs containing a BS/ΨBS compared to those that lacked these sequences ( Figure 2H , I; RNAs 3 and 7 + CA ) . In contrast , addition of CA changed τ2 but not A2 on RNAs that lacked the BS/ΨBS ( Figure 2H , I; RNA 7 and ±CA ) . Similar results were also obtained using WCE prepared from strains in which the small subunit of the CBC was genetically deleted ( Figure 2—figure supplement 3 and Supplementary file 4 ) . Thus , addition of CA only decreases U1 lifetimes on RNAs lacking a BS/ΨBS and removal of the BS/ΨBS only decreases U1 lifetimes in the presence of CA . However , the BS/ΨBS additionally changes A2 independent of the presence or absence of CA . The above results show that molecular features of the RNA other than the 5' SS can influence U1 binding . It is likely this occurs by interactions between U1 and ECPs— protein-protein contacts with CBC associated at the 5' cap and BBP/Mud2 associated at the BS . We therefore tested whether or not the longest-lived U1 events were correlated with simultaneous binding of the transcript by ECPs . CBC is very abundant in WCE and has so far proven intractable for CoSMoS studies ( data not shown ) . However , BBP is amenable to C-terminal SNAPf tagging and binds surface-immobilized RNAs with sequence specificity ( Figure 2—figure supplement 1 ) . We engineered a yeast strain containing two DHFR-tags on U1 and a SNAPf tag on BBP to permit simultaneous visualization of U1 and BBP ( Supplementary file 4 ) . WCE prepared from this strain formed CC and retained splicing activity in vitro ( Figure 3—figure supplement 1 and Figure 3—figure supplement 2 ) . We then carried out 3-color CoSMoS assays to monitor U1 and BBP binding dynamics on immobilized pre-mRNAs that contained a consensus 5' SS and only the BS ( Figure 3A ) . U1 and BBP frequently colocalized to the same pre-mRNA ( Figure 3B , C and Figure 3—figure supplement 3 ) . Not all U1 or BBP binding events resulted in colocalization of the other component: U1 often bound without BBP and vice versa . However , colocalized complexes of U1 , BBP , and pre-mRNA could persist from a few seconds to several minutes ( Figure 3C ) . Since long-lived U1 and BBP complexes are dependent on the presence of a 5' SS ( Figure 1 ) or BS ( Figure 2—figure supplement 1 ) , respectively , it is likely that many of these co-complexes of U1 , BBP , and pre-mRNA represent the spliceosome E complex with U1 engaged with the 5' SS and BBP bound at the BS . To test whether or not the longest-lived U1 binding events are correlated with BBP occupancy , we measured the dwell times of U1 molecules colocalizing with BBP and compared those to U1 dwell times that showed no evidence for coincident BBP binding ( Figure 3D ) . The mean lifetime of U1 molecules colocalizing with BBP was significantly longer ( 182 . 3 ± 35 . 6 s ) compared to U1 molecules binding in the absence of BBP ( 73 . 6 ± 32 . 2 s ) , while no difference was seen in a randomized control ( Figure 3—figure supplement 4 ) . This supports the notion that simultaneous binding of the transcript by BBP promotes a functional change in U1 that promotes stable association and long-lived binding . In these three-color experiments , CBC could still potentially interact with the RNA and U1 independent of BBP . When BBP was not simultaneously present with U1 on the RNA , the influence of the CBC alone likely resulted in the ~74 s mean U1 lifetime . To confirm this , we compared this value to calculated mean U1 lifetimes for interactions on RNAs under conditions in which only the CBC was allowed to bind ( RNA 7; 82 . 6 ± 11 . 0 s ) or when binding of both BBP and CBC was prevented ( RNA 7 , +CA; 27 . 4 ± 2 . 6 s ) using two-color experiments and data described in Figure 2 . Only under the former condition is the average U1 lifetime similar to what we observed in the three-color experiments in the absence of simultaneous BBP association . Thus , U1 lifetimes when BBP binding is either not observed on the RNA ( three-color experiments ) or not permitted due to a BS mutation ( two-color experiments ) are likely due to the influence of the CBC and are much longer than U1 binding in the absence of both the CBC and BBP . These CoSMoS experiments also revealed the lifetime and dynamics of the spliceosome E complex . Consistent with the multi-exponential kinetics observed with both U1 ( Figure 1G ) and BBP ( Figure 2—figure supplement 1G ) , the distribution of E complex lifetimes was also best described by a function containing two exponential terms ( τ1 of ~20 s and τ2 of ~165 s , Figure 3E , Figure 3—figure supplement 5 , and Supplementary file 5 ) . This indicates that multiple types of E complexes exist that contain both U1 and BBP . The longest-lived complexes that contribute to τ2 may be the same as those that result in commitment and are observable by native PAGE . We next analyzed the pathways of E complex assembly and disassembly . E complex assembly can proceed by either initial U1 or BBP binding ( Figure 3—figure supplement 6 ) . After E complex formation , we observed frequent redefinition of the 5' SS and BS as U1 or BBP spontaneously dissociated and another molecule subsequently bound ( Figure 3C ) . For example , we noted the formation of E complexes that would persist for several minutes before loss of U1 followed by rebinding of a different U1 molecule . This resulted in redefinition of the 5' SS while definition of the BS was maintained by BBP . The alternative pathway of BS redefinition was also observed as was switching between 5' SS and BS redefinition on the same pre-mRNA molecule . In both cases , analysis of fluorescence trajectories supported recruitment of only a single U1 molecule and a single BBP molecule to the vast majority of pre-mRNAs at any one time ( Figure 3—figure supplement 7 ) . This signifies that redefinition requires prior release of either U1 or BBP from the 5ʹ SS or the BS , respectively . Since ATP was not included in these experiments , these redefinition events are unlikely to require ATPase activity . Analysis of the fates of individual E complexes revealed that the loss of either U1 or BBP was nearly equally probable ( Figure 3F ) . However , pre-mRNAs that lost U1 were twice as likely to also lose BBP rather than redefine the 5' SS . In contrast , pre-mRNAs that lost BBP were nearly twice as likely to redefine the BS as they were to subsequently lose U1 . This suggests that in vitro E complex formation results in different degrees of obligation to the 5' SS and BS . In sum , these experiments reveal that E complex formation is dynamic and results in little commitment to any particular 5' SS/BS pair . Our data support distinct short- and long-lived U1 interactions influenced by both 5' SS sequence as well as ECPs . Structures of the human U1 snRNP place the U1-C protein ( yeast Yhc1 ) at the site of duplex formation between the snRNA and 5' SS ( Kondo et al . , 2015 ) , and there is strong genetic and biochemical evidence to support links between Yhc1 , ECPs , and stability of the snRNA/5' SS duplex ( Schwer and Shuman , 2014; Abovich and Rosbash , 1997; Hage et al . , 2009; Schwer and Shuman , 2015; Zhang and Rosbash , 1999 ) . Therefore , it is possible that Yhc1 also plays a critical role in modulating U1 binding kinetics . We tested this by engineering a yeast strain that permits fluorescent labeling of the U1 snRNP with SNAPf tags as well as exchange of Yhc1 alleles by plasmid shuffling ( Supplementary files 4 and 6 ) . This allowed us to prepare WCEs containing fluorescent U1 in which every U1 molecule contained a Yhc1 protein harboring specific amino acid mutations near the site of snRNA/5' SS duplex formation . We introduced two well-characterized alleles of Yhc1 that have minimal impact on yeast viability but are predicted to have opposing impacts on U1 stability during splicing ( Figure 4A ) . Yhc1-L13F permits bypass of Prp28 function during spliceosome activation , presumably due to destabilization of the U1 snRNA/5' SS duplex at this stage ( Chen et al . , 2001 ) . In contrast , Yhc1-D36A reinforces the need for Prp28 , possibly due to stabilization of this same duplex ( Schwer and Shuman , 2014 ) . As expected , yeast containing SNAPf-tagged U1 and these Yhc1 alleles were viable , and WCE prepared from these strains maintained high levels of splicing in vitro ( Figure 3—figure supplement 1 ) . We first measured binding dynamics of the mutant U1 molecules on RNAs containing a consensus 5' SS and BS/ΨBS in the absence of CA ( Figure 4B , RNA 3 –CA ) . U1 binding dynamics were minimally perturbed by inclusion of the Yhc1-L13F mutation ( Figure 4C–F and Figure 4—figure supplement 1 ) . We observed only a slight reduction of τ2 and A2 with Yhc1-L13F even though this mutation eliminates the ATP-dependence of U1 release during spliceosome activation . On the other hand , Yhc1-D36A showed little change in the long-lived parameters and instead more than doubled the short-lived parameter , τ1 ( Figure 4D ) . When the BS/ΨBS were removed and CA added ( Figure 4B , RNA 7 + CA ) , the impact of the Yhc1-L13F mutation became more striking . Under these conditions , we were unable to observe binding of U1 to RNAs even though they contained a strong , consensus 5’ SS ( Figure 4C and Figure 4—figure supplement 1 ) . These data agree with previous analysis of the L13F mutation using short RNA fragments ( Du et al . , 2004 ) and suggest that ECPs become essential for recruitment of U1 to pre-mRNAs when U1/5' SS pairing is destabilized by the Yhc1-L13F mutation . U1 molecules containing Yhc1-D36A were still able to bind RNAs after the BS/ΨBS were removed and CA was added ( Figure 4C ) . However , the increase in τ1 seen in the presence of ECPs was no longer apparent ( Figure 4D–F ) . Together data from the Yhc1-L13F and -D36A mutations show that Yhc1 is likely directly involved in both short- and long-lived U1 binding . Importantly , the impact of Yhc1 mutation on U1 binding can be modified by the presence or absence of ECPs . Finally , we tested whether or not additional potential base pairing interactions between U1 and the 5' SS could counteract the Yhc1-L13F mutation and enable cap- and BS/ΨBS-independent recruitment . Indeed , Yhc1-L13F U1 readily colocalized with RNAs containing a hyperstabilized 5' SS even in the absence of the BS/ΨBS and in the presence of CA ( Figure 5A , B ) . Unexpectedly , Yhc1-L13F U1 binding was much more stable than WT U1 under these conditions and persisted in some cases for tens of minutes ( Figure 5C , D and Figure 5—figure supplement 1 ) . Analysis of the dwell times revealed a dramatic increase in both τ2 and A2 but only when the RNAs contained the hyperstabilized 5' SS and the influence of ECPs was mitigated ( Figure 5E , F and Supplementary file 2 ) . These observations are consistent with previous reports of the L13F mutation facilitating additional base pairing interactions between U1 and the 5' SS using a fragment of the RP51A pre-mRNA that lacked the BS sequence ( Du et al . , 2004 ) . These extremely long-lived Yhc1-L13F U1 binding events on hyperstabilized RNAs could only be completely suppressed by simultaneously permitting both BBP and the CBC to bind . Individually including the BS/ΨBS sequence or omitting CA did not result in a significant decrease in τ2 ( Figure 5—figure supplement 2 ) . Both conditions were required to reduce τ2 to values observed with WT U1; although , the BS/ΨBS alone influenced A2 . This is in agreement with our previous observation that hyperstabilization leads to an increase in WT U1 dwell time only after the influence of ECPs was mitigated ( Figure 1I , J vs . Figure 2E , F ) . How ECPs mask the effects of hyperstabilization is not clear but could involve altered snRNP or RNA conformations that do not favor additional pairing interactions . While observations with Yhc1-L13F may be specific to this particular mutation , it illustrates how ECPs can tune U1 interactions in different ways depending on the composition of U1 and the 5ʹ SS to either enhance binding or promote release .
In the absence of ECPs , U1 recruitment is strictly 5' SS-dependent , and the dwell times exhibit a bi-exponential distribution of short- and long-lived complexes ( Figure 2C–F ) . Two simple kinetic schemes that could account for this observation are a two-step binding pathway ( Figure 6A ) or independent pathways that form two , non-interconverting U1/5' SS complexes ( Figure 6—figure supplement 1 ) . While it is difficult to completely exclude the presence of independent pathways without knowledge of the forward and backward rates occurring at each step , we favor the two-step binding mechanism since it is supported by multiple aspects of our single molecule data . First , all U1/5' SS interactions in the absence of ECPs are dependent on the 5' SS sequence ( Figure 2C–F ) , the Yhc1 protein ( Figure 4 ) , and the snRNA ( Figure 1—figure supplement 3 ) . This indicates that both short- and long-lived events are likely occurring at the site of snRNA/5' SS duplex formation . A two-step binding mechanism does not require the formation of unrelated complexes at this critical location , as would the other model . Second , stabilization or destabilization of either the short- or long-lived complexes impacts both the complex lifetime as well as amplitude . For example , increasing the number of potential base pairs at the 5' SS in the absence of ECPs results in a simultaneous increase of more than 2-fold in both τ2 and A2 without changing the binding event frequency ( Figure 2C–F ) . This can be rationalized by two-step binding in which short-lived complexes are more likely to transition into long-lived complexes when additional pairing is possible , thus resulting in the decrease in A1 and increase in A2 . Concerted changes in lifetimes and amplitudes are not as easily compatible with independent binding pathways ( Figure 6—figure supplement 1 ) since the pathways are not kinetically coupled to one another . Finally , even under our most stabilizing conditions for U1 interaction , we still observe short-lived events ( Figure 5D ) . This is expected if formation of the short-lived complex is a necessary precursor to more stable binding . The molecular and functional differences between the initial and stable complexes are not clear . Results with the hyperstabilized 5' SS suggest that the transition between the two will involve a change in the snRNA/5' SS pairing region , as has been previously proposed ( Du et al . , 2004; McGrail et al . , 2008 ) . Our data do not preclude the possibility that short- or long-lived complexes may originate from altered interactions with other factors that transiently associate with U1 [for example , U2 ( Das et al . , 2000 ) or allow us to determine if different complexes promote subsequent steps in splicing to different degrees . Previous work showed that longer U1 lifetimes were correlated with intron loss ( Hoskins et al . , 2011 ) . However , further experiments will be needed to determine the origin of those increased lifetimes and if they came about as a direct consequence of formation of the long-lived U1 complexes detected here . While the initial complex is short-lived , it is dependent on the presence of a strong 5' SS in the RNA . This property is distinct from the proposed U1 scanning complex ( McGrail et al . , 2008 ) , which is predicted to involve snRNA nucleotides other than those that pair with the 5' SS and may not recognize the 5' SS itself . It is possible that scanning occurs prior to the 5' SS dependent interactions that we observe . Our single molecule data show that U1 still interacts dynamically with RNAs even while BBP is simultaneously bound ( Figure 3C ) . This indicates that the spliceosome E complex is readily disassembled without ATP and only a subset of complexes are stable enough to result in commitment—in agreement with previous identification of the committed and uncommitted ‘δ' complex by Ruby ( Ruby , 1997 ) . Moreover , our kinetic description of E complex lifetimes characterizes its heterogeneity and reveals the presence of both short- and long-lived E complexes ( Figure 3E ) . We do not know if short-lived E complexes can transition into those that are long-lived . It is possible that E complexes of diverse stability originate from different levels of functional engagement of U1 and/or ECPs with the pre-mRNA or with one another . One consequence of dynamic and reversible E complex formation is that it permits rapid definition and redefinition of the 5' SS and BS that are subsequently used by the spliceosome . While BBP association with the pre-mRNA stimulates stable U1 binding ( Figure 6A ) , the frequent release and rebinding of BBP while U1 remains bound suggests that U1 stabilization could be triggered by BS sequences not ultimately used by the spliceosome . Such a strategy could facilitate efficient spliceosome assembly by anchoring U1 at the 5' SS while delaying BS definition until the optimal sequence for pairing with U2 is found . This flexibility may be important for regulating splicing , particularly on long introns which may contain many potential BS sequences . In the presence of ATP , it is likely that redefinition competes with spliceosome assembly at a particular 5' SS/BS pair . Thus , modulation of both E complex and other splicing factor ( e . g . , U2 ) binding kinetics may be important to either enhance or suppress redefinition events and permit proper 5' SS and BS selection . While it has often been noted that alternative splicing correlates with low complementarity between the 5' SS and the U1 snRNA , it is less clear what factors lead to discrimination between splice sites of equal strength ( Roca et al . , 2013 ) . This discrimination could arise from trans-acting factors tuning U1 binding at particular locations to either stabilize or destabilize snRNP association . Our data indicate that yeast ECPs function in this capacity to facilitate U1 recruitment to pre-mRNAs containing weak splice sites ( Figure 1D ) , modulate U1 interactions occurring at consensus splice sites ( Figure 2 ) , and destabilize U1 binding if excessive complementarity is present ( Figures 1 , 2 and 5 ) . Tuning of U1 interactions is thus a general feature of both weak and strong splice sites , and the lifetime of a U1/pre-mRNA complex may not be apparent from base pairing potential alone . In our experiments , we focused on substrates containing only a single 5' SS; however , it is possible that tuning may also change how two or more potential 5' SS compete with one another within a single transcript . Extensive analysis has revealed a rich network of genetic and physical interactions between the CBC , BBP/Mud2 , and both the proteins and snRNA of U1 ( Chang et al . , 2012; Qiu et al . , 2012; Schwer and Shuman , 2014; Agarwal et al . , 2016; Schwer and Shuman , 2015; Schwer et al . , 2013; Zhang and Rosbash , 1999 ) . While the contribution RNA conformation plays in promoting U1 release is not yet known , the single molecule and genetic data suggest U1 tuning is occurring through physical connections between components of the E complex . The measured E complex lifetimes ( Figure 3E ) allow us to define that any such physical interactions resulting in a cross-intron bridging complex ( Abovich and Rosbash , 1997 ) between U1 at the 5' SS and BBP/Mud2 at the BS are transient and likely to dissociate with a rate constant of ~0 . 4 min−1 or greater . Tuning U1 interactions with cognate RNAs may be critical for processes other than spliceosome assembly including those that are splicing-independent such as telescripting ( Berg et al . , 2012 ) . While stable association of U1 promotes assembly , failure to release U1 during spliceosome activation inhibits splicing since the U6 snRNA must also pair to the 5' SS ( Chiou et al . , 2013; Staley and Guthrie , 1999 ) . A two-step model for U1 binding ( Figure 6A ) permits a reversible change in affinity that could be used to decrease U1 stability during activation . The more weakly-bound U1 complex may then facilitate U1/U6 exchange . Kinetic data obtained with the L13F and D36A mutants of Yhc1 support this idea ( Figure 4 ) . These mutations bypass ( L13F ) ( Chen et al . , 2001 ) or reinforce ( D36A ) ( Schwer and Shuman , 2014 ) the need for the Prp28 ATPase to promote activation , and U1’s containing these mutations would be predicted to associate with the 5' SS less ( L13F ) or more ( D36A ) strongly than WT . Indeed , we found this prediction to be true but the effects were modulated by ECPs . Other components of the spliceosome known to interact with U1 [ ( e . g . , Prp8; ( Li et al . , 2013 ) ] may function similarly to tune U1 binding during later stages of splicing . Genetic experiments have elucidated functional redundancy between the CBC and BBP/Mud2 with respect to their ability to complement mutations in U1 ( Qiu et al . , 2012; Chang et al . , 2012; Schwer and Shuman , 2014; Agarwal et al . , 2016; Schwer and Shuman , 2015; Schwer et al . , 2013 ) . We were also able to observe some functional redundancy for these factors at the single molecule level , as both ECPs could extend U1 lifetimes ( Figures 1 and 2 ) . Whether or not both ECPs accomplish this by a similar mechanism is not yet clear . However , the BS exerts an additional and unexpected effect on U1 by increasing the relative abundance ( A2 ) of long-lived complexes . We rationalize this by BBP/Mud2 bound at the BS impacting the transition between initial and stable U1 binding in the two-step model ( Figure 6A ) . This model could also explain differences in genetic lethality observed when snRNA mutations are combined with Mud2Δ or Cbp20 mutations , with the former more often being lethal than the latter ( Schwer et al . , 2013 ) . While BBP/Mud2 and CBC each help to anchor U1 to transcripts , their molecular influences on U1 are distinct . One consequence of this observation is that transcripts destined for splicing ( i . e . , those containing introns ) are more competitive for U1 compared to those which only contain 5' SS sequences or only bound by ECPs ( Figure 6B ) . We propose that CBC initially facilitates sequence-independent U1 recruitment to transcripts , which biases U1 occupancy towards Pol II transcription products that may contain introns . Since the 5' end of the snRNA is not required for transient , ECP-dependent interactions ( Figure 1—figure supplement 3 ) , initial recruitment may not involve pairing between the snRNP and transcript . U1 could then subsequently engage with a cognate 5' SS sequence , at which point it may or may not switch into a more stably bound conformation . If the transcript also contains a downstream BS bound by BBP/Mud2 , this promotes recruitment as well as the switch to stable binding . The features of this model are consistent with in vivo data obtained in both yeast and humans which have detected splicing-independent recruitment of U1 and the influences of the CBC and BS on U1 occupancy ( Görnemann et al . , 2005; Lacadie and Rosbash , 2005; Patel et al . , 2007; Spiluttini et al . , 2010 ) . Notably , the two-step binding mechanism ( Figure 6A ) , in which BBP/Mud2 influences U1’s transition between initial and stable complex formation , results in a coupling of 5' SS and BS recognition . Stable U1 binding can be associated with multiple landmarks found on the transcript including the cap , 5' SS , and BS—all indicators of an intron-containing pre-mRNA . Rapid identification of specific nucleic acid sequences is a challenge faced by many cellular RNPs including the ribosome , the RNA-induced silencing complex ( RISC ) , and bacterial CRISPR machineries in addition to spliceosome snRNPs . While mechanisms of ribosome scanning remain to be elucidated ( Archer et al . , 2016; Hinnebusch , 2014 ) , recent structural and single-molecule data indicate that both Cas9 and Argonaute share common strategies for finding particular sequences of DNA or RNA ( Chandradoss et al . , 2015; Salomon et al . , 2015; Sternberg et al . , 2014 ) . These RNP machineries utilize ‘seed’ or PAM sequences to facilitate rapid scanning of nucleic acids in a step distinct from that of formation of extended base pairing interactions . These two-step mechanisms enable efficient target identification and likely also limit the lifetimes of non-productive interactions . Our data support a similar two-step mechanism for U1 snRNP recruitment to 5ʹ SS in which short-lived interactions give way to longer-lived complexes . Whether or not U1 also contains a ‘seed’ region that nucleates binding is not yet clear; however , the U1-C/Yhc1 protein is positioned to potentially participate in such a process . This could explain the origin of our observed kinetics as well as provide U1 an efficient means of searching for potential 5ʹ SS .
Capped [32P]-labeled RP51A pre-mRNAs for in vitro splicing and trace [32P]-labeled RP51A substrates for single molecule assays were made by in vitro transcription with T7 RNA polymerase in the presence of [α-32P] UTP and G ( 5’ ) ppp ( 5’ ) G RNA cap analog ( NEB ) . Trace [32P]-labeled RP51A was fluorescently labeled by splinted ligation with a biotinylated 2’-O-methyl oligonucleotide derivatized with a single Alexa Fluor 488 ( Alexa488 , ThermoFisher Scientific ) or Cy5 ( GE Life Sciences ) fluorophore as previously described . Yeast strains containing fast SNAP ( SNAPf ) ( Sun et al . , 2011 ) tags on the U1 snRNP proteins Prp40 and Snp1 or BBP ( Supplementary file 4 ) were prepared by homologous recombination as previously described ( Hoskins et al . , 2016 ) . The SNAPf-tagged Yhc1 deletion strain and Yhc1 plasmids ( Supplementary file 4 and 6 ) were kind gifts of Dr . Magda Konarska and Dr . Beate Schwer , respectively . The triple-labeled strain containing DHFR tags on Prp40 and Snp1 and a SNAPf tag on BBP was prepared from strain yAAH3 ( Hoskins et al . , 2011 ) and by SNAPf tagging the Msl5 ( BBP ) gene by homologous recombination as previously described . Alleles of Yhc1 were exchanged by plasmid shuffling and selection for loss of the WT allele on dropout plates containing 1 mg/mL 5-FOA . Yeast whole cell extract ( WCE ) was prepared as previously described ( Crawford et al . , 2007 ) . SNAPf-tagged proteins were labeled by incubation of the lysate for 30 min at room temperature with the fluorophore ( e . g . , benzylguanine-Dy549/SNAP-Surface 549 , New England Biolabs ) before gel filtration . A fluorophore concentration of 1 . 1 µM was used to label SNAPf tags . SNAPf-tagged proteins derivatized with fluorophores were visualized by denaturing polyacrylamide gel electrophoresis ( SDS-PAGE ) followed by imaging fluorescence on a LAS 4000 or Typhoon fluorescence imager ( GE Life Sciences ) . Splicing assays were carried out as previously described using 40% WCE and 0 . 2–0 . 5 nM substrate ( Crawford et al . , 2007 ) . Splicing assays with WCE used in single molecule experiments also contained the oxygen scavengers and triplet quenchers described below . [32P]-labeled RNAs were visualized by denaturing PAGE followed by phosphorimaging . Data were analyzed using ImageQuant software ( GE Lifesciences ) . Commitment complex assays were performed as previously described with the following modifications ( Seraphin and Rosbash , 1989 ) . 50 µL standard splicing reactions containing 40% WCE were depleted of ATP by the addition of 2 mM glucose and incubating at 25°C for 20 min . [32P]-labeled RP51A was prepared from a template truncated at the DdeI site within the RP51A gene . The RNA ( 0 . 5 nM ) was then added to the reaction and incubated for 20 min at 16°C . 10 µL of the reaction was added to 10 µL of ice cold buffer R [2 mM MgOAc , 50 mM HEPES pH 7 . 5 , 1 mg/mL tRNA , 50-fold molar excess of cold RP51A pre-mRNA; ( Price et al . , 2014 ) ] and incubated for 10 min on ice . 5 µL of loading dye ( 2 . 5x TBE , 50% v/v glycerol , 0 . 3% w/v bromophenol blue , and 0 . 3% w/v xylene cyanol dyes ) was added to the solution before it was loaded on a 0 . 5x TBE , 3% 60:1 acrylamide , 0 . 5% w/v agarose , 2% v/v glycerol gel . The 26 cm gel was run in 0 . 5x TBE at 12V for 24 hr at 4°C . The gel was then dried before phosphorimaging . Oligonucleotide directed RNase H digestion of the 5ʹ end of the U1 snRNA was carried out by addition of 6 . 5 µg of a DNA oligo complimentary to the 5ʹ end of the U1 snRNA ( 5ʹ-CTTAAGGTAAGTAT-3ʹ ) and 0 . 048 U/µL RNase H ( Invitrogen ) to a 100 µL splicing reaction and incubating at 30°C for 30 min ( Du and Rosbash , 2001 ) . Part of the reaction was quenched by adding 20 µL of the reaction solution to 180 µL of splicing dilution buffer ( 0 . 1 M Tris pH 7 . 5 , 1% w/v SDS , 0 . 3 M NaOAc , 150 mM NaCl , 1 mM EDTA ) and placed on ice . The remaining reaction solution was immediately used for single molecule assays after the addition of triplet state quenchers and oxygen scavengers . Primers complementary to nucleotides 114–135 ( 5ʹ-GACCAAGGAGGTTGCATCAATG-3ʹ ) of the U1 snRNA and nucleotides 100–121 of the ( 5ʹ-GCCAAAAAATGTGTATTGTAA-3ʹ ) of the U2 snRNA were end labeled using [γ-32P] ATP ( PerkinElmer ) . Total yeast RNA was isolated by phenol extraction and ethanol precipitation from the quenched U1 ablation reaction and resuspended in 10 µL of water . The extracted RNA ( 1 µL ) was added to 1 . 5 µL of oligo mix ( 0 . 4 pmol/µL [γ-32P] U1 oligo , 0 . 4 pmol/µL [γ-32P] U2 oligo , 1 . 25M KCl , and 50 mM Tris pH 8 . 0 ) and placed on ice for 3 min . Primers were annealed by heating the mixture to 90°C and then cooling on ice for 3 min each . The solution was then heated to 45°C for 5 min before addition of 6 . 5 µL of RT mix [150 mM Tris pH 8 . 0 , 50 mM MgCl2 , 50 mM DTT , 2 . 5 mM dNTPs , 0 . 3 µL SuperScript III Reverse Transcriptase ( Invitrogen ) ] . Primer extension products were analyzed by 6% denaturing polyacrylamide gel electrophoresis . The gel was then dried before phosphorimaging . Cleaning of slides and coverslips and assembly and passivation of flow cell chambers were carried out as previously described ( Crawford et al . , 2007 ) . Two-color CoSMoS assays were carried out in splicing buffer with the addition of PCD/PCA as an oxygen scavenging system and trolox as previously described . Three-color CoSMoS experiments additionally included cyclooctatetraene , propyl gallate , and 4-nitrobenzylacohol as triplet quenchers as previously described ( Hoskins et al . , 2011 ) . ATP was depleted from WCE by addition of 2 mM glucose prior to the experiment . Identical U1 and BBP binding kinetics were observed in the absence and presence of additional exogenous , purified hexokinase [added to final concentration of 5 U/mL , Roche 11426362001 ( 1500 U/mL , ~450 U/mg protein ) ; Supplementary file 2 and 3] . Therefore , only glucose addition and the endogenous hexokinase activity present in the extract was used to deplete ATP . A custom-built , objective-based micromirror TIRFM ( Larson et al . , 2014 ) was used to image the interactions between fluorescently labeled , surface-tethered RNAs and fluorescently labeled U1 or BBP molecules . Two-color and three-color CoSMoS experiments were carried out as previously described with the following modifications ( Hoskins et al . , 2011 ) . Laser powers were set between 200 and 500 µW for both the 532 and 633 nM lasers in two- and three-color experiments and 1 . 5 mW for the 488 nm laser in three-color experiments . U1 snRNPs and BBP were imaged with a 1 s exposure at 5 s intervals between each frame . Surface immobilized RNAs were imaged with a 1 s exposure every 60 frames , except in three-color experiments during which RNAs were imaged only at the beginning and end of the experiment . Photobleaching analysis was carried out as described by varying the laser power in each experiment and measuring U1 and BBP dwell times ( Hoskins et al . , 2016 ) . Under these imaging conditions , the contribution of photobleaching to U1 and BBP the fitted parameters describing the distribution of dwell times was minimal . Drift correction was achieved by tracking changes in position of the surface tethered RP51A throughout the durations of the experiment and was seldom more than one pixel in any direction during the duration of each experiment . Auto focusing was carried out using a 785 nm laser immediately before imaging of the immobilized RNAs in all experiments . Mapping between the <635 nm and >635 fluorescence fields of view ( FOV ) was achieved using a reference data set generated using fluorescent beads which emit light at multiple wavelengths . Data analyses were done as previously described ( Hoskins et al . , 2011; Shcherbakova et al . , 2013 ) . In brief , fluorescence signal from surface tethered pre-mRNAs were used to select areas of interest ( AOIs ) . AOIs were mapped from the >635 nm FOV to the <636 nm FOV and pixel intensity was integrated for each AOI using custom MATLAB software . Peaks in the fluorescence intensity from the <635 nm FOV were manually inspected to confirm the presence of a colocalized spot in the AOI . The distributions of observed dwell times for each subcomplex were displayed by constructing probability density plots in which the dwell times were binned and each bin divided by the product of the bin width and total number of events . Error bars for each bin were calculated as the error of a binomial distribution as previously described ( Hoskins et al . , 2011 ) . Distributions described by one or two exponential terms were fit by maximum likelihood methods to exponential probability density functions as previously described using Equations 1 or 2 , respectively ( Hoskins et al . , 2011 ) . In all equations , tm represents the time between consecutive frames; tmax represents the duration of the experiment ( 30 min ) ; A1 and A2 the fitted amplitudes; and τ1 and τ2 represent the fitted parameters . Errors in the fit parameter were determined by bootstrapping 1000 random samples of the data and determining the standard deviation of the resultant values . ( 1 ) [ ( A1⋅ ( e−tmτ1−e−tmaxτ1 ) ) ]−1⋅[A1τ1e−tτ1] ( 2 ) [ ( A1⋅ ( e−tmτ1−e−tmaxτ1 ) ) + ( ( 1−A1 ) ⋅ ( e−tmτ1−e−tmaxτ1 ) ) ]−1⋅[A1τ1e−tτ1+1−A2τ2e−tτ2] | Our genes contain coded instructions for making the molecules in our bodies , but this information must be extensively processed before it can be used . The instructions from each gene are first copied into a molecule called a pre-mRNA , before a process known as splicing removes certain sections to form a mature mRNA molecule . Splicing can remove different sections of the pre-mRNA to make different mRNA molecules from the same gene depending on the current needs of the cell . Splicing is controlled by a combination of proteins and other molecules , collectively called the spliceosome . A part of the spliceosome called U1 recognizes the start of pre-mRNA sections that need to be removed , which is referred to as the five-prime splice site ( or “5’ SS” for short ) . The attachment of U1 to such a site allows other molecules to also attach to the pre-mRNA , which eventually assemble a spliceosome . The very first steps in this process involve U1 and a set of other proteins that create what is called the “Early” or “E” complex . Although there are many molecules involved in the E complex , it was not known how they interact with each other and how this affects which splice sites are used for splicing in different cells . Using advanced microscopy , Larson and Hoskins examined individual U1 molecules from yeast cells while the molecules formed E complexes and identified two different ways U1 can bind to five-prime splice sites . One process involved U1 attaching to pre-mRNA for a short time , whilst the other involved a longer association between U1 and pre-mRNA . Sometimes U1 could also transition between the first process and the second . The results showed that other parts of the E complex affected which process was used at different sites by affecting the type or duration of U1’s attachment . All U1 particles use the same components to attach to splice sites in all pre-mRNAs , but the most used splice sites are not always those that are predicted to have the strongest attachments to U1 . This work helps to reveal how other proteins involved in splicing influence this effect , altering U1’s ability to attach to pre-mRNAs to suit each new situation . This also allows cells to change gene splicing to fit different situations . Many genes in our bodies rely on splicing and understanding this process in detail could be the key to diagnosing and treating a range of different illnesses . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2017 | Dynamics and consequences of spliceosome E complex formation |
Cell migration requires the cyclical assembly and disassembly of focal adhesions . Adhesion induces phosphorylation of focal adhesion proteins , including Cas ( Crk-associated substrate/p130Cas/BCAR1 ) . However , Cas phosphorylation stimulates adhesion turnover . This raises the question of how adhesion assembly occurs against opposition from phospho-Cas . Here we show that suppressor of cytokine signaling 6 ( SOCS6 ) and Cullin 5 , two components of the CRL5SOCS6 ubiquitin ligase , inhibit Cas-dependent focal adhesion turnover at the front but not rear of migrating epithelial cells . The front focal adhesions contain phospho-Cas which recruits SOCS6 . If SOCS6 cannot access focal adhesions , or if cullins or the proteasome are inhibited , adhesion disassembly is stimulated . This suggests that the localized targeting of phospho-Cas within adhesions by CRL5SOCS6 and concurrent cullin and proteasome activity provide a negative feedback loop , ensuring that adhesion assembly predominates over disassembly at the leading edge . By this mechanism , ubiquitination provides a new level of spatio-temporal control over cell migration .
During development , wound healing and cancer invasion , migrating cells need to move between other cells and through the dense extracellular matrix ( ECM ) . Cells can attach to and pull on the ECM by using integrins ‒ transmembrane receptors that link ECM outside the cell to focal adhesions ( FAs ) and the actin cytoskeleton inside the cell ( Alexander et al . , 2008; Hynes , 2002; Pelham and Wang , 1997; Petrie et al . , 2012; Puklin-Faucher and Sheetz , 2009 ) . FAs are dynamic assemblies containing many proteins held together by dense networks of protein-protein interactions ( Kanchanawong et al . , 2010; Zaidel-Bar et al . , 2007a ) . Nascent FAs ( often called focal complexes ) initiate when talin and other proteins associate with β integrin tails to stabilize an active integrin conformation and stimulate binding to the ECM ( Calderwood et al . , 1999; Tadokoro et al . , 2003 ) . Talin then binds actin and vinculin and actin flow exerts forces that create additional binding sites for vinculin , which in turn recruits more FA proteins and more actin ( del Rio et al . , 2009; Jiang et al . , 2003 ) . In this way , the force generated by actin flow , resisted by the ECM , creates a positive feedback loop to stabilize and grow the adhesion ( Case and Waterman , 2015 ) . In concert , force from the FA acts on actin filaments to induce the formation of contractile stress fibers and actin arcs ( Burridge and Wittchen , 2013; Livne and Geiger , 2016; Roca-Cusachs et al . , 2013 ) . The contraction of stress fibers and actin arcs provides motive power to advance the cell body . As the cell body moves forwards over an FA , the force vector is redirected and the FA remodels or disassembles , allowing the FA proteins to recycle through the cytosol for reuse at the leading edge ( Wehrle-Haller , 2012 ) . Inhibition of Rho kinase or myosin relaxes actomyosin tension and induces rapid FA disassembly ( Chrzanowska-Wodnicka and Burridge , 1996; Volberg et al . , 1994 ) . These findings support a mechanical model in which increased force drives FA assembly and reduced force stimulates disassembly . FA dynamics are also regulated by signaling through protein kinases and small GTPases . ECM binding stimulates integrin-associated focal adhesion kinase ( FAK ) and Src-family kinases ( SFKs ) . These tyrosine kinases phosphorylate several FA proteins , helping to build the FA by creating phosphotyrosine ( pY ) sites that bind to the SH2 domains of additional FA proteins , as well as by activating RhoA ( Parsons et al . , 2010; Zaidel-Bar et al . , 2003 , 2007b ) . Thus , FA size increases when endogenous SFKs are activated , consistent with a positive role for SFKs during FA assembly ( Thomas et al . , 1995 ) . However , other evidence suggests that SFKs stimulate FA disassembly . For example , FA size increases and FA turnover is inhibited in fibroblasts with mutations in Src , FAK , or a variety of Src/FAK substrates ( Ilic et al . , 1995; Ren et al . , 2000; Volberg et al . , 2001; Webb et al . , 2004 ) . Furthermore , FAs grow when kinase-inactive Src is over-expressed ( Fincham and Frame , 1998 ) , and activated mutant Src weakens integrin-cytoskeletal interactions and accelerates FA disassembly ( Felsenfeld et al . , 1999; Fincham and Frame , 1998; Fincham et al . , 1995; Galbraith et al . , 2002 ) . Therefore , SFKs may regulate both FA assembly and disassembly , perhaps by phosphorylating different substrates . Cas ( p130Cas , BCAR1 ) is an FA protein that is phosphorylated by Src when integrins engage the ECM ( Fonseca et al . , 2004; Nojima et al . , 1995; Petch et al . , 1995; Vuori and Ruoslahti , 1995 ) . Cell motility and invasion are stimulated when Cas is over-expressed and inhibited when Cas is deleted ( Brabek et al . , 2005; Cary et al . , 1998; Honda et al . , 1999 , 1998; Huang et al . , 2002; Patwardhan et al . , 2006; Sanders and Basson , 2005 ) . At the molecular level , phosphorylation of Cas by SFKs induces binding to adaptor proteins Crk/CrkL and activation of the Rac1 and Rap1 GTPases ( Hasegawa et al . , 1996; Sakai et al . , 1994; Tanaka et al . , 1994; Vuori et al . , 1996 ) . Forced relocalization of CrkL to FAs stimulates migration via DOCK1 and Rac1 ( Li et al . , 2003 ) . A SFK/Cas pathway stimulates protrusive activity , inhibits stress fibers , and stimulates FA disassembly ( Webb et al . , 2004 ) . Therefore , Cas presents a paradoxical case where phosphorylation is induced by integrin engagement but stimulates FA disassembly and migration . Interestingly , Cas is mechanosensitive . Cytoskeletal tension induces Cas tyrosine phosphorylation ( Janostiak et al . , 2011; Tamada et al . , 2004 ) . Mechanical extension of purified Cas promotes its phosphorylation by SFKs in vitro ( Sawada et al . , 2006 ) . Thus integrin signaling and cytoskeletal tension may increase the level of phosphotyrosine Cas ( pYCas ) and stimulate FA disassembly whilst simultaneously unfolding talin to promote FA assembly . This raises the question of how FA assembly and disassembly are balanced in such a way that assembly predominates near the leading edge and disassembly occurs under the cell body . Here we provide evidence that CRL5SOCS6 , a cullin-RING ubiquitin E3 ligase composed of cullin-5 ( Cul5 ) , Rbx2 , elonginBC and suppressor of cytokine signaling 6 ( SOCS6 ) , inhibits pYCas-dependent FA turnover at the front of migrating epithelial cells . We find that SOCS6 is recruited to pYCas in FAs at the leading edge , and that localization of SOCS6 to leading edge FAs slows their disassembly , dependent on cullin and proteasome activity and on binding of SOCS6 to Cas and CRL5 . SOCS6 is not present in adhesions under the cell body , and sustained Cas signaling may facilitate their disassembly . The results suggest that CRL5SOCS6 targets pYCas within adhesions at the front of the cell , thereby allowing FAs to grow and provide anchorage for stress fibers .
To investigate FA dynamics , we wounded monolayers of MCF10A cells that express low levels of EYFP-vinculin and allowed the cells to migrate in the absence of EGF . TIRF microscopy was used to detect EYFP puncta . We previously noted that migrating MCF10A cells contained large FAs and prominent stress fibers along the leading edge , which were lost following Cul5 knockdown ( Teckchandani et al . , 2014 ) . TIRF microscopy of migrating EYFP-vinculin-expressing cells also revealed large FAs at the cell front , but also detected many much smaller FAs all across the ventral surface ( Figure 1a , left ) . Large FAs were absent from migrating cells that had been treated with Cul5 siRNA ( Figure 1a , right , see Figure 1—figure supplement 1a for knockdown efficiency ) . To quantify FA size , we divided the cell into front and back with a line ~ 6 µm from the leading edge ( Figure 1—figure supplement 1b ) , and quantified the size distribution of front FAs of control and Cul5-deficient cells . While the number of large FAs at the front of control cells was not sufficient to affect the mean or median FA size , approximately 3 . 6% of FAs at the front of control cells were larger than 200 pixels ( 5 µm2 ) , but <0 . 5% of FAs at the front of Cul5-deficient cells were this large ( Figure 1b ) . 10 . 7554/eLife . 17440 . 003Figure 1 . Cul5 stabilizes focal adhesions at the front of migrating cells . Focal adhesion dynamics of control and Cul5-deficient MCF10A cells , migrating into a scratch wound in EGF-deficient medium , monitored using EYFP-vinculin . ( a ) FAs visualized using TIRF microscopy of control and Cul5-deficient cells expressing EYFP-vinculin . ( b ) Histogram of FA sizes at the front of control ( gray ) and Cul5-deficient ( blue ) cells . All structures greater than 0 . 05 μm2 ( two pixels ) were quantified . The inset shows structures greater than 5 μm2 on an expanded scale . ( c ) Rainbow color representation of FA appearance and disappearance . FAs are colored according their presence during the time course , from blue to red . Only the front region of the cell is shown . ( d ) Individual frames from regions boxed in c . Arrowheads and arrows indicate stable and dynamic FAs , respectively . ( e ) Automated curve fitting to intensity/time plots for representative FAs from the front of control and Cul5-deficient cells . R2: Pearson’s correlation coefficient squared . ( f ) FA assembly and disassembly rate constants from the front ( circles ) and back ( squares ) of control ( gray ) and Cul5-deficient ( blue ) cells . Mean and standard error of median rates from each of 4–6 time-lapse movies are indicated . *p<0 . 05; ***p<0 . 001 . Student’s t-test , two tailed , unequal variance . ( f ) Histogram of disassembly rate constants at the front and back of control ( gray ) and Cul5-deficient ( blue ) cells . Mean and standard error of biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 00310 . 7554/eLife . 17440 . 004Figure 1—figure supplement 1 . Quantification of FA dynamics . ( a ) Characterization of cells . Western blot and RT-PCR data showing expression of EYFP-vinculin , Cas and Cul5 in MCF10A cells . ( b ) Definition of cell front and back . Red line indicates the position of the mask , ~6 µm from the cell front at zero time that was used to define the cell front and back . Note that the area included in the cell front increases over time as the leading edge advances . ( c ) Automated segmentation of EYFP-vinculin fluorescence into individual FAs . Different colors outline each FA . ( d ) Time series for a specific FA . Images recorded at 2 min intervals . ( e ) Plot of FA mean intensity against time , with automated curve fitting to estimate first-order rate constants for assembly and disassembly . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 00410 . 7554/eLife . 17440 . 005Figure 1—figure supplement 2 . FA size is independent of FA assembly and disassembly rate constants . Plots of assembly and disassembly rate constants against mean area ( pixels ) for FAs at the front and back of control ( gray ) and Cul5-deficient ( blue ) cells . There is a slight but insignificant correlation between FA size and disassembly rates for Cul5-deficient cells but complete independence for control cells . Only structures >2 pixels ( 0 . 05 um2 ) were included in the analysis . Note that small FAs in control cells assemble and disassemble much slower than similarly sized FAs in Cul5-deficient cells . R: Pearson’s correlation coefficient . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 005 To investigate whether different-sized FAs resulted from different dynamics , we recorded time lapse movies . FAs at the front of Cul5-deficient cells were notably more dynamic than those at the front of control cells ( Video 1 and 2 ) . To illustrate this difference , FAs were color-coded based on their presence in different frames , starting with blue at time zero and advancing through teal , green , orange , red and crimson by 100 min ( Figure 1c ) . Many FAs at the front of control cells were white , indicating that they were present during the entire movie , while most FAs at the front of Cul5-deficient cells were singly colored , indicating they were short-lived . Blue FAs , present at the start , marked the original leading edge , and red FAs , born at the end , marked the ending leading edge position . The contrast between stable FAs in control cells and transient FAs in Cul5-deficient cells was also apparent from examination of individual frames from the movies ( Figure 1d ) . 10 . 7554/eLife . 17440 . 006Video 1 . EYFP-vinculin in migrating MCF10A cells ( siCon ) imaged every 2 min on a 100× TIRF objective , starting approximately 6 hr after wounding . Length of movie 120 min . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 00610 . 7554/eLife . 17440 . 007Video 2 . EYFP-vinculin in migrating MCF10A cells ( siCul5 ) imaged every 2 min on a 100× TIRF objective , starting approximately 6 hr after wounding . Length of movie 120 min . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 007 To quantify FA dynamics , movies were analyzed using FA analysis software ( FAAS ) ( Berginski et al . , 2011 ) . This software does not distinguish FAs from the smaller focal complexes , so we refer to all EYFP structures larger than 0 . 05 µm2 ( two pixels ) as FAs . We extracted first-order rate constants for the assembly and disassembly phases of each FA ( Figure 1—figure supplement 1c–e ) . The rate constants for assembly and disassembly were generally lower for front FAs of control cells than front FAs of Cul5-deficient cells ( Figure 1e ) . However , as expected , individual FAs were very heterogeneous . To control for technical or biological variation between experiments , we determined the median assembly and disassembly rate constants of many FAs in each of four to six experiments and then averaged these parameters across independent experiments . We found that FAs at the front and back of control cells assembled and disassembled with similar rate constants ( 0 . 012–0 . 013 min−1 for assembly; 0 . 008–0 . 009 min-1 for disassembly ) , suggesting that the front FAs were not larger because of increased assembly or decreased disassembly ( Figure 1f , Table 1 ) . Indeed , there was no correlation between mean FA area and assembly or disassembly rate constants ( Figure 1—figure supplement 2 ) , as expected for first-order reactions . Removing Cul5 had no effect on the assembly or disassembly of back FAs ( Figure 1f , Table 1 ) . At the front , however , removing Cul5 stimulated both assembly and disassembly approximately two-fold , to ~0 . 029 min−1 for assembly and ~0 . 021 min−1 for disassembly ( Figure 1f , Table 1 ) . Similar assembly and disassembly rates were measured in cells in which Cul5 was stably knocked down with shRNA targeting a different sequence , so the effect is unlikely to be off-target ( see below , Figure 3—figure supplement 1 ) . The increase in disassembly of leading edge FAs was also apparent from histograms of disassembly rate constants: the percentage of FAs with disassembly rates <0 . 02 min−1 shifted from >60% to <10% when Cul5 was removed ( Figure 1g ) . While we were unable to directly measure the lifetimes of the large FAs at the front of control cells , most persisted for longer than the 3 hr duration of the movies , suggesting lifetimes exceeding 3 hr . In contrast , front FAs of Cul5-deficient cells had an average 24 min half time for assembly , 36 min stable phase , and 33 min half time for disassembly , for a total 93 min average lifetime . The longer persistence of FAs at the front of control cells may contribute to their increased size . 10 . 7554/eLife . 17440 . 008Table 1 . Summary of first-order rate constants for FA assembly and disassembly . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 008Conditions Region Assembly Disassembly # of movies # of FAs ( X 10−2 min−1 ) ( X 10−2 min−1 ) siConfront1 . 3 ( ±0 . 2 ) 0 . 8 ( ±0 . 03 ) 464siCul5front2 . 9 ( ±0 . 1 ) p=0 . 00022 . 1 ( ±0 . 9 ) p=0 . 000046263siConback1 . 2 ( ±0 . 2 ) 0 . 9 ( ±0 . 1 ) 4667siCul5back1 . 3 ( ±0 . 05 ) 1 . 3 ( ±0 . 5 ) p=0 . 04 6811siConfront1 . 1 ( ±0 . 1 ) 0 . 8 ( ±0 . 1 ) 5142siCul5front3 . 0 ( ±0 . 4 ) p=0 . 008 1 . 9 ( ±0 . 08 ) p=0 . 00008 4444siCasfront1 . 6 ( ±0 . 3 ) 0 . 7 ( ±0 . 1 ) 6151siCul5+siCasfront1 . 1 ( ±0 . 2 ) 0 . 8 ( ±0 . 07 ) 5137siConback0 . 8 ( ±0 . 04 ) 0 . 7 ( ±0 . 09 ) 5632siCul5back0 . 9 ( ±0 . 08 ) 0 . 8 ( ±0 . 08 ) 4505siCasback0 . 8 ( ±0 . 05 ) 0 . 6 ( ±0 . 08 ) 6484siCul5+siCasback0 . 7 ( ±0 . 1 ) 0 . 5 ( ±0 . 05 ) 5541shConfront1 . 0 ( ±0 . 1 ) 1 . 0 ( ±0 . 1 ) 4105shCul5front2 . 0 ( ±0 . 1 ) p=0 . 0012 2 . 0 ( ±0 . 09 ) p=0 . 0015 4217shCasfront0 . 9 ( ±0 . 1 ) 1 . 0 ( ±0 . 06 ) 4126shCul5+shCasfront1 . 0 ( ±0 . 2 ) 0 . 8 ( ±0 . 07 ) 4152shConback0 . 9 ( ±0 . 06 ) 0 . 7 ( ±0 . 8 ) 4701shCul5back0 . 8 ( ±0 . 04 ) 0 . 9 ( ±0 . 1 ) 4497shCasback0 . 6 ( ±0 . 06 ) 0 . 9 ( ±0 . 1 ) 4559shCul5+shCasback0 . 7 ( ±0 . 1 ) 0 . 8 ( ±0 . 4 ) 4608siConfront1 . 4 ( ±0 . 1 ) 0 . 9 ( ±0 . 06 ) 5185siSOCS6front1 . 7 ( ±0 . 2 ) 1 . 4 ( ±0 . 03 ) p=0 . 0004 5201siConback0 . 9 ( ±0 . 09 ) 0 . 7 ( ±0 . 04 ) 5620siSOCS6back0 . 8 ( ±0 . 07 ) 0 . 8 ( ±0 . 01 ) 5469siConfront1 . 4 ( ±0 . 2 ) 1 . 0 ( ±0 . 1 ) 4254siSOCS6 ( alternate ) front1 . 7 ( ±0 . 08 ) 1 . 9 ( ±0 . 08 ) p=0 . 0007 4179siConback1 . 1 ( ±0 . 02 ) 1 . 0 ( ±0 . 06 ) 4640siSOCS6 ( alternate ) back1 . 2 ( ±0 . 05 ) 1 . 2 ( ±0 . 1 ) 4600Confront1 . 0 ( ±0 . 09 ) 1 . 0 ( ±0 . 06 ) 476MLN4924front2 . 0 ( ±0 . 08 ) p=0 . 00016 1 . 6 ( ±0 . 03 ) p=0 . 0004 4111Conback0 . 6 ( ±0 . 07 ) 0 . 6 ( ±0 . 05 ) 4182MLN4924back1 . 0 ( ±0 . 2 ) 0 . 7 ( ±0 . 04 ) 4293Rate constants are reported as mean ± SEM Microtubules ( MTs ) are required for FA disassembly and are observed repeatedly engaging with FAs during disassembly ( Bershadsky et al . , 1996; Kaverina et al . , 1999 , 1998; Stehbens et al . , 2012 ) . We tested whether CRL5-regulated FA disassembly requires MTs . We treated cells with nocodazole to disrupt MTs and then removed nocodazole and assayed adhesion disassembly during MT regrowth ( Ezratty et al . , 2005 ) . We found that nocodazole stabilized FAs in both control and Cul5-deficient cells ( Figure 2a ) . However , FAs disappeared more rapidly from Cul5-deficient cells than control cells upon nocodazole removal ( Figure 2a , b ) . Stress fibers also disappeared more rapidly from Cul5-deficient cells ( Figure 2 ) . Accelerated FA disassembly in Cul5-deficient cells was not due to faster MT regrowth after nocodazole removal because the MT cytoskeleton fully recovered in less than 15 min whether or not Cul5 was present ( Figure 2—figure supplement 1a ) . Indeed , MT plus ends , tracked with fluorescent plus-end binding-protein EB1 , reached FAs in the ventral membrane within minutes of removing nocodazole ( Figure 2—figure supplement 1b ) . Therefore , Cul5 slows the disassembly of FAs after MTs have regrown , implying that Cul5 controls the speed of disassembly after MT targeting . 10 . 7554/eLife . 17440 . 009Figure 2 . Cul5 regulates microtubule-dependent FA disassembly . Control and Cul5-deficient HeLa cells were plated on collagen IV-coated coverslips , serum-starved overnight , and incubated with nocodazole for 3 hr to induce microtubule disassembly and stabilize FAs . Cells were fixed at various times after nocodazole removal and stained for F-actin and paxillin . ( b ) Cells lacking large FAs and prominent stress fibers were scored as percent of total cells . Mean and standard deviation of three biologically independent experiments . ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 00910 . 7554/eLife . 17440 . 010Figure 2—figure supplement 1 . Cul5 does not regulate microtubule growth . ( a ) HeLa cells were fixed at 0 or 15 min after reversal of nocodazole block and stained for tyrosinated tubulin . Deconvolution microscopy , ventral section . Cul5 did not affect the complete regrowth of microtubules . ( b ) HeLa cells expressing EYFP-EB1 , a microtubule + end tracking protein , and mChSOCS6 , a marker for FAs ( see below , Figure 4 ) were incubated in nocodazole and imaged live by TIRF microscopy . EB1 was diffusely localized when nocodazole was present , but was detected in puncta that approached or co-localized with FAs two minutes after nocodazole washout , suggesting that microtubules regrow within 2 min of nocodazole reversal in cells that contain Cul5 . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 01010 . 7554/eLife . 17440 . 011Figure 2—figure supplement 2 . Adhesion turnover in Cul5-deficient cells is not regulated by clathrin-mediated endocytosis . ( a ) HeLa cells were treated with control , Cul5 or clathrin heavy chain ( CHC ) siRNA and subject to the nocodazole washout assay for FA disassembly as in Figure 2 . Mean and standard error of two biologically-independent experiments . CHC was required for the delayed FA disassembly in Cul5-proficient cells but was largely dispensable for the accelerated disassembly in Cul5-deficient cells . ( b ) Localization of CHC in migrating MCF10A shCul5 cells . Immuofluorescence for CHC ( red ) , F-actin ( green ) and nucleus ( blue ) in migrating control and Cul5-deficient MCF10A cells . Deconvolution microscopy , ventral section . Note the low numbers of clathrin-coated structures in the leading lamellipodium of Cul5-deficient cells . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 011 MT targeting is reported to stimulate FA disassembly by clathrin-dependent integrin endocytosis ( Ezratty et al . , 2009; Ezratty et al . , 2005 ) . Accordingly , depleting clathrin heavy chains ( CHC ) strongly inhibited FA disassembly in control cells ( Figure 2—figure supplement 2a ) . However , CHC depletion only partly inhibited the more rapid FA disassembly in Cul5-deficient cells ( Figure 2—figure supplement 2a ) , suggesting that Cul5 inhibits a distinct mechanism of disassembly that is clathrin-independent . Consistent with this interpretation , clathrin-coated pits were missing from the leading lamellipodium of migrating Cul5-deficient cells ( Figure 2—figure supplement 2b ) , as recently reported for other cell types ( Kural et al . , 2015 ) . This suggests that Cul5 regulates FA disassembly through mechanisms unrelated to clathrin-mediated endocytosis . Since we previously identified pYCas as a CRL5 substrate ( Teckchandani et al . , 2014 ) and since pYCas is required for FA disassembly in migrating fibroblasts ( Webb et al . , 2004 ) , we tested whether Cul5-stimulated disassembly requires Cas . We found that removing Cas suppressed the increase in FA disassembly at the front of migrating Cul5-deficient MCF10A cells but had no effect on FAs at the back or F control cells ( Figure 3a , b ) . These effects were unlikely to be due to off-target effects of Cas siRNA because similar results were obtained with cells stably knocked down for Cul5 and Cas using shRNA ( Figure 3—figure supplement 1 ) . Consistent with a Cul5-Cas pathway , Cas knockdown restored MT-dependent FA disassembly in Cul5-deficient HeLa cells in the nocodazole washout assay , but had no effect on control cells ( Figure 3c , d ) . Therefore , Cas is required for increased FA disassembly when Cul5 is absent . 10 . 7554/eLife . 17440 . 012Figure 3 . Cul5-mediated FA turnover at the leading edge is regulated by Cas . Focal adhesion dynamics of MCF10A cells migrating into a scratch wound in EGF-deficient medium , monitored using EYFP-vinculin and TIRF microscopy . ( a ) Rainbow color representation of FA appearance and disappearance at the front of control , Cul5-deficient , Cas-deficient , and Cul5-Cas-deficient cells . ( b ) FA assembly and disassembly rate constants from the front and back . Mean and standard error of median rates from each of 4–6 time-lapse movies are indicated . ***p<0 . 001 . Student’s t-test , two tailed , unequal variance . ( c , d ) Nocodazole washout assay as in Figure 2 . ( c ) Images ( deconvolution microscopy , ventral section ) and ( d ) quantification of FA disassembly in cells depleted for Cul5 and Cas . Cas was required for the accelerated FA disassembly in Cul5-deficient cells . Mean and standard error of three biologically-independent experiments . ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 01210 . 7554/eLife . 17440 . 013Figure 3—figure supplement 1 . Specificity controls for Cul5 and Cas knockdown . Focal adhesion dynamics of stable vector , shCul5 , shCas and shCul5/shCas EYFP-vinculin-expressing MCF10A cells migrating into a scratch wound in EGF-deficient medium . FA assembly and disassembly rate constants from the front 6 µm and back of leading cells . Mean and standard error of median rates from each of four time-lapse movies are indicated . **p<0 . 01 . Student’s t-test , two tailed , unequal variance . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 013 Removal of Cas also suppressed the increased FA assembly of Cul5-deficient cells ( Figure 3a , b ) , consistent with the observation that FA and stress fiber assembly is slowed in cas mutant fibroblasts ( Antoku et al . , 2008; Honda et al . , 1998 ) , potentially due to decreased signaling through a Crk/ C3G/Rap1 signaling pathway ( Li et al . , 2002; Voss et al . , 2003 ) . SOCS6 is the substrate receptor through which CRL5 binds pYCas ( Teckchandani et al . , 2014 ) . Accordingly , depleting SOCS6 stimulated FA disassembly at the front but not back of migrating MCF10A cells ( Figure 4a , b ) . This was unlikely to be due to an off-target effect , because a different pool of SOCS6 siRNA had the same effect ( Figure 4—figure supplement 1 ) . SOCS6 depletion also stimulated FA disassembly in the nocodazole washout assay and Cas knockdown inhibited this effect ( Figure 4c–e ) . Normal rates of FA disassembly were rescued by re-expression of wildtype SOCS6 , but not by mutant SOCS6LCQQ that cannot bind elongin B and the rest of the CRL5 complex ( Figure 4—figure supplement 2a ) . Moreover , expression of CasFF , a mutant that does not bind SOCS6 ( Teckchandani et al . , 2014 ) , stimulated FA disassembly regardless of the presence or absence of Cul5 , while wildtype Cas only stimulated FA disassembly when Cul5 was absent and Cas15F , a mutant that cannot signal downstream , did not stimulate FA disassembly even when Cul5 was absent ( Figure 4—figure supplement 2b ) . These results suggest that SOCS6 inhibits FA disassembly dependent on binding to both Cas and CRL5 . However , unlike Cul5 , SOCS6 did not impact FA assembly rates ( Figure 4b ) , suggesting that , while an increase in pYCas ( and other SOCS6 targets ) is sufficient to stimulate disassembly , increased pYCas does not accelerate assembly . Other SOCS proteins may target different pY substrates for ubiquitination by CRL5 to regulate FA assembly . 10 . 7554/eLife . 17440 . 014Figure 4 . FA disassembly in migrating cells is regulated by SOCS6 . Focal adhesion dynamics of MCF10A cells migrating into a scratch wound in EGF-deficient medium , monitored using EYFP-vinculin and TIRF microscopy . ( a ) Rainbow color representation of FA appearance and disappearance at the front of control and SOCS6-deficient cells . ( b ) FA assembly and disassembly rate constants from the front ( circles ) and back ( squares ) of control ( gray ) and SOCS6-deficient ( green ) cells . Mean and standard error of median rates from each of five time-lapse movies are indicated . ***p<0 . 001 . Student’s t-test , two tailed , unequal variance . ( c–e ) Nocodazole washout assay as in Figure 2 . ( c ) Staining for paxillin and F-actin in control and SOCS6-deficient HeLa cells 30 min after nocodazole washout reveals accelerated disassembly in SOCS6-deficient cells . Deconvolution microscopy , ventral section . ( d ) Quantification of FA disassembly in cells depleted for SOCS6 . Mean and standard error of three biologically-independent experiments . *p<0 . 05 . ( e ) Quantification of FA disassembly in cells depleted for SOCS6 and Cas . Cas was required for the accelerated FA disassembly in SOCS6-deficient cells . Mean and standard error of three biologically-independent experiments . ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 01410 . 7554/eLife . 17440 . 015Figure 4—figure supplement 1 . Specificity of SOCS6 knockdown in MCF10A cells . To control for off-target effects , MCF10A cells were treated with an independent pool of siRNAs against SOCS6 before analysis as in Figure 4b . ***p<0 . 001 . Student’s t-test , two tailed , unequal variance . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 01510 . 7554/eLife . 17440 . 016Figure 4—figure supplement 2 . Regulation of FA disassembly requires SOCS6 interaction with CRL5 and Cas . Microtubule-dependent FA disassembly was assayed in transfected HeLa cells , scoring the percent of cells lacking FAs at 30 min after nocodazole reversal . ( a ) Rescue of normal FA disassembly by expression of siRNA-resistant mouse SOCS6 wildtype ( WT ) but not by SOCS box mutant LCQQ . ( b ) Expression of CasFF , a mutant that does not bind SOCS6 , stimulates FA disassembly independent of Cul5 , while wildtype Cas only stimulates disassembly when Cul5 is absent and Cas15F , which is unable to bind downstream signaling molecules , is inactive . Mean and standard deviation of two experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 016 CRL5 may regulate FA dynamics by directly interacting with substrates in FAs or by altering the steady-state concentrations of FA proteins in the cytoplasm . We investigated whether CRL5 localizes to FAs . Available antibodies revealed endogenous Cul5 in the nucleus and throughout the cytosol ( data not shown ) . Cul5 was not detectably enriched in FAs . In addition , none of the available SOCS protein antibodies we tested were sufficiently sensitive or specific to detect endogenous SOCS proteins . However , when we transiently transfected HeLa cells with T7 epitope-tagged SOCS6 , the tagged protein was enriched in FAs relative to the rest of the ventral membrane ( Figure 5a and Figure 5—figure supplement 1 ) . Enrichment was slight relative to the cytosol , suggesting a low affinity of binding and the potential for rapid exchange . We investigated the basis for SOCS6 association with FAs . Because FAs contain pYCas and because SOCS6 binds pYCas ( Petch et al . , 1995; Teckchandani et al . , 2014 ) , we tested whether pYCas is required for SOCS6 localization . FA localization of SOCS6 was inhibited in cells treated with Cas siRNA or with the SFK inhibitor SU6656 ( Figure 5a ) . Moreover , a point mutation in the phosphotyrosine-binding SH2 domain of SOCS6 inhibited binding to pYCas ( Figure 5—figure supplement 2a ) and prevented FA localization ( Figure 5a ) . Therefore , pYCas is required for SOCS6 localization to FAs . 10 . 7554/eLife . 17440 . 017Figure 5 . SOCS6 localizes to adhesion sites dependent on pYCas . ( a ) SOCS6 localization to FAs requires Cas , SFK activity , and a functional SH2 domain . HeLa cells were transiently transfected with T7-tagged wildtype ( WT ) or SH2 domain mutant ( R407K ) SOCS6 , treated with or without Cas siRNA , and plated on collagen IV , serum-starved , and incubated with nocodazole for 3 hr to stabilize FAs . One sample was treated with SFK inhibitor SU6656 ( 10 μM ) during nocodazole treatment . Fixed cells were stained with antibodies to T7 ( green ) and FAK ( magenta ) . Images were collected from the ventral plane using TIRF microscopy , and from the entire cell using epifluorescence . ( b , c ) SOCS6 and pYCas localization in migrating MCF10A cells . ( b ) MCF10A cells stably expressing mCherry-tagged SOCS6 ( mChS6 ) were allowed to reach confluence , wounded , and allowed to migrate in the absence of EGF . 5 µM MLN4924 was added at the time of wounding and washed off 6 hr later . Cells were fixed 2 hr later and stained with antibodies against vinculin ( magenta ) . Images were collected using TIRF microscopy . The mean ratio of mCherry to vinculin integrated intensity was calculated for 135 FAs at the front and 528 FAs at the back in ~12 cells in two separate experiments . ( c ) MCF10A cells were allowed to reach confluence , wounded , and allowed to migrate in the absence of EGF . Cells were stained with antibodies against vinculin ( magenta ) and pYCas ( green ) . Images were collected using TIRF microscopy . The mean ratio of pYCas to vinculin integrated intensity was calculated for 153 FAs at the front and 896 FAs at the back in ~16 cells in two separate experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 01710 . 7554/eLife . 17440 . 018Figure 5—figure supplement 1 . Focal adhesion localization of SOCS6 . HeLa cells were transiently transfected to express T7-epitope tagged mouse SOCS6 or vector ( truncated GFP ) , plated on collagen IV , serum-starved , and incubated with nocodazole to stabilize FAs . Fixed cells were stained with antibodies to T7 ( green ) and FAK ( magenta ) . Images were collected from the ventral plane using TIRF microscopy , and from the entire cell using epifluorescence . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 01810 . 7554/eLife . 17440 . 019Figure 5—figure supplement 2 . Characterization of SOCS6 mutants . ( a ) Transient expression of T7-tagged wildtype ( WT ) or mutant SOCS6 in HeLa cells . Anti-T7 antibody immunoprecipitates and cell lysates were analyzed by Western blotting with antibodies to Cas , ElonginB and T7 . Mutation of the SOCS box ( LC to QQ ) prevents binding of SOCS6 to ElonginB ( CRL5 ) but not Cas , whereas mutation of the SH2 domain ( R407K ) prevents binding of SOCS6 to Cas but not ElonginB . ( b , c ) Stable expression of mCherry-tagged ( WT ) or SOCS box mutant ( LCQQ ) in MCF10A cells . ( b ) Steady state levels of mChSOCS6 and Cas in two independent cell lines expressing mChSOCS6WT and one line expressing mChSOCS6LCQQ . Note that mChSOCS6WT inhibits expression of endogenous Cas while mChSOCS6LCQQ is poorly expressed . ( c ) Partial rescue of Cas levels in cells expressing mChSOCS6WT by treatment with MLN4924 for 6 hr and washout for 2 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 01910 . 7554/eLife . 17440 . 020Figure 5—figure supplement 3 . Cul5 knockdown increases pYCas in FAs . ( a ) HeLa cells , stably knocked down for Cul5 with shRNA , were plated on collagen IV , serum-starved , and treated with nocodazole to stabilize FAs . Cells were stained for paxillin ( magenta ) and pYCas ( green ) . The mean ratio of pYCas to paxillin intensity ( integrated ) was calculated for 1343 FAs in ~20 control and 1502 FAs in ~20 Cul5-deficient cells . ( b ) MCF10A cells , stably knocked down for Cul5 with shRNA , were allowed to reach confluence , wounded , and allowed to migrate in the absence of EGF . Cells were stained for vinculin ( magenta ) and pYCas ( green ) . The mean ratio of pYCas to vinculin intensity ( integrated ) was calculated for >700 FAs ( front ) and >2000 FAs ( back ) of ~12 cells for each condition . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 020 To test whether SOCS6 also localizes to front FAs in migrating MCF10A cells , we stably expressed wildtype mCherry-tagged SOCS6 ( mChSOCS6WT ) in MCF10A cells . Unfortunately , stable expression of mChSOCS6WT down-regulated Cas and , perhaps as a consequence , mChSOCS6WT was not detected in FAs ( Figure 5—figure supplement 2b and data not shown ) . We attempted to express the mChSOCS6LCQQ mutant , which does not stimulate Cas degradation . However , this mutant was only expressed at low level in MCF10A cells , perhaps due to instability or toxicity of the mutant ( Figure 5—figure supplement 2a , b ) . Therefore , we tested whether treating mChSOCS6WT-MCF10A cells with MLN4924 , which inhibits the neddylation process required for CRL activity ( Enchev et al . , 2015; Petroski et al . , 2005; Soucy et al . , 2009 ) , might allow recovery of Cas levels and localize mChSOCS6WT to FAs . We found empirically that incubation with 5 µM MLN4924 for 6 hr followed by normal media for 2 hr allowed Cas levels to recover without disturbing FA stability ( Figure 5—figure supplement 2c ) . Under these conditions , mChSOCS6WT co-localized with vinculin in FAs ( Figure 5b ) . Remarkably , mChSOCS6WT was approximately 1 . 6-fold more abundant in FAs in the front 6 µm than FAs in the rest of the cell . We tested whether leading edge FAs are also enriched for pYCas . The ratio of pYCas to vinculin was approximately 3-fold higher in FAs in the front 6 µm than back ( Fonseca et al . , 2004 ) ( Figure 5c ) . These results suggest that pYCas is enriched in FAs at the front of migrating cells and that SOCS6 preferentially associates with these FAs , as expected if SOCS6 binds to pYCas . Since both Cas and SOCS6 exchange between the cytosol and adhesions , SOCS6 could regulate Cas levels or activity in either location . Inhibition of SOCS6 or Cul5 expression slows Cas turnover , causing a 2–3-fold increase in the steady-state level of Cas ( Teckchandani et al . , 2014 ) ( Figure 1—figure supplement 1a ) . Accordingly , phosphorylated Cas levels in FAs increased approximately 3-fold when Cul5 expression was inhibited , especially in FAs in the front 6 µm of the cell ( Figure 5—figure supplement 3 ) . Since Cas exchanges rapidly between the cytosol and FAs ( Janostiak et al . , 2011; Machiyama et al . , 2014 ) , the increase in pYCas in FAs could be secondary to the increased Cas in the cytosol . Moreover , the chronic increase in pYCas in adhesions could accelerate adhesion turnover regardless of whether SOCS6 targets Cas in the cytosol or in FAs . We therefore tested whether SOCS6 could regulate FA turnover if it was sequestered away from adhesion sites . To this end , we constructed mChS6mito , a fusion of mCherry , SOCS6 and a sequence that targets the mitochondrial outer membrane ( Bear et al . , 2000 ) . mChS6mito still bound to Cul5 and to Cas ( Figure 6—figure supplement 1a ) , but localized to mitochondria instead of FAs ( Figure 6—figure supplement 1b ) . In the nocodazole washout assay , mChS6 rescued FA disassembly in SOCS6-depleted cells but mChS6mito did not ( Figure 6—figure supplement 1c ) . These results suggested that SOCS6 needs to access FAs in order to inhibit their disassembly . We used two approaches to test whether CRL5SOCS6 represses Cas during FA disassembly . First , we used an optogenetic approach . We used the light-regulated association between the CIBN and CRY2 proteins ( Kennedy et al . , 2010 ) to sequester SOCS6 away from adhesion sites during the disassembly process . We introduced two plasmids into HeLa cells from which endogenous SOCS6 had been removed with siRNA . One plasmid encoded CRY2 fused to mCherry-tagged SOCS6 ( CRY2mChS6 ) , and the second encoded CIBN fused to GFP and the mitochondrial targeting sequence ( CIBNGFPmito ) ( Figure 6a ) . In the dark , CRY2mChS6 co-localized with FAK in FAs ( Figure 6b left ) . In the light , however , CRY2mChS6 co-localized with CIBNGFPmito at the mitochondria and was not detected in FAs ( Figure 6b right ) . We then tested whether CRY2mChS6 could rescue FA disassembly using the nocodazole washout assay . Transfected cells were kept in the dark during growth and nocodazole treatment , so that CRY2mChS6 could diffuse and maintain normal levels of Cas . We then subjected the cells to light or dark conditions and removed nocodazole . The results showed that light inhibits FA disassembly ( Figure 6c ) . This implies that SOCS6 must have access to FAs during the disassembly process , and that the steady state level of Cas in the cytoplasm only plays a minor role . 10 . 7554/eLife . 17440 . 021Figure 6 . SOCS6 , Cullin and proteasome activity are required during FA disassembly ( a ) Light-regulated localization of SOCS6 . ( b ) Localization of CRY2mChS6 in nocodazole-treated HeLa cells . In the dark , CRY2mChS6 was detected in the cytosol as well as co-localizing with FAK in FAs . However , under blue light illumination , CRY2mChS6 co-localized with CIBNGFPmito at the mitochondria and was not detected in FAs . Entire cell: z-projection , Deconvolution microscopy; ventral section: single plane , TIRF . ( c ) SOCS6 slows FA disassembly only if it has access to FAs during the disassembly process . Cells were plated on collagen IV-coated coverslips , serum-starved overnight , and incubated with nocodazole for 3 hr to induce microtubule disassembly and stabilize FAs . Cells were either kept in the dark during all steps of the assay or illuminated with blue light for 30 min before and during nocodazole washout ( ‘light’ ) . Cells expressing both CRY2mChS6 and CIBNGFPmito were identified by epifluorescence and FA disassembly scored by immunofluorescence for FAK . Mean and standard deviation of three biologically independent experiments . *p<0 . 05; ***p<0 . 001 . ( d ) Brief treatment with Cullin inhibitor MLN4924 or proteasome inhibitor epoxomicin stimulates FA disassembly . Nocodazole washout assay as in Figure 2 . Cells were treated with 5 µM MLN4924 or 10 µM epoxomicin 1 hr before washout . Mean and standard deviation of three biologically independent experiments . ***p<0 . 001DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 02110 . 7554/eLife . 17440 . 022Figure 6—figure supplement 1 . Restricting the access of SOCS6 to FAs speeds FA disassembly . ( a ) Biochemical properties of mitochondrial-targeted SOCS6 . Plasmids encoding mCherry-SOCS6 ( mChS6 ) , mCherry-SOCS6-mito ( mChS6mito ) or mCherry-mito ( mChXmito ) were transiently transfected into HeLa cells together with ( left ) T7-tagged Cul5KR ( mutated to inhibit Cullin activity ) or ( right ) HA-tagged Cas . Following 30 min stimulation with 2 mM pervanadate to inhibit tyrosine phosphatases , cell lysates were analyzed by immunoprecipitation and Western blotting . ( b ) Cytosolic mChS6 rescues normal rates of FA disassembly but mitochondrion-associated mChS6mito does not rescue . Sample images and quantification from a nocodazole washout assay . Mean and standard error of three biologically-independent experiments . ***p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 022 As an independent approach to assess the timing of CRL5SOCS6 action , we again used the neddylation inhibitor , MLN4924 , which inhibits CRL5 and other cullin-RING E3s ( Soucy et al . , 2009 ) . MT-dependent FA disassembly was stimulated when MLN4924 was added an hour before and during 30 min of nocodazole removal ( Figure 6d ) . A similar result was obtained with epoxomicin , a proteasome inhibitor ( Meng et al . , 1999 ) ( Figure 6d ) . The results suggest that one or more CRLs and the proteasome need to be active during the disassembly process . Moreover , two hours of treatment with MLN4924 also increased FA assembly and disassembly rates at front but not back of migrating MCF10A cells ( Figure 7a , b ) . This short exposure to MLN4924 increased pYCas levels in front but not back FAs and had no effect on total Cas levels ( Figure 7c ) . Even though the turnover of many cell proteins is altered by MLN4924 and epoxomicin , the results are consistent with the hypothesis that the activities of Cullins and the proteasome are required during the process of FA disassembly . Combined with the optogenetics results , we suggest that SOCS6 regulates FA turnover locally by direct interaction with pYCas in or near FAs , and not by binding to or altering the steady state concentration of Cas or other proteins in the cytosol . 10 . 7554/eLife . 17440 . 023Figure 7 . FA turnover is inhibited by Cullin and proteasome activity . ( a , b ) Focal adhesion dynamics of control and MLN4924-treated MCF10A cells migrating into a scratch wound in EGF-deficient medium , monitored using EYFP-vinculin and TIRF microscopy . 5 µM MLN4924 was added to cells 2 hr before imaging . ( a ) Rainbow color representation of FA appearance and disappearance at the front of control and MLN4924-treated cells . ( b ) FA assembly and disassembly rate constants from the front ( circles ) and back ( squares ) of control ( gray ) and MLN4924-treated ( purple ) cells . Mean and standard error of median rates from each of four time-lapse movies are indicated . ( c , d ) MLN4924 increases pYCas locally in FAs at the leading edge but does not increase overall Cas levels . ( c ) Western blot of total cell lysate . 5 µM MLN4924 was added 4 hr before lysis . Cas was not increased on MLN4924 treatment . ( d ) MCF10A cells were allowed to reach confluence , wounded , and allowed to migrate in the absence of EGF . 5 µM MLN4924 was added to cells 4 hr before fixing . Cells were stained with antibodies against vinculin and pYCas . Deconvolution microscopy , ventral section . The mean ratio of pYCas to vinculin intensity ( integrated ) was calculated for >500 FAs at the front and >1800 FAs at the back in ~16 cells for each condition . ( e ) Summary model . Cas is phosphorylated specifically in FAs at the front of the cell . SOCS6 binds to pYCas in the FAs and locally inhibits pYCas via CRL5 and the proteasome . In cells lacking SOCS6 or Cul5 , or in which SOCS6 cannot access FAs or cullins or the proteasome is inhibited , then pYCas stimulates FA disassembly , dependent on microtubules ( MTs ) . In other parts of the cell , slow FA disassembly may occur by MT and clathrin-dependent mechanisms . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 023 The results suggest a model in which FA disassembly is slowed at the cell front by pYCas-dependent binding of SOCS6 . SOCS6 recruits CRL5 and ubiquitinates pYCas , which is then removed through proteasomal degradation ( Figure 7d ) . If SOCS6 or Cul5 is absent , if SOCS6 is held away from FAs , or if cullin or proteasomal inhibitors are present , then pYCas stimulates FA disassembly . A different , clathrin-dependent process prevails under the cell body . SOCS6 traffic thus controls the balance between FA assembly and disassembly in different parts of the cell .
The coordination between FA assembly and disassembly is critical for cell migration . We find that Cul5 , SOCS substrate receptors , and the proteasome regulate FA dynamics in epithelial cells . While Cul5 regulates both assembly and disassembly , one of the CRL5 substrate receptors , SOCS6 , is required for disassembly but not assembly , implying that other SOCS proteins are involved in FA assembly . SOCS6 inhibits FA disassembly by binding to the FA protein Cas and to the rest of the CRL5 complex . Our results suggest that CRL5SOCS6 , in conjunction with the proteasome , locally inhibits the SFK-Cas pathway in leading edge FAs to regulate their stability ( Figure 7d ) . We hypothesize that Cas ubiquitination and degradation occurs close to or in the FA , although we have not shown this directly . Without CRL5SOCS6 or proteasome activity , FAs become unstable , stress fibers are lost and the lamellipodium becomes unstable . Therefore , feedback control through CRL5SOCS6 organizes FA and actin dynamics at the wound edge and ensures a normal leading lamellipodium . The environment may further regulate SOCS6 , CRL5 or proteasome levels or activity , providing additional inputs to modulate the mode and speed of migration . This mechanism may be important in some cancers , where decreased SOCS6 expression has been associated with aggressive disease and increased metastasis ( Fang et al . , 2015; Letellier et al . , 2014; Li et al . , 2015; Wu et al . , 2013; Zhu et al . , 2013 ) . In migrating fibroblasts , FA turnover depends on Src and various Src substrates , including FAK , paxillin and Cas ( Fincham et al . , 1998; Webb et al . , 2004 ) . However , this mechanism has not been reported in other cell types . We found that the SFK-Cas pathway is also required for the rapid FA turnover at the leading edge of epithelial cells in which CRL5SOCS6-dependent ubiquitination has been inhibited . This suggests that removing restraints imposed by CRL5SOCS6 from migrating epithelial cells unveils a latent mechanism more similar to that in migrating fibroblasts , in which Src and Cas , and potentially also FAK and paxillin , stimulate FA turnover . We and others have observed that Cas phosphorylation is greatest in the large FAs at the front of normal cells and around the edge of spreading cells ( Fonseca et al . , 2004; Sawada et al . , 2006 ) . Moreover , SOCS6 is specifically recruited to the leading edge adhesions by binding to pYCas , and our optogenetics and inhibitor experiments imply that CRL5SOCS6 and the proteasome act at the FAs to counteract pYCas and to slow disassembly . Towards the rear , Cas is less highly phosphorylated , there is less SOCS6 , and disassembly is slow regardless of SOCS6 or Cul5 activity . This suggests that the regulation of SOCS6 localization is critical for selective stabilization of FAs at the leading edge as cells migrate forwards . Inhibition of SOCS6-CRL5 did not induce FA disassembly in cells that lack MTs . This suggests that MTs trigger disassembly by a pathway that is inhibited by SFKs and Cas . MT-induced FA disassembly requires the MT motor kinesin-1 , presumably to mediate anterograde traffic along MTs towards the FA ( Krylyshkina et al . , 2002 ) . Indeed , MTs deliver a protein complex including IQSEC1 , an activator of the Arf6 small GTPase , to FAs ( Yue et al . , 2014 ) . Arf6 , together with clathrin , dynein , and clathrin adaptors such as Dab2 and Numb , stimulates integrin β1 endocytosis and MT-dependent adhesion turnover in keratinocytes and HeLa cells ( Chao et al . , 2009; Ezratty et al . , 2009; Ezratty et al . , 2005; Nishimura et al . , 2007; Yue et al . , 2014 ) . This suggests that MTs deliver endocytic machinery and that integrin endocytosis stimulates FA disassembly . MTs are also involved in weakening the integrin-ECM connection by delivering matrix metalloproteases to FAs ( Stehbens et al . , 2014 ) . However , our evidence suggests that MT-dependent integrin endocytosis is not involved in the SFK-Cas FA disassembly pathway . Clathrin-coated pits are absent from the ventral surface of the leading edge of motile glioblastoma cells ( Kural et al . , 2015 ) and were absent from the long leading lamellipodium of Cul5-deficient MCF10A cells . Moreover , clathrin knockdown inhibited the slow FA disassembly in normal HeLa cells but not the rapid FA disassembly in Cul5-deficient HeLa cells . Therefore , the MT-dependent signal for FA disassembly in Cul5-deficient cells is unlikely to involve clathrin-dependent endocytosis . The exact nature of the MT trigger for Cas-CRL5SOCS6 regulated disassembly remains to be determined . Other studies have implicated ubiquitination in cell migration and FA dynamics . RING protein XRNF185 , which destabilizes paxillin via the proteasome , stimulates mesodermal cell migration during convergent extension in Xenopus embryos ( Iioka et al . , 2007 ) . It is not clear whether paxillin ubiquitination occurs locally in this system . On the other hand , paxillin ubiquitination is also stimulated by ubiquitin ligase RNF5 , but RNF5 does not induce paxillin degradation and it slows , rather than stimulates , cell migration ( Didier et al . , 2003 ) . Other authors reported that TRIM5 stimulates paxillin turnover and increases FA disassembly . This is the opposite of the effect of CRL5SOCS6 in inhibiting disassembly ( Uchil et al . , 2014 ) . Ubiquitination may also be involved in an autophagy ( ATG ) pathway for FA disassembly ( Kenific et al . , 2016 ) . Inhibiting autophagy slows FA assembly and disassembly and increases FA lifetime and size . While it is unclear how autophagy stimulates FA assembly , the evidence suggests that an ATG cargo receptor , NBR1 , recruits autophagosomes to FAs to stimulate FA disassembly . Interestingly , NBR1 contains a ubiquitin-binding domain , suggesting that ubiquitination may localize NBR1 to FAs during disassembly . Perhaps our most surprising finding is that SOCS6 needs to be present in FAs , and cullins and the proteasome need to be active , to regulate FA dynamics . This implies that the localized ubiquitination and degradation of pYCas regulates FA disassembly . Localized degradation of talin by the protease calpain also triggers FA disassembly ( Bhatt et al . , 2002; Franco et al . , 2004 ) . However , calpain is an endopeptidase that creates specific cleavage products , and the talin fragments remain in the FA and regulate FA stability ( Huang et al . , 2009 ) . In contrast , ubiquitin-dependent proteolysis typically requires translocation of the ubiquitinated protein into the proteasome ( Ravid et al . , 2008 ) . Removal of Cas by this mechanism would be expected to open up sites in the FA where other Cas molecules could bind . It is known that most of the Cas in a FA exchanges with Cas from the cytosol with a very rapid ( ~20 s ) rate constant ( Janostiak et al . , 2011; Machiyama et al . , 2014 ) . It seems remarkable that ubiquitination and degradation of pYCas would affect the kinetics of FA disassembly if new Cas molecules could replace degraded molecules with such speed . New approaches will be required to measure the exact sequence of events – Cas binding , phosphorylation , SOCS6 binding , ubiquitination and proteolysis – in FAs during cell migration .
PCR amplification utilized proof-reading Phusion or Herculase enzymes and PCR amplified regions and junctions were sequenced . pMSCVpuroEYFP-vinculin was made by amplifying murine vinculin with NotI and EcoRI primers from pLenti-H1_CAG_EYFP_C2_mVinculin ( Antoku et al . , 2008 ) and insertion into pMSCVpuroEYFP-C1 . T7-tagged SOCS and Cul5 constructs have been previously described ( Simo et al . , 2013; Teckchandani et al . , 2014 ) . LCQQ and R407K mutants were made by Quikchange mutagenesis ( Agilent ) . pCAGHAmCas and pSGTSrcYF have been described previously ( Arnaud et al . , 2003; Teckchandani et al . , 2014 ) . pLenti CAG EYFPN3 EB1 is an unpublished construct kindly provided by Susumu Antoku . Cas WT , 15F and FF mutants were described previously ( Teckchandani et al . , 2014 ) and moved into pCAG-T7 by Carissa Pilling . pmChS6mito was made as follows . Murine SOCS6 was amplified with BamHI and EcoRI primers from pCAGT7SOCS6 ( Simo et al . , 2013 ) and cloned into Bgl2-EcoRI cut pmChXmito ( CAG enhancer/promoter ) kindly provided by Russell McConnell and Frank Gertler ( Bear et al . , 2000 ) . The 47 aa mitochondrial targeting sequence is derived from the ActA protein of Listeria monocytogenes . To make pmChS6 , pmChS6mito was cut with EcoRI and a self-complementary oligo , AATTGATGACGTCATC , inserted , abolishing the EcoRI site and adding tandem TGA stop codons . pBabePuromChSOCS6 was made by amplifying mChXS6 with BamHI and EcoRI primers and insertion into pBabePuro ( Morgenstern et al . , 1990 ) . The LCQQ mutant was made by Quikchange mutagenesis ( Agilent ) . pCRY2mChS6 and pCIBNGFPmito were made as follows: pCRY2PHR-mCherryN1 and pCIBN ( deltaNLS ) -pmGFP were obtained from Addgene ( Hughes et al . , 2012 ) . pCRY2PHR-mCherryN1 was cut with NheI and BsrGI and CRY2PHR-mCherryN1 cloned into pmChS6 which had been cut similarly , removing mCh but leaving S6 and the pCAG backbone , to create pCRY2mChS6 . pCIBN ( deltaNLS ) -pmGFP was cut with NheI and AgeI and CIBN ( deltaNLS ) cloned into pGFPXmito ( Bear et al . , 2000 ) which had been cut similarly , creating pCIBNGFPmito . MCF10A cells were cultured in DMEM/F12 ( Thermo Fisher Scientific ) supplemented with 5% horse serum ( Thermo Fisher Scientific ) , 20 ng/ml EGF ( Thermo Fisher Scientific ) , 0 . 5 µg/ml hydrocortisone ( Sigma-Aldrich , St . Louis , MO ) , 0 . 1 µg/ml cholera toxin ( EMD Millipore , Billerica , MA ) , 10 µg/ml insulin , and penicillin/streptomycin both at 100 U/mL ( Thermo Fisher Scientific ) . MCF10A cells stably expressing Cul5 shRNA , Cas shRNA , pMXpuroII or pLXSH empty vectors have been previously described ( Teckchandani et al . , 2014 ) . MCF10A cells stably expressing EYFP-vinculin from the MSCV promoter were prepared by retrovirus infection with pMSCVpuroEYFP-vinculin and selected with puromycin . MCF10A cells stably knocked down for Cas , Cul5 or both were infected with pLenti-H1_CAG_EYFP_C2_mVinculin and selected by FACS . MCF10A cells stably expressing mCherry-SOCS6 wildtype ( WT ) or LCQQ were prepared similarly by retrovirus infection with pBabePuromChSOCS6 constructs and selected with puromycin . Recombinant retroviruses were packaged using HEK 293 T cells , and infection carried out by standard protocols . HeLa cells were cultured in DMEM supplemented with 10% FBS and penicillin/streptomycin both at 100 U/ml . Recombinant retroviruses containing pMXpuroII shScrm or shCul5 were packaged using HEK 293 T cells , and infections were carried out by standard protocols . After selection with 2 µg/ml puromycin cells were maintained in the presence of antibiotic . The identities of the MCF10A and HeLa cells were confirmed by STR DNA profiling . Mycoplasma testing showed that the MCF10A cells were mycoplasma free while the HeLa cells had low level contamination that could be reduced with Primocin ( 100 µg/ml , two weeks , InvivoGen , San Diego , CA ) . Primocin-treated Hela cells showed the same Cul5-regulated FA disassembly as untreated cells ( compare Figure 4—figure supplement 2band Figure 2 ) , suggesting that low level mycoplasma contamination did not affect our results The following antibodies were used: mouse anti-paxillin ( BD Biosciences , San Jose , CA ) , rabbit anti-Cas , rabbit anti-FAK , rabbit anti-GAPDH and mouse anti-tubulin ( Santa Cruz Biotechnology , Inc . , Santa Cruz , CA ) , mouse anti-vinculin ( Sigma-Aldrich ) , rabbit anti-phospho-p130 Cas ( pTyr165 ) ( Cell Signaling Technology , Beverly , MA ) ; mouse anti-T7 ( EMD Millipore ) , rat anti-tyr-tubulin and mouse anti-clathrin ( Abcam , Cambridge , MA ) , mouse anti-HA ( HA . 11 ) ( BioLegend , Dedham , MA ) and rabbit anti-HA ( Bethyl , Montgomery , TX ) . Mouse monoclonal to mCherry was courtesy of Ben Hoffstrom ( FHCRC antibody facility ) and Jihong Bai- . Target sequences and sources of siRNAs and shRNAs are shown in Table 2 . For siRNA transfection , EYFP vinculin-expressing MCF10A cells were resuspended in growth media and added directly to wells containing 50 pmol pooled siRNA oligonucleotides and Lipofectamine2000 ( Thermo Fisher Scientific ) on days 1 and 3 for analysis on day 5 . Day three transfections were done in uncoated 35 mm fluorodishes ( glass thickness 0 . 17 mm ) ( World Precision Instruments , Sarasota , FL ) . 10 . 7554/eLife . 17440 . 024Table 2 . Target sequences and sources of si and shRNA constructs . DOI: http://dx . doi . org/10 . 7554/eLife . 17440 . 024ReagentTargetSourceCul5 shRNA5'-GCTGCAGACTGAATTAGTAG-3' ( Teckchandani et al . , 2014 ) Cul5 siRNA pool5'-GACACGACGTCTTATATTA-3' 5'-CGTCTAATCTGTTAAAGAA-3' 5'-GATGATACGGCTTTGCTAA-3' 5'-GTTCAACTACGAATACTAA-3'GE Dharmacon , Lafayette , COCas shRNA5'-GGTCGACAGTGGTGTGTA-3' ( Teckchandani et al . , 2014 ) Cas siRNA pool5'-AAGCAGTTTGAACGACTGGA-3' 5'-CTGGATGGAGGACTATGACTA-3' 5'-CCAGGAATCTGTATATATTTA-3' 5'-CAACCTGACCACACTGACCAA-3'*QiagenSOCS6 siRNA pool5'-CAGCTGCGATATCAACGGTGA-3' 5'-TAGAATCGTGAATTGACATAA-3' 5'-CGGGTACAAATTGGCATAACA-3' 5'-TTGATCTAATTGAGCATTCAA-3'QiagenSOCS6 siRNA pool ( alternate ) †5'-GAACATGTGCCTGTCGTTA-3' 5'-GAAAGTATGCGCTGTCATT-3' 5'-TTTAAGCTTGAGCTTTCGCTC-3'GE Dharmacon Thermo Fisher ScientificCHC siRNA pool5'-GAAAGAATCTGTAGAGAAATT-3' 5'-GCAATGAGCTGTTTGAAGATT-3' 5'-TGACAAAGGTGGATAAATTTT-3' 5'-GGAAATGGATCTCTTTGAATT-3'GE DharmaconsiConsh Scrm5'-AATTCTCCGAACGTGTCACGT-3' 5'-TCGAGCGAGGGCGACTTAACC-3Qiagen this paper*Also targets mouse Cas . Not used in rescue experiments†Used in Figure 4—figure supplement 1 . Knockdown experiments in HeLa cells were performed as described ( Teckchandani et al . , 2014 ) . Briefly , cells were transfected with 50 pmol pooled siRNA oligonucleotides using Oligofectamine ( Thermo Fisher Scientific ) on days 1 and 3 , transferred to collagen IV-coated glass coverslips on day four and analyzed on day 5 . For rescue and light-activation experiments , DNA was transfected on day 4 , cells were transferred to coated coverslips on day five and analyzed approximately 8 hr later . HeLa cells were grown in 6-well plates to near confluence and transfected with Lipofectamine 2000 ( Thermo Fisher Scientific ) . Cells were transferred to collagen IV-coated coverslips or fluorodishes the next day and imaged 24 hr later . MCF10A cells were grown to confluence in uncoated fluorodishes or glass coverslips . Cell monolayers were starved overnight in the assay medium ( DMEM/F12 with 2% horse serum , 0 . 5 µg/ml hydrocortisone , 0 . 1 µg/ml cholera toxin , 10 µg/ml insulin and 1 ng/ml Mitomycin C ( Sigma-Aldrich ) ) , wounded by scratching the surface with a P200 micropipette tip , and the medium replaced with fresh assay medium . For live cell imaging EYFP vinculin-expressing MCF10A cells were recorded every 2 min for up to 180 min , starting approximately 6 hr after wounding . If required , 5 µM MLN4924 ( Active Biochem , Maplewood , NJ ) was added 2 hr before recording . A 100×/1 . 49 CFI Apo TIRF oil immersion objective ( pixel size 0 . 16 µm ) on a Nikon Ti fully automated inverted microscope equipped with Perfect Focus , and a stage top incubator with temperature and CO2 control . Images were recorded using an Andor iXon X3 EMCCD camera . Images were acquired using the Nikon NIS Elements software and stacks were assembled using ImageJ . Slight drifts were corrected using the ImageJ registration tool Stackreg . A line was drawn within ~6 µm of the front separating the cell into ‘front’ and ‘back’ ( Figure 1—figure supplement 1b ) . Rainbow color representations were prepared using Image/Hyperstacks/Temporal-Color Code option in Fiji ( http:// fiji . sc ) . To detect mChSOCS6 in the front of migrating cells , 5 µM MLN4924 was added at the time of wounding , removed after 6 hr and cells were fixed 2 hr later . To detect pYCas in the front of migrating cells , cells were fixed 6 hr after wounding . Fixed cells were visualized using either TIRF or a 100× NA 1 . 4 oil objective ( pixel size 0 . 064 µm ) on a DeltaVision IX71 microscope ( Olympus ) equipped with an HQ2 CCD camera ( Olympus ) . Images were acquired and deconvolved using SoftWorx ( Applied Precision ) . Deconvolved images from single planes corresponding to the ventral surfaces of the cells or flattened z projections were analyzed using ImageJ ( National Institutes of Health ) . FA disassembly experiments were performed as described by ( Ezratty et al . , 2005 ) with the following changes . Serum-starved HeLa cells grown at sub-confluent density ( ~50% ) on glass coverslips or fluorodishes coated with collagen IV ( 2 µg/ml ) were treated with 4 µg/ml nocodazole ( Sigma-Aldrich ) in DMEM with 0 . 5% BSA and 20 mM HEPES ( pH 7 . 1 ) for 3 hr to completely depolymerize MTs . After three washes in PBS , cells were left in DMEM with 0 . 5% BSA and 20 mM HEPES to allow MT regrowth . 10 µM SU6656 ( Sugen , San Francisco , CA ) was added with Nocodazole . 5 µM MLN4924 and 10 µM epoxomicin ( Sigma-Aldrich ) were added 1 hr before and during washout . To score FA disassembly , cells were visualized using a 100× NA 1 . 4 oil or 60× NA 1 . 42 oil objective ( pixel size 0 . 11 µm ) on a DeltaVision IX71 microscope ( Olympus ) as described above . FAs below 0 . 5 μm2 were not distinguishable from noise and were excluded . Cells lacking FAs and prominent stress fibers were scored as percent of total cells . All figures , except Figure 2—figure supplement 2 , show data from three biologically independent experiments . In every experiment , 20–30 cells were analyzed for each condition . For live imaging of EYFPEB1 and mChSOCS6 during nocodazole treatment and washout , 100×/1 . 49 CFI Apo TIRF oil immersion objective , Perfect Focus , and a stage top incubator with temperature and CO2 control were used . Images were acquired using the Nikon NIS Elements software . To detect SOCS proteins in FAs , HeLa cells were transfected with T7-SOCS2 , T7-SOCS6 or T7 vector , seeded and serum-starved as described above , treated with 4 µg/ml nocodazole in DMEM with 0 . 5% FBS and 20 mM HEPES ( pH 7 . 1 ) for 3 hr , fixed and imaged using a 100×/1 . 49 CFI Apo TIRF oil immersion objective in both TIRF and wide-field modes . SOCS6 was localized in migrating MCF10A cells using cells stably expressing pBabePuromChSOCS6WT . HeLa cells were transfected with siSOCS6 on days 1 and 3 as described above . On day 4 , CRY2mChS6 and CIBNGFPmito were transfected using Lipofectamine2000 . On day 5 , cells were plated on collagen IV-coated fluorodishes . Approximately- 8 hr later they were treated with 4 µg/ml nocodazole for 3 hr in the dark . Washout was either done in the dark or cells were flashed with blue light 30 min before and during nocodazole washout . Blue light illumination was performed using the apparatus described ( Hughes et al . , 2012 ) , with six , 350 mW Royal Blue LEDs positioned 7 cm above a 6-well plate and a duty cycle of 50 ms on , 12 s off . Cells were fixed in the dark under red LED illumination . To confirm that SOCS6 was removed from FAs after blue light illumination , cells were imaged using a 100×/1 . 49 CFI Apo TIRF oil immersion objective . To score FA disassembly and image SOCS6 in mitochondria cells were visualized using a 100× NA 1 . 4 oil oil objective on a DeltaVision IX71 microscope . Cells were fixed in formalin at 25°C for 20 min or in methanol for 5 min at −20°C and rehydrated in TBS to visualize microtubules . After permeabilizing with 0 . 1% Triton X-100 in PBS for 5 min at 25°C , cells were washed in PBS and blocked for 1 hr in 5% normal goat serum/2% BSA in PBS before primary antibody was added for either for 3–4 hr at 25°C or overnight at 4°C . Coverslips were rinsed in PBS before the addition of Alexa Fluor 350- , Alexa Fluor 488- , Alexa Fluor 568- or Alexa Fluor 647-conjugated secondary antibodies , diluted 1:1000 ( for deconvolution microscopy ) or 1:500 ( for TIRF microscopy ) , for 1 hr at 25°C . Alexa Fluor-tagged phalloidin was used to visualize actin . After several PBS rinses , coverslips were mounted in ProLong Gold solution ( for deconvolution microscopy ) or left in PBS ( for TIRF microscopy ) . To determine FA assembly and disassembly rates , ‘front’ and ‘back’ time-lapse movies were uploaded on to the focal adhesion analysis server ( FAAS ) ( Berginski et al . , 2011 ) ( http:// faas . bme . unc . edu/ ) . The first frame of every movie was used for thresholding . The server returned visualizations showing each frame with every adhesion numbered and outlined ( Figure 1—figure supplement 1c ) , as well as images of individual FAs tracked through time ( Figure 1—figure supplement 1d ) . These visualizations were used to verify that the adhesions were correctly detected and tracked . To qualify for analysis , an adhesion had to be detected in at least five sequential frames ( 10 min ) and had to be larger than 0 . 05 µm2 . Once these parameters were set up , movies were submitted for automated analysis . The server calculated the mean EYFP-vinculin intensity for each FA through time , plotted intensity against time and automatically fitted linear models to the log-transformed time series of intensity values to calculate assembly/disassembly rate constants . FAs with P values less than 0 . 05 were omitted from the analysis . For each condition , many focal adhesions were measured ( Table 1 ) . The median rate constants in each of 4–6 experiments were determined and reported . RNA was extracted and cDNA synthesized . The abundance of Cul5 and SOCS6 RNA was measured by qPCR using QuantiTect SYBR green PCR kit ( Qiagen ) , the 7900HT Real Time PCR System and SDS software ( Applied Biosystems ) . The following primers were used: Cul5 forward , 5’-TTTTATGCGCCCGATTGTTTTG-3’ Cul5 reverse 5’-TTGCTGGGCCTTTATCATCCC-3’ SOCS6 forward 5’-ATCACGGAGCTATTGTCTGGA-3’ SOCS6 reverse 5’-CTGACTCTCATCCTCGGGGA-3’ GusB forward 5’-AGCGTGGAGCAAGA-3’ GusB reverse 5’-ATACAGATAGGCAG-3’ pmChS6mito and control plasmids were co-transfected with pSGTSrcYF and pCAGHACas or empty vector , into HeLa cells with Lipofectamine 2000 ( Invitrogen ) . Cells were stimulated with 2 mM pervanadate for 30 min before lysis in TX100 lysis buffer . Samples were immunoprecipitated with rabbit anti-HA antibodies and protein A beads . Western blots were probed with mouse anti-HA or anti-mCherry . In parallel , pmChS6mito and control plasmids were co-transfected with pCAGT7Cul5KR or empty vector into HeLa cells with Lipofectamine 2000 . Cells were lysed and samples were immunoprecipitated with mouse anti-T7 antibody and protein A beads . Blots were probed with mouse anti-T7 or mouse anti-mCherry . | Animal cells can move in the body , for example to heal a wound , by protruding a leading edge forwards , attaching it to the surroundings and then pulling against these new attachments while disassembling the older ones . Mechanical forces regulate the assembly and disassembly of these attachments , known as focal adhesions , and so do signals from outside the cell that are transmitted to the adhesions via specialized proteins . However , it was not clear how the assembly and disassembly of adhesions is coordinated . CRL5 is a ubiquitin ligase , an enzyme that can mark other proteins for destruction . Cells migrate more quickly if CRL5 is inhibited , and so Teckchandani and Cooper set out to uncover whether CRL5 affects the assembly and disassembly of focal adhesions . The experiments showed that human cells lacking a crucial component of the CRL5 complex , SOCS6 , disassemble adhesions faster than normal cells , but only at their leading edge and not at the rear . Teckchandani and Cooper also found that SOCS6 localizes to the leading edge by binding to a focal adhesion protein called Cas . Shortly after the attachments assemble , the Cas protein becomes tagged with a phosphate group and then acts to promote the adhesion to disassemble . Further experiments indicated that Cas was marked by the CRL5 complex and possibly destroyed while in or very close to the leading edge adhesions , slowing their disassembly . Together , these findings suggest that by binding Cas , SOCS6 regulates the turnover of adhesions , specifically by inhibiting disassembly and allowing adhesions to grow at the leading edge . Since SOCS6 is not present in adhesions outside of the leading edge , this may help explain how the older adhesions are disassembled . Future studies could next focus on the exact sequence of events that occur in focal adhesions after the CRL5 complex binds to Cas as the cell migrates . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2016 | The ubiquitin-proteasome system regulates focal adhesions at the leading edge of migrating cells |
Phospholipid membranes form cellular barriers but need to be flexible enough to divide by fission . Phospholipids generally contain a saturated fatty acid ( FA ) at position sn1 whereas the sn2-FA is saturated , monounsaturated or polyunsaturated . Our understanding of the impact of phospholipid unsaturation on membrane flexibility and fission is fragmentary . Here , we provide a comprehensive view of the effects of the FA profile of phospholipids on membrane vesiculation by dynamin and endophilin . Coupled to simulations , this analysis indicates that: ( i ) phospholipids with two polyunsaturated FAs make membranes prone to vesiculation but highly permeable; ( ii ) asymmetric sn1-saturated-sn2-polyunsaturated phospholipids provide a tradeoff between efficient membrane vesiculation and low membrane permeability; ( iii ) When incorporated into phospholipids , docosahexaenoic acid ( DHA; omega-3 ) makes membranes more deformable than arachidonic acid ( omega-6 ) . These results suggest an explanation for the abundance of sn1-saturated-sn2-DHA phospholipids in synaptic membranes and for the importance of the omega-6/omega-3 ratio on neuronal functions .
Although it is common knowledge that polyunsaturated fatty acids ( PUFAs ) especially omega-3 FAs are important for health , the underlying mechanisms are not fully understood ( Bazinet and Layé , 2014; Marszalek and Lodish , 2005; Stillwell and Wassall , 2003 ) . PUFAs act through three different states: as free molecules , as precursors of biological mediators , or as esters in membrane phospholipids . The third form results from the activity of acyl transferases , which selectively incorporate defined fatty acids into phospholipids ( Harayama et al . , 2014; Shindou et al . , 2013 ) . This allows cells to control the acyl chain profile of their phospholipids , which varies tremendously among organisms , tissues and cells , and even among organelles ( Harayama et al . , 2014; Hulbert , 2003; Shindou et al . , 2013 ) . Interestingly , the FA diversity in phospholipids applies mostly to the sn2 position of the glycerol backbone , hence resulting in asymmetric phospholipids containing a saturated FA at position sn1 and an unsaturated FA at position sn2 ( Hanahan et al . , 1960; Lands and Merkl , 1963; Tattrie , 1959; Yabuuchi and O'Brien , 1968 ) . For example , the brain is enriched in phospholipids with PUFAs , notably in the case of phosphatidylethanolamine ( PE ) and phosphatidylserine ( PS ) ( Tam and Innis , 2006; Yabuuchi and O'Brien , 1968 ) . Moreover , an interesting pattern has been detected in neurons , where the axon tip is enriched in phosphatidylcholine ( PC ) molecules containing arachidonate ( AA or 20:4 omega-6 ) or docosahexaenoate ( DHA or 22:6 omega-3 ) at the expense of less unsaturated PC species ( Yang et al . , 2012 ) . Thus , phospholipids with PUFAs are found at very high concentration in synaptic vesicles , where they account for up to 70 mol% of the phospholipid pool ( Takamori et al . , 2006 ) . Retinal discs also show very high concentrations of phospholipids containing PUFAs ( Boesze-Battaglia and Schimmel , 1997; Rice et al . , 2015 ) . These striking enrichments suggest that the fatty acyl chain profile of phospholipids could impact on the properties of cellular membranes . We previously showed that phospholipids with the sn2 PUFA DHA facilitate the membrane shaping and fission activities of dynamin and endophilin ( Pinot et al . , 2014 ) . These proteins are involved in the formation of endocytic vesicles by assembling into spirals around the neck of membrane buds ( Antonny et al . , 2016; Boucrot et al . , 2015; Farsad et al . , 2001; Slepnev and De Camilli , 2000; Sundborger et al . , 2011 ) . Physical manipulations , molecular dynamics simulations and biochemical measurements revealed that DHA-containing phospholipids decrease membrane-bending rigidity by adapting their conformation to membrane curvature , hence providing an advantage for membrane deformation and fission by the dynamin endophilin complex ( Pinot et al . , 2014 ) in contrast to more rigid membranes , which are less prone to fission ( Morlot et al . , 2012 ) . More generally , the flexibility of polyunsaturated phospholipids along the membrane normal ( z direction ) might soften various mechanical stresses in the membrane ( Barelli and Antonny , 2016 ) . The activity of dynamin and endophilin was previously determined on extreme membrane compositions: the phospholipids were either saturated-monounsaturated ( 16:0-18:1 ) or saturated-DHA ( 18:0-22:6 ? 3 ) ( Pinot et al . , 2014 ) . However , others PUFAs are found in phospholipids ( Harayama et al . , 2014; Tam and Innis , 2006; Yabuuchi and O'Brien , 1968 ) . The most common are 18:2 omega-6 ( linoleate ) , 18:3 omega-3 ( linolenate ) , and 20:4 omega-6 ( arachidonate ) , which differ by the number of double bonds and their position along the chain ( Figure 1A ) . Here , we present a comprehensive study of the impact of phospholipid unsaturation on the mechanical activities of dynamin and endophilin where we varied both the degree of FA unsaturation and the combination of FAs at position sn1 and sn2 of phospholipids considering that most natural phospholipids have an asymmetric FA distribution ( Hanahan et al . , 1960; Lands and Merkl , 1963; Tattrie , 1959; Yabuuchi and O'Brien , 1968 ) . The analysis reveals that the combination of an sn1 saturated acyl chain and an sn2 polyunsaturated acyl chain solves the conundrum between making a membrane very permissive to vesiculation while maintaining a proper control of membrane permeability .
GTP hydrolysis in the dynamin spiral occurs by a mutual nucleophilic attack between dynamin molecules from adjacent rungs ( Chappie et al . , 2011 ) . Consequently , the rate of GTP hydrolysis depends on dynamin self-assembly and , in effect , increases from negligible values in solution where dynamin is not polymerized , to rates in the range of 2 to 5 s-1 on optimal membrane templates where dynamin forms spirals ( Stowell et al . , 1999 ) . We reasoned that GTPase measurements should provide a robust , although indirect , assay to survey a comprehensive library of liposomes made of phospholipids of defined acyl chains for their permissibility to the mechanical activity of dynamin . Dynamin was purified from rat brain , which contains mostly the neuron-specific dynamin-1 isoform , which has a higher membrane curvature generating activity than dynamin-2 , the ubiquitous isoform ( Liu et al . , 2011 ) . This screen could also be performed in the presence of proteins that cooperate with dynamin ( e . g . BAR domains ) . Thereafter , the most interesting membrane parameters could be further analyzed by more direct assays of dynamin mechanical activity ( e . g . EM observations or assays with Giant Unilamellar Vesicles ( GUVs ) ) , which are difficult to standardize for large screens . This second round of analysis is important because conditions exist where dynamin readily self-assembles and undergoes fast GTP hydrolysis and yet does not efficiently promote membrane vesiculation ( Neumann and Schmid , 2013; Stowell et al . , 1999 ) . We prepared large unilamellar vesicles ( extrusion 400 nm ) made of five lipids: PC , PE , PS , phosphatidylinositol ( 4 , 5 ) bisphosphate ( PI ( 4 , 5 ) P2 ) and cholesterol ( Figure 1A ) . The relative amount of these lipids was kept constant and was chosen to be compatible with the recruitment of both dynamin , which interacts with PI ( 4 , 5 ) P2 , and of BAR domain proteins , which interacts with negatively charged lipids ( e . g . PS and PI ( 4 , 5 ) P2 ) . However , PI ( 4 , 5 ) P2 was present at low density ( 1 mol %; close to physiological values ) to amplify the need for other facilitating factors such as cooperation with endophilin and membrane flexibility . The only variable in the liposome formation was the acyl chain profile of PC , PE and PS , which accounted for 99% of total phospholipids . Using commercially available or custom lipids , we systematically changed the acyl chain profile in two ways ( Supplementary file 1 and Figure 1A ) . First , we gradually increased the length and unsaturation level of both the sn1 and sn2 acyl chains of PC , PE and PS according to the series 14:0-14:0 , 18:1-18:1 ( omega-9 ) , 18:2-18:2 ( omega-6 ) , 20:4-20:4 ( omega-6 ) , and 22:6-22:6 ( omega-3 ) . Considering that most physiological phospholipids have different acyl chains at positions sn1 and sn2 ( Hanahan et al . , 1960; Lands and Merkl , 1963; Tattrie , 1959; Yabuuchi and O'Brien , 1968 ) , we performed a second series in which we maintained a saturated ( 18:0 ) acyl chain at position sn1 and solely changed the sn2 chain ( 18:0-18:1 , 18:0-18:2 , 18:0-20:4 , 18:0-22:6 ) . These asymmetric combinations are most frequent in mammalian lipids . The parallel increase in acyl chain length and unsaturation enabled all lipid mixtures to be fluid above 20°C ( Huang , 2001 ) . The result of this comprehensive analysis is shown in Figure 1B ( typical GTPase experiments are provided in Figure 1—figure supplement 1A ) . Despite the identical composition of all liposomes in term of lipid polar head groups , the rate of GTP hydrolysis by dynamin varied up to 300-fold indicating that dynamin is very sensitive to the acyl chain content of the lipid membrane on which it acts . Two parameters emerged: acyl chain asymmetry and acyl chain unsaturation . First , the GTPase activity of dynamin increased dramatically ( x 20 ) on membranes made of symmetric diunsaturated phospholipids compared to asymmetric saturated-unsaturated phospholipids . Second , the GTPase activity of dynamin increased with the unsaturated level of phospholipids ( 18:1 < 18:2 < 20:4 < 22:6 ) , a trend that was observed in both symmetric and asymmetric phospholipid series . How to explain the spectacular effect of symmetric diunsaturated phospholipids on the GTPase activity of dynamin ? By negative staining electron microscopy , we observed that dynamin alone extensively deformed 22:6-22:6 liposomes , whereas 18:0-22:6 liposomes were largely unaffected ( Figure 1C and Figure 1—figure supplement 1B ) . The analysis was performed either in the presence of GTPγS , where dynamin self-assembles into stable spirals on membranes , or in the presence of GTP , where dynamin spirals further constrict to promote liposome vesiculation into round profiles of ≈ 20 nm in diameter . Liposomes containing phospholipids with two polyunsaturated acyl chains appeared exceptionally malleable as compared to liposomes made of phospholipids with one saturated and one polyunsatured acyl chain . Deformation of spherical liposomes is necessarily accompanied by a diminution of volume of the encapsulated solution . For example , the COPI coat is more efficient at making vesicles from liposomes that have been permeabilized with a pore-forming toxin ( Manneville et al . , 2008 ) . Because a previous study reported that membranes made of 18:2-18:2-PC or 18:3-18:3 PC showed a two to three fold higher water permeability than membranes made of 18:0-18:1-PC or 18:0-18:2-PC ( Olbrich et al . , 2000 ) , we suspected that membranes with dipolyunsaturated phospholipids might be very permissive to deformation by dynamin due to higher permeability . To assess the permeability of our artificial membranes , we combined molecular dynamics simulations with various measurements . The simulations were performed at the all-atom scale on bilayers containing 2 × 144 phospholipids with the same composition as that used in the experiments . To evaluate water permeability , we determined the number of water molecules that visit the hydrophobic part of the membrane during a period of 100 ns ( Figure 2A ) . These movements were subdivided into two classes: events in which the water molecule fully crosses the lipid bilayer; events in which the water molecule enters into the hydrophobic region of the bilayer and then exists on the same side . In saturated ( 14:0-14:0 ) membranes , only few water movements were detected . In membranes made of phospholipids with one unsaturated FA , the number of moving water molecules increased up to 10-fold with the level of acyl chain unsaturation ( 18:1 < 18:2 < 20:4 < 22:6 ) ( Figure 2A ) . Importantly , membranes with phospholipids containing two unsaturated FA showed a further 2 to 3-fold increase in the number of moving water molecules as compared to membranes with asymmetric saturated-unsaturated phospholipids . This increase occurred whatever the acyl chain considered ( e . g . 20:4-20:4 vs 18:0-20:4 or 22:6-22:6 vs 18:0-22:6 ) . If membranes with symmetric polyunsaturated phospholipids are more hydrated than membranes with asymmetric saturated-unsaturated phospholipids , this should influence the fluorescence of polarity-sensitive dyes at the membrane interface . To investigate this possibility , we used a recently synthesized push-pull pyrene ( PA ) . This probe , similarly to the popular probe Laurdan , changes its emission maximum as a function of membrane hydration and solvent relaxation , which are parameters linked to lipid order ( Niko et al . , 2016 ) . PA showed a gradual red shift in emission when the number of double bonds in the sn2 acyl chain increased ( 18:0-18:1 < 18:0-18:2 < 18:0-20:4 < 18:0-22:6 ) ( Figure 2B and Figure 2—figure supplement 1 ) . However , replacing asymmetric saturated-polyunsaturated phospholipids with dipolyunsaturated phospholipids caused a much larger red shift ( Figure 2B and Figure 2—figure supplement 1; e . g . 18:0-22:6 << 22:6-22:6 ) , suggesting that duplication of the PUFA in phospholipids dramatically increased membrane hydration . Considering the importance of maintaining ion gradients across biological membranes , we next assessed the permeability of our liposomes to the oxoanion dithionite ( [S2O4]2- ) by an NBD quenching assay . When added to liposomes , dithionite ( MW = 128 Da ) immediately quenches the fraction of ( C16:0-C16:0 ) PE-NBD that is present in the outer leaflet . This fraction is about 50% but varies depending on factors like membrane curvature or the presence of multi-lamellar liposomes ( Kamal et al . , 2009 ) . Thereafter , dithionite slowly quenches the remaining PE-NBD molecules . This process occurs either by penetration of dithionite into the liposomes or because of PE-NBD flip-flop . Previous work established that dithionite entry is about 1000 times faster than lipid flip-flop ( Armstrong et al . , 2003 ) . Therefore , the slow phase in PE-NBD quenching experiments should reflect dithionite permeability . Figure 2C shows that membrane permeability to dithionite modestly increased with the level of polyunsaturation in asymmetric phospholipids ( 18:0-18:1 < 18:0-18:2 < 18:0-20:4 < 18:0-22:6 ) and that liposomes with symmetric polyunsaturated phospholipids showed a > 10 fold higher permeability . This effect was particularly evident for 20:4-20:4 and 22:6-22:6 membranes: 95% of PE-NBD was quenched after 300 s incubation with dithionite as compared to 55–60% in the case of 20:4 and 18:0-22:6 membranes . Note that the liposomes used in the dithionite experiments were the same as that used in the dynamin experiments ( see Figure 1B ) to allow a direct comparison between the two assays . However , a drawback of liposomes obtained by extrusion through large pore size filters ( here 400 nm ) , is the presence of multi-lamellar species ( Kamal et al . , 2009 ) . For such species , dithionite has to cross several bilayers to fully quench all PE-NBD molecules . This effect probably explains why the second phase of PE-NBD quenching , although quite fast in the case of 20:4-20:4 and 22:6-22:6 liposomes , was not complete; a small percentage ( ≈ 5% ) of NBD signal remained unquenched after 300 s incubation ( Figure 2C ) . Despite these limitations , these experiments suggest that the presence of two polyunsaturated acyl chains in phospholipids strongly compromise membrane impermeability to ions . Last , we visualized the permeability of GUVs to the large fluorescent solute Alexa A488 maleimide ( MW = 720 Da ) . This compound was added externally to the GUVs , which were imaged by fluorescence microscopy ( Figure 2D ) . Again , the difference between symmetric and asymmetric polyunsaturated phospholipids was clear-cut . GUVs containing 55 mol% 22:6-22:6 phospholipids were about 10 times more permeable to Alexa A488 maleimide than GUVs containing 55 mol% 18:0-22:6 phospholipids . Altogether , these experiments revealed a remarkable correlation between the ability of dynamin alone to readily vesiculate membranes and the permeability of these membranes to water and even to large or charged solutes . If membranes with symmetric diunsaturated phospholipids appear exceptionally prone to vesiculation by dynamin , their high permeability to ions and large solutes disqualify them for the formation of selective membrane barriers . The GTPase assay of Figure 1B and the membrane permeability experiments of Figure 2 suggest that asymmetric saturated-polyunsaturated phospholipids offer a compromise between low permeability and dynamin activity . Even though the intrinsic GTPase activity of dynamin on such liposomes was low , it increased about 10 times in the presence endophilin-A1 ( Figure 1B ) . In cells , endophilin works in close partnership with dynamin both through protein/protein interactions and through the ability of the two proteins to form membrane-deforming spirals ( Boucrot et al . , 2015; Farsad et al . , 2001; Sundborger et al . , 2011 ) . The BAR domain of endophilin is followed by an SH3 domain , which interacts with the proline-rich region of dynamin leading to cooperative BAR/dynamin membrane recruitment ( Farsad et al . , 2001; Meinecke et al . , 2013; Sundborger et al . , 2011 ) . In addition , membrane deformation by BAR domains facilitates dynamin self-assembly , which by itself is a relatively weak membrane deforming protein and which preferentially self-assembles on pre-curved membranes ( Neumann and Schmid , 2013; Roux et al . , 2010 ) , unless the membranes are particularly deformable as observed with symmetric unsaturated phospholipids ( Figure 1C ) . These considerations and the fact that asymmetric polyunsaturated phospholipids are much more frequent than symmetric ones in biological membranes prompted us to focus on reconstitutions in which both dynamin and endophilin were present and acted on membranes with asymmetric phospholipids . Under such conditions , the exact nature of the sn2 acyl chain appeared very important: significant differences in the dynamin GTPase activity were observed between acyl chain combinations that are chemically very close ( e . g . 18:0-20:4 vs 18:0-22:6 ) ( Figure 1A ) . To better analyze these differences , we repeated the GTPase assay under conditions where we gradually increased the amount of asymmetric polyunsaturated phospholipids at the expense of 18:0-18:1 phospholipids ( Figure 3 ) . Two omega-3 combinations ( 18:0-18:3 and 18:0-22:6 ) and two omega-6 combinations ( 18:0-18:2 and 18:0-20:4 ) were included in the analysis to evaluate the importance of the double bond position . All polyunsaturated phospholipids facilitate dynamin GTPase activity . However , the effect of 18:0-22:6 phospholipids surpassed by 2 to 4-fold that observed with the less complex asymmetric polyunsaturated phospholipids ( 18:0-18:2 , 18:0-18:3 and 18:0-20:4 ) ( Figure 3 ) . Dynamin GTPase activity plateau at about 60 mol% of 18:0-22:6 phospholipids , close to the amount of polyunsaturated lipids in synaptic vesicles ( Takamori et al . , 2006 ) . To check that the effect of asymmetric polyunsaturated phospholipids was not restricted to our particular conditions , we modified several parameters . First , we varied the % of PI ( 4 , 5 ) P2 . In a background of 18:0-18:1 phospholipids , the GTPase activity of dynamin in the presence of endophilin was low and increased with the % of PI ( 4 , 5 ) P2 ( from 0% to 5%; Figure 3—figure supplement 1A ) . In a background of 18:0-18:2 , 18:0-18:3 or 18:0-20:4 phospholipids , the GTPase activity was much higher and required not more than 1 mol% PI ( 4 , 5 ) P2 . Strikingly , the activity of dynamin was almost maximal with 18:0-22:6 phospholipids even in the absence of PI ( 4 , 5 ) P2 ( Figure 3—figure supplement 1A ) . Next , we replaced endophilin by SNX9 , another BAR-domain containing protein that interacts with dynamin . Both endophilin and SNX9 increased dynamin activity much more efficiently on membranes containing 18:0-22:6 phospholipids or 18:0-20:4 than 18:0-18:1 phospholipids ( Figure 3—figure supplement 1B ) . In addition , SNX9 was more efficient than endophilin for assisting dynamin activity in agreement with a previous study ( Neumann and Schmid , 2013 ) . All these experiments converge towards the same conclusions . First , all asymmetric saturated-polyunsaturated phospholipids favor dynamin GTPase activity , notably under conditions close to physiological conditions ( low concentration of PI ( 4 , 5 ) P2 , low protein concentration , cooperation with BAR-domain proteins ) . Second , docosahexaenoic acid ( 22:6 ) , which is the most polyunsaturated species of the omega-3 family , surpasses all other tested species including arachidonate ( 20:4 ) , the most polyunsaturated species of the omega-6 family . 18:0-20:4 and 18:0-22:6 phospholipids are abundant in specialized membranes ( e . g . synaptic vesicles; Takamori et al . , 2006 ) . Considering the importance of the omega-6/omega-3 ratio for health , we next focused on these acyl chain combinations and used 18:0-18:1 membranes as negative control . By transmission electron microscopy ( TEM ) we observed that both 18:0-20:4 and 18:0-22:6 liposomes but not 18:0-18:1 liposomes , became extensively deformed after incubation with dynamin , endophilin and GTPγS or GTP . With GTPγS , membrane tubulation dominated ( Figure 4A ) . The tubes were surrounded by a protein spiral with a pitch of ~20 nm characteristic of the endophilin-dynamin complex ( Farsad et al . , 2001; Pinot et al . , 2014; Sundborger et al . , 2011 ) ( Figure 4B and Figure 4—figure supplement 1A ) . However , the tubes formed from 18:0-22:6 membranes were significantly thinner than the tubes formed from 18:0-20:4 membranes ( Figure 4B ) . With GTP present , liposome vesiculation dominated ( Figure 4A ) . The size distribution of the membrane profiles was different between 18:0-20:4 and 18:0-22:6 phospholipids ( Figure 4C ) . With 18:0-20:4 phospholipids , there was a remaining peak of large membrane profiles ( R > 50 nm ) , which coexisted with a peak of small vesicles ( R < 50 nm ) . With 18:0-22:6 phospholipids , the liposomes were almost fully transformed into small vesicles . In addition , the vesicles formed from 18:0-22:6 membranes were slightly smaller than that formed from 18:0-20:4 membranes ( Figure 4C ) . Note that after short incubation with GTP , some tubes were observed both with 18:0-20:4 and 18:0-22:6 liposomes ( Figure 4—figure supplement 1B ) . These tubes were not straight as with GTPγS but showed constrictions , suggesting snapshots in the process of membrane fission ( black arrows in Figure 4—figure supplement 1B ) Considering the technical limitations caused by spontaneous membrane fission on TEM grids ( Danino et al . , 2004 ) , we next performed a GUV shrinking assay ( Meinecke et al . , 2013 ) . In these experiments , dynamin , endophilin and GTP were added to GUVs containing 55 mol% of polyunsaturated phospholipids , which were pre-stabilized in buffer at the reaction temperature ( 37°C ) . Dynamin , endophilin and GTP caused GUV consumption over time for both 18:0-20:4 and 18:0-22:6 membranes ( Figure 5A ) but not for the control GUVs that contained only 18:0-18:1 phospholipids ( Figure 5—figure supplement 1A and B ) . After 1 hr of incubation , the difference between 18:0-20:4 and 18:0 22:6 phospholipids was significant , as we detected a larger population of shrunk 18:0-22:6 GUVs and a higher amount of intact 18:0-20:4 GUVs ( Figure 5B ) . The difference in GUV shrinking between 18:0-20:4 and 18:0-22:6 phospholipids was already evident after 15 min and increased over time ( Figure 5C ) . To better understand the advantage provided by asymmetric polyunsaturated phospholipids on membrane deformation and fission , we conducted MD simulations on lipid bilayers . In the coarse-grained mode , we considered large membrane patches and imposed a pulling force to deform them into a tube , which might undergo fission ( Baoukina et al . , 2012; Pinot et al . , 2014 ) ( Figure 6A and B ) . This approach is informative in terms of membrane mechanics , but the 1:4 scale of the MARTINI force field ( one elementary bead for 3 to 4 bonded atoms ) makes the depiction of the chemistry of polyunsaturated phospholipids quite imprecise . Nevertheless , we could construct PC bilayers in which acyl chains made of 4 or five beads approximate the series 18:0-18:1 , 18:0-18:2 , 18:0-20:4 and 18:0-22:6 ( Figure 6A and B ) . In the all-atom mode , we considered membrane patches of 2 × 144 phospholipids with the same composition as that used in the experiments . This approach is limited to flat membranes but enables an accurate description of the slight chemical differences between polyunsaturated acyl chains ( Figure 6C–E ) . For all coarse-grained membranes tested , applying a constant force above a threshold of 175 kJ mol−1 nm−1 induced the formation of a tube , which grew by a fast protrusion phase followed by a linear phase as previously observed ( Baoukina et al . , 2012 ) . Increasing the degree of phospholipid polyunsaturation ( 18:0-18:1 < 18:0-18:2 < 18:0-20:4 < 18:0-22:6 ) accelerated the linear phase and resulted in the formation of longer and thinner tubes ( Figure 6A and Figure 6—figure supplement 1A and B ) . Because the bending energy of a membrane tube is proportional to the ratio between tube length and radius ( Eb = πΚbL/R ) , we plotted L/R as a function of the applied force ( Figure 6A ) . At t = 200 ns and for F = 200 KJ mol−1 nm−1 , L/R = 6 , 12 , 16 and 20 nm/nm for 18:0-18:1 , 18:0-18:2 , 18:0-20:4 and 18:0-22:6 tubes , respectively . Considering that these tubes should have stored the same curvature energy , these changes in L/R suggested inverse changes in membrane bending rigidity: 18:0-18:2 , 18:0-20:4 and 18:0-22:6 membranes had relative values of Κb equal to 50% , 35% and 30% of Κb for 18:0-18:1 membranes , respectively . During the time of the simulations ( 200 ns ) , we observed fission events for some tubes formed from 18:0-20:4 and 18:0-22:6 membranes but not from 18:0-18:1 or 18:0-18:2 membranes ( Figure 6B , Figure 6—figure supplement 1A and Video 1 ) . Although the number of simulations did not allow us to establish robust statistics , we noticed that the force threshold at which fission occurred was lower for 18:0-22:6 membranes than for 18:0-20:4 membranes ( Figure 6B and Figure 6—figure supplement 1B ) . Moreover , fission occurred sooner for 18:0-22:6 tubes as compared to 18:0-20:4 tubes . Thus , the coarse-grained simulations agreed well with the experiments: the propensity of membranes to undergo deformation and fission correlates with the unsaturation level of the phospholipid sn2 acyl chain . For all-atom bilayers , we focused on parameters informative for the tendency of the phospholipid acyl chains to depart from the straight conformation . This tendency allows phospholipids to adopt different shapes and , consequently , to reduce the stress induced by membrane curvature ( Pinot et al . , 2014 ) . We determined ( i ) the speed at which the terminal CH3 group moves along the membrane normal ( z velocity ) vs membrane plane ( x velocity ) ; ( ii ) the frequency of FA torsions ( when the acyl chain displays an angle <100° ) , ( iii ) the number of protrusions of the terminal CH3 group above the glycerol during 100 ns; and ( iv ) the density of lipid packing defects , that is interfacial regions where aliphatic carbons are directly accessible to the solvent . Figure 6C–E and Figure 6—figure supplement 2 show that whatever the parameter considered , the calculated value always increased with the polyunsaturation level of the sn2 chain , with 22:6 clearly surpassing all other polyunsaturated FAs . In contrast , the behavior of the sn1 18:0 chain was relatively constant and appeared poorly dependent on the nature of the neighboring sn2 chain . Altogether , these various analyses show that the main effect of having an sn2 polyunsaturated chain in phospholipids is to increase the probability of fast movements of along the z-axis . To determine whether the membrane features endowed by the polyunsaturated acyl chain depend on its esterification at position sn2 as observed in natural lipids , we performed molecular dynamics simulations on phospholipid bilayers in which we swapped the sn1 and sn2 acyl chains ( Figure 6—figure supplement 3 ) . In all atom simulations , we observed that the rough features distinguishing the saturated and the polyunsaturated acyl chains remained after having permutated their position . These include z velocity , acyl chain torsions , and number of protrusions ( Figure 6—figure supplement 3A ) . However , measurements of the acyl chain density across the bilayer indicated that acyl chain swapping modified the mean position of the saturated and polyunsaturated acyl chains across the bilayer ( Figure 6—figure supplement 3B ) . This effect probably resulted from the tilted orientation of glycerol , which makes the sn1 and sn2 positions not equivalent in term of z coordinates . In natural phospholipids , the density profile of the sn1 saturated FA showed a peak in the bilayer center whereas the sn2 polyunsaturated FA showed a characteristic dip . This shift indicates that the sn1 saturated FA tail invaded the central region of the bilayer left vacant by the sn2 polyunsaturated FA , which goes up ( Eldho et al . , 2003 ) . With swapped 22:6-18:0 phospholipids , this difference in density disappeared and the membrane appeared thinner than with natural 18:0-22:6 phospholipids ( Figure 6—figure supplement 3B ) . Thus , the relative esterification position of the saturated and polyunsaturated FAs in natural lipids facilitates compensatory z movements where the polyunsaturated FA explores the interfacial region while the saturated FA explores the bilayer center . In coarse-grained simulations , the propensity of the membrane to undergo tubulation and fission increased with the level of phospholipid polyunsaturation ( 18:1-18:0 < 20:4-18:0 ≈ 22:6-18:0 ) , that is the same trend as that observed on classical sn1-saturated-sn2-unsaturated membranes ( Figure 6—figure supplement 3C ) . In addition , acyl chain swapping did not significantly modify membrane fission ( Figure 6—figure supplement 3C and Video 2 ) .
Although a wealth of information is available on the interactions between endocytic proteins and specific lipids ( Puchkov and Haucke , 2013 ) , the role of the hydrophobic membrane matrix has been poorly investigated . In vivo , manipulating the acyltransferases that are responsible for the large differences in the acyl chain profile of differentiated cells is challenging and is just starting to emerge ( Hashidate-Yoshida et al . , 2015; Rong et al . , 2015 ) . In vitro , purified lipids are generally available from disparate sources ( e . g . egg PC , brain PS ) , implying different acyl chain profiles . Synthetic lipids provide the best alternative but the most affordable ones generally display symmetric acyl chain combinations . This is exemplified by DOPS ( 18:1-18:1 PS ) , which has allowed spectacular advances in our understanding of the structure of the dynamin spiral ( Chappie et al . , 2011 ) , but is very rare in mammalian membranes ( Yabuuchi and O'Brien , 1968 ) . Overall , the membrane templates on which dynamin and its partners have been studied are generally ill defined in terms of acyl chain profiles . Our comprehensive analysis indicates that acyl chain asymmetry and acyl chain polyunsaturation have major effects on the mechanical activity of dynamin . A few studies have established that polyunsaturated phospholipids considerably modify the properties of membranes ( Armstrong et al . , 2003; Eldho et al . , 2003; Garcia-Manyes et al . , 2010; Huang , 2001; Olbrich et al . , 2000; Rawicz et al . , 2000 ) . For LUVs , a high dithionite permeability of membranes containing 18:3-18:3 phospholipids has been reported ( Armstrong et al . , 2003 ) . For GUVs , micropipette manipulations indicate that the presence of at least one polyunsaturated acyl chain results in a drop of the membrane bending modulus whereas the presence of two polyunsaturated acyl chains causes a jump in water permeability ( Olbrich et al . , 2000; Rawicz et al . , 2000 ) . These pioneer studies were performed on membranes made of a single lipid ( PC ) with a limited combination of acyl chains and in the absence of mechanically active proteins . By using lipid mixtures covering a larger spectrum of acyl chain profiles and by including membrane shaping/fission proteins , our study underlines the importance of both phospholipid polyunsaturation and phospholipid acyl chain asymmetry in membrane mechanics . Depending on its acyl chain profile , a membrane can be either very resistant or very permissive to dynamin-mediated membrane vesiculation despite harboring the proper repertoire of polar head groups for protein recruitment . However , these manipulations can also cause large changes in membrane permeability . Our analysis uncovers a narrow chemical window that allows phospholipid membranes to be both highly deformable and still impermeable to small solutes . Membranes with asymmetric saturated-polyunsaturated phospholipids such as 18:0-20:4 or 18:0-22:6 phospholipids are much less leaky than membranes with symmetrical 20:4-20:4 or 22:6-22:6 phospholipids but can still be readily vesiculated by dynamin provided that BAR-domain proteins are present . Evidently , these features are advantageous for membranes such as synaptic membranes that undergo super-fast endocytosis ( Watanabe and Boucrot , 2017 ) . Furthermore , the fact that membranes with 18:0-22:6 phospholipids are systematically more permissive to the mechanical activity of dynamin and endophilin than membranes with 18:0-20:4 phospholipids is of interest given the importance of the omega-6/omega-3 ratio for health and notably for brain function . The distinctive chemical feature of polyunsaturated acyl chains is the presence of saturated carbons ( CH2 ) sandwiched between two unsaturated ones ( =CH-CH2-CH= ) . Rotational freedom around these CH2 groups is exceptionally high as compared to rotation around the CH2 groups of monounsaturated or saturated acyl chains ( Feller et al . , 2002 ) . Our MD simulations indicate that motions of the acyl chain along the normal of the membrane ( z movements ) increase in speed and in amplitude with the level acyl chain polyunsaturation ( 18:2 < 18:3 < 20:4 < 22:6 ) . Such movements should allow phospholipids to readily adapt their conformation to membrane curvature ( Barelli and Antonny , 2016; Pinot et al . , 2014 ) , hence explaining the gradual decrease in membrane bending rigidity . Concurrently , the presence of a neighboring saturated acyl chain should secure lipid packing and prevents the passage of small molecules . Whether this model also accounts for the facilitation of the fission step per se remains , however , difficult to assess . This step involves a change in membrane topology for which rare events such as protrusions of the terminal CH3 groups could be decisive as they could nucleate bilayer merging or favor friction effects by proteins ( Simunovic et al . , 2017 ) . Other variations in the acyl chain content of mammalian phospholipids will deserve further investigations . First , we did not consider C22:5 acyl chains ( omega-6 or omega-3 ) , which are closer to C22:6 than C20:4 ( Eldho et al . , 2003 ) . Although not abundant , C22:5 acyl chains are present in brain phospholipids ( Yabuuchi and O'Brien , 1968 ) . Second , we did not study the influence of the linkage between the acyl chains and glycerol . Plasmalogens , which form a large subclass of PE in the brain , harbor a sn1 saturated acyl chain that is bound to the glycerol through an ether-vinyl bond . Interestingly , ether-vinyl phospholipids considerably decrease the permeability of model membranes to ions because these lipids pack more tightly than their ester counterparts ( Zeng et al . , 1998 ) . The influence of plasmalogens on membrane flexibility and fission remains to be investigated . Last , we only partially addressed the bias observed in natural phospholipids , where saturated and polyunsaturated acyl chains are preferentially esterified on different positions of the glycerol backbone ( sn1 and sn2 , respectively ) . Our simulations suggest that some general traits provided by the combination of one saturated acyl chain and one unsaturated acyl chain are preserved when the acyl chains are swapped between the sn1 and sn2 positions; notably the fact that membrane deformation and fission is facilitated by the level of polyunsaturation ( 18:1-18:0 < 20:4-18:0 ≈ 22:6-18:0 ) . Testing this hypothesis by in vitro reconstitutions will require considerable efforts in lipid synthesis since swapped phospholipids are not commercially available . The abilities to vesiculate and to act as selective barriers are two fundamental properties of cellular membranes . Without membrane vesiculation , a cell cannot divide; without selective permeability , it cannot control the concentration of its nutrients . Experiments aimed at mimicking the emergence of primitive membranes have illuminated how these properties need to be finally balanced . Single chain amphiphilic molecules ( e . g . fatty acids ) , the most plausible building blocks for primitive membranes , can self-assemble into bilayers , which spontaneously vesiculate ( Bruckner et al . , 2009 ) . However , these bilayers are very leaky to even large solutes ( in the 103 Da range ) . Later , the shift from single chain to dual chain lipids has probably allowed primitive cells to reduce the general permeability of their membrane , thereby imposing an evolutionary pressure for the emergence of specialized transporters ( Budin and Szostak , 2011 ) . The experiments presented here suggest that phospholipids with one saturated and one polyunsaturated acyl chain , which are absent in many eukaryotes ( e . g . yeast ) but abundant in some highly differentiated cells ( e . g . neurons , photoreceptors , sperm ) provide a solution to an early dilemma in evolution: finding the right balance between efficient membrane vesiculation without loss in membrane permeability . Moreover , the fact that saturated-DHA ( omega-3 ) phospholipids are systematically better for membrane vesiculation than many other saturated-polyunsaturated phospholipids , including saturated-arachidonate ( omega-6 ) , is informative considering the importance of the omega-6/omega-3 ratio for health .
Proteins were purified as described ( Pinot et al . , 2014; Stowell et al . , 1999 ) . Dynamin was purified from rat brain using a recombinant amphiphysin-2 SH3 domain as an affinity ligand . Brain extracts were incubated with 10 mg ml-1 glutathione-S-transferase-tagged amphiphysin-2 SH3 domain on glutathione–agarose beads at 4°C . After extensive washing of the matrix in buffer A ( 100 mM NaCl , 20 mM HEPES , pH 7 . 3 , 1 mM dithiothreitol ( DTT ) ) , dynamin was eluted in 3 ml buffer B ( 1 . 2 M NaCl , 20 mM HEPES , pH 6 . 5 , 1 mM DTT ) , and dialysed overnight into 200 mM NaCl , 20 mM HEPES , 20% glycerol . Full-length mouse endophilin A1 in pGEX-6p1 ( gift of A . Schmidt ) was expressed in E coli for 3 hr at 37°C after induction with 1 mM IPTG . Cells were lysed in 50 mM Tris pH 7 . 4 , 150 mM NaCl using a French press in the presence of antiproteases and spun at 40 , 000 rpm for 30 min at 4°C . The supernatant was incubated with glutathione-Sepharose 4B beads followed by extensive washes in lysis buffer . PreScission protease was directly added to the beads at 4°C overnight under gentle agitation to cleave the fusion protein . Endophilin was recovered in supernatant and further purified on a Superdex 200 column in 20 mM Tris pH 7 . 4 , 300 mM KCl , 5 mM imidazole , 1 mM DTT . Lipids were purchased from Avanti Polar Lipids as chloroform solutions ( see Key resources table ) . These included the following species of phosphatidylcholine ( PC ) , phosphatidylethanolamine ( PE ) and phosphatidylserine ( PS ) : 14:0-14:0 , 18:0-18:1 , 18:1-18:1 , 18:0-18:2 , 18:2-18:2 , 18:0-18:3 , 18:0-20:4 , 20:4-20:4 , 18:0-22:6 and 22:6-22:6 . Note that 18:0-18:3 phospholipid species were custom-made lipids from Avanti . Phosphatidylinositol ( 4 , 5 ) bisphosphate ( PIP2 ) was from natural source ( brain ) . Submicrometer liposomes used for biochemical experiments and for electron microscopy were prepared by extrusion . A lipid film containing phospholipids and cholesterol at the desired molar ratio ( see Table S1-3 in supplementary file 1 ) was formed in a rotary evaporator and hydrated at a final lipid concentration of 1 mM in a freshly degassed HK buffer ( 50 mM Hepes pH 7 . 2 , 120 mM K Acetate ) supplemented with 1 mM DTT . The suspension was submitted to five cycles of freezing and thawing and stored at −20°C under argon to avoid lipid oxidation . Calibrated liposomes were obtained by extrusion through 400 or 100 nm polycarbonate filters using a hand extruder ( Avanti Polar Lipids ) . The size distribution of the liposomes was determined by dynamic light scattering at a final concentration of 0 . 1 mM lipids in HK buffer . All liposome suspensions were used within 1–2 days after extrusion . Special care was taken to minimize lipid oxidation by using fleshly degassed buffer ( supplemented with 1 mM DTT ) and by storing the liposome suspensions under argon . Giant unilamellar vesicles were generated by electroformation as described ( Pinot et al . , 2014 ) with the following modifications . Lipid mixtures ( 0 . 5 mg/ml; see Table S4 in supplementary file 1 ) in chloroform were deposited on indium tin oxide coated glass slides at 50°C to prevent lipid de-mixing and dried under vacuum for 1 hr to remove all solvents . After this step , sucrose 250 mM osmotically equilibrated with buffers was added to the chamber . GUVs were electroformed ( Angelova et al . , 1992 ) with Vesicle Prep Pro ( Nanion Technologies GmbH , Munich , Germany ) , applying an AC electric field with 3 V and 5 Hz for 218 min at 37°C . GTP hydrolysis in dynamin was measured using a colorimetric assay ( Leonard et al . , 2005 ) . The sample ( 60 µl ) initially contained 400 nm extruded liposomes ( 0 . 1 mM ) of defined composition ( see Supplementary file 1 ) in HK buffer supplemented with 2 . 5 mM MgCl2 , 1 mM DTT and 500 µM GTP . Just before measurement , endophilin ( 0 . 6 µM ) was added and the reaction was initiated by the addition and mixing of 0 . 3 µM dynamin . At the indicated times ( 15 , 45 , 75 , 120 , 180 , 240 and 360 s ) , aliquots ( 7 . 5 µl ) were withdrawn and immediately mixed with a drop of EDTA ( 5 µl , 250 mM ) in a 96 well plate . At the end of the experiment , 150 µl of a malachite green stock solution was added to each well and the absorbance at 650 nm was measured using a microplate reader and compared to that of a standard curve of phosphate ( 0–200 µM ) in order to determine the concentration of GTP hydrolyzed by dynamin . Fluorescence spectra of the PA probe with liposomes was performed as described ( Niko et al . , 2016 ) . The sample ( 600 µl ) initially contained 0 . 1 mM extruded liposomes ( 100 nm ) of defined composition ( see Supplementary file 1 ) . After 5 min incubation of the liposomes solution with 1 µM PA probe at 37°C , a fluorescence emission spectrum ( 450–700 nm; bandwidth 1 nm ) was recorded upon excitation at 430 nm ( bandwidth 5 nm ) . All spectra were corrected for the corresponding blank ( suspension of liposomes without the probe ) . The extent of dithionite quenching of the NBD-labeled PE was performed as described ( Angeletti and Nichols , 1998 ) . Briefly , the sample ( 600 µl ) that initially contained 400 nm extruded liposomes ( 0 . 1 mM ) of defined composition ( see Supplementary file 1 ) were let equilibrating in HK buffer 5 min at 37°C with 600 rpm stirring . After 30 s of fluorescence measurements ( excitation 505 nm , bandwidth 1 nm; emission 540 nm , bandwidth 10 nm ) , NBD quenching was started by adding 10 mM dithionite and the reaction was followed during 5 min at 37°C with 600 rpm stirring . The percentage of NBD quenching was calculated by the equation: Quenching NBD ( % ) = ( Fi – F0 ) / ( FT – F0 ) x 100 where F0 corresponds to the fluorescence of the vesicles at time 0–30 s; Fi is the fluorescence after a certain period of incubation with dithionite , and FT is the maximum quenching that corresponds to the fluorescence value obtained after addition of 0 . 1% Triton X-100 . Mixtures containing liposomes , dynamin , endophilin and nucleotides were prepared in HK buffer supplemented with 2 . 5 mM MgCl2 and 1 mM DTT ( final volume 50 µl ) . For the tubulation experiments in presence of the non-hydrolyzable analog GTPγS , vesicles were incubated for 5 or 15 min at room temperature . For the fission experiments in presence of GTP , vesicles were incubated for 30 min at room temperature . Thereafter , an EM grid was put on the protein-liposome mixture for 5 min , rinsed with a droplet of 100 mM Hepes ( pH 7 . 0 ) for 1 min , and then stained with 1% uranyl acetate . The grid was observed in a JEOL JEM1400 transmission electron microscope equipped with a MORADA SIS camera . To determine the size distribution of the liposomes or of the protein-liposome profiles , 500 to 1000 profiles for each condition and from three independent experiments were analyzed using the ellipse tool of the NIH Image J software . The apparent radius was calculated as R= ( A/π ) 1/2 where A is the apparent area of the profile . All experiments were performed with 0 . 5 µM dynamin , 1 µM endophilin , 500 µM nucleotide and 0 . 1 mM lipids . GUV permeability was studied in 18:0-22:6 and 22:6-22:6 liposomes using a previously developed assay with some modifications ( Jiménez-Rojo et al . , 2014 ) . After GUVs stabilization soluble Alexa488 was externally added to follow the entrance of the probe over time . After 15 min incubation , vesicles were imaged by confocal microscopy and permeability was quantified using the following equation: Permeability ( % ) =Iin/Iex x 100 where Iin is the average of the fluorescence inside the individual GUV and Iex is the average of external fluorescence of the probe in solution . Membrane fission induced by dynamin and endophilin in the presence of GTP was followed indirectly by monitoring the size of GUVs over time since the vesicles produced by the proteins are too small to be optically resolved . We used a previously developed assay with some modifications ( Meinecke et al . , 2013 ) . After GUVs stabilization , dynamin , endophilin and GTP were added and incubated for 1 hr at 37°C before and imaging by confocal microscopy . All experiments were performed with 0 . 5 µM dynamin , 1 µM endophilin , 500 µM nucleotide and 0 . 1 mM lipids . Shrinking percentage was calculated by the equation: Shrinking ( % ) =100 x A0/Ai where A0 is the vesicles area at time 0 and Ai is vesicles area after a defined period of incubation . All-atom simulations were performed with GROMACS 5 ( Abraham et al . , 2015 ) software and CHARMM36 ( Klauda et al . , 2010 ) force field . The various systems were built with the Charmm-Gui tool ( Lee et al . , 2016 ) with 33% Cholesterol , 1% PI ( 4 , 5 ) P2 , and with 30% PS , 20% PE , and 16% PC , harboring defined acyl chains ( 18:0-18:1 , 18:0-18:2 , 18:0-18:3 , 18:0-20:4 , 18:0-22:6 , 14:0-14:0 , 18:2-18:2 , 20:4-20:4 , 22:6-22:6 ) . Lipids not present in the Charmm-Gui database ( 18:0-18:2 , 18:0-18:3 , 18:2-18-2 and 22:6-22:6 and swapped lipids ) were built by adding unsaturations to related lipids ( i . e . 18:0-18:1 , 18:1-18:1 or 22:1-22:1 ) . Note that one of this lipid ( 18:0-18:2 ) is now present in the database and has the same topology as the one used here . The bilayers contained 2 × 144 phospholipids with counter ions to neutralize the system and with 120 mM NaCl . The simulation parameters were those of Charmm-Gui under semi isotropic conditions within the NPT ensemble: x and y directions were coupled , whereas z direction was independent . Periodic boundaries applied to all directions . We first equilibrated the membranes for 200 ps using the standard Charmm-Gui six-step process during which constraints on lipids were gradually released . Next , an additional equilibration step was performed to equilibrate the TIP3P model of water . All simulations were equilibrated using the Berendsen thermostat and barostat at 303 K and 1 bar , respectively , except for 14:0-14:0 bilayers , which were equilibrated at 310 K . Lipids and water+ions were coupled separately . Production runs were performed with the V-rescale thermostat at 303 K except for 14:0-14:0 bilayers ( 310 K ) . The Parrinello-Rahman thermostat was used to stabilize the pressure at 1 bar with a time constant of 5 ps and a compressibility of 4 . 5 × 10−5 bar−1 ( Parrinello and Rahman , 1981 ) . Again , lipids and water+ions were coupled separately . The time step was set at 2 fs . Bond lengths were constrained using the P-LINCS algorithm ( Hess , 2008 ) . Cutoff was fixed at 1 . 2 nm for the Lennard-Jones and electrostatic interactions . The smooth particle-mesh was used to evaluate the electrostatic interactions . Frames were saved every 10 ps . Trajectory analyses were performed from 400 ns simulations from which we discarded the first 100 ns in order to rule out processes that are not at equilibrium . The remaining 300 ns trajectory was divided in 3 blocks of 100 ns to determine the standard deviation . Frames were analyzed every 100 ps except for the velocity and permeability analysis for which we used 10 ps frames . Coarse-grained simulations were performed with GROMACS 4 . 5 ( Hess et al . , 2008 ) using the Martini force field ( Wassenaar et al . , 2015 ) . The systems were built with the Charmm-Gui tool adapted to coarse-grained simulations ( Qi et al . , 2015 ) . In all simulations , we varied the acyl chains composition while keeping the PC polar head constant . We used four lipids to approximate the asymmetric lipids 18:0-18:1 , 18:0-18:2 , 18:0-20:4 , and 18:0-22:6 . Note that the coarse-grained simplification does not distinguish C16:0 from C18:0 , C20:4 from C20:5 , and C22:6 from C22:5 . We built coarse-grained models of swapped phospholipids from natural phospholipids having acyl chains of the same length . The systems contained 18 000 lipids and were solvated with a 100 nm thick layer of water . The sytems were equilibrated with the standard Charmm-Gui six-step process . Production runs were performed with the V-rescale thermostat at 303K . The Berendsen barostat was used to stabilize the pressure at 1 bar with a time constant of 4ps and compressibility of 5 × 10–5 bar−1 ( Berendsen et al . , 1984 ) . The different membranes were simulated under a semi-isotropic condition and the periodic boundaries were applied in all directions . Lipids and water/ions were coupled separately . The time step was fixed at 20 fs and the cutoff for the Lennard-Jones and electrostatic interactions was set at 1 . 2 nm . The smooth particle-mesh was used to evaluate the electrostatic interactions . To simulate membrane deformation and fission , we applied a force perpendicular to the initially flat bilayer ( Baoukina et al . , 2012 ) . The force ( from 175 to 250 KJ mol−1 nm−1 ) was applied to the center of mass of a lipid patch of radius = 3 nm , in which lipids were restrained in the lateral ( x , y ) directions . The simulations were performed for 200 ns and were repeated two to three times under most conditions . For further information on all molecular dynamics simulations , refer to the Gromacs mdp files ( supplementary files 2 and 3 ) . To evaluate membrane permeability , we counted the number of water molecule ( s ) that have visited the center of membrane ( corresponding to 65% of the thickness ) during 100 ns time frames . These water molecules were separated in two classes: those that fully crossed the bilayer and those that entered the hydrophobic region and then exited from the same side . Results were normalized to the total number of water molecules . The velocity rate of the terminal methyl group of the acyl chains was calculated from the sum of distances traveled by each methyl group in the x or in the z direction every 10 ps . For protrusions , we calculated the number of events during 100 ns blocks where the CH3 terminal group of the acyl chain reached a z position above the central carbon of glycerol from the same lipid . An acyl chain torsion corresponds to an angle below 100° between carbons that have relative positions of n-2 , n and n+2 along the acyl chain . Packing defect analysis was performed as previously described ( Vamparys et al . , 2013 ) . This membrane scanning procedure allows the detection of aliphatic atoms that are directly accessible to the solvent and that are either <1 Å ( shallow defect ) or >1 Å ( deep defect ) below the nearest glycerol . | Surrounding each living cell is a membrane that is mainly made of fat molecules called phospholipids . Similar membranes also surround many of the structures inside cells . It is important for life that these membranes are impermeable to many molecules; for example , they do not allow ions to cross them freely . The membranes also need to be flexible and allow cells to form different shapes . Flexible membranes also allow cells to move molecules around and to divide to produce new cells . Each phospholipid includes two long chains of atoms called fatty acids . There are many fatty acids but they are typically grouped into saturated and unsaturated based on their chemical structures . The omega-3 and omega-6 fats are both groups of unsaturated fatty acids that are found in brain cells . Many phospholipids in cell membranes contain one saturated and one unsaturated fatty acid but it is not clear why . By studying fat molecules in the laboratory and combining this with simulations , Manni et al . have now examined the effects of fatty acids on membranes . The investigation showed that phospholipids with both saturated and unsaturated fatty acids strike a balance between impermeable and flexible membranes . More unsaturated fatty acids make more flexible membranes but they are too permeable to be used in cells . The experiments also revealed that omega-3 unsaturated fats aid flexibility more than omega-6 . This finding may help to explain why the relative amounts of omega-3 and -6 are so important in the membranes of brain cells . The connection between the fats we eat and the fatty acids in our cells is complex . Yet , findings like these serve to remind us that we need a balanced diet of different fats to keep all our cells healthy . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2018 | Acyl chain asymmetry and polyunsaturation of brain phospholipids facilitate membrane vesiculation without leakage |
Many animals change their body color for visual signaling and environmental adaptation . Some dragonflies show wax-based color change and ultraviolet ( UV ) reflection , but the biochemical properties underlying the phenomena are totally unknown . Here we investigated the UV-reflective abdominal wax of dragonflies , thereby identifying very long-chain methyl ketones and aldehydes as unique and major wax components . Little wax was detected on young adults , but dense wax secretion was found mainly on the dorsal abdomen of mature males of Orthetrum albistylum and O . melania , and pruinose wax secretion was identified on the ventral abdomen of mature females of O . albistylum and Sympetrum darwinianum . Comparative transcriptomics demonstrated drastic upregulation of the ELOVL17 gene , a member of the fatty acid elongase gene family , whose expression reflected the distribution of very long-chain methyl ketones . Synthetic 2-pentacosanone , the major component of dragonfly’s wax , spontaneously formed light-scattering scale-like fine structures with strong UV reflection , suggesting its potential utility for biomimetics .
Many organisms exhibit a variety of body color patterns for visual communication and environmental adaptation . The diversity of the color patterns encompasses the ultraviolet ( UV ) range , reflecting the fact that many animals can detect UV light as well as green , blue and/or red light ( Osorio and Vorobyev , 2008 ) . UV reflection has been reported from numerous organisms and may be important not only for protection against UV-induced damage but also for visual signaling ( Silberglied , 1979; Eaton and Lanyon , 2003; Paul and Gwynn-Jones , 2003; Lee , 2007 ) . Previous studies on biological UV reflection have focused on its optical properties and structural bases , such as multilayer surface structures ( Sun et al . , 2013 ) . In some plants and insects , the production and secretion of wax on their surface has been reported to increase UV reflection ( Clark and Lister , 1975; Pope , 1979; Holmes and Keiller , 2002; Kakani et al . , 2003 ) . Dragonflies ( including damselflies ) are colorful , large-eyed , diurnal and actively flying insects , whose body colors often differ markedly between sexes , developmental stages , and closely related species ( Tillyard , 1917; Corbet , 1999; Futahashi et al . , 2012; Futahashi , 2016; Futahashi , 2017; Bybee et al . , 2016 ) . Because dragonflies are able to perceive UV light ( Bybee et al . , 2012; Futahashi et al . , 2015 ) , it seems plausible that UV color also plays important roles in their mate recognition , in male–male competition and in other ecological characteristics such as habitat selection and behavioral differences . Several studies have reported that the presence of a pruinose wax layer on the body surface accounts for UV reflection patterns in dragonflies ( Silberglied , 1979; Robertson , 1984; Hilton , 1986; Gorb , 1995; Harris et al . , 2011 ) , but the biochemical properties and molecular composition of the wax , and the genes involved in wax production by dragonflies , are totally unknown . Here , we mainly focus on the white-tailed skimmer dragonfly ( Orthetrum albistylum ) , which is one of the most common dragonfly species in Japan ( Sugimura et al . , 2001; Ozono et al . , 2012 ) . As sexual maturity is reached , adult males of O . albistylum show wax-based body color change from light brown to blueish white , whereas adult females remain brownish throughout most of their lifetime , although very aged females become slightly whitish ( Figure 1A ) ( Sugimura et al . , 2001; Ozono et al . , 2012 ) . Notably , androchrome females , whose body color is very similar to that of adult males , have been recorded , though very rarely , in the field ( Figure 1A ) ( Sugimura et al . , 2001; Ozono et al . , 2012 ) . Androchrome females can be distinguished from very aged females because the dorsal abdomen is more whitish than the ventral abdomen after the semimature stages ( Sugimura et al . , 2001; Ozono et al . , 2012 ) . In this study , we investigated the ultrastructure , reflectance , wettability , chemical composition , self-organization , and biosynthesis pathway of surface wax in O . albistylum and allied dragonfly species . We found that , during the maturation process , adult males secrete a strongly light-scattering wax layer onto their body surface , thereby increasing their visibility not only in the blue and green wavelength ranges but also in the UV range . Chemically , the UV-reflective surface wax consisted of very long-chain methyl ketones and aldehydes , which have not been previously identified as major wax components . Comparative transcriptomics identified a gene encoding a member of the elongation of very long-chain fatty acids ( ELOVL ) protein family , whose expression was strongly correlated with the distribution of very long-chain methyl ketones on the surface of dragonflies . Notably , chemically synthesized 2-pentacosanone , the major component of the surface wax , spontaneously formed scale-like fine structures that strongly reflected UV light . These results provide a previously undescribed molecular and structural basis for wax-based body color change and UV reflection that has ecological and applied relevance .
We compared the wax-based body color changes and UV reflection patterns of adult insects of O . albistylum using a high-sensitivity camera with a UV filter . UV reflection was hardly detected on the body surface of immature males and females ( Figure 1B , D , G and I; Video 1 ) . As sexual maturation proceeded , males accumulated whitish wax , mainly on their dorsal abdomen , which strongly reflected UV ( Figure 1C , F , H and K; Video 1 ) . It should be noted that in mature females , UV-reflective whitish wax was secreted on the ventral abdomen only ( Figure 1D , F , I and K; Video 1 ) . As adult aging proceeded further , not only males but also females developed pruinose wax on the entire body surface ( Figure 1A ) , which resulted in considerable UV reflection even in females . Optical measurements of reflectance on both the dorsal and ventral abdominal regions were used for quantitative evaluation of sex- and stage-dependent changes in the adult insects of O . albistylum . Immature males and females mainly reflected light of above 500 nm in wavelength , and did not exhibit remarkable UV reflection ( Figure 2A and D; Figure 2—figure supplement 1A and D ; Figure 2—source data 1; Figure 2—source data 2 ) . In mature males , reflectance increased , in particular of light of wavelengths below 600 nm , resulting in strong UV reflection on the dorsal abdomen and moderate UV reflection on the ventral abdomen ( Figure 2B; Figure 2—figure supplement 1B ; Figure 2—source data 3; Figure 2—source data 4 ) . In mature females , by contrast , moderate UV reflection was observed on the ventral abdomen only ( Figure 2E; Figure 2—figure supplement 1E; Figure 2—source data 3; Figure 2—source data 4 ) . In aged males and females , reflectance increased to some extent on both the dorsal and the ventral sides of the abdomen ( Figure 2C and F;Figure 2—figure supplement 1C and F ; Figure 2—source data 5; Figure 2—source data 6 ) . Micro-spectrometry of small areas ( 10 μm x 10 μm ) on the dorsal abdomen of a mature male indicated that the surface wax is responsible for overall reflectance , in particular in the short wavelength range including UV ( Figure 2G; Figure 2—source data 7 ) : strong reflection was found in the wax-covered white micro-areas ( Figure 2G; a , c , d , e and f ) , whereas little reflection was detected in the blackish micro-areas where the surface wax was lost ( Figure 2G; b , g and h ) . Sex- and stage-dependent changes in the surface fine structure of O . albistylum , with special attention to the surface wax , were observed by scanning electron microscopy ( SEM ) . In mature males , the dorsal abdomen was covered with scale- or plate-like fine structures ( 2–3 µm wide , 50 nm thick ) , which represented the secreted wax layer ( Figure 3A–D and K ) . The depth of the wax layer reached up to 6 μm from the cuticle surface ( Figure 3D ) . In mature females and immature individuals , by contrast , only tiny nanopillar-like structures ( 100 nm wide , 200–300 nm high ) were seen on the dorsal abdomen ( Figure 3E–J and L ) . In both mature males and females , on the other hand , small plate-like structures ( up to 2 μm wide ) were observed on the ventral abdomen , which presumably represented the pruinose wax secretions ( Figure 3O and P ) that were not conspicuous in immature males and females ( Figure 3M and N ) . Here , we suggest that these fractal surface structures , consisting of randomly arranged fine wax platelets , are responsible for the whitish structural color that results from light scattering , as observed on the dorsal abdomen of mature males and the ventral abdomen of mature males and females . The idea that the secreted wax layer produces structural color was confirmed by a simple experiment: the whitish color disappeared when light scattering was disturbed by acetone application , and the whitish color instantly recovered upon evaporation of the applied acetone ( Video 2 ) . Note that the sizes of the wax platelets ( up to 2–3 μm ) are larger than the wavelength of UV and visible light and thus these platelets are capable of light scattering , while the sizes of the nanopillar structures ( up to 200–300 nm ) are smaller than the wavelength of UV light and thus incapable of light scattering ( Vukusic and Sambles , 2003 ) , and thus might cause a descrease in the reflectivity of the surface ( Kinoshita and Yoshioka , 2005 ) . Specialized glands or structures for wax secretion have been characterized in diverse insects ( Pope , 1979; Ammar et al . , 2015 ) , but such wax-producing structures have not been described from dragonflies . Transmission electron microscopic ( TEM ) observations of the abdominal ultrathin sections of O . albistylum identified a number of fine ducts penetrating the cuticle layer in both immature and mature individuals ( Figure 3Q–X ) . On the dorsal abdomen of mature males in particular , the cuticular ducts were well-developed and full of electron-dense material , probably reflecting the active wax secretion there ( Figure 3S , arrowheads ) . To investigate the biochemical properties and molecular composition of dragonfly wax , the surface wax of O . albistylum was tested for solubility in organic solvents with reference to surface fine structure and wettability . We found that the secreted wax is insoluble in ethanol ( Figure 4B ) , partially soluble in hexane ( Figure 4C and E ) , and completely soluble in chloroform ( Figure 4D and F ) . The untreated abdominal surface exhibited strong water repellency ( Figure 4A ) , but removal of the wax by hexane or chloroform treatment resulted in drastically reduced water repellency or increased wettability ( Figure 4C and D ) . As reported in a variety of plants and insects ( Hadley , 1981 ) , the strong water repellency conferred by the surface wax may be important for in reducing water loss from dragonflies that have an aerial lifestyle . On the basis of these results , we extracted the surface wax of O . albistylum with hexane or chloroform , and analyzed its chemical composition by gas chromatography and mass spectrometry . In the hexane extract from the dorsal abdomen of mature males , only three very long-chain methyl ketones , namely 2-pentacosanone ( C25H50O ) , 2-heptacosanone ( C27H54O ) and 2-nonacosanone ( C29H58O ) , were identified ( Figure 5A; Figure 5—figure supplement 1 ) . In the chloroform extract from the dorsal abdomen of mature males , in addition to the three very long-chain methyl ketones , four very long-chain aldehydes , namely tetracosanal ( C24H48O ) , hexacosanal ( C26H52O ) , octacosanal ( C28H56O ) , and triacosanal ( C30H60O ) , were detected ( Figure 5B; Figure 5—figure supplement 1 ) . It should be noted that , in mature males , the very long-chain methyl ketones were dominant on the dorsal abdomen whereas the very long-chain aldehydes were dominant on the ventral abdomen ( Figure 5B and C ) . By contrast , only the four very long-chain aldehydes were identified in the chloroform extract from the ventral abdomen of mature females ( Figure 5E ) : neither the very long-chain methyl ketones nor the very long-chain aldehydes were found in the chloroform extract from the dorsal abdomen of mature females ( Figure 5D ) . Such very long-chain methyl ketones and aldehydes have been identified , although not as major components , in the surface wax of plants ( Kolattukudy and Walton , 1972; Post-Beittenmiller , 1996; Yamamoto et al . , 2008; Kunst and Samuels , 2003 ) and in the skin lipids of snakes ( Ahern and Downing , 1974 ) . Similar ketones were characterized as sex pheromones of cockroaches ( Nishida et al . , 1976 ) and snakes ( Parker and Mason , 2014 ) . In the light of previous studies on the wax secretions of various insects , in which hydrocarbons , long-chain esters , alcohols , and/or free fatty acids were identified as major components ( Brown , 1975; Blomquist and Jackson , 1979; Hadley , 1981 ) , it seems that the chemical composition of the dragonfly’s abdominal UV-reflective surface wax is unique among insects . We chemically synthesized 2-pentacosanone , the main UV-reflective wax component identified from the dorsal abdomen of mature males of O . albistylum ( Figure 5F ) , and attempted to recrystallize it on glass plates using three methods , namely dropping , quenching , and slow cooling ( see Materials and methods ) . The dropping method yielded numerous wax platelets randomly arranged on the substratum ( Figure 6B and F ) , which were reminiscent of the fine structure of the surface wax on the dorsal abdomen of mature males ( Figure 6A and E ) . By contrast , the quenching method and the slow cooling method resulted in larger wax platelets ( Figure 6C , D , G and H ) , which looked structurally dissimilar to dragonfly surface wax ( Figure 6A and E ) . The 2-pentacosanone sheets made by the three methods showed qualitatively similar reflectance patterns across UV to visible range , which were also similar to the reflectance pattern shown by dragonfly surface wax ( Figure 6I–L-source data 1 ) . Notably , however , the 2-pentacosanone sheets made by the dropping method yielded stronger light reflectance and lower wettability than those made by the quenching method and the slow-cooling method ( Figure 6I–P ) , probably because the smaller wax platelets made by the dropping method better mimic the fine structure of dragonfly surface wax . These results strongly suggest that the light-scattering nanostructure that is spontaneously formed by the secreted very long-chain methyl ketones , including 2-pentacosanone , should play a pivotal role in the formation of the wax layer that strongly reflects UV and visible light on the dorsal abdomen of mature males of O . albistylum . Diverse dragonfly species are known to secrete whitish or bluish wax on their body surface ( Tillyard , 1917; Corbet , 1999; Sugimura et al . , 2001; Ozono et al . , 2012 ) . We investigated the wax production , reflectance , and chloroform-extracted wax composition of the abdominal body surface of three other dragonfly species , Orthetrum melania , Sympetrum darwinianum , and Crocothemis servilia ( Figure 7 ) . The blue-tailed skimmer dragonfly ( O . melania ) , which is closely related to O . albistylum , prefers shady habitats in contrast to O . albistylum that tends to form territories in sunny places ( Sugimura et al . , 2001; Ozono et al . , 2012 ) . Mature males of O . melania develop bluish wax mainly on the dorsal abdomen , whereas mature females do not secrete wax even on the ventral abdomen ( Figure 7A ) . In O . melania , UV reflection was observed only on the wax-bearing mature males ( Figure 7E , I and L; Figure 7—figure supplement 1A and D; Figure 7—source data 1; Figure 7—source data 2; Video 3 ) . In the red dragonfly S . darwinianum , only mature females secrete whitish wax on the ventral abdomen ( Figure 7B ) , where UV reflection was clearly detected ( Figure 7F , J and M; Figure 7—figure supplement 1B and E; Figure 7—source data 3; Figure 7—source data 4; Video 3 ) . The scarlet dragonfly C . servilia develops little wax on its body surface throughout its lifetime ( Figure 7C and D ) , and no UV reflection was detected in either sex ( Figure 7G , H , K and N; Figure 7—figure supplement 1C and F; Figure 7—source data 5; Figure 7—source data 6 ) . These reflectance data indicated that ( i ) the presence of the secreted wax on the body surface accounts for the UV-reflecting body regions in all dragonfly species , ( ii ) the levels of the UV reflection vary among dragonfly species and also across different body regions , and ( iii ) the dorsal abdomen of mature males of O . albistylum exhibits the strongest UV reflection ( Figures 2 and 7 ) . The wax composition also varied among dragonfly species ( Figure 8 ) . For example , only very long-chain aldehydes were detected from mature males of O . melania ( Figure 8A and C ) , whereas only a very long-chain aldehyde , tetracosanal , was identified from the ventral abdomen of mature females of S . darwinianum ( Figure 8B and C ) . Very long-chain methyl ketones were detected only in regions where UV reflection was strong in O . albistylum ( Figure 8C ) . Here we point out that the UV reflection and wax production patterns observed in these dragonflies seem to reflect , at least to some extent , their environmental and behavioral characteristics . In general , body pigmentation and/or wax production are conspicuous in mature dragonflies , especially in reproductively active territorial males ( Tillyard , 1917 ) , which may be related to aspects of mate recognition , male–male competition , UV protection and anti-desiccation protection that are fatally important for them ( Corbet , 1999 ) . In the closely related Orthetrum species , O . albistylum dominates sunny habitats and shows stronger UV reflection than O . melania , which prefers shady habitats: mature males of O . albistylum form a dense whitish wax layer on their dorsal abdomen ( Figure 1C and F; Figure 9A ) whereas mature males of O . melania develop a relatively thin bluish wax layer ( Figure 7A; Figure 9B ) ; mature females of O . albistylum wear pruinose wax on their ventral abdomen ( Figure 1F ) whereas mature females of O . melania do not ( Figure 7A ) . Similar patterns are found in other Orthetrum species: mature males of O . luzonicum prefer sunny habitats and develop whitish wax ( Figure 9C ) , whereas mature males of O . glaucum are associated with shady habitats and form bluish wax ( Figure 9D ) . It is intriguing to ask why the mature females of some dragonflies , such as in O . albistylum ( Figure 1F and K ) and S . darwinianum ( Figure 7B and F ) , produce UV-reflective wax only on their ventral abdomen . We point out that , in these species , males form territories in sunny places , wait for females that fly in , and chase and copulate with them on nearby plants ( Figure 1F; Figure 7B ) . Therefore , these species usually mate in sunny places , where the female’s ventral abdomen is exposed to direct sunshine for an extended period ( Figure 1F; Figure 7B ) . By contrast , O . melania usually mates in shady places ( Figure 7A ) and C . servilia quickly mates during flight for only a few seconds ( Figure 7D ) , and the female’s ventral abdomen develops no wax and exhibits no UV reflection in these species ( Figure 7A , D , E and H ) . On the basis of these observations , we speculatively suggest that the female’s ventral wax might protect the ventral abdomen , which is less pigmented , less sclerotized and containing ovaries , against UV damage . What are the molecular mechanisms that underlie the production of dragonfly surface wax of unique chemical composition ? Notably , rarely discovered gynandromorphic dragonflies consistently exhibit discontinuous surface wax patterns ( Figure 9E and F ) ( Sugimura et al . , 2001; Karube and Machida , 2015 ) , suggesting that de novo wax production may be regulated in a cell-autonomous manner . In an attempt to gain insights into the molecular basis of dragonfly wax production , we performed RNA sequencing using samples of the dorsal and ventral abdominal epidermis dissected from immature , semimature , mature , mature-aged , and aged individuals of both sexes of O . albistylum . The adult maturity was judged on the basis of the amount of wax and the wing condition . In addition , we were able to examine a mature androchrome female ( Figure 10A ) . Figure 10B summarizes the RNA sequencing data . A total of 7790 genes whose maximum fragments per kilobase of transcript per million mapped reads ( FPKM ) values were greater than 10 , 1708 exhibited doubled expression levels or significantly higher expression in the dorsal epidermis of mature males ( with wax ) when compared with that in the dorsal epidermis of mature females ( no wax ) ; 518 genes exhibited doubled expression levels or significantly higher expression in the dorsal epidermis of an androchrome female ( with wax ) when compared with that in the dorsal epidermis of mature females ( no wax ) ; and 305 genes were commonly identified in these two categories as upregulated genes associated with wax production . Of these 305 genes , 26 genes were highly expressed ( FPKM >100 ) , of which five genes exhibited extremely high expression levels ( FPKM >1 , 000 ) ( Figure 10C; Figure 10—figure supplement 1; accession nos . LC416763- LC416767 ) . Among the extremely upregulated genes , we identified a gene belonging to the elongation of very long-chain fatty acids ( ELOVL ) protein family that was drastically and specifically ( more than 100-fold ) expressed in the dorsal epidermis of semimature and mature males and the androchrome female ( Figure 10C and D; Figure 10—figure supplement 1 ) . ELOVL proteins catalyze the elongation of fatty acids with acyl chains longer than 18 carbon atoms ( Kihara , 2012 ) and also of hydrocarbons ( Chertemps et al . , 2007 ) . ELOVL proteins are conserved from yeast to mammals , and 3 , 7 or 20 ELOVL family protein genes are identified in the genomes of yeast , mammals or the fruit fly Drosophila melanogaster , respectively ( Figure 11A ) ( Szafer-Glusman et al . , 2008; Kihara , 2012 ) . In the transcriptomic data of O . albistylum and also in the draft genome data of the scarce chaser dragonfly ( Ladona fulva ) ( BCMHGSC: I5K , GenBank accession no . APVN00000000 . 2 ) , we identified 17 ELOVL genes in total ( Figure 11A; accession nos . BR001497-BR001513 , LC416747-LC416763 ) , all of which contained a conserved HXXHH motif ( Figure 11B ) . In addition to the drastically upregulated ELOVL gene ( =ELOVL17 ) mentioned above , two ELOVL genes exhibited notable upregulation patterns ( Figure 11C ) . ELOVL14 was highly expressed in the dorsal epidermis of semimature and mature males , and also in both the dorsal and the ventral epidermis of the androchrome female ( Figure 10E; Figure 11 ) . In aged females , ELOVL17 and ELOVL14 genes were slightly upregulated ( Figure 10D and E ) , which may account for the slight wax secretion of aged females of O . albistylum ( see Figure 1A ) . Meanwhile , the ELOVL3 gene was preferentially expressed in the ventral abdomen of immature females ( Figure 10F ) , which may be relevant to the preferential accumulation of very long-chain aldehydes on the ventral abdomen of mature females of O . albistylum ( see Figure 5E ) . These results strongly suggest that these ELOVL genes are involved in production of the surface wax of dragonflies , which mainly consists of very long-chain methyl ketones and aldehydes . To confirm this idea , we attempted to knock-down the expression level of the ELOVL17 gene by injection of small interfering RNA followed by electroporation , a technique established in other dragonfly species ( Okude et al . , 2017 ) . After employing the RNAi treatment , we expected that the wax production on the abdominal surface of mature males would be suppressed . Unfortunately the electroporation damaged the adult cuticle and caused high mortality of the treated insects , so we failed to observe the phenotypes expected for the RNAi experiment . Hence , the precise functions of the dragonfly’s ELOVL gene products remain to be verified in future studies , for which the establishment of a stable laboratory rearing system for O . albistylum and the successful application of genome-editing technology in this species are anticipated . In addition to the ELOVL17 gene , the Acyl-CoA Delta ( 11 ) desaturase , ferritin , NPC intercellular transporter 2 , and uncharacterized protein ( LC416764- LC416767 ) genes exhibited extremely high expression in wax-producing regions , although their stage and region specificity was not prominent compared to those of the ELOVL17 gene . The biological roles of these genes also deserve future studies . In this study , we found that mature males of O . albistylum exhibit strong light reflection , including the reflection of light in the UV range , because of a previously uncharacterized mechanism , namely very long-chain methyl ketone production . Plausibly , differences in wax production between sexes , stages and species are important for signal communications between dragonflies , and may reflect their habitats and behavior . It should be noted that synthesized 2-pentacosanone , a major component of very long-chain methyl ketone , reproduced the strong reflection , surface fine structure , and water repellency . Considering that UV reflective materials can be applied in the fields of cosmetics and painting , and that O . albistylum has been traditionally used as medicine in Asia ( Corbet , 1999 ) , the dragonflies’ UV reflective wax may have the potential to be utilized as a novel eco-friendly biological material .
Adult insects of O . albistylum , O . melania , S . darwinianum , and C . servilia were collected at Tsukuba , Ibaraki , Japan , or at Imizu , Toyama , Japan . Photos of UV reflection were taken in the field using a high-sensitivity camera ( Sony A7S , IDAS UV-VIS mod ) and a UV filter ( IDAS-U ) . In order to investigate the wax-based color change quantitatively , the dorsal and ventral parts of abdominal segment 5 were surgically divided and used for reflection measurements . The reflections from small areas ( diameter 6 mm ) were taken using a spectrometer ( HR2000+ , Ocean Optics ) , and those of micro areas ( 10 × 10 µm ) were carried out using a micro-spectrometer ( CRAIC Technologies ) equipped with an upright microscope ( Eclipse E-400; Nikon ) . The specimens were epi-illuminated with a 75 W Xenon arc lamp ( Nikon ) , and the measurements were obtained from an approximately 10 µm x 10 µm area . The reflected spectral radiances were converted to relative reflectance by normalization with a white reflectance standard ( Spectralon USRS-99–010 , Labsphere ) . The microstructural changes resulting from wax secretion were examined by scanning electron microscopy ( SEM ) and transmission electron microscopy ( TEM ) . For SEM observation , the dissected dorsal and ventral parts of abdominal segment 5 were coated with a 2–3 nm osmium layer using hollow-cathode plasma chemical vapor deposition ( HPC-1SW; Vacuum Device ) . They were then observed under a scanning electron microscope ( H-4800; Hitachi ) with an accelerating voltage of 5 kV . For TEM observation , dissected dorsal and ventral parts of abdominal segment 5 were prefixed for 12 hr in 2% glutaraldehyde and 2% paraformaldehyde in 0 . 1 M cacodylate buffer ( pH7 . 2 ) , post-fixed with 1% osmium tetroxide for 2 hr in 0 . 1 M cacodylate buffer , and embedded in Quetol 812-Araldite regin mixture ( Nisshin EM ) . Ultrathin sections ( approximately 70 nm thick ) were cut perpendicular to the anterior–posterior axis on an ultra-microtome ( UC7; Leica ) with a diamond knife ( DiATOME ) , stained with 2% uranyl acetate for 5 min followed by lead citrate solution for 3 min ( Sigma-Aldrich ) , and observed under a transmission electron microscope ( JEM-1400; JEOL , 100 KV ) . To elucidate the biochemical properties of dragonfly wax , surface wettability and solubility were examined . Surface wettability was evaluated on the basis of the contact angle of water micro-droplet on the samples . Each sample was fixed on a glass substrate , and a micro-droplet of distilled water ( about 1 . 0 nL ) was placed on the surface of the sample . The shape of droplet was recorded immediately using a high-speed camera ( HAS-220; Ditect ) with a microscopic contact angle meter ( MCA-3; Kyowa Interface Science ) . Wax solubility was analyzed by treating the dissected abdominal segment 5 with hexane or chloroform for 30 min . To identify the molecular composition of dragonfly wax , wax samples were extracted from living specimens with chloroform or hexane . To avoid contamination from internal lipids during gas chromatography/mass spectrometry ( GC-MS ) analysis , the solvent was carefully pipetted several times onto the abdominal surface of the living individuals . The extracts were analyzed by GC-MS using a 6890N GC coupled with 5973 MSD ( Agilent ) in the split-less mode , using a DB-5MS fused silica column ( 30 m x 0 . 25 mm i . d . , 0 . 25 µm film thickness , Agilent ) with helium as the carrier gas at a flow rate of 1 . 0 mL/min at a temperature programmed to change from 80°C ( 1 min ) to 320°C at a rate of 15 °C/min and then held for 3 min . The mass spectrometer was operated in the scan mode with 70 eV ionazation voltage as electron ionization . Histological examinations , wettability tests , and GC-MS analyses were conducted using different samples . To confirm whether the very long-chain methyl ketones form the scale-like fine structures , 2-pentacosanone , the major component of dragonfly wax was chemically synthesized from 1-tetracosanol via 1-tetracosanal and 2-pentacosanol . Pyridinium chlorochromate ( PCC , 1 . 29 g , 5 . 97 mmol ) was added to a suspension of 1-tetracosanol ( 395 mg , 1 . 11 mmol ) and powdered molecular sieves 4A ( 2 . 5 g ) in dry CH2Cl2 ( 35 mL ) , and stirred for 4 hr at room temperature . The mixture was filtered through Celite and washed with diethyl ether . The combined filtrate and washings were filtered through florisil ( 15 g ) , washed with diethyl ether ( 200 mL ) and concentrated in vacuo . The residue was chromatographed on silica gel ( 15 g ) and concentrated in vacuo to give a white solid of 1-tetracosanal ( 290 mg , 0 . 82 mmol , 74% , GC tR = 23 . 7 min , MS m/z ( % ) = 352 ( M+ , 2 ) , 334 ( 18 ) , 96 ( 78 ) , 82 ( 100 ) , 57 ( 93 ) , 43 ( 72 ) ) . A solution of 1-tetracosanal ( 176 mg , 0 . 50 mmol ) in dry tetrahydrofuran ( THF , 10 mL ) was cooled in ice bath . When the temperature reached 0°C , 1 . 4 M CH3MgBr in THF:toluene 1:3 ( 1 mL , 1 . 4 mmol ) was added dropwise , and stirred for 1 . 5 hr at 0°C and for 1 hr at room temperature . The reaction was quenched with saturated NH4Cl ( 5 mL ) , and the product was extracted with hexane ( 3 × 20 mL ) . The organic layer was dried with anhydrous magnesium sulfate and concentrated in vacuo . The residue was chromatographed on silica gel ( 15 g , ethyl acetate/hexane , 1:5 ) to give a white solid of 2-pentacosanol ( 106 mg , 0 . 29 mmol , 58% ) . PCC ( 335 mg , 1 . 56 mmol ) was added to a suspension of 2-pentacosanol ( 257 mg , 0 . 70 mmol ) and powdered molecular sieves 4A ( 1 . 0 g ) in dry CH2Cl2 ( 20 mL ) , and stirred for 3 hr at room temperature . The mixture was filtered through Celite and washed with diethyl ether . The combined filtrate and washings were filtered through florisil ( 15 g ) , washed with diethyl ether ( 200 mL ) and concentrated in vacuo . The residue was recrystallized from hexane to give a white solid of 2-pentacosanone ( 202 mg , 0 . 55 mmol , 76% ) . Nuclear magnetic resonance ( NMR ) spectra of 2-pentacosanal and 2-pentacosanone were measured with Bruker AV-400 III Spectrometer ( 400 MHz ) using tetramethylsilane ( TMS ) as an internal standard . Each sample was dissolved in CDCl3 and 1H spectrum was acquired . 1H-NMR of 2-pentacosanal ( CDCl3 , 400 MHz ) was as follows: δ 3 . 79 ( 1H , sex , J = 6 . 0 Hz , H-2 ) , δ 1 . 18 ( 3H , d , J = 6 . 0 Hz , H-1 ) , δ 0 . 88 ( 3H , t , J = 6 . 0 Hz , H-25 ) . 1H-NMR of 2-pentacosanone ( CDCl3 , 400 MHz ) is as follows: δ 2 . 41 ( 2H , d , J = 7 . 6 Hz , H-3 ) , δ 2 . 13 ( 3H , s , H-1 ) , δ 0 . 88 ( 3H , t , J = 6 . 4 Hz , H-25 ) . Biomimetic wax surfaces were composed of micro crystals of 2-pentacosanone . Heated 2-pentacosanone was recrystallized on gold-coated glass plates by different cooling processes: 1 ) continuous dropping of micro fused material of 1 . 0 µL under room temperature , 2 ) quenching from the melting state , and 3 ) maintenance near melting point at 64°C from the melting state . Micro-spectrometry , surface fine structure , and wettability were analyzed using a micro-spectrometer ( CRAIC Technologies ) , a scanning electron microscope ( H-4800; Hitachi ) , or a high-speed camera ( HAS-220; Ditect ) with a microscopic contact angle meter ( MCA-3; Kyowa Interface Science ) , respectively , as described above . To investigate the genes involved in wax production , total RNA samples were extracted from the freshly dissected abdomens of O . albistylum using an RNeasy mini kit ( Qiagen ) or a Maxwell 16 LEV Simply RNA Tissue kit ( Promega ) . RNA sequencing was performed as described previously ( Futahashi et al . , 2015 ) . Using 1 µg of total RNA per sample as template , cDNA libraries were constructed using TruSeq RNA Sample Preparation Kits v2 ( Illumina ) and sequenced by HiSeq2000 , Hiseq2500 , or MiSeq ( Illumina ) . The sequence data were deposited in the DNA Data Bank Japan Sequence Read Archive ( accession numbers are shown in Supplementary file 1 ) . The raw reads were subjected to de novo assembly using the Trinity program ( Grabherr et al . , 2011 ) implemented in the MASER pipeline ( Kinjo et al . , 2018 ) . After automatic assembling , we checked and manually corrected the sequences of genes that are highly expressed in mature males using the Integrative Genomics Viewer ( Thorvaldsdóttir et al . , 2013 ) as reported previously ( Futahashi , 2017 ) . After revising the sequences , sequence read mapping was performed using the BWA-MEM program ( Li , 2013 ) implemented in the MASER pipeline , whereby transcript expression levels were estimated in terms of fragments per kilobase per million reads ( FPKM ) values . ELOVL genes of L . fulva were obtained by a tBLASTn search against the draft genome sequence ( APVN00000000 . 2 ) ( https://www . hgsc . bcm . edu/ ) . To construct the molecular phylogeny of ELOVL family genes , deduced amino-acid sequences were aligned using the Clustal W program implemented in MEGA7 ( Kumar et al . , 2016 ) . Molecular phylogenetic analyses were conducted by the neighbor-joining method and the maximum-likelihood method using MEGA7 , and by the Bayesian method using MrBayes version 3 . 1 . 2 ( Ronquist et al . , 2012 ) . Bootstrap values for neighbor-joining and maximum likelihood phylogenies were obtained by 1000 resampling . In total , 3750 trees were generated for each Bayesian analysis ( ngen = 500 , 000 , samplefreq = 100 , burn in = 1250 ) . | Humans have often looked to nature for answers to problems . Living things has evolved for millions of years to deal with life’s challenges , and so engineers and inventors faced with similar challenges can also take inspiration from the natural world . Several plants and animals , for instance , reflect ultraviolet light . This ability may protect them from some of the damaging effects of sunlight; materials with similar properties would have a range of uses , including as coatings on windows that protect our homes and furniture or as cosmetics that protect ourselves in the same way . Some dragonflies – including the white-tailed skimmer , which is particularly common in Japan – are partly coated with a wax that reflects both ultraviolet and visible light . These insects can also see ultraviolet light , which means it is likely that they also use the reflective wax to send visual signals to one another . However , the biochemistry of this wax and the genes involved in its production remained poorly understood . Futahashi et al . have now found that the dragonfly wax consists mostly of very long-chain molecules known as methyl ketones and aldehydes; neither of which are a common components of other waxes . The wax was found in distinct patches on the bodies of adults; these patches were colored white with a hint of blue , while the rest of the dragonfly was mostly brown . Looking at gene activity in different parts of the dragonflies showed that a gene called ELOVL17 is much more active in the wax-coated areas . This gene encodes an enzyme that makes long-chain molecules , and its activity closely matched the distribution of the especially long-chain methyl ketones on the dragonflies’ surface . Futahashi et al . then synthesized the major component of the surface wax – specifically , a chemical called 2-pentacosanone – in the laboratory , and saw that it spontaneously formed fine , scale-like structures that strongly reflected ultraviolet light . Further work is now needed to explore the potential applications of this bio-inspired wax , and to understand exactly what the dragonflies use it for in the wild . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"evolutionary",
"biology"
] | 2019 | Molecular basis of wax-based color change and UV reflection in dragonflies |
Although the human gut microbiome plays a prominent role in xenobiotic transformation , most of the genes and enzymes responsible for this metabolism are unknown . Recently , we linked the two-gene ‘cardiac glycoside reductase’ ( cgr ) operon encoded by the gut Actinobacterium Eggerthella lenta to inactivation of the cardiac medication and plant natural product digoxin . Here , we compared the genomes of 25 E . lenta strains and close relatives , revealing an expanded 8-gene cgr-associated gene cluster present in all digoxin metabolizers and absent in non-metabolizers . Using heterologous expression and in vitro biochemical characterization , we discovered that a single flavin- and [4Fe-4S] cluster-dependent reductase , Cgr2 , is sufficient for digoxin inactivation . Unexpectedly , Cgr2 displayed strict specificity for digoxin and other cardenolides . Quantification of cgr2 in gut microbiomes revealed that this gene is widespread and conserved in the human population . Together , these results demonstrate that human-associated gut bacteria maintain specialized enzymes that protect against ingested plant toxins .
The human gut microbiome extends the metabolic capabilities of the human body , extracting energy from otherwise indigestible dietary polysaccharides , synthesizing essential vitamins and amino acids , and modifying endogenous compounds . This microbial community also extensively metabolizes xenobiotics , including synthetic and natural product-based drugs , food additives , and environmental toxins ( Koppel et al . , 2017; Spanogiannopoulos et al . , 2016 ) . Previous work has focused on cataloguing the xenobiotics subject to transformation by gut microbes , their downstream products , and the extensive inter-individual variation in these activities ( Koppel et al . , 2017; Spanogiannopoulos et al . , 2016 ) . However , reduction of these complex metabolic networks to mechanism has been limited by our lack of knowledge about the microbial enzymes responsible for xenobiotic biotransformation . Digoxin , a cardenolide used to treat heart failure and arrhythmia , represents a valuable test case for our ability to elucidate the biochemical and evolutionary underpinnings of gut microbial xenobiotic metabolism . It has been known for decades that human gut bacteria reduce digoxin to the inactive metabolite dihydrodigoxin , decreasing drug efficacy and toxicity ( Saha et al . , 1983; Lindenbaum et al . , 1981a; Lindenbaum et al . , 1981b ) . Screening hundreds of gut bacterial strains from humans that excreted high levels of dihydrodigoxin revealed only two isolates that were capable of metabolizing digoxin , both of which were strains of the anaerobic , low abundance bacterium Eggerthella lenta ( Saha et al . , 1983 ) . However , the presence of E . lenta in the gut microbiome cannot accurately predict this reactivity , as patients colonized by E . lenta still show marked variation in dihydrodigoxin production ( Mathan et al . , 1989; Alam et al . , 1988 ) . We identified two mechanisms underlying this discrepancy: strain-level variations in the E . lenta population and inhibition of bacterial drug metabolism by dietary amino acids . Digoxin induces expression of a 2-gene operon encoded by the type strain of E . lenta ( DSM 2243 ) , which we named the cardiac glycoside reductase ( cgr ) operon ( Haiser et al . , 2013 ) . The cgr operon was absent in two E . lenta strains that did not metabolize digoxin ( ‘non-reducing’ strains ) and cgr operon presence and abundance predicted the extent of drug inactivation by human gut microbial communities in ex vivo incubations ( Haiser et al . , 2013 ) . Furthermore , germ-free mice that had been mono-colonized by a reducing ( cgr+ ) strain of E . lenta had lower serum levels of digoxin than mice colonized by a non-reducing ( cgr- ) strain , and dietary arginine efficiently blocked digoxin reduction by the cgr operon in cgr+ E . lenta-colonized mice ( Haiser et al . , 2013; Haiser et al . , 2014 ) . Although it is tempting to consider applying these insights to develop novel microbiome-based diagnostics and co-therapies ( Spanogiannopoulos et al . , 2016 ) , multiple critical questions remained unaddressed . Our original studies were entirely based on the E . lenta type strain , isolated in 1938 from a rectal cancer biopsy ( Moore et al . , 1971 ) ; thus , the presence of cgr +E . lenta in the modern-day human gastrointestinal tract was unclear . Although we had associated the cgr operon with digoxin reduction , the minimal genetic machinery necessary and sufficient for this biotransformation had not been determined . Perhaps most importantly , the specificity of the digoxin-reducing enzyme ( s ) for cardenolides and their ability to accept additional endogenous or ingested substrates remained unclear . Here , we used a combination of comparative genomics , heterologous expression , biochemistry , and metagenomics to address these long-standing questions . We uncovered a highly conserved cluster of genes that co-occurs with the cgr operon , representing a single genetic locus predictive of digoxin metabolism . We demonstrated that a single protein encoded by this locus , Cgr2 , is sufficient for digoxin reduction and is widespread in human gut microbiomes . Cgr2 is a novel oxygen-sensitive reductase that requires flavin adenine dinucleotide ( FAD ) and at least one [4Fe-4S] cluster . Surprisingly , we found that Cgr2 only accepts digoxin and other cardenolides , prompting the provocative hypothesis that the gut microbiome provides a first-line of protection against ingested toxins analogous to that of host enzymes expressed in the intestinal epithelium and liver . Finally , this work establishes a generalizable framework for mechanistic investigations of gut microbial xenobiotic metabolism that will enhance our understanding of the complex dietary , host , and microbial factors that impact pharmacology and toxicology , and provide a stronger foundation for translational studies in patient populations .
E . lenta strains vary in their ability to reduce digoxin ( Haiser et al . , 2013; Haiser et al . , 2014 ) ; however , our prior attempts at identifying the minimal genetic machinery necessary for metabolism were limited by the availability of just a single strain capable of this activity ( E . lenta DSM 2243 ) . Through public repositories and isolation of novel strains , we curated , sequenced , and annotated a collection of 25 E . lenta and closely related Coriobacteriia strains ( Bisanz et al . , 2018 ) . These bacteria were isolated from 22 individuals in 6 countries across three continents spanning the years of 1938–2015 ( Figure 1—source data 1 ) . We used liquid chromatography-tandem mass spectrometry ( LC-MS/MS ) to quantify the biotransformation of digoxin to dihydrodigoxin by each strain and identified seven additional strains capable of drug inactivation ( Figure 1A ) . Culturing experiments were performed using 10 µM of digoxin , which is within the estimated range ( 0 . 4–14 . 6 µM ) of therapeutic concentrations of the drug in the gut prior to absorption by host epithelial cells ( Schiller et al . , 2005 ) . The digoxin metabolizing strains did not exhibit a significant phylogenetic signal ( Figure 1B; p=0 . 275 , K = 0 . 049 , Blomberg’s K test ) , suggesting that this phenotypic trait has been gained ( or lost ) multiple times over the course of E . lenta evolution . Machine learning ( random forests ) analysis of orthologous gene cluster presence/absence across the strain collection revealed a single genetic locus with 100% discriminative value between metabolizers and non-metabolizers ( Figure 1C ) . This locus , referred to hereafter as the cgr gene cluster , includes the previously identified 2-gene cgr operon ( cgr1 and cgr2 ) and six neighboring genes termed cac ( cgr-associated cluster ) genes ( Figure 1D ) . The cac genes include a putative LuxR type transcriptional regulator ( Cac3 ) , a predicted flavin-dependent fumarate reductase ( Cac4 ) , three proteins of unknown function ( Cac1 , Cac5 , Cac6 ) , and a short protein ( Cac2 ) that is conserved in both cgr- and cgr+ strains of E . lenta . The 10 . 4 kb cgr gene cluster was highly conserved between strains with an average global nucleotide identity of 99 . 95 ± 0 . 05% ( mean ± standard deviation ( SD ) ) . Both cgr+ and cgr- strains share a short 174 bp hypothetical gene ( cac2 ) that is conserved with 100% global nucleotide identity in cgr+ strains , while cgr- strains are 90 . 20–91 . 37% identical to cgr+ cac2 . This conservation and genomic context may be indicative of multiple translocations in the region creating the cgr-associated gene cluster although obvious markers of recent translocation of the cgr gene cluster are absent . Multiple lines of evidence suggested that the cgr operon encodes the enzymes responsible for digoxin metabolism . Of the eight genes in the cgr gene cluster , only three show primary sequence homology to reductases: cgr1 , cgr2 , and cac4 . RNA sequencing ( Haiser et al . , 2013 ) demonstrated that the cgr operon ( cgr1 and cgr2 ) is highly upregulated ( >165 fold ) in response to digoxin , whereas cac4 is not significantly induced ( 1 . 3-fold change relative to vehicle controls , p=0 . 83 ) ( Figure 1E ) . The remainder of the cgr-associated cluster is largely transcriptionally dormant during exponential growth both with and without the presence of digoxin ( <6 normalized counts per gene ) and is therefore unlikely to be linked to digoxin metabolism . Initial annotations of Cgr1 and Cgr2 suggested both proteins might mediate digoxin reduction . Cgr1 is a putative membrane-anchored protein that belongs to the cytochrome c3 superfamily ( Pfam 14537 ) and is predicted to harbor covalently bound heme groups ( CXXCH motif ) . It most closely resembles the NapC/NirT ( NrfH ) family of proteins that transfer electrons from the membrane quinone pool to associated reductases , facilitating reduction of terminal electron acceptors such as nitrite and sulfite ( Kemp et al . , 2010; Kern et al . , 2008 ) . We also identified a close homolog of Cgr1 ( Elen_2528 ) in E . lenta DSM 2243 ( 91 . 75% amino acid identity , BLASTP ) that is a component of the E . lenta core genome ( 99 . 39 ± 0 . 81% global identity mean ± SD ) . The presence of this highly similar protein in both metabolizing and non-metabolizing strains further indicates that Cgr1 is involved in a more general function ( e . g . electron transfer , membrane docking ) rather than direct reduction of digoxin . On the other hand , Cgr2 is unique to the genomes of cgr+ E . lenta , and the closest homologs of Cgr2 display <28% amino acid identity . Cgr2 is a homolog of flavin adenine dinucleotide ( FAD ) -dependent fumarate reductases ( Pfam 00890; Interpro 003953/027477 ) and is predicted to undergo secretion via the twin arginine translocation ( Tat ) pathway . Taken together , these observations led us to hypothesize that Cgr1 and Cgr2 form a membrane-associated complex that catalyzes reduction of the α , β-unsaturated butyrolactone of digoxin . Heterologous expression of Cgr1 and Cgr2 in the model Actinobacterium Rhodococcus erythropolis L88 ( Mitani et al . , 2005; Nakashima and Tamura , 2004a , 2004b ) allowed us to test whether these proteins were sufficient for digoxin reduction . After inducing protein expression , cultures were incubated with 10 µM of digoxin and dihydrodigoxin production was quantified by LC-MS/MS ( Figure 2A ) . The Cgr2 expressing strains showed a significant increase in dihydrodigoxin levels relative to empty vector controls ( Figure 2A; Figure 2—source data 1 ) . In contrast , no activity was observed for the strain expressing only Cgr1 , although this could be due to a lack of protein , as no overexpression of Cgr1 could be detected in either clarified lysates or membrane fractions . These results show that Cgr2 is sufficient for digoxin reduction in R . erythropolis cells , and endogenous redox active proteins and/or metabolites in this heterologous host may fulfill the putative function of Cgr1 as an electron donor . Having identified Cgr2 as the critical reductase enzyme , we next aimed to reconstitute its activity in vitro . Examining multiple tagged versions and truncations of Cgr2 in R . erythropolis revealed that a Cgr2 ( –48aa ) -NHis6-tagged construct lacking the Tat secretion signal gave the highest yield and purity ( Figure 2—figure supplement 1A ) . This construct , hereafter referred to as ‘wild-type’ Cgr2 , was used for all in vitro studies . Computational analysis of Cgr2 predicted that it would bind flavin through a Rossmann fold ( Dym and Eisenberg , 2001 ) , and the three motifs required for cofactor binding are conserved in all 8 Cgr2 sequences ( Figure 2B ) . However , Cgr2 did not co-purify with flavin . Moreover , Cgr2 obtained from initial purifications was thermally unstable ( melting temperature <37°C ) , was prone to degradation during cell lysis ( Figure 2—figure supplement 1 ) , and displayed low activity for digoxin reduction ( Figure 2C ) ; Figure 2—source data 2 ) . Together , these observations indicated that an essential cofactor was likely missing . We also noticed that purified Cgr2 was light brown in color , suggesting the presence of a metallocofactor . Certain flavin-dependent reductases use metallocofactors to mediate the transfer of electrons to the active site , including cytochromes c in soluble enzymes and oxygen-sensitive iron-sulfur ( [Fe-S] ) clusters in membrane-bound enzymes ( Kern et al . , 2008; Iverson et al . , 2002; Leys et al . , 1999 ) . The combination of the brown color and the presence of 16 cysteines in the mature Cgr2 sequence led us to hypothesize that this enzyme contained one or more [Fe-S] clusters . However , we were unable to detect any canonical [2Fe-2S] , [3Fe-4Fe] , or [4Fe-4S] cluster binding motifs within the Cgr2 sequence ( Figure 2—source data 3 ) ( Zhang et al . , 2010; Nakamaru-Ogiso et al . , 2002; Lee et al . , 2004; Pandelia et al . , 2011; Schnackerz et al . , 2004; Leech et al . , 2003; Gorodetsky et al . , 2008; Lee et al . , 2010; Weiner et al . , 2007; Klinge et al . , 2007; Dickert et al . , 2002; Conover et al . , 1990; Schneider and Schmidt , 2005; Iwasaki et al . , 2000; Banci et al . , 2013; Dailey and Dailey , 2002; Jung et al . , 2000 ) . We therefore sought to determine whether Cgr2 required an [Fe-S] cluster to catalyze digoxin reduction . Attempts to chemically reconstitute [Fe-S] cluster formation by incubating Cgr2 with iron and sulfide under anaerobic conditions greatly improved protein stability ( Figure 2—figure supplement 1B–D ) . In addition to performing this reconstitution step , including FAD in assay mixtures dramatically increased the digoxin reduction activity of purified protein ( Figure 2C ) . Even after reconstitution , Cgr2 required excess FAD for maximal activity , suggesting that Cgr1 could be important for enhancing FAD binding , as has been observed for proteins whose FAD-binding site is predicted to occur at the interface of two domains ( Kleven et al . , 2015 ) . Having demonstrated that [Fe-S] reconstitution was essential for activity , we next attempted to determine the exact nature of this metallocofactor . Prior to reconstitution , purified Cgr2 contained between 0 . 2–0 . 6 equivalents of iron and sulfide , and its ultraviolet-visible ( UV-vis ) spectrum displayed absorption features consistent with the presence of low levels of [Fe-S] clusters ( Ayala-Castro et al . , 2008 ) ( Figure 2D ) . After reconstitution , Cgr2 exhibited a broad peak around 400 nm ( Figure 2D ) that decreased in absorbance upon addition of an excess amount of the reducing agent sodium dithionite ( Figure 2—figure supplement 2A ) . These spectral properties are characteristic of redox-active [Fe-S] clusters . Exposure of reconstituted Cgr2 to oxygen led to [Fe-S] cluster decomposition as evidenced by a decrease in the absorbance at 400 nm ( Figure 2D ) , demonstrating that the [Fe-S] cluster ( s ) of Cgr2 are oxygen-sensitive . Although these UV-vis experiments indicated the presence of [Fe-S] cluster ( s ) in Cgr2 , they could not define the precise structures of these cofactors . To more definitively characterize these metallocofactors , we turned to electron paramagnetic resonance ( EPR ) spectroscopy . This technique detects unpaired electrons and can differentiate between the various types of [Fe-S] clusters as well as provide information about cluster orientation and redox state . In the absence of a reducing agent , purified , unreconstituted Cgr2 does not have an EPR signal ( Figure 2—figure supplement 2B ) . Upon reduction with sodium dithionite , the EPR spectrum of Cgr2 exhibits a signal with axial symmetry and with principal g-components of 2 . 045 and 1 . 94 . This signal increases in intensity upon reconstitution of Cgr2 ( 10 K ) , and is barely detectable above 40 K ( Figure 2E ) . Both the principal g-values and the relaxation properties ( temperature dependence ) of this signal are characteristic of low-potential tetranuclear [4Fe-4S]1+ centers . These observations indicate that Cgr2 contains [4Fe-4S]2+ cluster that can undergo reduction to the corresponding [4Fe-4S]1+ state . These redox properties suggest that these [4Fe-4S] cluster ( s ) may participate in catalysis ( electron transfer ) . Next , we sought to determine how many [4Fe-4S] clusters were present in Cgr2 . Using a Cu2+-EDTA standard , we determined that purified , unreconstituted Cgr2 contains 0 . 02–0 . 03 [4Fe-4S]1+ clusters per protein monomer . After reconstitution with iron and sulfide , the intensity of the EPR signal increased to 0 . 13–0 . 25 [4Fe-4S]1+ clusters per Cgr2 . Though these data may suggest the presence of one [4Fe-4S] cofactor per Cgr2 , they do not exclude the possibility of multiple [Fe-S] centers . Indeed , the in vitro activity of Cgr2 increases upon reconstitution with increasing equivalents of iron and sulfide ( Figure 2—figure supplement 2C–D ) . Additional spectroscopic or structural characterization ( e . g . crystallography ) will be required to definitively determine the number of [Fe-S] cluster ( s ) present in Cgr2 . As motif analysis could not identify putative [4Fe-4S] cluster binding sites in Cgr2 , we attempted to use site-directed mutagenesis to reveal the cysteine residues required for cofactor assembly . Individually mutating each of the 16 cysteines present in wild-type Cgr2 to alanine revealed six residues that , when mutated , significantly decreased dihydrodigoxin production by both heterologously expressed and purified Cgr2 ( Figure 2—figure supplement 1A; Figure 2—figure supplement 3A–C; Figure 2—source data 4; Figure 2—source data 5 ) . EPR analysis of these six Cgr2 mutants revealed comparable levels of [4Fe-4S]1+ cluster incorporation relative to the wild-type enzyme ( Figure 2—figure supplement 3D ) , which may argue against the involvement of these cysteines in [4Fe-4S] cluster ligation . However , substitution of a single cysteine residue may not always be sufficient to prevent [4Fe-4S] cluster formation ( Iismaa et al . , 1991; Hewitson et al . , 2002; Martín et al . , 1990 ) . Alternatively , these six cysteines may be critical for protein structure ( e . g . through participating in disulfide formation ) or could coordinate another metal center not detectable in our spectroscopic experiments . Consistent with this latter proposal , we found that a range of divalent metal cations ( Fe2+ , Mn2+ , Mg2+ ) stimulated the activity of Cgr2 in vitro ( Figure 2—figure supplement 3E ) without altering protein stability . Additionally , Fe2+ stimulated the in vitro activity of only 3 out of 6 impaired mutants ( C158A , C187A , C327A ) ( Figure 2—figure supplement 3F ) . Notably , binding of digoxin to its target in human cells , Na+/K+ ATPase , is thought to be mediated by long-range electrostatic interactions between a Mg2+ ion and the electron rich , partially negatively charged oxygen atom of the unsaturated lactone ( Laursen et al . , 2015; Weigand et al . , 2014 ) . It is possible that the three remaining cysteine residues ( C82 , C265 , C535 ) could influence binding of a divalent metal cation that similarly positions or activates digoxin in the Cgr2 active site . To identify additional amino acids that may be important for Cgr2 function , we compared the Cgr2 sequences encoded within our collection of E . lenta genomes . Strikingly , only two cgr2 nucleotide variants were detected , which were validated by targeted Sanger sequencing . One of these variants is only found in the DSM 2243 type strain resulting in a conservative methionine ( M ) to valine ( V ) substitution at position 381 . The other results in either aromatic tyrosine ( Y ) as in the type strain DSM 2243 or neutral asparagine ( N ) at position 333 ( Figure 3A ) . We were also able to fully or partially reconstruct 14 additional cgr2 sequences using reads mapping to the cgr gene cluster from 96 gut microbiome datasets with a high abundance of E . lenta ( >1x coverage or >0 . 001 proportional abundance ) . These metagenome fragments confirmed the presence of both Y333 and N333 variants in a 9:5 ratio ( Figure 3A ) while the DSM 2243 M381 variant was not observed . To avoid biases against lower E . lenta coverage metagenomes , we also applied an assembly-free method based on calling variants from aligned reads ( Figure 3B ) . This uncovered 49 metagenomes with at least one read mapping over the variant position , confirming the bi-allelic nature with 15 Y333 and 34 N333 metagenomes . Nearly all metagenomes ( 41/42 ) with reads mapping to position 381 supported the valine residue suggesting that the DSM 2243 M381 variant is rare . Given that this analysis confirmed the highly conserved nature of the cgr locus , we analyzed the conservation of cgr2 in the context of the E . lenta pan-genome ( based on 24 sequenced isolates ) finding that it is at the 67th percentile of conservation . These results suggest that cgr2 sequence conservation is not unusual for this species , with the caveat that relatively few genomes were available for analysis ( Figure 3C ) . Given the ubiquity of variation at position 333 , we assessed its functional consequences by comparing the activity of the two Cgr2 variants in vivo and in vitro . E . lenta strains encoding the N333 variant show a trend towards a decreased ability to metabolize digoxin as compared to Y333-encoding strains ( Figure 3D; Figure 3—source data 1; p=0 . 052 Welch’s t-test ) . This decreased activity was more readily apparent following incubation of digoxin with Cgr2 proteins in vitro ( Figure 3E; Figure 3—source data 2 ) . While kinetic parameters for wild-type Cgr2 ( Y333 ) were KM = 94 . 6 ± 7 . 1 µM and a catalytic efficiency of 2 . 4 ± 0 . 8×103 M−1 s−1 , saturating Vmax conditions could not be reached for the N333 variant within the range of concentrations where digoxin is soluble ( ≤0 . 5 mM ) . Despite its lower activity , Cgr2 N333 converted digoxin to dihydrodigoxin at a comparable efficiency to Y333 after 4 . 5 hr ( Figure 3F; Figure 3—source data 3 ) . Compared to the activity of other FAD-dependent reductases towards their native substrates , the Y333 Cgr2 variant is less efficient for digoxin reduction ( Kemp et al . , 2010; Rohman et al . , 2013; Morris et al . , 1994; Bogachev et al . , 2012 ) ( Figure 3—source data 4 ) . This decreased activity could arise from impaired cofactor binding or reconstitution , or inefficient electron transfer in vitro in the absence of Cgr1 . Alternatively , these results could indicate that digoxin is not the endogenous substrate of Cgr2 . To systematically test for additional Cgr2 substrates , we assessed the enzyme’s activity toward 28 small molecules using a colorimetric assay ( Figure 4; Figure 4—figure supplement 1 ) . These molecules were selected based on their chemical similarity to digoxin and their relevance in the context of the human gut . Cgr2 displayed robust activity only toward cardenolides , the family of plant toxins that includes the pharmaceutical agents digoxin and digitoxin as well as ouabain , which is used as an arrowhead poison ( Michalak et al . , 2017 ) . The cardenolide aglycones digoxigenin and ouabagenin were metabolized at a significantly faster rate than their glycosylated forms digoxin and ouabain , respectively ( **p<0 . 01 , Student’s t test ) . The isolated lactone 2 ( 5H ) -furanone was minimally processed by Cgr2 , indicating that an intact steroid core is important for activity . However , the qualitatively similar rates observed for reduction of the various cardenolides demonstrates that the enzyme tolerates differences in the number and position of hydroxyl groups on the steroid scaffold . Overall , these results suggest that Cgr2 activity is restricted to cardenolide toxins and does not extend to other structurally related endogenous or exogenous compounds . Additionally , neither fumarate nor any of the metabolized cardenolides conferred a growth advantage to cgr2+ E . lenta in minimal or rich medias , suggesting that these compounds are not used as alternative terminal electron acceptors ( Figure 4—figure supplement 2 ) . The inability of Cgr2 to reduce fumarate , a common electron acceptor used during bacterial anaerobic respiration , led us to revisit the original annotation of Cgr2 as a ‘fumarate reductase’ ( Saunders et al . , 2009 ) . To more systematically assess the relationship between Cgr2 and biochemically characterized reductases , we constructed a sequence similarity network ( SSN ) using the 5000 most similar sequences from the UniProtKB protein database . Within the network , there were seven enzymes that had been biochemically characterized ( UniProtKB IDs: Q07WU7 , Q9Z4P0 , 8CVD0 , P71864 ) , biochemically and structurally characterized ( PDB IDs: 1D4D , 1E39 ) , or genetically characterized ( UniProtKB ID: Q7D5C1 ) ( Figure 5—source data 1 ) ( Leys et al . , 1999; Bogachev et al . , 2012; Brzostek et al . , 2005; Doherty et al . , 2000; Knol et al . , 2008; Rothery et al . , 2003; Pealing et al . , 1992; Dobbin et al . , 1999 ) . At all thresholds at which Cgr2 remained connected to other protein sequences , all characterized enzymes within the SSN were co-clustered , precluding the resolution of unique biochemical functions at this cutoff ( Figure 5A ) . At higher alignment thresholds that separated these characterized enzymes into discrete isofunctional clusters , Cgr2 was always present as a ‘singleton’ with no links to other protein sequences ( Figure 5B ) . Our SSN also revealed that reductase enzymes are widespread among human gut bacteria , with three validated enzymatic activities and 113 distinct clusters with uncharacterized biochemical functions detected in sequenced gut bacterial genomes . We validated our SSN by aligning the biochemically characterized enzymes within the network with additional co-clustered sequences to assess conservation of essential active site residues ( Figure 5—figure supplements 1–3 ) . Comparing the sequence of Cgr2 with sequences of biochemically characterized reductases revealed that Cgr2 lacks active site residues required for the activity of fumarate reductases ( 6/7 divergent residues ) , urocanate reductases ( 4/5 divergent residues ) , and ketosteroid dehydrogenases ( 3/5 divergent residues ) ( Figure 5—figure supplement 4A–D ) ( Leys et al . , 1999; Rohman et al . , 2013; Bogachev et al . , 2012; Knol et al . , 2008; Reid et al . , 2000 ) . Individually mutating the two residues shared between Cgr2 and ketosteroid dehydrogenases confirmed that one amino acid involved in substrate binding ( G536 backbone ) was also important for the activity of Cgr2 but the other ( Y532 ) was not ( Figure 5—figure supplement 4E ) . Together , the location of Cgr2 within the SSN and the differences in its sequence indicate that this enzyme is distinct from characterized bacterial reductases and may use a unique set of residues to catalyze cardenolide reduction . To assess the broader relevance of this cardenolide-metabolizing enzyme , we quantified the prevalence , conservation , and genomic context of cgr2 in the human gut microbiome . We mined gut microbiome datasets from 1872 individuals sampled in 6 countries across three continents ( Nayfach et al . , 2015 ) . Analysis of the E . lenta pan-genome led to the discovery of a single copy marker gene ( referred to here as elnmrk1 ) conserved in all sequenced strains that serves as a proxy for E . lenta abundance in both sequencing and quantitative PCR ( qPCR ) assays ( Bisanz et al . , 2018 ) . The abundance of elnmrk1 was significantly associated with E . lenta abundance across the individuals ( R2 = 0 . 973 , p<2 . 2e-16 ) . Using this marker gene , we detected E . lenta in 41 . 5% of subjects at abundances of −3 . 5 ± 0 . 58 mean ± SD log10 ( elnmrk1 copies/cell ) ( Figure 6A ) . Cgr2 was detectable in 48 . 5% of the E . lenta-positive individuals , while detection occurred in 27 . 7% of all subjects ( −3 . 7 ± 0 . 64 mean ± SD log10 ( cgr2 copies/cell ) ) ( Figure 6A ) , and the abundance of E . lenta and cgr2 was significantly associated ( rho = 0 . 455 , p<0 . 001; Figure 6B ) . The distributions were skewed towards participants with less cgr2 than expected based on the abundance of E . lenta [ ( skew = −0 . 773 , p=2 . 5e-8 , n = 375 , D’Agostino skewness test of log10 ( cgr/elnmrk1 ) with quantifiable elnmrk1 and cgr2] , consistent with prior data suggesting that many individuals are colonized by a mixture of cgr2+ and cgr2- E . lenta strains ( Haiser et al . , 2013 ) . These results were validated by qPCR in an independent set of 158 individuals ( 228 samples ) from multiple sites in the USA and Germany ( Figure 6—source data 1 ) revealing a similar skew towards higher E . lenta abundances versus cgr2 ( skew = −0 . 65 , p<0 . 001 , n = 165 ) . Using this more sensitive detection method , we detected E . lenta in 81 . 6% of individuals ( 1 . 5e7 ± 3 . 5e6 copies/g feces ) and cgr2 in 74 . 7% ( 1 . 5e6 ± 3 . 5e6 copies/g feces ) ( Figure 6C–D ) . Similar to the sequence-based analysis , the outliers were skewed towards samples with less cgr2 than expected based on the abundance of E . lenta . Overall , both the qPCR- and metagenomic sequencing-based analyses show that E . lenta and cgr2 are widely distributed in the human microbiome .
For over three decades , the human gut bacterium E . lenta has been linked to cardiac drug inactivation ( Saha et al . , 1983 ) . However , the identity , specificity , and distribution of the enzymes responsible for this activity were unknown . In this work , we unambiguously show that the E . lenta protein , Cgr2 , inactivates cardenolides , including the pharmaceutical agents digoxin and digitoxin that have been used for over two centuries in the treatment of cardiac diseases . Although we cannot rule out the possibility that other enzymes are involved in digoxin inactivation , no other microbes have been discovered that possess this metabolic activity apart from cgr2+ E . lenta ( Saha et al . , 1983 ) . Cgr2 represents a novel flavoprotein reductase , and contains oxygen-sensitive [4Fe-4S] cluster ( s ) and a divergent set of predicted active site residues . The failure of bioinformatic analyses to identify this essential [4Fe-4S] cluster highlights the need for additional structural and mechanistic studies of the cgr operon and other gut microbial enzymes involved in xenobiotic metabolism . Our working model is that Cgr1 and Cgr2 form a membrane-anchored , extracellular complex that mediates electron transfer from an electron donor ( e . g . the membrane quinone pool ) through multiple cytochromes c in Cgr1 to Cgr2 , which ultimately reduces the α , β-unsaturated γ-butyrolactone of digoxin and other cardenolides ( Figure 7A ) . We propose that the [4Fe-4S]2+ cluster ( s ) of Cgr2 sequentially transfer electrons to the FAD cofactor . The resulting hydride equivalent is then transferred to the cardenolide , and proton transfer generates the fully reduced γ-butyrolactone ( Figure 7B ) , yielding the therapeutically inactive metabolite dihydrodigoxin . While we have demonstrated that Cgr2 is necessary and sufficient for digoxin reduction in a heterologous host and in vitro using a chemical electron donor , additional proteins within the expanded cgr gene cluster may be important for digoxin reduction in vivo . Sequence analyses and transcriptional data suggest that Cgr1 is likely important for this metabolic activity in E . lenta . However , we were unable to observe overexpression or heme c incorporation into Cgr1 using a variety of heterologous constructs , hosts , and expression conditions , which may be due to an incompatibility of heterologous cytochrome c maturation factors and this protein ( Sanders and Lill , 2000; Thöny-Meyer , 2002 ) . The use of alternative heterologous systems that are more suitable for producing multi-heme cytochromes c ( Ozawa et al . , 2001; Kern and Simon , 2011 ) or the development of genetic tools in E . lenta would thus be required to obtain functional Cgr1 and determine its role in digoxin metabolism . In contrast to Cgr1 , the relevance of the Cac proteins to the Cgr proteins and digoxin metabolism is unclear . Apart from the putative LuxR-type regulator Cac3 , which was modestly upregulated in response to digoxin and may be involved in regulating transcription of the cgr operon , RNA sequencing does not show clear evidence for the expression of the other cac genes during growth in pure culture . While Cac4 is annotated as a secreted , FAD-dependent fumarate reductase , it lacks all known catalytic and binding residues for this enzyme class ( Figure 5—figure supplement 1 ) . Furthermore , Cac4 shares only 23% sequence identity to Cgr2 , and is thus likely to metabolize a different substrate than either of these enzymes . Cac6 is homologous to stomatin , prohibitin , flotillin , and HflK/C ( SPFH ) proteins , which are often associated with lipid rafts or functional microdomains in bacteria ( Bramkamp and Lopez , 2015 ) and could potentially interact with the Cgr enzymes , the substrates of the cgr operon , or additional steroidal substrates of E . lenta ( Ridlon et al . , 2006; Devlin and Fischbach , 2015 ) . Additional work is required to understand the biochemical function of the Cac proteins and whether they influence E . lenta metabolism of digoxin and/or other small molecules . We have demonstrated that E . lenta strains harboring the cgr operon are widespread in the human gut microbiome , which supports the high incidence of dihydrodigoxin production observed clinically . Strikingly , cgr2 and its associated genes are highly conserved , with two naturally occurring variants ( frequent: Y333/N333; infrequent: V381/M381 ) . This conservation is surprising given the strict specificity of Cgr2 towards cardenolides , which are ingested at very low concentrations to minimize toxicity in the context of cardiac therapy ( Smith et al . , 1969; Gheorghiade et al . , 2004 ) . These results , together with the overall high degree of conservation in the E . lenta pan-genome , suggest that much of the phenotypic variation within this species may be driven by gene gain/loss rather than by genetic polymorphisms . Data from other bacterial lineages suggests that this phenomenon may not be unique to Coriobacteriia ( Hao and Golding , 2006; Marri et al . , 2006; Marri et al . , 2007; Nowell et al . , 2014; Vos et al . , 2015 ) ; in Pseudomonas syringae , 1% amino acid divergence accumulates at the same time in hundreds or even thousands of genes ( Nowell et al . , 2014 ) . The high sequence conservation and levels of cgr operon transcription in response to digoxin exposure suggest that digoxin metabolism may provide a physiological benefit to E . lenta . However , as we could not observe any direct benefit of cardenolide metabolism for cgr+ E . lenta ( Figure 4—figure supplement 2 ) , we hypothesize that these bacteria may have evolved to protect the host from plant toxins and thus maintain a habitat for colonization . Although we cannot rule out an as-of-yet unidentified endogenous substrate of Cgr2 , our results suggest that similar to intestinal and hepatic enzymes , gut bacteria have co-evolved with the human host and maintain detoxification systems that can be rapidly and efficiently mobilized on demand . Additional studies are warranted that directly compare and contrast the gut microbiomes of herbivorous animals and insects that evolved under more constant exposure to cardenolides ( Agrawal et al . , 2012 ) , as these communities may represent a reservoir for evolutionary or functional homologs of cgr2 . While we do not yet understand the factors that maintain this gene in the E . lenta population in the absence of a direct selective pressure , our studies , coupled with ex vivo experiments with human fecal samples and in vivo experiments in mice ( Haiser et al . , 2013 ) , suggest that the metabolic activities of low abundance members of the gut microbiome can significantly influence host physiology . Our results also highlight an important consideration for ongoing efforts to predict and manipulate gut microbial metabolism , particularly in the context of therapeutics . Not only do E . lenta strains vary in the presence or absence of cgr2 , but we have also identified a naturally occurring , single amino acid substitution that causes a dramatic loss of activity in the Cgr2 enzyme . This result emphasizes the need for methods that can achieve nucleotide-level precision in mapping inter-individual differences in human gut microbiome gene content . Several clinical trials have recently investigated the use of digoxin for treating diverse cancers ( Lin et al . , 2014; Kayali et al . , 2011 ) , rheumatoid arthritis ( Huh et al . , 2011 ) and HIV-1 infection ( Wong et al . , 2013 ) . Paired studies of host genetics combined with nucleotide-resolution analyses of the gut microbiome are needed to test the feasibility of using microbial genetic information to predict drug bioavailability and improve treatment outcomes in these new disease contexts . Finally , our results demonstrate the feasibility of progressing from case studies of a clinically relevant gut microbial biotransformation ( Saha et al . , 1983 ) to identifying the responsible genes , enzymes , and biochemical mechanisms associated with metabolism . Isolation of individual xenobiotic metabolizing strains is a crucial first step towards uncovering the genetic and biochemical bases of these transformations . Starting from complex microbial communities ( e . g . human fecal samples ) , individual strains can be selectively enriched ( Kumano et al . , 2016; Martínez-del Campo et al . , 2015 ) or isolated ( Saha et al . , 1983; Bisanz et al . , 2018 ) to enable screening and identification of microbes that metabolize a xenobiotic of interest . The observation that many xenobiotic-processing or transporting genes are only upregulated in the presence of substrate can be leveraged to identify xenobiotic-metabolizing genes using techniques such as RNA-seq ( Haiser et al . , 2013 ) or native protein purification ( Kumano et al . , 2016 ) . Genetic and/or heterologous expression experiments can then help to validate the role of the identified genes and enzymes in xenobiotic transformation . Finally , the responsible genes can serve as candidate biomarkers to probe the distribution and potential for xenobiotic metabolism in ex vivo incubations or in relevant clinical populations . These studies also provide new insights into the chemistry made possible by complex host-associated microbial communities . The gut microbiome encodes over 3 million genes ( Qin et al . , 2010 ) and >50% have unknown functions ( Human Microbiome Project Consortium , 2012 ) . Our study of just one highly unique , clinically relevant gut microbial enzyme has illuminated underappreciated functional diversity within the broader flavin-dependent reductases , which is widespread among human gut microbes ( Figure 5 ) . As reductive transformations represent a major route by which gut microbes metabolize xenobiotics , these unique , putative reductase enzymes provide a promising starting point for identifying additional gut microbiome-xenobiotic interactions . The approaches described here , together with high-throughput methods to characterize protein subfamilies and advances in bacterial culturing and genetic tools ( Goodman et al . , 2011; Lim et al . , 2017 ) , are beginning to unlock the genetic ‘dark matter’ of the human gut microbiome as well as its critical role in the etiology and treatment of human disease .
Publically available genomes were retrieved from NCBI ( E . lenta DSM 2243 , PRJNA21093; E . lenta FAA1-3-56 PRJNA40023 ) . New isolates were sequenced as described elsewhere ( Bisanz , et al . 2018 ) . Genomes were assembled with SPAdes 3 . 11 . 1 ( 84 ) and annotated with Prokka 1 . 12 ( Seemann , 2014 ) . All E . lenta strains studied were identified as E . lenta based on 16S rRNA sequencing and were de-replicated at the strain level by considering a pairwise average nucleotide identity ( ANI ) >99 . 99% as the same strain ( github . com/widdowquinn/pyani ) . The maximum and minimum ANI between studied E . lenta strains were 97 . 9% and 99 . 6% respectively . The phylogenetic tree was prepared using a set of 400 conserved proteins ( Segata et al . , 2013 ) rooted on the Gordonibacter strains . Newly sequenced strains were included as part of Bioproject PRJNA412637 . Global nucleotide and amino acid identity and related statistics were determined via the Needleman-Wunsh implementation in the pairwiseAlignment function of Biostrings with percentage identity calculated as 100*identical positions/ ( aligned positions + internal gaps ) . Statistical and graphical analysis was carried out using R 3 . 4 . 0 . RNA sequencing reanalysis was carried out by mapping reads to the reference genome with Bowtie2 ( Langmead and Salzberg , 2012 ) , counting with HTSeq ( Anders et al . , 2015 ) , and differential expression analysis using DESeq2 ( Love et al . , 2014 ) . Original sequence data is available from the SRA with project identifier SRP018311 . For the purposes of comparative genomics , gene conservation was calculated by first clustering into orthologous clusters with proteinortho5 ( Lechner et al . , 2011 ) with a minimum 60% amino acid identity and 80% coverage . A presence/absence matrix de-replicated for co-occurring features was then used as the input for a random forest classifier ( randomForest 4 . 6–12 ) . Variable importance ( mean decrease GINI ) was used to extract the 15 most important features . A tool for this comparative genomic analysis is available as ElenMatchR ( jbisanz . shinyapps . io/elenmatchr; copy archived at https://github . com/elifesciences-publications/ElenMatchR ) ( Bisanz and Turnbaugh , 2018 ) with digoxin reduction available as a demonstration dataset . Eggerthella lenta and related strains were grown in BBL Brain Heart Infusion ( BHI ) media ( BD , Franklin Lakes , NJ ) supplemented with L-arginine ( Sigma-Aldrich , St . Louis , MO ) under an atmosphere of 2–5% H2 , 2–5% CO2 , and balance N2 . Strains were streaked onto BHI agar plates supplemented with 1% arginine ( w/v ) in an anaerobic chamber ( Coy Laboratory Products , Grass Lakes , MI ) . Individual colonies were inoculated into 16 × 125 mm Hungate tubes ( Chemglass Life Sciences , Vineland , NJ ) containing 5–10 mL of BHI supplemented with 1% arginine and grown at 37°C for 2–3 days . Cardiac glycoside substrates were dissolved at a concentration of 10 mM in dimethylformamide ( DMF ) and added to cultures at a final concentration of 10 µM . Starter cultures were diluted into 10 mL of BHI + substrate to a starting of OD600 of 0 . 05 and grown anaerobically at 37°C for 2 days . Experiments were performed in biological triplicate . For growth assays , E . lenta DSM 2243 was grown in either rich ( BHI ) or defined media . Basal media lacking terminal electron acceptors was prepared as previously described ( Löffler et al . , 2005 ) with the following modifications: yeast extract and tryptone were each added to 0 . 1% ( w/v ) , L-cysteine concentration was 0 . 4 mM , sodium sulfide was not added , and either 5% H2 or 10 mM sodium acetate were used as electron donors . Starter cultures were prepared as described above in BHI media supplemented with 1% arginine , and diluted 1:100 into media that had been supplemented with substrates ( dissolved in DMF ) to a final concentration of 10 µM . Cultures were grown anaerobically at 37°C in biological triplicate . OD600 measurements were recorded on a Genesys20 spectrophotometer ( Thermo Fisher Scientific , Waltham , MA ) . Bacterial cultures were centrifuged ( 10 min x 4000 rpm ) , 1 mL of supernatant was extracted three times with 1 mL of dichloromethane and the pooled organic fractions were concentrated using a rotary evaporator . Samples were resuspended in 1 mL of 50% methanol in water and diluted 10x prior to liquid chromatography-tandem mass spectrometry ( LC-MS/MS ) analysis . Metabolites were detected on an Agilent 6410 Triple Quad LC/MS using electrospray ionization in negative ion mode . The mass spectrometer settings were as follows: gas temperature ( 300°C ) , gas flow ( 10 L/min ) , nebulizer pressure ( 25 psi ) , capillary voltage ( 4000 V ) , and chamber current ( 0 . 1 µA ) . Digoxin was monitored using a 779 . 4 → 649 . 3 m/z transition with a fragmentor voltage of 250V and collision energy of 52 , and dihydrodigoxin was monitored using a 781 . 4 → 521 . 3 m/z transition with a fragmentor voltage of 200V and collision energy of 20 . Standard curves were prepared using 0 . 01–1 µM samples of each compound . Digoxin was purchased from Sigma-Aldrich ( St . Louis , MO ) , and a dihydrodigoxin standard was obtained through chemical hydrogenation of digoxin as previously described ( Haiser et al . , 2013 ) . Liquid chromatography was performed on an Acclaim Polar Advantage II column with a flow rate of 0 . 125 mL/min , 5 µL sample injection , solvent A ( 10% methanol + 1 mM ammonium hydroxide ) and solvent B ( 100% methanol + 1 mM ammonium hydroxide ) , and a gradient: 70–100% B over 10 min , 100% B for 1 . 5 min , 100–70% B over 3 . 5 min , and 70% B for 7 min . E . lenta DSM 2243 was grown in 5 mL of BHI + 1% arginine at 37°C . After 2 days , the culture was pelleted and genomic DNA ( gDNA ) was purified with the UltraClean Microbial DNA Isolation Kit ( QIAGEN , Germantown , MD ) according to the manufacturer’s protocol . The cgr operon was amplified from 50 ng of gDNA in a 50 µL reaction volume with 0 . 5 µM of each primer ( Table 1 ) and Phusion High-Fidelity PCR master mix with HF buffer ( New England Biolabs , Ipswich , MA ) . The following thermocycling parameters were used: denaturation at 98°C for 3 min; 35 cycles of 98°C for 15 s , 71°C for 20 s , and 72°C for 75 s; and a final extension at 72°C for 5 min . The PCR reaction was analyzed by agarose gel electrophoresis ( 1% agarose gel ) , and the cgr amplicon was excised and purified with the Illustra GFX PCR DNA and Gel Band Purification kit ( GE Healthcare , Chicago , IL ) . Cgr1 and cgr2 variants were amplified in 20 µL PCR reactions using 1 ng of purified cgr operon as template , 0 . 5 µM primer pairs and Phusion High-Fidelity PCR master mix with HF buffer ( New England Biolabs , Ipswich , MA ) ( Table 1 ) . PCR conditions were as follows: denaturation at 98°C for 2 min; 35 cycles of 10 s at 98°C , 10 s at specified annealing temperature , and 72°C for the specified extension time; and a final extension at 72°C for 5 min . Cgr amplicons were digested in a 30 µL reaction with 1 . 5 µL of each restriction enzyme ( New England Biolabs , Ipswich , MA ) for 2 . 5 hr at 37°C . pTip vectors were similarly digested , and the linearized vector was excised from a 1% agarose gel and purified . Insert and vector pairs were ligated at a 1:3 ratio at room temperature for 2 hr with T4 DNA ligase ( New England Biolabs , Ipswich , MA ) . 2 . 5 µL of the ligation reaction was transformed into chemically competent One Shot Top10 E . coli cells ( Thermo Fisher Scientific , Waltham , MA ) and plated on LB agar with ampicillin . Plasmid inserts were sequenced using the primers listed in Table 1 . Site-directed mutagenesis was performed in 25 µL reactions using 200 ng of template DNA ( Cgr2 ( –48aa ) -NHis6 in pTipQC2 ) , 0 . 5 µM of each primer pair ( Table 2 ) , 0 . 5 mM dNTP , and 1 µL of Pfu Turbo polymerase AD ( VWR , Radnor , PA ) . The following thermocycling parameters were used: denaturation at 95°C for 1 min; 18 cycles of 95°C for 30 s , 65°C for 50 s , and 68°C for 22 min ( 2 min/kb ) ; and a final extension at 68°C for 7 min . The template plasmid was digested with 1 µL of DpnI ( New England Biolabs , Ipswich , MA ) for 1 hr at 37°C , and 2 µL of the reaction were transformed into chemically competent One Shot Top10 E . coli cells ( Thermo Fisher Scientific , Waltham , MA ) . All Rhodococcus strains and expression vectors were obtained from the National Institute of Advanced Industrial Science and Technology ( Tokyo , Japan ) . 40 ng of plasmid DNA were added to 400 µL of R . erythropolis L-88 electrocompetent cells in 30% PEG 1000 ( Sigma-Aldrich , St . Louis , MO ) in a 2 mm gap electroporation cuvette ( VWR , Radnor , PA ) . Cells were transformed in a MicroPulser electroporator ( Bio-Rad , Hercules , CA ) with a 2 . 5 kV pulse ( time constant ~4 . 8 – 5 . 2 ) , rescued with 0 . 6 mL of LB ( Lennox ) broth ( Alfa Aesar , Tewksbury , MA ) , and incubated for 4 hr at 28°C , 175 rpm . Cells were plated onto LB agar plates +antibiotic ( 17 µg/mL chloramphenicol for pTipQC plasmids; 8 µg/mL tetracycline for pTipQT plasmids ) and incubated at 28°C for 5–7 days . Single colonies were inoculated into 50–75 mL of LB +antibiotic ( 34 µg/mL chloramphenicol or 8 µg/mL tetracycline ) and grown for 3–5 days at 28°C , 175 rpm until reaching saturation . For gain of function studies , 50 mL of LB and antibiotic were inoculated to a starting OD600 of 0 . 2 and grown at 28°C , 175 rpm . Experiments were performed in biological triplicate . When cultures reached an OD600 of 0 . 6 ( ~6–8 hr ) , protein expression was induced with thiostrepton ( Sigma-Aldrich , St . Louis , MO ) at a final concentration of 0 . 01 µg/mL , and cultures were incubated at 15°C , 175 rpm . In cultures where Cgr1 was overexpressed , media was supplemented with the heme precursor δ-amino levulinic acid hydrochloride ( 50 µg/mL final concentration ) ( Frontier Scientific , Logan , Utah ) . After 16–20 hr , digoxin was added to cultures as a solution in DMF at a final concentration of 10 µM and incubated for either 7 days at 15°C , or 2 days at 28°C , 175 rpm . Culture supernatants were extracted and analyzed as previously described . For large-scale purifications , 2 L of LB-chloramphenicol in a 4 L baffled flask were inoculated to a starting OD600 of 0 . 02 and grown to an OD600 of 0 . 6 ( ~18–25 hr ) . Protein expression was induced with 0 . 01 µg/mL thiostrepton , and cultures were incubated at 15°C , 175 rpm for approximately 21 hr before harvesting cells by centrifugation ( 10 , 800 rpm x 20 min ) . Cell pellets were frozen and stored at –80°C . All protein purification steps were carried out at 4°C . Harvested cells were resuspended in 5 mL/g of cell pellet in lysis buffer ( 50 mM Tris , pH 8 , 1 mM MgCl2 , 25 mM imidazole ) containing Pierce EDTA-free protease inhibitor cocktail ( Thermo Fisher Scientific , Waltham , MA ) . Cells were passaged through a cell disruptor ( Avestin EmulsiFlex-C3 ) five times at 15 , 000–25 , 000 psi and centrifuged for 20 min at 13 , 000 rpm . The clarified lysate was incubated on a nutating mixer with 5–10 mL of HisPur Ni-NTA resin ( Thermo Fisher Scientific , Waltham , MA ) for 1 hr and then applied to a gravity flow column . The resin was washed with 50 mL of wash buffer ( 25 mM HEPES , 0 . 5 M NaCl , pH 8 , 25 mM imidazole ) and eluted with 25 mL of elution buffer ( 25 mM HEPES , 0 . 5 M NaCl , pH 8 , 200 mM imidazole ) . Eluted protein was concentrated using a 20 mL Spin-X UF 30 k MWCO PES spin filter ( Corning , Corning , NY ) to a volume of 1–2 . 5 mL , and then desalted on a Sephadex G-25 PD-10 desalting column ( GE Healthcare , Chicago , IL ) that had been equilibrated with desalting buffer ( 50 mM HEPES , 100 mM NaCl , pH 8 ) . Desalted protein was sparged with argon on ice for 30–45 min . Chemical reconstitution of [Fe-S] cluster ( s ) in Cgr2 was carried out at 4°C in an anaerobic chamber ( Coy Laboratory Products , Grass Lakes , MI ) under an atmosphere of 2% hydrogen and 98% nitrogen . A 30 µM solution of Cgr2 was prepared in reconstitution buffer ( 50 mM HEPES , 100 mM NaCl , pH 8 , and 2 mM dithiothreitol ( DTT ) ) . Fe ( NH4 ) 2 ( SO4 ) 2·6H2O ( Sigma-Aldrich , St . Louis , MO ) was added in four aliquots over 60 min , followed by addition of Na2S·9H20 ( Sigma-Aldrich , St . Louis , MO ) in four aliquots over 60 min to final concentrations of 0 . 24 or 0 . 375 mM ( 8 or 12 . 5 equivalents relative to Cgr2 ) , and stirred for 16–24 hr . The reaction was filtered through a 0 . 25 mm , 0 . 2 µM pore-size PES syringe filter ( VWR , Radnor , PA ) to remove precipitant and concentrated in a 6 mL Spin-X UF 30 k MWCO PES spin filter inside a 50 mL conical-bottom centrifuge tube with plug seal cap ( Corning , Corning , NY ) . The concentrated protein ( 1–2 . 5 mL ) was desalted on a PD-10 column into desalting buffer . Protein was aliquoted into 0 . 5 mL PP conical tubes with skirt ( Bio Plas ) , sealed inside 18 × 150 mm Hungate tubes ( Chemglass Life Sciences , Vineland , NJ ) and stored at –80°C . Protein concentration was determined by Bradford using Protein Assay Dye Reagent ( Bio-Rad , Hercules , CA ) and bovine serum albumin ( BSA ) ( Sigma-Aldrich , St . Louis , MO ) as a reference standard . Typical protein yields were ~20 mg/L of culture for both wild-type and point mutants of Cgr2 ( –48aa ) -NHis6 , ~8 mg/L for Cgr2 ( –48aa ) -CHis6 , and ~1 mg/L for Cgr2-CHis6 . The iron and sulfur content of Cgr2 samples ( protein concentrations between 20–50 µM ) was determined using previously reported colorimetric assays ( Craciun et al . , 2014 ) . Thermal denaturation assays of purified and reconstituted Cgr2 were prepared on ice in 0 . 2 mL skirted 96-well PCR plates ( VWR , Radnor , PA ) sealed with optical adhesive covers ( Life Technologies , Woburn , MA ) . Each reaction contained 10 µg of purified or reconstituted Cgr2 , Sypro Orange protein gel stain ( Thermo Fisher Scientific , Waltham , MA ) diluted 5000-fold , and buffer containing 100 mM buffering agent and 100 mM NaCl in a total volume of 30 µL . The following buffering agents were used: acetate/acetic acid for pH 4–6 , HEPES for pH 7 , Tris-HCl for pH 8–9 , and glycine-NaOH for pH 10 . For metal binding assays , metal salts ( Sigma-Aldrich , St . Louis , MO ) were dissolved in pH 8 buffer to generate 100 mM stock solutions and added to a final concentration of 48 µM ( 8 equivalents relative to Cgr2 ) . Data was collected on a CFX96 Touch Real-Time PCR machine ( Bio-Rad , Hercules , CA ) using the ‘FRET’ filter setting with FAM excitation and HEX emission channels ( 485 nm and 556 nm respectively ) . The following temperature-scan protocol was used: 25°C for 30 s , then ramp from 25°C to 100°C at a rate of 0 . 1 °C/ s . Gel filtration experiments were carried out on a Superdex 200 10/300 GL column ( GE Healthcare , Chicago , IL ) attached to a BioLogic DuoFlow chromatography system ( Bio-Rad , Hercules , CA ) . Experiments were carried out either aerobically or anaerobically inside a Coy anaerobic chamber ( Coy Laboratory Products , Grass Lakes , MI ) . 100 µL protein samples ( 50–100 µM ) were loaded onto the column at a rate of 0 . 2 mL/min for 1 mL followed by an isocratic flow of 0 . 33 mL/min for 30 mL with 50 mM HEPES , 100 mM NaCl , pH 8 . The molecular weight for Cgr2 ( –48aa ) -NHis6 is 55 . 7 kDa . A gel filtration standard ( Bio-Rad , Hercules , CA ) containing thyroglobulin ( 670 kDa ) , γ-globulin ( 158 kDa ) , ovalbumin ( 44 kDa ) , myoglobin ( 17 kDa ) , and vitamin B12 ( 1 . 35 kDa ) was used to determine the molecular weight of Cgr2-containing peaks . Cgr2 was diluted to a final concentration of 50–100 µM in UV-Star UV-transparent 96-well microplates ( Greiner Bio-One , Monroe , NC ) . The absorbance was measured between 250–750 nm using a PowerWave HT Microplate Spectrophotometer ( BioTek , Winooski , VT ) inside of an anaerobic glovebox ( Mbraun , Stratham , NH ) . Curves were baseline subtracted using respective absorbance values at 700 nm . To determine whether the [Fe-S] cluster ( s ) were redox active , Cgr2 was incubated with 10 equivalents of sodium dithionite ( Sigma-Aldrich , St . Louis , MO ) for 15 min at room temperature prior to taking additional absorption spectra . To assess the oxygen sensitivity of [Fe-S] cluster ( s ) , Cgr2 was taken out of the anaerobic chamber and exposed to oxygen , and the absorption spectra was measured aerobically on a PowerWave HT Microplate Spectrophotometer ( BioTek , Winooski , VT ) . Oxygen-exposed Cgr2 was then sparged for 30 min with argon ( on ice ) and brought back into the Mbraun glovebox for activity assays . All samples were prepared in 50 mM HEPES , 100 mM NaCl , pH 8 under oxygen-free conditions in an anaerobic glovebox ( Mbraun , Stratham , NH ) . For all EPR experiments the final concentration of Cgr2 was either 150 or 200 µM . When required , the samples were reacted with an excess of sodium dithionite ( 10–20 equivalents ) for 20–30 min at 22°C prior to freezing in liquid N2 . Spin quantification was carried out against a Cu2+-EDTA standard containing an equimolar concentration of CuSO4 in 10 mM EDTA ( 150 or 200 µM ) , under non-saturating conditions . Samples ( 250 µL ) were loaded into 250 mm length , 4 mm medium wall diameter Suprasil EPR tubes ( Wilmad LabGlass , Vineland , NJ ) and frozen in liquid N2 under oxygen-free conditions . EPR spectra were acquired on a Bruker E500 Elexsys continuous wave ( CW ) X-Band spectrometer ( operating at approx . 9 . 38 GHz ) equipped with a rectangular resonator ( TE102 ) and a continuous-flow cryostat ( Oxford 910 ) with a temperature controller ( Oxford ITC 503 ) . The spectra were recorded at variable temperatures between 10–40 K at a microwave power of 0 . 2 mW , using a modulation amplitude of 0 . 6 mT , a microwave frequency of 9 . 38 GHz , a conversion time of 82 . 07 ms , and a time constant of 81 . 92 ms . Methyl viologen ( paraquat ) dichloride hydrate ( Sigma-Aldrich , St . Louis , MO ) that had been reduced with sodium dithionite was used as an artificial electron donor ( Watanabe and Honda , 1982 ) to initiate anaerobic Cgr2-mediated reduction of digoxin in vitro . Assays were carried out at 25°C in an anaerobic glovebox ( Mbraun , Stratham , NH ) under an atmosphere of nitrogen and < 5 ppm oxygen . Reagents were brought into the glovebox as solids or sparged liquids and resuspended in anoxic buffer inside the chamber: flavin ( FAD or FMN ) and methyl viologen ( MV ) were resuspended in 50 mM HEPES , 100 mM NaCl , pH 7 to generate stock solutions of 1 mM and 50 mM respectively; sodium dithionite was resuspended in 50 mM HEPES , 100 mM NaCl , pH 8 to generate a stock solution of 25 mM; all substrates ( Figure 4—figure supplement 1 ) were dissolved in DMF to generate stock solutions of 25 mM , with the exception of sodium fumarate dibasic and urocanic acid which were dissolved in water . All substrates and reagents were purchased from Sigma-Aldrich ( St . Louis , MO ) except for the bufadienolides ( Enzo Life Sciences , Farmingdale , NY ) and prostaglandins ( Cayman Chemicals , Ann Arbor , MI ) . The final assay mixture ( 100 µL ) contained 5 µM Cgr2 , 50 µM flavin , 0 . 375 mM MV , 0 . 25 mM dithionite , and was initiated by addition of 0 . 5 mM substrate . For metal activation studies , metal salts were dissolved in pH 7 buffer ( 1 mM ) and added to a final concentration of 40 µM . Assays were prepared in a 96-well polysterene microplate ( Corning , Corning , NY ) and activity was continuously monitored by measuring the absorbance at 600 nm on a PowerWave HT Microplate Spectrophotometer ( BioTek , Winooski , VT ) ; a decrease in the absorbance at 600 nm corresponded to MV oxidation coupled to substrate reduction . For endpoint assay , reactions were quenched in methanol , diluted to a final concentration of 1 µM in 50% methanol , and analyzed by LC-MS/MS as previously described . Kinetic assays were performed in an anaerobic glovebox ( Mbraun , Stratham , NH ) at 25°C . Reactions were run in triplicate ( 200 µL ) in assay buffer containing 5 µM Cgr2 , 500 µM FAD , 1 . 5 mM MV , and 1 mM sodium dithionite , and were initiated by addition of digoxin as a solution in DMF to a final concentration of 0 . 01 , 0 . 025 , 0 . 05 , 0 . 1 , 0 . 2 , 0 . 25 , 0 . 3 , and 0 . 5 mM . 20 µL reaction aliquots were quenched in 180 µL of ice-cold methanol in Costar flat bottom polysteryene 96-well plates ( Corning , Corning , NY ) . The plates were sealed with adhesive aluminum foil for 96-well plates ( VWR , Radnor , PA ) , brought out of the anaerobic chamber , and further diluted ( 50-fold ) into 50% methanol . Digoxin and dihydrodigoxin standard curves were prepared in the full reaction matrix and processed identically such that final concentrations ( after 500x total dilution ) generated a standard curve between 0 . 01–1 µM . Plates were centrifuged ( 4000 rpm x 10 min , 4°C ) and 200 µL of each reaction were transferred to a 0 . 5 mL PP 96-well plate ( Agilent Technologies , Santa Clara , CA ) sealed with EPS easy piercing seals ( BioChromato , San Diego , CA ) . The reactions were monitored by LC-MS/MS as previously described , except that samples were directly injected ( no column ) , and isocratic flow was used with 75% methanol with 1 mM ammonium hydroxide . The chemical similarity of all substrates was assessed using the ChemMine software ( http://chemminetools . ucr . edu ) ( Backman et al . , 2011 ) . Substrates were imported into ChemMine in SMILES format . The hierarchical clustering tool was used to generate a heatmap visualizing the structural distance matrix between each substrate and digoxin . The full length Cgr2 protein sequence from E . lenta DSM 2243 was used as a query for BLASTP ( Atlschul et al . , 1997 ) using the NCBI non-redundant protein sequence database ( search performed 9/26/17 ) . Cgr2 was also used to query the HHPred prediction tool ( https://toolkit . tuebingen . mpg . de/#/tools/hhpred ) to identify additional remote protein homologs using hidden Markov models ( Alva et al . , 2016 ) . The PDB_mmCIF70_27_Aug database was used ( search performed 9/26/17 ) . A SSN was generated using the EFI-EST tool ( http://efi . igb . illinois . edu/efi-est/ ) ( Gerlt et al . , 2015 ) . The full length ( native ) Cgr2 protein sequence was used as an input to generate a network with the 5000 most similar sequences from the UniProtKB protein database . An initial alignment score cutoff of 10−66 generated a SSN with 2018 nodes ( with 100% identity ) and 317 , 130 edges . The SSN was imported into Cytoscope v 3 . 2 . 1 and visualized with the ‘Organic layout’ setting . Seven characterized enzymes were present within the network ( UniProtKB IDs: fumarate reductases: P83223 , P0C278 , Q07WU7 , Q9Z4P0; urocanate reductase: Q8CVD0; 3-oxosteroid-1-dehydrogenase: P71864; Q7D5C1 ) . The alignment score cutoff was increased to e-value <10−130 , until enzymes with known functions separated into putatively isofunctional clusters . At this threshold , Cgr2 appears as a singleton . The network shown in Figure 5A was generated with a cutoff of e-value <10−50 , a threshold at which nearly all protein sequences form one cluster . Multiple sequence alignments were generated in Geneious and visualized in Jalview ( clustalx coloring ) . To validate that the clusters in the SSN likely contained isofunctional proteins , Cgr2 was aligned with characterized enzymes and additional selected proteins within the corresponding clusters of the SSN , and the alignment was analyzed for the presence of conserved active site residues involved in substrate binding , activation and proton transfer ( Leys et al . , 1999; Rohman et al . , 2013; Bogachev et al . , 2012; Knol et al . , 2008; Reid et al . , 2000 ) . E . lenta and cgr2 prevalence were determined using the copy number abundance ( gene copies/cell ) as derived from Metaquery2 ( Nayfach et al . , 2015 ) using the median abundance from individuals with repeated sampling . E . lenta abundance was determined from a single copy E . lenta marker gene described elsewhere ( elnmrk1 ) ( Bisanz et al . , 2018 ) . Matches were required to have a minimum 90% nucleotide identity and query/target coverage . Reconstruction of metagenomic cgr2 sequences was carried out by quality trimming reads from 96 metagenomes with >0 . 001 proportional abundance of E . lenta or >1 fold coverage using default sliding window settings with Trimmomatic ( Bolger et al . , 2014 ) and extracting reads which mapped to the cgr cluster and associated intergenic space ( 2957889 . . 2968387 ) in the reference DSM 2243 assembly with Bowtie 2 . These were assembled and annotated as above . Alignments to Cgr2 in metagenomic coding sequences were filtered by a global alignment identity of >80% to position 333 ± 60 residues . For assembly-free variant calling , reads were filtered for a minimum mapping quality of 10 and a pileup was created ( SAMtools ) . 49 metagenomes had at least one read mapping to the variant position ( 2959294 ) . Variants were called when > 50% of reads at a site supported an alternative sequence from the reference . Conservation of nucleotide sequence in isolates was independently confirmed via Sanger sequencing ( GENEWIZ , San Francisco , CA , USA ) using the following primers: cgr2_fwd ( TGCAATCAAGACAACCACGA ) , cgr2_internal ( TCGGTGTACAACCACAATGC ) , and cgr2_rev ( GTTGCGCTGTGATTAGACTG ) . PCR was carried out with high-fidelity Q5 enzyme ( New England Biolabs , Ipswich , MA ) . To validate metagenomics inquiries , qPCR analysis with double-dye probes was carried out in a duplexed fashion using the following primers and probes: ElentaUni_F ( GTACAACATGCTCCTTGCGG ) , ElentaUni_R ( CGAACAGAGGATCGGGATGG ) , ElentaUni_Probe ( [6FAM]TTCTGGCTGCACCGTTCGCGGTCCA[BHQ1] ) , cgr2_F ( GAGGCCGTCGATTGGATGAT ) , cgr2_R ( ACCGTAGGCATTGTGGTTGT ) , and cgr2_probe ( [HEX]CGACACGGAGGCCGATGTCG[BHQ1] ) . Reactions were carried out in triplicate using 10 µL reactions with 200 nM primers and probes using BioRad Universal Probes Supermix on a BioRad CFX 384 thermocycler according to the manufacturer’s suggested settings for fast cycles with a 60 °C annealing temperature . The estimated assay detection limit based on spike-in experiments is 1 . 4 × 103 GE/g after accounting for DNA extraction . Human samples were collected for the purpose of microbiome analysis as part of the following registered studies: NCT03022682 , NCT01967563 , and NCT01105143 and approved by their respective institutional review boards . DNA was extracted with variable methods using either MoBio Power Soil ( QIAGEN , Germantown , MD ) , Qiagen Fast Stool ( QIAGEN , Germantown , MD ) , or Promega Wizard ( Promega , Madison , WI ) SV 96 kits . All statistical analysis was carried out using either Student’s t-test as implemented in Graphpad Prism version 7 ( La Jolla , CA , USA ) or R version 3 . 4 . 0 using appropriate base functions for Welch’s t-test , Pearson and Spearman correlations , and ANOVA with multcomp version 1 . 4–6 for Dunnett’s multiple comparison test . Graphing was carried out with Graphpad Prism and R using ggplot2 version 2 . 2 . 1 . Skewedness was calculated using the R package Moments version 0 . 14 . | Trillions of microbes live within the human gut and influence our health . In particular , these microbes can modify food and drugs into compounds ( metabolites ) that humans cannot produce on their own . These compounds are often beneficial to the human host , but in some cases – for example , if the modification alters how a drug works – can be detrimental . Digoxin is a toxic chemical produced by plants that , in low doses , can be used to treat heart conditions . It has been known for decades that the human gut bacterium Eggerthella lenta transforms digoxin into a metabolite that is an ineffective drug . Microbes use biological catalysts called enzymes to produce metabolites , but it was not known which enzymes enable E . lenta to modify digoxin . Using biochemical and genomic techniques , Koppel et al . now show that an enzyme called Cgr2 inactivates digoxin and other related plant toxins . Data about the gut microbes in nearly 1 , 900 people from three continents revealed that bacteria that can produce Cgr2 were present in the guts of more than 40% of the individuals , although often in low abundance . Further experiments did not reveal any obvious benefits that E . lenta gains from modifying digoxin . Instead , Koppel et al . propose that the bacteria carry out this modification to protect their human host from plant toxins . The results presented by Koppel et al . emphasise that the activities of gut microbes should be considered when designing new drugs or assessing how they work in the human body . The strategies used to identify Cgr2 could now be applied to discover other important gut microbe-drug interactions . Ultimately , this knowledge will help us to predict and control the activities of gut microbes in ways that could improve human health . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2018 | Discovery and characterization of a prevalent human gut bacterial enzyme sufficient for the inactivation of a family of plant toxins |
While traditional microbiological freshwater tests focus on the detection of specific bacterial indicator species , including pathogens , direct tracing of all aquatic DNA through metagenomics poses a profound alternative . Yet , in situ metagenomic water surveys face substantial challenges in cost and logistics . Here , we present a simple , fast , cost-effective and remotely accessible freshwater diagnostics workflow centred around the portable nanopore sequencing technology . Using defined compositions and spatiotemporal microbiota from surface water of an example river in Cambridge ( UK ) , we provide optimised experimental and bioinformatics guidelines , including a benchmark with twelve taxonomic classification tools for nanopore sequences . We find that nanopore metagenomics can depict the hydrological core microbiome and fine temporal gradients in line with complementary physicochemical measurements . In a public health context , these data feature relevant sewage signals and pathogen maps at species level resolution . We anticipate that this framework will gather momentum for new environmental monitoring initiatives using portable devices .
The global assurance of safe drinking water and basic sanitation has been recognised as a United Nations Millennium Development Goal ( Bartram et al . , 2005 ) , particularly in light of the pressures of rising urbanisation , agricultural intensification , and climate change ( Haddeland et al . , 2014; Schewe et al . , 2014 ) . Waterborne diseases represent a particular global threat , with zoonotic diseases such as typhoid fever , cholera , or leptospirosis resulting in hundreds of thousands of deaths each year ( Prüss et al . , 2002; Prüss-Ustün et al . , 2019 ) . To control for risks of infection by waterborne diseases , microbial assessments can be conducted . While traditional microbial tests focus on the isolation of specific bacterial indicator organisms through selective media outgrowth in a diagnostic laboratory , this cultivation process is all too often time consuming , infrastructure-dependent and lacks behind in automatisation ( Salazar and Sunagawa , 2017; Tringe and Rubin , 2005 ) . Environmental metagenomics , the direct tracing of DNA from environmental samples , constitutes a less organism-tailored , data-driven monitoring alternative . Such approaches have been demonstrated to provide robust measurements of relative taxonomic species composition as well as functional diversity in a variety of environmental contexts ( Almeida et al . , 2019; Bahram et al . , 2018; Tara Oceans coordinators et al . , 2015 ) , and overcome enrichment and resolution biases common to culturing ( Salazar and Sunagawa , 2017; Tringe and Rubin , 2005 ) . However , they usually depend on expensive stationary equipment , specialised operational training and substantial time lags between fieldwork , sample preparation , raw data generation and access . Combined , there is an increasing demand for freshwater monitoring frameworks that unite the advantages of metagenomic workflows with high cost effectiveness , fast technology deployability , and data transparency ( Gardy and Loman , 2018 ) . In recent years , these challenges have been revisited with the prospect of mobile DNA analysis . The main driver of this is the ‘portable’ MinION device from Oxford Nanopore Technologies ( ONT ) , which enables real-time DNA sequencing using nanopores ( Jain et al . , 2016 ) . Nanopore read lengths can be comparably long , currently up to ~2*106 bases ( Payne et al . , 2019 ) , which is enabled by continuous electrical sensing of sequential nucleotides along single DNA strands . In connection with a laptop for the translation of raw voltage signal into nucleotides , nanopore sequencing can be used to rapidly monitor long DNA sequences in remote locations . Although there are still common concerns about the technology's base-level accuracy , mobile MinION setups have already been transformative for real-time tracing and rapid data sharing during bacterial and viral pathogen outbreaks ( Boykin et al . , 2019; Chan et al . , 2020; Faria et al . , 2018; Faria et al . , 2017; Kafetzopoulou et al . , 2019; Quick et al . , 2015; Quick et al . , 2016 ) . In the context of freshwater analysis , a MinION whole-genome shotgun sequencing protocol has recently been leveraged for a comparative study of 11 rivers ( Reddington et al . , 2020 ) . This report highlights key challenges which emerge in serial monitoring scenarios of a relatively low-input DNA substrate ( freshwater ) , for example large sampling volumes ( 2–4 l ) and small shotgun fragments ( mean < 4 kbp ) . We reasoned that targeted DNA amplification may be a suitable means to bypass these bottlenecks and assess river microbiomes with nanopore sequencing . Here , we report a simple , cost-effective workflow to assess and monitor microbial freshwater ecosystems with targeted nanopore DNA sequencing . Our benchmarking study involves the design and optimisation of essential experimental steps for multiplexed MinION usage in the context of local environments , together with an evaluation of computational methods for the bacterial classification of nanopore sequencing reads from metagenomic libraries . To showcase the resolution of sequencing-based aquatic monitoring in a spatiotemporal setting , we combine DNA analyses with physicochemical measurements of surface water samples collected at nine locations within a confined ~12 km reach of the River Cam passing through the city of Cambridge ( UK ) in April , June , and August 2018 .
Using a bespoke workflow , nanopore full-length ( V1-V9 ) 16S ribosomal RNA ( rRNA ) gene sequencing was performed on all location-barcoded freshwater samples at each of the three time points ( Figure 1; Supplementary file 1; Materials and methods ) . River isolates were multiplexed with negative controls ( deionised water ) and mock community controls composed of eight bacterial species in known mixture proportions . To obtain valid taxonomic assignments from freshwater sequencing profiles using nanopore sequencing , twelve different classification tools were compared through several performance metrics ( Figure 2; Figure 2—figure supplement 1; Materials and methods ) . Our comparison included established classifiers such as RDP ( Wang et al . , 2007 ) , Kraken ( Wood and Salzberg , 2014 ) , and Centrifuge ( Kim et al . , 2016 ) , as well as more recently developed methods optimised for higher sequencing error rates such as IDTAXA ( Murali et al . , 2018 ) and Minimap2 ( Li , 2018 ) . An Enterobacteriaceae overrepresentation was observed across all replicates and classification methods , pointing towards a consistent Escherichia coli amplification bias potentially caused by skewed taxonomic specificities of the selected 16S primer pair 27F and 1492R ( Frank et al . , 2008; Figure 2b ) . Root mean square errors ( RMSE ) between observed and expected bacteria of the mock community differed slightly across all classifiers ( Figure 2c ) . Robust quantifications were obtained by Minimap2 alignments against the SILVA v . 132 database ( Quast et al . , 2013 ) , for which 99 . 68% of classified reads aligned to the expected mock community taxa ( mean sequencing accuracy 92 . 08%; Figure 2—figure supplement 2c ) . Minimap2 classifications reached the second lowest RMSE ( excluding Enterobacteriaceae ) , and relative quantifications were highly consistent between mock community replicates . Benchmarking of the classification tools on one aquatic sample further confirmed Minimap2's reliable performance in a complex bacterial community ( Figure 2d ) , although other tools such as MAPseq ( Matias Rodrigues et al . , 2017 ) , SPINGO ( Allard et al . , 2015 ) , or IDTAXA also produced highly concordant results – despite variations in memory usage and runtime over several orders of magnitude ( Figure 2—figure supplement 1b ) . Using Minimap2 classifications within our bioinformatics consensus workflow ( Figure 1—figure supplement 1; Materials and methods ) , we then inspected sequencing profiles of three independent MinION runs for a total of 30 river DNA isolates and six controls . This yielded ~8 . 3 million sequences with exclusive barcode assignments ( Figure 3a; Supplementary file 2 ) . Overall , 82 . 9% ( n = 6 , 886 , 232 ) of raw reads could be taxonomically assigned to the family level ( Figure 3b ) . To account for variations in sample sequencing depth , rarefaction with a cut-off at 37 , 000 reads was applied to all samples . While preserving ~90% of the original family level taxon richness ( Mantel test , R = 0 . 814 , p = 2 . 1*10−4; Figure 3—figure supplement 1a–b ) , this conservative thresholding resulted in the exclusion of 14 samples , mostly from the June time point , for subsequent high-resolution analyses . The 16 remaining surface water samples revealed moderate levels of microbial heterogeneity ( Figure 3b; Figure 3—figure supplement 1c ) : microbial family alpha diversity ranged between 0 . 46 ( June-6 ) and 0 . 92 ( April-7 ) ( Simpson index ) , indicating low-level evenness with a few taxonomic families that account for the majority of the metagenomic signal . Hierarchical clustering of taxon profiles showed a dominant core microbiome across all aquatic samples ( clusters C2 and C4 , Figure 4a ) . The most common bacterial families observed were Burkholderiaceae ( 40 . 0% ) , Spirosomaceae ( 17 . 7% ) , and NS11-12 marine group ( 12 . 5% ) , followed by Arcobacteraceae ( 4 . 8% ) , Sphingomonadaceae ( 2 . 9% ) , and Rhodobacteraceae ( 2 . 5% ) ( Figure 4b ) . Members of these families are commonly associated with aquatic environments; for example , major fractions of Burkholderiaceae reads originated from genera such as Limnohabitans , Rhodoferax , Polynucleobacter , or Aquabacterium ( Figure 4—figure supplement 1 ) , which validates the suitability of this nanopore metagenomics workflow . Hierarchical clustering additionally showed that two biological replicates collected at the same location and time point ( April samples 9 . 1 and 9 . 2 ) , grouped with high concordance; this indicates that spatiotemporal trends are discernible even within a highly localised context . Besides the dominant core microbiome , microbial profiles showed a marked arrangement of time dependence , with water samples from April grouping more distantly to those from June and August . Principal component analysis ( PCA ) illustrates the seasonal divergence among the three sampling months ( Figure 5a; Figure 5—figure supplement 1 ) . The strongest differential abundances along the seasonal axis of variation ( PC3 ) derived from Carnobacteriaceae ( Figure 5b ) , a trend also highlighted by taxon-specific log-normal mixture model decomposition between the two seasons ( April vs . June/August; p < 0 . 01; Materials and methods ) . Indeed , members of this bacterial family have been primarily isolated from cold substrates ( Lawson and Caldwell , 2014 ) . While a seasonal difference in bacterial composition can be expected due to increasing water temperatures in the summer months , additional changes may have also been caused by alterations in river hydrochemistry and flow rate ( Figure 6a; Figure 6—figure supplement 1; Supplementary file 1 ) . To assess this effect in detail , we measured the pH and a range of major and trace cations in all river water samples using inductively coupled plasma-optical emission spectroscopy ( ICP-OES ) , as well as major anions using ion chromatography ( Materials and methods ) . As with the bacterial composition dynamics , we observed significant temporal variation in water chemistry , superimposed on a spatial gradient of generally increasing sodium and chloride concentrations along the river reach ( Figure 6b–c ) . This spatially consistent effect is likely attributed to wastewater and agricultural discharge inputs in and around Cambridge city . A comparison of the major element chemistry in the River Cam transect with the world's 60 largest rivers further corroborates the likely impact of anthropogenic pollution in this fluvial ecosystem ( Gaillardet et al . , 1999; Figure 6d; Materials and methods ) . Freshwater sources throughout the United Kingdom have been notorious for causing bacterial infections such as leptospirosis ( Public Health England , 2016; Public Health England , 2019 ) . In line with the physicochemical profile of the River Cam , we therefore next determined the spatiotemporal enrichment of potentially important functional bacterial taxa through nanopore sequencing . We retrieved 55 potentially pathogenic bacterial genera through integration of species known to affect human health ( Jin et al . , 2018; Wattam et al . , 2017 ) , and also 13 wastewater-associated bacterial genera ( Global Water Microbiome Consortium et al . , 2019; Supplementary file 3 ) . Of these , 21 potentially pathogenic and 8 wastewater-associated genera were detected across all of the river samples ( Figure 7; Materials and methods ) . Many of these signals were stronger downstream of urban sections , within the mooring zone for recreational and residential barges ( location 7; Figure 1a ) and in the vicinity of sewage outflow from a nearby wastewater treatment plant ( location 8 ) . The most prolific candidate pathogen genus observed was Arcobacter , which features multiple species implicated in acute gastrointestinal infections ( Kayman et al . , 2012 ) . In general , much of the taxonomic variation across all samples was caused by sample April-7 ( PC1 explains 27 . 6% of the overall variance in bacterial composition; Figure 5—figure supplement 1a–b ) . Its profile was characterised by an unusual dominance of Caedibacteraceae , Halomonadaceae and others ( Figure 5—figure supplement 1c ) . Isolate April-8 also showed a highly distinct bacterial composition , with some families nearly exclusively occurring in this sample ( outlier analysis; Materials and methods ) . The most predominant bacteria in this sewage pipe outflow are typically found in wastewater sludge or have been shown to contribute to nutrient pollution from effluents of wastewater plants , such as Haliangiaceae , Nitospiraceae , Rhodocyclaceae , and Saprospiracea ( Nielsen et al . , 2012; Global Water Microbiome Consortium et al . , 2019; Figure 7 ) . Using multiple sequence alignments between nanopore reads and pathogenic species references , we further resolved the phylogenies of three common potentially pathogenic genera occurring in our river samples , Legionella , Salmonella , and Pseudomonas ( Figure 8a–c; Materials and methods ) . While Legionella and Salmonella diversities presented negligible levels of known harmful species , a cluster of reads in downstream sections indicated a low abundance of the opportunistic , environmental pathogen Pseudomonas aeruginosa ( Figure 8c ) . Along the course here investigated , we also found significant variations in relative abundances of the Leptospira genus , which was recently described to be enriched in wastewater effluents in Germany ( Numberger et al . , 2019; Figure 8d ) . Indeed , the peak of River Cam Leptospira reads fell into an area of increased sewage influx ( ~0 . 1% relative abundance; Figure 7 ) . The Leptospira genus contains several potentially pathogenic species capable of causing life-threatening leptospirosis through waterborne infections , however , also features close-related saprophytic and ‘intermediate’ taxa ( Vincent et al . , 2019; Wynwood et al . , 2014 ) . To resolve its complex phylogeny in the River Cam surface , we aligned Leptospira reads from all samples together with many reference sequences assigned to pre-classified pathogenic , saprophytic and other environmental Leptospira species ( Figure 8d; Supplementary file 4; Materials and methods ) . Despite the presence of nanopore sequencing errors ( Figure 2—figure supplement 2c ) and correspondingly inflated read divergence , we could pinpoint spatial clusters and a distinctly higher similarity between our amplicons and saprophytic rather than pathogenic Leptospira species . These findings were subsequently validated by targeted , Leptospira species-specific qPCR ( Supplementary file 5; Materials and methods ) , confirming that R9 . 4 . 1 nanopore sequencing quality is already high enough to yield indicative results for bacterial monitoring workflows at the species level .
Using a cost-effective , easily adaptable and scalable framework , we provide the first spatiotemporal nanopore sequencing atlas of bacterial microbiota throughout the course of a river . Our results suggest that this workflow allows for robust assessments of both , the core microbiome of an example fluvial ecosystem and heterogeneous bacterial compositions in the context of supporting physical ( temperature , flow rate ) and hydrochemical ( pH , inorganic solutes ) parameters . We show that the technology's current sequencing accuracy of ~92% allows for the designation of significant human pathogen community shifts along rural-to-urban river transitions , as illustrated by downstream increases in the abundance of pathogen candidates . Our assessment of bioinformatics workflows for taxonomic classification highlights current challenges with error-prone nanopore sequences . A number of recent reports feature bespoke 16S read classification schemes centred around a single software ( Acharya et al . , 2019; Benítez-Páez et al . , 2016; Kerkhof et al . , 2017; Nygaard et al . , 2020 ) , and others integrated outputs from two methods ( Cuscó et al . , 2018 ) . Through systematic benchmarking of twelve different classification tools , using matched mock community and river water datasets with respect to the SILVA v . 132 reference database , we lay open key differences in terms of these methods' read ( mis ) classification rates , consensus agreements , speed and memory performance metrics . For example , our results indicate that very fast implementations like Kraken 2 or Centrifuge yield less accurate classifications than slightly slower and more memory-demanding frameworks such as Minimap2 ( Figure 2; Figure 2—figure supplement 1 ) . Using Minimap2 , 16 . 2% of freshwater-derived sequencing reads were assigned to a bacterial species on average , thereby primarily encouraging automated analyses on the genus ( 65 . 6% assigned ) or family level ( 76 . 6% assigned ) . As nanopore sequencing quality continues to increase through refined pore chemistries , basecalling algorithms and consensus sequencing workflows ( Calus et al . , 2018; Karst et al . , 2021; Latorre-Pérez et al . , 2020; Rang et al . , 2018; Santos et al . , 2020; Zurek et al . , 2020 ) , future bacterial taxonomic classifications are likely to improve and advance opportunities for species discovery . We show that nanopore amplicon sequencing data can resolve the core microbiome of a freshwater body , as well as its temporal and spatial fluctuations . Common freshwater bacteria account for the vast majority of taxa in the River Cam; this includes Sphingomonadaceae , which had also been previously found at high abundance in source water from the same river ( Rowe et al . , 2016 ) . Our findings suggest that the differential abundances of Carnobacteriaceae most strongly contribute to seasonal loadings in the River Cam . Carnobacteriaceae have been previously associated with a range of low-temperature environments ( Lawson and Caldwell , 2014 ) , and we found these taxa to be more abundant in colder April samples ( mean 11 . 3°C , vs . 15 . 8°C in June and 19 . 1°C in August ) . This might help to further establish this family as an indicator for bacterial community shifts along with temperature fluctuations , albeit the influence of co-occurring hydrochemical trends such as water hardness , dissolved carbon or flow speed changes should also be noted ( Figure 6b–d; Figure 6—figure supplement 1d ) . Most routine freshwater surveillance frameworks focus on semi-quantitative diagnostics of only a limited number of target taxa , such as pathogenic Salmonella , Legionella and faecal coliforms ( Ramírez-Castillo et al . , 2015; Tan et al . , 2015 ) , whereas metagenomics approaches can give a complete and detailed overview of environmental microbial diversity . Besides nanopore shotgun-sequencing ( Reddington et al . , 2020 ) , our proof-of-principle analysis highlights that targeted full-length 16S rRNA gene MinION sequencing is a suitable complement to hydrochemical controls in pinpointing relatively contaminated freshwater sites , some of which in case of the River Cam had been previously highlighted for their pathogen diversity and abundance of antimicrobial resistance genes ( Rowe et al . , 2017; Rowe et al . , 2016 ) . Nanopore amplicon sequencing has here allowed us to reliably distinguish closely related pathogenic and non-pathogenic bacterial species of the common Legionella , Salmonella , Pseudomonas , and Leptospira genera . For Leptospira bacteria , which are of particular interest to communal stakeholders of the River Cam , we validated nanopore sequencing results through the gold standard qPCR workflow of Public Health England ( Supplementary file 5 ) . In order to also study the potential viability and functional implications of sequenced pathogen candidates for public health , we encourage future studies to combine nanopore based freshwater metagenomics with targeted follow-up measurements of living pathogens by established microbiological approaches , including species-specific isolation and subsequent culturing . A number of experimental intricacies should be addressed towards nanopore freshwater sequencing with our approach , mostly by scrutinising water DNA extraction yields , PCR biases and molar imbalances in barcode multiplexing ( Figure 3a; Figure 2—figure supplement 2a–b; Supplementary file 2 ) . Similar to challenges with other organic substrates , microbial raw DNA extraction protocols require careful pre-testing and optimisation towards the physicochemical composition of a given freshwater source , in order to avoid both taxonomic enrichment biases and drop-offs in total yield . One example lies in the optimisation of the filtrate volume – in this study , membrane DNA extraction from 400 mL River Cam water was sufficient to yield valuable insights , while as much as 10 , 000 mL were used in a previous study of the same river ( Rowe et al . , 2016 ) . Moreover , potentially dissolved inhibitory compounds for DNA extraction , sample cooling and storage chains should be thoroughly considered for larger and remote river monitoring projects . We witnessed that yield variations may bear negative effects on the molar balance of barcoded nanopore sequencing runs , as illustrated by elevated sample dropouts in June 2018 , emphasising the need for highly accurate concentration measurement and scaling when dozens of input DNA sources are pooled . Our study further highlights that MinION ( R9 . 4 . 1 ) flow cell throughput can fluctuate by an order of magnitude , altogether causing the exclusion of measurements upon application of a conservative read threshold . We reason that real-time selective nanopore sequencing could serve as a powerful means to improve barcode balances in the context of multiplexed 16S analyses ( Loose et al . , 2016 ) , albeit such approaches are yet undergoing computational optimisations ( Kovaka et al . , 2020; Payne et al . , 2020 ) . Our results show that it would already be theoretically feasible to obtain meaningful river microbiota from > 100 barcoded samples on a single nanopore flow cell , thereby enabling water monitoring projects involving large collections at costs below £20 per sample ( Supplementary file 6 ) . In line with this , ONT has already released several commercial 96-barcode multiplexing kits for PCR- and non-PCR-based applications , as well as the smaller ‘Flongle’ flow cell with considerably reduced cost as compared to the traditional MinION model . On the other hand , shotgun nanopore sequencing approaches may bypass pitfalls associated with amplicon sequencing , namely taxon-specific primer biases ( Frank et al . , 2008 ) , 16S rRNA copy number fluctuations between species ( Darby et al . , 2013 ) or the omission of functionally relevant sequence elements . In combination with sampling protocol adjustments , shotgun nanopore sequencing could moreover be used for the serial monitoring of eukaryotic microorganisms and viruses in freshwater ecosystems ( Reddington et al . , 2020 ) . Since the commercial launch of the MinION in 2015 , a wide set of microbial nanopore sequencing applications in the context of rRNA gene ( Benítez-Páez et al . , 2016; Cuscó et al . , 2018; Kerkhof et al . , 2017; Nygaard et al . , 2020 ) and shotgun ( Leggett et al . , 2020; Nicholls et al . , 2019; Reddington et al . , 2020; Stewart et al . , 2019 ) metagenomics have attracted the interest of a growing user community . Two independent case studies have recently provided decomposition analyses of faecal bacterial pathogens in MinION libraries derived from river and spring waters in Montana , USA ( Hamner et al . , 2019 ) and Kathmandu Valley , Nepal ( Acharya et al . , 2019 ) . Although it is to be expected that short-read metagenomics technology continues to provide valuable environmental insights , as illustrated through global cataloguing efforts of ocean ( Tara Oceans coordinators et al . , 2015 ) and wastewater ( Global Water Microbiome Consortium et al . , 2019 ) microbiomes , due to their large sizes and fixed costs these traditional platforms remain unfeasible for the monitoring of remote environments – especially in low-resource settings . We reason that the convenience of MinION handling and complementary development of portable DNA purification methods ( Boykin et al . , 2019; Gowers et al . , 2019 ) will allow for such endeavours to become increasingly accessible to citizens and public health organisations around the world , ultimately democratising the opportunities and benefits of DNA sequencing .
We monitored nine distinct locations along a 11 . 62 km reach of the River Cam , featuring sites upstream , downstream and within the urban belt of the city of Cambridge , UK . Measurements were taken at three time points , in two-month intervals between April and August 2018 ( Figure 1; Supplementary file 1a ) . To warrant river base flow conditions and minimise rain-derived biases , a minimum dry weather time span of 48 hr was maintained prior to sampling ( Fisher et al . , 2015 ) . One litre of surface water was collected in autoclaved DURAN bottles ( Thermo Fisher Scientific , Waltham , MA , USA ) , and cooled to 4°C within 3 hr . Two bottles of water were collected consecutively for each time point , serving as biological replicates of location 9 ( samples 9 . 1 and 9 . 2 ) . We assessed various chemical , geological and physical properties of the River Cam ( Figure 6; Figure 6—figure supplement 1; Supplementary file 1b-c ) . In situ water temperature was measured immediately after sampling . To this end , we linked a DS18B20 digital temperature sensor to a portable custom-built , grid mounted Arduino nano v3 . 0 system . The pH was later recorded under temperature-controlled laboratory conditions , using a pH edge electrode ( HI-11311 , Hanna Instruments , Woodsocket , RI , USA ) . To assess the dissolved ion concentrations in all collected water samples , we aerated the samples for 30 s and filtered them individually through a 0 . 22 µM pore-sized Millex-GP polyethersulfone syringe filter ( MilliporeSigma , Burlington , MA , USA ) . Samples were then acidified to pH ~2 , by adding 20 µL of 7M distilled HNO3 per 3 mL sample . Inductively coupled plasma-optical emission spectroscopy ( ICP-OES , Agilent 5100 SVDV; Agilent Technologies , Santa Clara , CA , USA ) was used to analyse the dissolved cations Na+ , K+ , Ca2+ , Mg2+ , Ba2+ , Li+ , as well as Si and SO42- ( as total S ) ( Supplementary file 1b ) . International water reference materials ( SLRS-5 and SPS-SW2 ) were interspersed with the samples , reproducing certified values within 10% for all analysed elements . Chloride concentrations were separately measured on 1 mL of non-acidified aliquots of the same samples , using a Dionex ICS-3000 ion chromatograph ( Thermo Fisher Scientific , Waltham , MA , USA ) ( Supplementary file 1b ) . Long-term repeat measurements of a USGS natural river water standard T-143 indicated precision of more than 4% for Cl- . However , the high Cl- concentrations of the samples in this study were not fully bracketed by the calibration curve and we therefore assigned a more conservative uncertainty of 10% to Cl- concentrations . High calcium and magnesium concentrations were recorded across all samples , in line with hard groundwater and natural weathering of the Cretaceous limestone bedrock underlying the river catchment ( Figure 6a ) . There are no known evaporite salt deposits in the river catchment , and therefore the high dissolved Na+ , K+ , and Cl- concentrations in the River Cam are likely derived from anthropogenic inputs ( Rose , 2007; Figure 6c–d ) . We calculated bicarbonate concentrations through a charge balance equation ( concentrations in mol/L ) : conc ( HCO3- ) = conc ( Li+ ) + conc ( Na+ ) + conc ( K+ ) + 2*conc ( Mg2+ ) + 2*conc ( Ca2+ ) - conc ( Cl- ) - 2*conc ( S2- ) . The total dissolved solid ( TDS ) concentration across the 30 freshwater samples had a mean of 458 mg/L ( range 325–605 mg/L ) which is relatively high compared to most rivers , due to ( 1 ) substantial solute load in the Chalk groundwater ( particularly Ca2+ , Mg2+ , and HCO3- ) and ( 2 ) likely anthropogenic contamination ( particularly Na+ , Cl- , and SO42- ) . The TDS range and the major ion signature of the River Cam is similar to other anthropogenically heavily-impacted rivers ( Gaillardet et al . , 1999 ) , exhibiting enrichment in Na+ ( Figure 6d ) . Overall , ion profiles clustered substantially between the three time points , indicating characteristic temporal shifts in water chemistry . PC1 of a PCA on the solute concentrations [µmol/L] shows a strong time effect , separating spring ( April ) from summer ( June , August ) samples ( Figure 6b ) . We highlighted the ten most important features ( i . e . features with the largest weights ) and their contributions to PC1 ( Figure 6c ) . We integrated sensor data sets on mean daily air temperature , sunshine hours and total rainfall from a public , Cambridge-based weather station ( Figure 6—figure supplement 1a–c; Supplementary file 1c ) . Similarly , mean gauged daily Cam water discharge [m3s−1] of the River Cam was retrieved through publicly available records from three upstream gauging stations connected to the UK National River Flow Archive ( https://nrfa . ceh . ac . uk/ ) , together with historic measurements from 1968 onwards ( Figure 6—figure supplement 1d ) . Within 24 hr of sampling , 400 mL of refrigerated freshwater from each site was filtered through an individual 0 . 22 µm pore-sized nitrocellulose filter ( MilliporeSigma , Burlington , MA , USA ) placed on a Nalgene polysulfone bottle top filtration holder ( Thermo Fisher Scientific ) at −30 mbar vacuum pressure . Additionally , 400 mL de-ionised ( DI ) water was also filtered . We then performed DNA extractions with a modified DNeasy PowerWater protocol ( Qiagen , Hilden , Germany ) . Briefly , filters were cut into small slices with sterile scissors and transferred to 2 mL Eppendorf tubes containing lysis beads . Homogenisation buffer PW1 was added , and the tubes subjected to ten minutes of vigorous shaking at 30 Hz in a TissueLyser II machine ( Qiagen ) . After subsequent DNA binding and washing steps in accordance with the manufacturer's protocol , elution was done in 50 µL EB . We used Qubit dsDNA HS Assay ( Thermo Fisher Scientific ) to determine water DNA isolate concentrations ( Supplementary file 2a ) . DNA extracts from each sampling batch and DI water control were separately amplified with V1-V9 full-length ( ~1 . 45 kbp ) 16S rRNA gene primers , and respectively multiplexed with an additional sample with a defined bacterial mixture composition of eight species ( Pseudomonas aeruginosa , Escherichia coli , Salmonella enterica , Lactobacillus fermentum , Enterococcus faecalis , Staphylococcus aureus , Listeria monocytogenes , Bacillus subtilis; D6305 , Zymo Research , Irvine , CA , USA ) ( Figure 2 ) , which was previously assessed using nanopore shotgun metagenomics ( Nicholls et al . , 2019 ) . We used common primer binding sequences 27F and 1492R , both coupled to unique 24 bp barcodes and a nanopore motor protein tether sequence ( Supplementary file 7 ) . Full-length 16S rDNA PCRs were performed with 30 . 8 µL DI water , 6 . 0 µL barcoded primer pair ( 10 µM ) , 5 . 0 µL PCR-buffer with MgCl2 ( 10x ) , 5 . 0 µL dNTP mix ( 10x ) , 3 . 0 µL freshwater DNA extract , and 0 . 2 µL Taq ( Qiagen ) under the following conditions: Amplicons were purified from reaction mixes with a QIAquick purification kit ( Qiagen ) . Two rounds of alcoholic washing and two additional minutes of drying at room temperature were then performed , prior to elution in 30 µL 10 mM Tris-HCl pH 8 . 0 with 50 mM NaCl . After concentration measurements with Qubit dsDNA HS , twelve barcoded extracts of a given batch were pooled in equimolar ratios , to approximately 300 ng DNA total ( Supplementary file 2b ) . We used KAPA Pure Beads ( KAPA Biosystems , Wilmington , MA , USA ) to concentrate full-length 16S rDNA products in 21 µL DI water . Multiplexed nanopore ligation sequencing libraries were then made by following the SQK-LSK109 protocol ( Oxford Nanopore Technologies , Oxford , UK ) . R9 . 4 . 1 MinION flow cells ( Oxford Nanopore Technologies ) were loaded with 75 µl of ligation library . The MinION instrument was run for approximately 48 hr , until no further sequencing reads could be collected . Fast5 files were basecalled using Guppy ( version 3 . 15 ) and output DNA sequence reads with Q > 7 were saved as fastq files . Various output metrics per library and barcode are summarised in Supplementary file 2c . In collaboration with Public Health England , raw water DNA isolates of the River Cam from each location and time point were subjected to the UK reference service for leptospiral testing ( Supplementary file 5 ) . This test is based on quantitative real-time PCR ( qPCR ) of 16S rDNA and LipL32 , implemented as a TaqMan assay for the detection and differentiation of pathogenic and non-pathogenic Leptospira spp . from human serum . Briefly , the assay consists of a two-component PCR; the first component is a duplex assay that targets the gene encoding the outer membrane lipoprotein LipL32 , which is reported to be strongly associated with the pathogenic phenotype . The second reaction is a triplex assay targeting a well-conserved region within the 16S rRNA gene ( rrn ) in Leptospira spp . Three different genomic variations correlate with pathogenic ( PATH probe ) , intermediate ( i . e . those with uncertain pathogenicity in humans; INTER probe ) and non-pathogenic Leptospira spp . ( ENVIRO probe ) , respectively . The described data processing and read classification steps were implemented using the Snakemake workflow management system ( Köster and Rahmann , 2012 ) and are available on Github - together with all necessary downstream analysis scripts to reproduce the results of this manuscript ( https://github . com/d-j-k/puntseq; Urban et al . , 2020; copy archived at swh:1:rev:1408d508c807b88e0989a5252c5d904072dc3c4a ) . Reads were demultiplexed and adapters trimmed using Porechop ( version 0 . 2 . 4 , https://github . com/rrwick/porechop ) . The only non-default parameter set was '--check_reads' ( to 50 , 000 ) , to increase the subset of reads to search for adapter sets . Next , we removed all reads shorter than 1 . 4 kbp and longer than 1 . 6 kbp with Nanofilt ( version 2 . 5 . 0 , https://github . com/wdecoster/nanofilt ) . We assessed read statistics including quality scores and read lengths using NanoStat ( version 1 . 1 . 2 , https://github . com/wdecoster/nanostat ) , and used Pistis ( https://github . com/mbhall88/pistis ) to create quality control plots . This allowed us to assess GC content and Phred quality score distributions , which appeared consistent across and within our reads . Overall , we obtained 2 , 080 , 266 reads for April , 737 , 164 for June , and 5 , 491 , 510 for August , with a mean read quality of 10 . 0 ( Supplementary file 2c ) . We used twelve different computational tools for bacterial full-length 16S rDNA sequencing read classification: ToolVersionCommandsBLASTN ( Altschul et al . , 1990; Camacho et al . , 2009 ) v . 2 . 9 . 0+blastn -task ‘blastn’ -db silva . fa -query Cam16S . fa -out Cam16S . out -outfmt '6'Centrifuge ( Kim et al . , 2016 ) v . 1 . 0 . 4centrifuge -x centrifuge_silva -U Cam16S . fq -S Cam16S . out --report-file Cam16S . reportIDTAXA ( Murali et al . , 2018 ) Implemented in R DECIPHER v . 2 . 10 . 2 ( Wright , 2016 ) load ( ‘SILVA_SSU_r132_March2018 . RData’ ) IdTaxa ( Cam16S . fa , trainingSet , strand = ‘both’ , threshold = 0 ) Kraken 2 ( Wood et al . , 2019; Wood and Salzberg , 2014 ) v . 2 . 0 . 7kraken2 --db kraken2_silva --output Cam16S . out --report Cam16S . report Cam16S . faMAPseq ( Matias Rodrigues et al . , 2017 ) v . 1 . 2 . 3mapseq Cam16S . fa silva . fa > Cam16S . outMegaBLAST ( Camacho et al . , 2009; Morgulis et al . , 2008 ) v . 2 . 9 . 0+blastn -task ‘megablast’ -db silva . fa -query Cam16S . fa -out Cam16S . out -outfmt '6'Minimap2 ( Li , 2018 ) v . 2 . 13-r852-dirtyminimap2 -ax map-ont -L silva . mmi Cam16S . fa > Cam16S . samMothur ( Schloss et al . , 2009 ) v . 1 . 43 . 0align . seqs ( candidate = Cam16S . fa , template = mothur . silva . nr_v132 . align , processors = 1 , ksize = 6 , align = needleman ) QIIME 2 ( Bolyen et al . , 2019 ) v . 2019 . 7qiime feature-classifier classify-consensus-blast --i-query Cam16S . qza --i-reference-reads silva . qza --i-reference-taxonomy silva_tax . qza --o-classification Cam16S . outRDP ( Wang et al . , 2007 ) Implemented in R DADA2 v . 1 . 12 . 1 ( Callahan et al . , 2016 ) assignTaxonomy ( seqs = Cam16S . fa , refFasta = silva_nr_v132_train_set . fa . gz’ , tryRC = T , outputBootstraps = T , minBoot = 0 ) SINTAX ( Edgar , 2016 ) Implemented in VSEARCH v . 2 . 13 . 3 ( Rognes et al . , 2016 ) vsearch -sintax Cam16S . fa -db silva . udb -tabbedout Cam16S . out -strand both -sintax_cutoff 0 . 5SPINGO ( Allard et al . , 2015 ) v . 1 . 3spingo -d silva . fa -k 8 -a -i Cam16S . fa > Cam16S . out This study was designed to enable freshwater microbiome monitoring in budget-constrained research environments . Although we had access to basic infrastructure such as pipettes , a PCR and TissueLyser II machine , as well a high-performance laptop , we wish to highlight that the total sequencing consumable costs were held below £4000 ( Supplementary file 6a ) . Individual processing and sequencing costs ranged at ~£75 per sample ( Supplementary file 6b ) . With the current MinION flow cell price of £720 , we estimate that per-sample costs could be further reduced to as low as ~£20 when barcoding and pooling ~100 samples in the same sequencing run ( Supplementary file 6c ) . Assuming near-equimolar amplicon pooling , flow cells with an output of ~5 , 000 , 000 reads can yield well over 37 , 000 sequences per sample and thereby surpass this conservative threshold applied here for comparative river microbiota analyses . | Many water-dwelling bacteria can cause severe diseases such as cholera , typhoid or leptospirosis . One way to prevent outbreaks is to test water sources to find out which species of microbes they contain , and at which levels . Traditionally , this involves taking a water sample , followed by growing a few species of ‘indicator bacteria’ that help to estimate whether the water is safe . An alternative technique , called metagenomics , has been available since the mid-2000s . It consists in reviewing ( or ‘sequencing’ ) the genetic information of most of the bacteria present in the water , which allows scientists to spot harmful species . Both methods , however , require well-equipped laboratories with highly trained staff , making them challenging to use in remote areas . The MinION is a pocket-sized device that – when paired with a laptop or mobile phone – can sequence genetic information ‘on the go’ . It has already been harnessed during Ebola , Zika or SARS-CoV-2 epidemics to track the genetic information of viruses in patients and environmental samples . However , it is still difficult to use the MinION and other sequencers to monitor bacteria in water sources , partly because the genetic information of the microbes is highly fragmented during DNA extraction . To address this challenge , Urban , Holzer et al . set out to optimise hardware and software protocols so the MinION could be used to detect bacterial species present in rivers . The tests focussed on the River Cam in Cambridge , UK , a waterway which faces regular public health problems: local rowers and swimmers often contract waterborne infections , sometimes leading to river closures . For six months , Urban , Holzer et al . used the MinION to map out the bacteria present across nine river sites , assessing the diversity of species and the presence of disease-causing microbes in the water . In particular , the results showed that optimising the protocols made it possible to tell the difference between closely related species – an important feature since harmful and inoffensive bacteria can sometimes be genetically close . The data also revealed that the levels of harmful bacteria were highest downstream of urban river sections , near a water treatment plant and river barge moorings . Together , these findings demonstrate that optimising MinION protocols can turn this device into a useful tool to easily monitor water quality . Around the world , climate change , rising urbanisation and the intensification of agriculture all threaten water quality . In fact , access to clean water is one of the United Nations sustainable development goals for 2030 . Using the guidelines developed by Urban , Holzer et al . , communities could harness the MinION to monitor water quality in remote areas , offering a cost-effective , portable DNA analysis tool to protect populations against deadly diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"microbiology",
"and",
"infectious",
"disease"
] | 2021 | Freshwater monitoring by nanopore sequencing |
AMPA receptor ( AMPAR ) function is modulated by auxiliary subunits . Here , we report on three AMPAR interacting proteins—namely CKAMP39 , CKAMP52 and CKAMP59—that , together with the previously characterized CKAMP44 , constitute a novel family of auxiliary subunits distinct from other families of AMPAR interacting proteins . The new members of the CKAMP family display distinct regional and developmental expression profiles in the mouse brain . Notably , despite their structural similarities they exert diverse modulation on AMPAR gating by influencing deactivation , desensitization and recovery from desensitization , as well as glutamate and cyclothiazide potency to AMPARs . This study indicates that AMPAR function is very precisely controlled by the cell-type specific expression of the CKAMP family members .
AMPARs mediate the majority of fast excitatory transmission in the central nervous system and play a key role in brain plasticity . AMPAR function is controlled by a multitude of auxiliary subunits ( Yan and Tomita , 2012 ) . These include TARPs ( Tomita et al . , 2003 ) , cornichons ( Schwenk et al . , 2009 ) , Sol-1 ( Zheng et al . , 2004 ) and SynDIG1 ( Kalashnikova et al . , 2010 ) . Recently , we identified a novel AMPAR auxiliary subunit , CKAMP44 , and characterized its modulation of AMPAR gating properties in CA1 and dentate gyrus neurons ( Khodosevich et al . , 2014; von Engelhardt et al . , 2010 ) . Unlike other auxiliary subunits , CKAMP44 contains an N-terminal cystine-knot domain that in other proteins , e . g . growth factors ( McDonald and Hendrickson , 1993 ) , was shown to stabilize the globular structure of the protein . The different auxiliary subunits exhibit distinct modulatory profiles . Since auxiliary subunits are differentially expressed in the brain , the specific combination in a particular cell type is likely to govern the AMPAR response to glutamate , as is the case for dentate gyrus granule cells , which express TARP γ-8 and CKAMP44 . Both proteins increase the number of AMPARs on the cell surface , decrease the deactivation rate and increase glutamate affinity . However , they differ in the influence that they extend on AMPAR desensitization , recovery from desensitization , long-term and short-term potentiation ( Khodosevich et al . , 2014 ) . Here , based on homology with CKAMP44 , we report on three novel CKAMP44-like proteins that were named CKAMP39 , CKAMP52 and CKAMP59 and , together with CKAMP44 , constitute the CKAMP family . Like CKAMP44 , the newly identified CKAMPs are all single transmembrane domain proteins that possess an extracellular cystine-knot domain and an intracellular domain ending with a PDZ type II motif . Notably , novel CKAMP family members bind to GluA1 and GluA2 and modify AMPAR-mediated currents in heterologous expression systems .
To investigate whether CKAMP44 has homologues in rodents , we searched the genomic databases using either the complete sequence of the CKAMP44 gene or the CKAMP44 cystine-knot domain sequence as a reference . We found three genes with a high degree of similarity to CKAMP44 and named them according to the predicted molecular weight of their corresponding protein products—CKAMP39 , CKAMP52 and CKAMP59 ( Figure 1A ) . Due to the similarity in their peptide sequences ( especially the cystine-knot motif ) , we classified these four proteins as the CKAMP family members . Like CKAMP44 , all CKAMP proteins comprise a signal peptide , N-terminal cystine-knot extracellular domain , transmembrane domain and a large intracellular C-terminal domain , which terminates in a PDZ type II motif ( Figure 1A , B ) . The predicted signal peptides vary in length , being 36 , 30 and 22 amino acids for CKAMP39 , CKAMP52 and CKAMP59 , respectively ( Figure 1B ) . Strikingly , cystine-knot domains exhibit a high degree of similarity and all eight cysteines are conserved among the family members ( Figure 1B ) . Although cystine-knot containing proteins differ much in their function ( Heinemann and Leipold , 2007; McDonald and Hendrickson , 1993; Zhang et al . , 2011 ) , the purpose of cysteine-knots is similar , i . e . to form a tightly packed globular domain made of β-strands with several variable loops for stable protein-protein interaction . Thus , although sequences of the extracellular domains of CKAMP proteins have little homology beyond the cystine-knot core , it is likely that the extracellular region of all four proteins exhibit a similar β-strand structure . Each of the proteins possesses a predicted single short transmembrane domain ( 18–19 amino acids ) , and novel CKAMP family members have ~80% homology with the transmembrane domain of CKAMP44 ( Figure 1B ) . Notably , a ~20 amino acid stretch immediately downstream of the transmembrane domain contains a conserved arginine-rich motif ( Figure 1B ) . This is of particular interest , since this region of CKAMP44 was shown to be necessary for interaction with GluA1 , and CKAMP44 mutants with a deletion of only 6 amino acids in this region did not bind to GluA1 when overexpressed in HEK293/T17 cells ( Khodosevich et al . , 2014 ) . Finally , another conserved region in the protein sequence amongst CKAMP family members is the PDZ type II motif located at the very end of the C-terminal domain ( Figure 1B ) . Thus , all CKAMP proteins end with the same EVTV stretch . Furthermore , another 5 amino acids upstream of the PDZ motif are also almost identical in CKAMP proteins . The PDZ motif is of functional importance as it mediates the interaction of CKAMP44 with PSD95 , which allows anchoring of AMPARs within synapses of dentate gyrus granule cells ( Khodosevich et al . , 2014 ) . At the time of CKAMP family identification , the genes of the CKAMP family members had not yet been recorded in the Refseq database . Currently , based on their similarity to Shisa genes , the Ckamp genes are termed in the genome database as Shisa6 ( CKAMP52 ) , Shisa7 ( CKAMP59 ) , Shisa8 ( CKAMP39 ) and Shisa9 ( CKAMP44 ) . Shisa 1–3 proteins were studied in embryonic development of Xenopus laevis , where they were found to inhibit Wnt and FGF signaling by retention of their receptors in ER ( Nagano et al . , 2006; Yamamoto et al . , 2005 ) . However , CKAMP family members differ significantly from Shisa2-5 proteins ( Figure 1—figure supplement 1 , Shisa1 protein does not have a mouse homolog ) and form a separate cluster on the phylogenetic tree ( Figure 1C , see also Pei and Grishin , 2012 ) . There are ~70 amino acids that are largely conserved in the CKAMP cluster , but not in the Shisa cluster ( Figure 1—figure supplement 1 ) . Furthermore , Shisa proteins are much shorter than CKAMPs , being 197-295 and 399-541 amino acids , respectively . Interestingly , the GluA1-interacting region of CKAMP44 exhibits a high degree of similarity to other CKAMPs , but not to that of Shisa proteins ( Figure 1—figure supplement 1 ) . Finally , Shisa proteins do not contain a PDZ binding motif at their C-terminus . Based on these considerations , we propose that CKAMP members constitute a protein family distinct from the Shisa protein family ( Figure 1C ) . We amplified open reading frames ( ORFs ) for the three novel CKAMP proteins using mouse brain-derived mRNA and confirmed the sequence of the corresponding proteins . Since at least 30 clones per CKAMP family member were analyzed , we were able to estimate the relative expression levels of different CKAMP splice isoforms in the brain . While CKAMP39 had only one splice variant , both CKAMP52 and CKAMP59 had two splice isoforms ( Figure 2—figure supplement 1 ) that differed in protein coding sequences . In the subsequent experiments , we utilized the most abundant versions of CKAMPs , i . e . CKAMP52 and CKAMP59 lacking exon 3 and exon 4 , respectively ( note that alignments in Figure 1B and Figure 1—figure supplement 1 are performed for exon 3- and exon 4-lacking versions ) . Exon 3 of CKAMP52 and exon 4 of CKAMP59 encode 32 and 17 intracellular amino acids , respectively , downstream of the AMPAR-interacting domain and upstream of the PDZ motif . Interestingly , there are also two splice variants of CKAMP44 ( von Engelhardt et al . , 2010 ) , but alternatively spliced mRNA was not reported for mouse Shisa family members . Previously , we demonstrated that CKAMP44 is expressed exclusively in the brain ( von Engelhardt et al . , 2010 ) . Based on gene expression database BioGPS , we found that the new members of the CKAMP family are also expressed exclusively in the brain ( Figure 2—figure supplement 2 ) . In situ hybridization of adult mouse brain sections with oligo probes against CKAMP39 , CKAMP52 and CKAMP59 mRNAs revealed that each of the novel CKAMP family proteins exhibited a region-specific expression pattern within the brain ( Figure 2A ) . CKAMP39 expression was restricted to two brain regions , namely the cerebellum and olfactory bulb , which were also the only brain regions with significant CKAMP39 expression according to the BioGPS database ( Figure 2—figure supplement 2 ) . Both CKAMP52 and CKAMP59 were expressed in the hippocampus , but CKAMP59 was also expressed in the cortex and olfactory bulb , whereas CKAMP52 was expressed in the cerebellum and septum . CKAMP39 is absent and CKAMP52 is barely detectable in the brain of embryonic day 17 ( E17 ) mice . In contrast , there is a strong signal for CKAMP59 already prenatally . Postnatally , there is little change in the expression pattern of any of these CKAMPs , except for an upregulation of CKAMP39 and CKAMP52 in the cerebellum and olfactory bulb and a modest downregulation of CKAMP59 in the thalamus and brainstem ( Figure 2A ) . To determine whether CKAMP39 , CKAMP52 and CKAMP59 interact with AMPARs , we inserted a flag-tag at the non-conserved C-terminal part of the proteins , and co-expressed flag-tagged CKAMPs along with GluA1 or GluA2 in HEK293/T17 cells . With an anti-flag antibody , GluA1 and GluA2 co-immunoprecipitated from protein samples of HEK293/T17 cells co-expressing flag-tagged CKAMP39 , CKAMP52 or CKAMP59 ( Figure 2B ) , showing that all novel CKAMPs bind to GluA1 and GluA2 in a heterologous expression system . All CKAMPs had two bands on Western blot , indicating their likely glycosylation that was shown previously for CKAMP44 ( von Engelhardt et al . , 2010 ) . To characterize the functional consequences of the interaction between the CKAMP family members and AMPARs , we performed electrophysiological experiments employing Xenopus Laevis oocytes . To investigate how gating properties are modulated by the novel CKAMPs , we performed fast perfusion patch-clamp recordings on outside-out macropatches pulled from oocytes expressing either GluA1 or GluA2 ( Q ) alone , or with CKAMP39 or CKAMP52 . An analysis of the effect of CKAMP59 on AMPAR gating in oocytes was not possible , as this auxiliary subunit was not sufficiently expressed in this heterologous expression system as revealed by the absence of detectable protein in Western-blot analysis ( Figure 3A ) . Consistently , no significant change in AMPAR gating was observed for the co-expression of GluA1 or GluA2 ( Q ) with CKAMP59 . In contrast , CKAMP39 and CKAMP52 exhibited protein levels comparable to CKAMP44 ( Figure 3A ) , and influenced AMPAR gating properties differentially . Neither CKAMP39 nor CKAMP52 modulated the GluA1-mediated deactivation time constant ( τdeact ) , but both increased τdeact of GluA2 ( Q ) -mediated currents ( Figure 3B and Supplementary file 1A—table 1 ) . Both proteins also had no influence on the GluA1-mediated desensitization time constant ( τdes ) , but significantly reduced τdes of GluA2 ( Q ) -mediated currents ( Figure 3C ) . There was a trend towards increased steady-state current amplitude ( as a percentage of maximal current ) during the 500 ms glutamate application in the oocyte patches for GluA1 with CKAMP52 ( see below for the significant effect on steady-state currents in HEK293/T17 cells ) , and a significant reduction in steady-state current amplitude of GluA2 ( Q ) -mediated currents by CKAMP39 ( Figure 3C and Supplementary file 1A—table 1 ) . The time constant of recovery from desensitization ( τrecovery ) of GluA1- and GluA2 ( Q ) -mediated currents was increased by CKAMP39 , whereas there was only a small increase and decrease of τrecovery of GluA1- and GluA2 ( Q ) -mediated currents , respectively , when co-expressing CKAMP52 ( Figure 3D and Supplementary file 1A—table 1 ) . We previously showed that CKAMP44 increases glutamate potency ( von Engelhardt et al . , 2010 ) . A comparable decrease in glutamate EC50 was observed when GluA1 or GluA2 was expressed with CKAMP39 or CKAMP52 . The most dramatic change was seen when expressing GluA2 together with CKAMP52; the EC50 was more than 10 fold smaller for CKAMP52-bound GluA2 compared to GluA2 alone ( Figure 4A ) . CKAMP39 and CKAMP44 , but not CKAMP52 , influenced not only the potency of glutamate , but also that of the AMPAR desensitization blocker cyclothiazide ( CTZ ) . Thus , there was an increase in the CTZ EC50 when co-expressing GluA1 with CKAMP39 , and GluA2 with CKAMP39 or CKAMP44 ( Figure 4B and Supplementary file 1B—table 2 ) . CKAMP59 is well expressed in HEK293/T17 cells as revealed by Western blot analysis ( Figure 2B ) in contrast to the oocytes . Hence , to investigate the influence of CKAMP59 on AMPAR-mediated currents , we expressed this auxiliary subunit along with GluA1 or GluA2 ( Q ) in HEK293/T17 cells . To be able to compare the influence of CKAMP59 with that of the other auxiliary subunits , we also investigated AMPAR-mediated currents in HEK293/T17 cells that co-expressed CKAMP39 or CKAMP52 . In contrast to the other auxiliary subunits , CKAMP59 did not modulate GluA1- or GluA2 ( Q ) -mediated current kinetics . However , there was a significant reduction in GluA2 ( Q ) -mediated current amplitude . Co-expression of CKAMP39 and CKAMP52 also reduced AMPAR-mediated current amplitude ( GluA1-current amplitude was reduced only by CKAMP39 ) ( Figure 5A and Supplementary file 1C—table 3 ) . A cell surface biotinylation assay performed in HEK293/T17 cells co-transfected with AMPARs and CKAMPs showed that all CKAMP family members , except for CKAMP52 , lead to a reduction in surface expression of GluA1 and GluA2 protein . These results can be accounted for by a reduction of total expression of GluA1 and GluA2 , and by reduced GluA2 forward trafficking or stabilization of GluA2 on the cell surface as indicated by the reduced ratio of surface to total protein . Interestingly , CKAMP52 increased the ratio of surface to total GluA1 expression , suggesting that this auxiliary protein exerts an opposite influence on forward trafficking or stabilization of GluA1 and GluA2 ( Figure 5—figure supplement 1 and Supplementary file 1D—table 4 ) . The strong reduction of surface AMPAR expression precluded an analysis of current kinetics by fast-application of glutamate onto outside-out patches . Thus , we performed an analysis using fast-application of glutamate onto whole HEK293/T17 cells instead . The expected solution exchange time is considerably slower when using whole cells instead of outside-out patches ( Barberis et al . , 2008 ) . Nevertheless , the analysis allowed us to draw conclusions about the influence of CKAMP family members on AMPAR kinetic properties . Thus , CKAMP39 and CKAMP52 modulated AMPAR gating in HEK293/T17 cells similarly to what we observed in oocytes with an increase in GluA2 ( Q ) τdeact by CKAMP52 ( Figure 5B and Supplementary file 1C—table 3 ) , a decrease in GluA2 ( Q ) τdes by CKAMP39 , an increase in GluA1 steady-state current amplitude by CKAMP52 , and a reduction of the GluA2 ( Q ) steady-state current amplitude by CKAMP39 ( Figure 5C and Supplementary file 1C—table 3 ) . Recovery from desensitization was analyzed in HEK293/T17 cells with a protocol that differed from that used in oocyte experiments , where we applied two 100 ms glutamate pulses with different interpulse intervals . In HEK293/T17 cell experiments , we applied two 1ms glutamate pulses . The rational was to probe whether CKAMP39 influences AMPAR recovery from desensitization also when glutamate is applied only for very short time periods , thus mimicking the short presence of glutamate in the synaptic cleft . Indeed , CKAMP39 slowed the recovery from desensitization of GluA1- and GluA2 ( Q ) -mediated currents also when tested with this modified protocol ( Figure 5D and Supplementary file 1C—table 3 ) . The effect was comparable to that of CKAMP44 , which modulates synaptic short-term plasticity in dentate gyrus granule cell synapses by slowing recovery from desensitization ( Khodosevich et al . , 2014 ) . There were some differences in the modulation of GluA2 ( Q ) -mediated currents in HEK293/T17 cells and oocytes . Thus , CKAMP39 increases τdeact in oocytes , but not in HEK293/T17 cells . In addition , τdes was reduced by co-expression with CKAMP52 in oocytes , but not in HEK293/T17 cells . Finally , the steady-state current amplitude was increased by CKAMP52 in HEK293/T17 cells , but not in oocytes . In conclusion , the new CKAMP family members display very distinct modulatory effects on AMPAR gating . Like the prototypical TARP auxiliary subunits ( Kato et al . , 2008 ) , the CKAMP family affects AMPAR in a subunit-specific manner .
In recent years , several labs have employed large proteomic screens to search for AMPAR interacting proteins , which resulted in the identification of new AMPAR auxiliary ( or auxiliary-like ) subunits , such as CKAMP44 ( von Engelhardt et al . , 2010 ) , cornichons ( Schwenk et al . , 2009 ) and GSG1L ( Schwenk et al . , 2012; Shanks et al . , 2012 ) . In this study , we searched a genomic and transcriptomic databases , and identified three new proteins that , together with CKAMP44 , form the CKAMP family of AMPAR auxiliary-like proteins . Presumably , evolutionarily the CKAMPs and Shisa proteins descend from the same protein family . The Shisa proteins were shown to be involved in fibroblast growth factor receptor maturation and degradation during embryogenesis in Xenopus laevis oocytes ( Nagano et al . , 2006; Yamamoto et al . , 2005 ) . All CKAMPs exhibited a significant homology in the region of CKAMP44 that is necessary for AMPAR binding ( Khodosevich et al . , 2014 ) . Thus , it is likely that , similar to CKAMP44 , a stretch of amino acids immediately downstream of the transmembrane regions of novel CKAMP proteins is involved in interaction with AMPARs . All CKAMPs possess an extracellular cystine-knot motif that was shown to be important for modulation of AMPAR gating , and an identical intracellular PDZ-domain binding motif ( Khodosevich et al . , 2014 ) . All CKAMPs bind to GluA1 and GluA2 and modify GluA1 and/or GluA2 currents in two heterologous systems , making them likely candidates for AMPAR auxiliary subunits in vivo . Importantly , despite their structural similarities , the members of the CKAMP family differ enormously in their modulation of AMPAR-mediated currents . Only the function of CKAMP39 resembles that of CKAMP44 ( Khodosevich et al . , 2014 ) , and indeed this subunit also displays the highest homology with CKAMP44 . CKAMP39 had a pronounced effect on the recovery from desensitization , similar to CKAMP44 , suggesting that it might also modulate synaptic short-term plasticity as was observed for CKAMP44 in dentate gyrus granule cells ( Khodosevich et al . , 2014 ) . The reduction of surface AMPAR number by all three CKAMPs in HEK293/T17 cells was unexpected , since CKAMP44 has the opposite effect on AMPARs of granule cells , possibly by promoting trafficking of AMPARs to the cell surface . The reduction in surface AMPAR number by the novel CKAMP family members was mainly due to a reduction in total AMPAR number in the cell . However , changes in the ratio of surface/total AMPAR number indicate that the novel CKAMP family members also influence the forward trafficking or stability of AMPARs on the cell surface . The fact that some parameters of GluA2 ( Q ) -mediated currents were modulated only in HEK293/T17 cells , others only in oocytes , cannot be explained by differences in AMPAR-composition , since the same GluA2 ( Q ) -flip version of this subunit was expressed in both expression systems . However , one explanation could be the different recording conditions used for HEK293/T17 cells and oocytes . Thus , there were differences in room temperature ( 17oC versus 22oC ) , patch size ( macropatches of oocytes versus lifted whole HEK293/T17 cells ) and the holding potential ( -70 versus -120 mV ) . Overall , the novel CKAMP family members greatly expand the pool of AMPAR auxiliary-like proteins expressed in the brain . CKAMP52 was also identified as an AMPAR interacting protein in recent proteomic screens ( Schwenk et al . , 2012; Shanks et al . , 2012 ) , showing that at least this CKAMP family member interacts with AMPARs in the brain . It is not clear why CKAMP39 and CKAMP59 were not identified as AMPAR interacting proteins , but it is possible that their expression level is too low or that their interaction is too loose for identification in the proteomic screens . The highly diverse expression pattern of CKAMP family members together with their unique biophysical profile make them strong candidates for region-specific AMPAR modulation . Future studies should unravel how different CKAMPs influence synaptic function by modulating expression and gating kinetics of AMPARs . Yet another effect of CKAMP family members might be an influence on neuron morphology as described for CKAMP44 and TARP γ-8 , which increase spine number by augmenting surface AMPAR expression ( Khodosevich et al . , 2014 ) .
Signal peptides were analyzed using SignalP 4 . 1 software ( Petersen et al . , 2011 ) ( http://www . cbs . dtu . dk/services/SignalP/ ) . Transmembrane domains were identified using TMHMM 2 . 0 software ( Krogh et al . , 2001 ) ( http://www . cbs . dtu . dk/services/TMHMM/ ) and PredictProtein ( Yachdav et al . , 2014 ) ( www . predictprotein . org ) . Alignment of protein sequences and phylogenetic analysis was performed using Jalview ( Troshin et al . , 2011; Waterhouse et al . , 2009 ) ( www . jalview . org ) . The alignments of CKAMP proteins and of CKAMP and Shisa proteins were done using ProbCons ( Do et al . , 2005 ) and Clustal ( Larkin et al . , 2007 ) tools , respectively . The phylogenetic tree of Shisa proteins and CKAMPs was obtained utilizing average distance methods . Expression of CKAMP family members in different tissues was retrieved from gene expression BioGPS database ( http://biogps . org/ ) , GNF1M dataset for CKAMP39 and MOE430 dataset for CKAMP52 and CKAMP59 . RNA was isolated from the whole mouse brain and cDNA was synthesized as described before ( Khodosevich et al . , 2013 ) . We used the following primers to amplify CKAMP39 , CKAMP52 and CKAMP59: C39F = 5’TAGGATCCGCCACCATGGAGCGCGCTGGGGCGCGGGGACAG C39R = 5’GCACTAGTCTAGACCGTGACCTCGGCTTTGC C52F = 5’ATGGATCCGCCACCATGGCGCTGCGCCGCCTCCTG C52R = 5’CGACTAGTTCACACGGTCACTTCAGTCTTGCTGGC C59F = 5’TAGGATCCGCCACCATGCCGGCCCTGCTGCTGCTC C59R = 5’GCCACTAGTTCAGACAGTCACTTCGTTCTTGCTG CKAMPs were amplified from cDNA using LA Taq polymerase with GC-rich buffer ( Clontech-Takara Bio , France ) for 5 cycles . PCR products were purified and re-amplified using PfuUltraII ( Agilent , USA ) for 30 more cycles . The resulting PCR products were digested by BamHI and SpeI and ligated into the pRK5 vector or an AAV vector with the human synapsin promoter , IRES and EGFP ( pAAV-Syn-IRES-EGFP ) ( von Engelhardt et al . , 2010 ) . CKAMP ORFs were sequenced and only those that corresponded to sequences from the genomic database were used in the subsequent experiments . To estimate the abundance of different splice-variants in the brain , at least 30 clones per each CKAMP family member were analyzed . To generate flag-tagged versions of CKAMPs , we inserted a flag-tag sequence close to the non-conserved terminal region of CKAMPs ( upstream to PDZ type II motif ) . The insertions were made using site-directed mutagenesis via two consecutive rounds of extension PCR , re-ligating flag-tagged PCR fragments using BspEI/SalI for CKAMP39 and XhoI/SalI for CKAMP52 and CKAMP59 . The primers that we utilized for flag-tag insertion are shown below: C39ctf-for = 5’CATCCGGAGGACTTGCCTGCGTTGC C39ctf-rev1 = 5’CTTGTCATCGTCATCCTTGTAATCGATATCATGATCTTTATAATCACCG CTCAGGTGCCGGGGTCCTC C39ctf-rev2 = 5’TCTGTCGACTCTAGTCTAGACCGTGACCTCGGCTTTGCTGTTGGTGT GCTTGTCATCGTCATCCTTGTAATCG C52ctf-for = 5’GCCTCGAGCGCGCCTGGTGTCTCAG C52ctf-rev1 = 5’CTTGTCATCGTCATCCTTGTAATCGATATCATGATCTTTATAATCAC CAGTGCGCAGGTGCTGGGGCAG C52ctf-rev2 = 5’TCTGTCGACTCTAGTTCAGACAGTCACTTCGTTCTTGCTGGCGTG CTTGTCATCGTCATCCTTGTAATCG C59ctf-for = 5’CACTTCCTCGAGAACGGCCACGCAG C59ctf-rev1 = 5’CTTGTCATCGTCATCCTTGTAATCGATATCATGATCTTTATAATCA CCTGTGTAGCAGGTGTGATGGC C59ctf-rev2 = 5’TCTGTCGACTCTAGTTCACACGGTCACTTCAGTCTTGCTGGCGTG CTTGTCATCGTCATCCTTGTAATCG For immunoprecipitation experiments , an STR-tested and authenticated HEK293/T17 cell line was used ( American Type Culture Collection , CRL-11268 , ATCC , USA ) . All cell cultures were tested for mycoplasma contamination prior to experiments using PCR Mycoplasma Test Kit I/C ( PK-CA91-1024 , PromoCell GmbH , Germany ) . Cell lines utilized in the study are not mentioned in the list of commonly misidentified cell lines maintained by the International Cell Line Authentication Committee . Both Western-blot and immunoprecipitation were performed as previously described ( Khodosevich et al . , 2014 ) . Briefly , HEK293/T17 cells were co-transfected with CKAMP44 , CKAMP39 , CKAMP52 , CKAMP59 ( in pRK5 ) or EGFP expression ( pEGFP-C1 ) plasmids together with a GluA1 or GluA2 expression plasmid ( in pRK5 ) . Two days post-transfection , protein was collected and affinity-purified with an agarose-bound flag antibody ( Anti-FLAG M2 Affinity Gel , Sigma-Aldrich , Germany ) as previously described ( Khodosevich et al . , 2014 ) . For immunoprecipitation we used 350 μg of total protein . Immunoprecipitated proteins were eluted by 50 μl of 3Xflag peptide solution ( Sigma-Aldrich , Germany ) . Denatured whole protein ( 6-10 μg ) and immunoprecipitated ( 10-25 μl ) samples were separated by SDS-PAGE and transferred onto PVDF membranes that were probed with the mouse anti-flag M2 antibody ( 1:2000 , F1804 , Sigma-Aldrich , Germany ) and rabbit GluA1 ( 1:1000 , Santa Cruz , Germany ) or mouse GluA2 antibody ( MAB397 , 1:500 , Millipore , Billerica , MA , USA ) For cell surface biotinylation assay , HEK293/T17 cells were co-transfected with pRK5-CKAMP39 , -CKAMP52 , -CKAMP59 or pAcGFP1-Mem ( Clontech-Takara Bio , France ) plasmids together with pRK5-GluA1-flip or pRK5-GluA2 ( Q ) -flip plasmids using Lipofectamine 2000 reagent ( Invitrogen , Germany ) . Forty-eight hours after transfection , cells were washed once with ice-cold PBS ( pH 8 . 0 ) . Cells were biotinylated at room temperature for 30 min with 0 . 8 mM solution of EZ-Link Sulfo-NHS-SS-Biotin reagent ( Pierce , Rockford , IL ) prepared in PBS ( pH 8 . 0 ) . Cells were subsequently washed with 50 mM Tris ( pH 8 . 0 ) to quench any non-reacted biotinylation reagent , and twice with ice-cold PBS ( pH 8 . 0 ) to remove excess biotinylation reagent . To capture biotinylated surface proteins , total protein was collected and affinity-purified with EZview Red Streptavidin Affinity Gel ( Sigma-Aldrich , Germany ) using manufacturer's protocol . Denatured total ( 5 μl ) and biotinylated surface protein ( 10 μl ) samples were separated by SDS-PAGE , and transferred onto PVDF membranes that were probed with mouse anti-GluA1 antibody ( 1:1500 ) or mouse anti-GluA2 antibody ( 1: 500 ) ( MAB2263 and MAB397 , respectively , both from Millipore , Billerica , MA , USA ) . For loading control , mouse anti-beta Actin antibody ( 1:4000 , MA5-15739 , Thermo Fisher Scientific , Rockford , USA ) was used . Surface protein was normalized for equal protein concentration . Relative quantification of surface GluA1 or GluA2 expression was carried out by densitometry of western blots using ImageJ software ( http://imagej . nih . gov/ij ) . HEK293/T17 cell lines stably expressing GluA1-flip or GluA2 ( Q ) -flip were grown and maintained using standard protocols . For electrophysiological recordings , cells were transfected using Lipofectamine 2000 ( Invitrogen , Germany ) and pRK5-CKAMP39 , -CKAMP52 or -CKAMP59 together with pEGFP-C1 ( Clontech-Takara Bio , France ) or pEGFP-C1 alone . Cells were recorded 24–72 h post-transfection . Fast application of glutamate onto lifted HEK293/T17 cells was performed as described ( Jonas and Sakmann , 1992 ) using theta glass tubing mounted onto a piezo translator . AMPAR-mediated currents were evoked by a 1 ms glutamate pulse for analyzing amplitude and deactivation , by a 100 ms glutamate pulse for analyzing desensitization and steady-state current amplitude , and by two 1ms pulses with 10 , 30 , 100 , 300 , 1000 and 3000 inter-event interval for analyzing recovery from desensitization . Application pipettes were tested by perfusing solutions with different salt concentrations through the two barrels onto open patch pipettes and recording current changes with 1 and 100 ms transitions of the application pipette . Only application pipettes with 20–80% rise times below 100 µs and with a reasonable symmetrical on- and offset were used . However , the expected solution exchange time is considerably slower with the use of whole cells instead of outside-out patches ( Barberis et al . , 2008 ) . The application solution contained ( in mM ) : 135 NaCl , 10 HEPES , 5 . 4 KCl , 1 . 8 CaCl2 , 1 MgCl2 , 5 glucose ( pH 7 . 2 ) . Whole-HEK293/T17 cell recordings were performed at room-temperature using pipettes pulled from borosilicate glass capillaries with a resistance of 3– 5 MΩ when filled with the following solution ( in mM ) : 120 Cs-gluconate , 10 CsCl , 8 NaCl , 10 HEPES , 10 phosphocreatine-Na , 0 . 3 GTP , 2 MgATP , 0 . 2 EGTA ( pH 7 . 3 , adjusted with NaOH ) . Liquid junction potentials were not corrected . AMPAR-current deactivation and desensitization were fitted with two exponentials , and the weighted tau ( 𝜏w ) was calculated as 𝜏w = ( 𝜏fx× af ) + ( 𝜏sx× as ) , where af and as are the relative amplitudes of the fast ( 𝜏f ) and slow ( 𝜏s ) exponential components . The in situ hybridization was done as described before ( von Engelhardt et al . , 2015 ) . Briefly , horizontal brain sections from adult C57Bl/6 mice were cut on the cryostat ( Leica Microsystems , Germany ) and hybridized with one of the following radiolabeled oligodeoxyribonucleotide probes: Ckamp39ins1 = 5’TGAGAAGTTCTGTCAGTGTCCTGGTCACCGTGCGCCGAGC Ckamp52ins1 = 5’AATGTCAGCCAGAGCCCTGTGGATGTTCATCTCTCGCGGA Ckamp59ins1 = 5’GCGGCATAGCACGCCAGTCGAGGTTGGAGGGCTTCATGGTGTT The oligodeoxyribonucleotide probes were 3′ end-labeled by terminal deoxynucleotidetransferase and ( a ) -33P-dATP ( Hartmann Analytic , Germany ) . Brain sections were then hybridized over night in , 4 x× SSC ( Saline-sodium citrate buffer , 0 . 6 M NaCl , 0 . 06 M sodium citrate ) , 50% formamide , 10% dextrane and 1 pg/μl labeled oligodeoxyribonucleotide probes at 42◦° C and subsequently washed at 55◦°C for 30 min , dehydrated and exposed to Kodak R X-omat AR film for 1 week . Stage V–VI Xenopus laevis oocytes were prepared , injected with cRNA as previously described ( Priel et al . , 2005 ) . Whole-cell two electrode voltage clamp ( TEVC ) recordings were used for estimation of current amplitude before proceeding to patch-clamp recordings and for determination of EC50 values for glutamate and CTZ . TEVC recordings were performed at 17oC , at holding potential of −70mV , using GeneClamp500 connected to digidata1322A and pCLAMP8 . 2 ( Axon Instruments ) . Data was analyzed by pCLAMP8 . 2 and ORIGIN 8 ( Origin Lab Corp . ) for estimation of the respective EC50s . For outside-out macropatch recordings the vitelline membrane was removed using forceps . Recordings were performed at 17oC , at membrane potential of -−120mV , using Axopatch 200B amplifier connected to digidata1322A and pCLAMP8 . 2 ( Axon Instruments , Foster City , CA ) . For rapid solution exchanges , a double-barrel glass ( theta tube ) mounted on a piezoelectric translator ( Burleigh , Fishers , NY ) was used as previously described ( Priel et al . , 2005 ) . Patch electrodes were fabricated from borosilicateglass with a low resistance of 0 . 3–1 MΩ . Receptor deactivation and desensitization were measured by applying glutamate ( 10 mM ) for 1 ms and 500 ms , respectively . Recovery from desensitization was estimated with the two-pulse protocol in which a constant 100 ms application of glutamate ( 10 mM ) was followed by a 100 ms test pulse applied at different time intervals . Western-blot analysis was done as previously described ( Priel et al . , 2005 ) on protein homogenates from 10 oocytes for each sample . Blots were probed with anti-Flag antibody ( 1:2000; monoclonal anti-FLAG M2 , Sigma-Aldrich , Israel ) and visualized using ChemiDoc XRS system ( Bio-Rad Laboratories ) . GluA1-flip and GluA2 ( Q ) -flip were injected at 1ng cRNA/oocyte and CKAMP39 , CKAMP44 & CKAMP52 were injected at 1 , 3 & 5 ng cRNA/oocyte , respectively . At these conditions , CKAMP39 , CKAMP44 & CKAMP52 exhibited comparable protein expression levels ( Figure 3A ) without a significant influence on total current amplitude compared to oocytes expressing the AMPAR alone ( Supplementary file 1A ) , thereby allowing better comparison between the CKAMPs in modulating AMPAR properties . Higher amounts of CKAMP cRNA injections caused a reduction in total current amplitude manifested by a reduction in total protein expression as revealed by Western-blot analysis with anti-GluA1 and anti-GluA2/3 antibodies , respectively ( not shown ) . Data are presented as mean ± standard error of the mean ( SEM ) and as median ± interquartile range ( IQR ) . Statistical differences between groups were examined by ANOVA , followed by Bonferroni test when the values showed a normal distribution , or by Kruskall-Wallis One Way ANOVA , followed by Dunn’s method for multiple comparisons for non-Gaussian distributed values . Normality of data distribution was tested by Kolmogorov-Smirnov test and equal variance by Bartlett’s test . Statistical analysis was performed using ORIGIN 8 ( Origin Lab Corp . ) or the GraphPad Prism version 5 . 00 , GraphPad Software , San Diego , CA , USA , www . graphpad . com . P values < 0 . 05 were considered statistically significant ( * = p < 0 . 05 , ** = p < 0 . 01 , *** = p < 0 . 001 , **** = p < 0 . 0001 ) . | The brain processes and transmits information through large networks of cells called neurons . A neuron can pass the information it receives to other neurons by releasing chemicals called neurotransmitters across junctions known as synapses . These chemicals bind to receptor proteins on the surface of the neighboring neuron , which triggers changes that affect the activity of this neuron . Glutamate is the most commonly used neurotransmitter in the brain and binds to receptor proteins called AMPA receptors . If a neuron frequently sends glutamate across a particular synapse , the number of AMPA receptors in the second neuron will increase in response . This makes signaling across the synapse easier – a process known as synaptic strengthening . The ability to change the strength of synapses is important for learning and memory . Proteins called auxiliary subunits also bind to AMPA receptors and regulate their properties , and hence also affect the strength of the synapse . For instance , some auxiliary subunits increase the number of AMPA receptors at the synapse , while others have an effect on how the receptor protein works . In 2010 , researchers identified a new auxiliary protein called CKAMP44 that modifies AMPA receptor activity . Now , Farrow , Khodosevich , Sapir , Schulmann et al . – including some of the researchers involved in the 2010 study – have identified three other auxiliary proteins that are similar to CKAMP44 . Collectively , these four proteins are termed the CKAMP family . The sequences of all four proteins were found to share many common features , especially in the regions that bind to the AMPA receptors . Like CKAMP44 , the new members of the CKAMP family are only present in the brain , although each protein is produced in different brain regions . Further investigation revealed that each member of the CKAMP family affects the AMPA receptor channels in a different way . Taken together , Farrow et al . ’s results suggest that the different CKAMP family members allow the activity of the AMPA receptors to be precisely controlled . The next challenge is to understand in more detail how each CKAMP family member influences how AMPA receptors work . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Auxiliary subunits of the CKAMP family differentially modulate AMPA receptor properties |
The classical drug development pipeline necessitates studies using animal models of human disease to gauge future efficacy in humans , however there is a low conversion rate from success in animals to humans . Non-alcoholic fatty liver disease ( NAFLD ) is a complex chronic disease without any established therapies and a major field of animal research . We performed a meta-analysis with meta-regression of 603 interventional rodent studies ( 10 , 364 animals ) in NAFLD to assess which variables influenced treatment response . Weight loss and alleviation of insulin resistance were consistently associated with improvement in NAFLD . Multiple drug classes that do not affect weight in humans caused weight loss in animals . Other study design variables , such as age of animals and dietary composition , influenced the magnitude of treatment effect . Publication bias may have increased effect estimates by 37-79% . These findings help to explain the challenge of reproducibility and translation within the field of metabolism .
Interventional studies in animals are an integral component of drug development . If a disease can be suitably modelled in an animal , then the therapeutic response to a treatment observed in animals should inform its potential efficacy in humans ( Howells et al . , 2014 ) . However , there is a well-documented translational gap between preclinical studies and subsequent outcomes in humans ( Hackam and Redelmeier , 2006; Landis et al . , 2012; Perel et al . , 2007 ) . Multiple factors contribute to this , including bias within study design ( Macleod et al . , 2015 ) , insufficiently powered preclinical studies ( Macleod et al . , 2005 ) , and biological differences between species ( Mestas and Hughes , 2004; Rangarajan and Weinberg , 2003 ) . Systematic analyses of preclinical studies have found that publication bias may account for at least a third of the estimate of efficacy in trials ( Henderson et al . , 2015; Sena et al . , 2010; van der Worp et al . , 2010 ) . In addition , other variables of animal model design can influence the magnitude of the treatment response ( Watzlawick et al . , 2019 ) and reporting of model design is often incomplete ( Flórez-Vargas et al . , 2016 ) . These findings are highly relevant in the context of the ‘reproducibility crisis’ ( Baker , 2016; von Herrath et al . , 2019 ) as well as having ethical implications for the use of animals in research that is not of optimum quality ( Prescott and Lidster , 2017 ) . Non-alcoholic fatty liver disease ( NAFLD ) is a highly active field of animal research ( Brenner , 2018; Farrell et al . , 2019 ) . NAFLD is a common condition characterised by increased liver fat ( hepatic steatosis ) that may progress to inflammation in the form of non-alcoholic steatohepatitis ( NASH ) and fibrosis ( Sanyal , 2019 ) . Cirrhosis , end-stage liver disease , and hepatocellular carcinoma develop in a small proportion of patients . However , due to the high prevalence of obesity , NAFLD is the second most common indication for liver transplant in the United States ( Younossi et al . , 2018 ) , predicted to overtake hepatitis C virus . NAFLD is intricately related with insulin resistance and therefore usually coexists with other features of the metabolic syndrome , such as type 2 diabetes and its recognised complications including cerebrovascular disease , coronary artery disease , and chronic kidney disease ( Byrne and Targher , 2015 ) . There are currently no approved pharmacological therapies for NAFLD ( Chalasani et al . , 2018 ) . Several Phase three trials are ongoing ( Ratziu et al . , 2019 ) , but many interventions that appeared to have substantial efficacy in preclinical models have failed to be replicated in humans ( Budas et al . , 2016; Harrison et al . , 2018; STELLAR-3 and STELLAR-4 Investigators et al . , 2020; Sanyal et al . , 2014 ) . These studies have used a wide range of preclinical NAFLD models , including genetically modified animals ( e . g . leptin deficient ob/ob mice ) , hypercaloric diets ( e . g . high-fat diet ) , and toxic insults ( e . g . streptozocin injections ) , all of which may be used in varying combinations and with different parameters ( Anstee and Goldin , 2006 ) . It is not known if , or which of , these variables influence treatment response to therapeutic agents in preclinical models of NAFLD , and which models are better predictors of response in humans . Therefore , we performed a meta-analysis of interventional rodent studies of NAFLD to describe which drug classes were associated with improvement in NAFLD and whether any study characteristics ( or biases ) were linked to the magnitude of effect .
We used random-effects meta-analysis to estimate the mean difference ( MD ) in hepatic triglyceride ( TG ) content between intervention and control groups ( Figure 2A ) . The overall mean difference in hepatic TG content was −29 . 9% ( 95% CI −33% , −27% ) with considerable between-study heterogeneity ( I2 = 90% ( 95% CI 89% , 90% ) , PQ <1×10−300 ) . Exclusion of outliers minimally affected the overall estimate ( −30 . 2% ( 95% CI −33% , −27% ) , Figure 2—source data 1 ) . For comparison , a relative decline of liver fat by ≥30% , as measured by magnetic resonance imaging proton-density fat fraction ( MRI-PDFF ) , has been determined as the reduction required to achieve histological response in humans with NAFLD ( Jayakumar et al . , 2019; Loomba et al . , 2020; Stine et al . , 2020 ) . We hypothesised that much of this heterogeneity would be due to the different drug class interventions , with some classes having a greater effect than others . On meta-analysis using drug class as a subgroup , 22/28 ( 79% ) of drug classes demonstrated a significant reduction in hepatic TG ( i . e . the upper limit of their 95% CI was negative ) . If we were to use ≥30% reduction as a benchmark for clinical significance ( analogous to change in MRI-PDFF ) , only 3/28 ( 11% ) of drug classes passed this cut-off: fibrates , omega-3 polyunsaturated fatty acids ( mixtures ) , and DPP-4 inhibitors . The 95% CI of 24/28 drug classes overlapped with the CI of the overall effect estimate . Two drug classes , thiazolidinediones and vitamin E , were found to have a smaller mean reduction in hepatic TG and two classes had a greater reduction: fibrates and mixtures of omega-3 polyunsaturated fatty acids ( PUFA ) . However , ‘PUFA mixtures’ was a comparatively broad drug class , and many PUFA mixtures included eicosapentaenoic acid ( EPA ) or docosahexaenoic acid ( DHA ) , which individually showed no significant reduction in hepatic TG . There remained substantial or considerable heterogeneity within drug class subgroups ( PQ <0 . 05 for 21/28 drug classes , Figure 2—source data 1 ) . In order to investigate whether this heterogeneity was due to variation between individual drugs within classes we repeated the meta-analysis with subgroup by individual drugs ( Figure 2—figure supplement 1 ) . There was sufficient data for meta-analysis of 28 individual drugs ( from the original 28 drug classes ) . 22/28 ( 79% ) individual drugs were found to have a significant reduction in hepatic TG . Vitamin E was associated with a smaller mean reduction in hepatic TG than the 95% CI of the overall estimate , whilst fenofibrate was the only drug with a greater mean difference than the overall estimate . There remained considerable heterogeneity within subgroups for 20/28 drugs ( I2 = 75–100% , PQ <0 . 05 ) . We then performed univariable meta-regression to investigate which variables accounted for the heterogeneity in results ( Figure 2—source data 1 ) . Though individual drug used was the single variable that accounted for most heterogeneity ( adj R2 = 4 . 9% , p=0 . 02 ) , the majority of variation in results was unaccounted . An association was also observed for weight difference ( adj R2 = 3 . 3% , p=6 . 4×10−4 ) , where greater weight loss in the intervention group was associated with a greater reduction in hepatic TG . This association was stronger after removal of NAFLD models that induce weight loss ( e . g . methionine-choline deficient diet ( MCD ) , Figure 2B ) and similar results were obtained for difference in fasting insulin levels ( Figure 2C ) . When these study characteristics were combined for multivariable meta-regression using an unbiased method , 10 variables were predicted to substantially contribute to the variation in hepatic TG difference ( Table 1 ) . In final model 1 , weight difference was the only variable to be significantly associated with MD in hepatic TG ( p=0 . 003 ) . Including drug used in model two was able to account for all heterogeneity in results ( Figure 2—source data 1 ) in a small subset of cohorts ( k = 42 ) , though neither of these models were significantly predictive of outcome following permutation tests ( p-value*>0 . 05 ) . Given that meta-regression implicated weight loss and improved insulin sensitivity in results , we explored how these traits were distributed by drug class ( Figure 3A ) . Including all available data , we observed that 12/33 ( 36% ) drug classes showed a significant reduction in weight ( i . e . the upper limit of their 95% CI was below 1 , Figure 3—source data 1 ) . 17/32 ( 53% ) and 15/25 ( 60% ) of drug classes were associated with reductions in fasting glucose ( Figure 3B ) and insulin ( Figure 3—figure supplement 1A ) , respectively . There was a positive correlation between weight , glucose , and insulin differences ( Figure 3—figure supplement 1B ) . In addition , there was a negative correlation between weight difference and study duration or the age of mice at the end of intervention , that is longer studies ( or those in older mice ) were associated with greater weight loss in interventional groups . We then explored whether these results showed study distribution ( publication ) bias or were heavily influenced by individual outliers ( Figure 2—figure supplement 2 ) . There was an uneven distribution of studies with a bias towards a reduction in hepatic TG , which was supported by Egger’s test ( β = - . 83 [95% CI −1 . 3 , −0 . 4] , p=2 . 2×10−4 ) . Using the trim-and-fill method to account for this bias , we estimated that the true overall mean difference in hepatic TG would be −18 . 7% ( 95% CI −21% , −16% ) , over a third smaller than the original estimate . Whilst hepatic TG was the most widely reported measure , histological assessment of disease is considered the gold standard for patients with NAFLD . Therefore , we performed a meta-analysis of MD in steatosis grade ( Figure 4A ) . The overall MD in steatosis was −0 . 7 ( 95% CI −0 . 8 , −0 . 5 ) again with considerable heterogeneity ( I2 = 94% ( 95% CI 93% , 95% ) , PQ <1×10−300 ) . Compared to hepatic TG , fewer drug classes were identified to be associated with a significant reduction in steatosis grade ( 8/22 , 36% ) , though again fibrates showed the largest effect size . Similar results were obtained when performing subgrouping by individual drugs , rather than classes ( Figure 4—source data 1 ) . Univariable meta-regression found a marked association between difference in plasma glucose levels and MD in steatosis grade ( Figure 4B , adj R221% , p=2 . 4×10−6 ) . Similar associations were observed for difference in weight and insulin levels , particularly after removal of weight-loss inducing models ( Figure 4C ) . In addition , the sex of animals ( adj R27% , p=0 . 01 ) and genetic background were associated with MD in steatosis grade ( Figure 4—source data 1 ) . When factors were combined in multivariable meta-regression ( Table 1 ) , a model using sex , fasting glucose difference , and fat ( %kcal ) in diet accounted for 92% of variability in a small subset of cohorts ( k = 19 ) , which remained robust after a multiple permutation test ( p-value*=0 . 03 ) . 9/16 ( 56% ) drug classes were associated with a reduction in MD of lobular inflammation ( Figure 5A ) . Again there was considerable heterogeneity within drug classes and when subgrouping by individual drugs ( Figure 5—source data 1 ) . Univariable meta-regression identified an association with difference in weight ( Figure 5B , adj R215% , p=4 . 0×10−4 ) , as had been observed for steatosis grade and hepatic TG content . In addition , an association was found for fat %kcal in diet and MD in lobular inflammation: a higher %kcal fat in diet was associated with a smaller difference in lobular inflammation ( Figure 5C , adj R221% , p=1 . 7×10−5 ) , indicating that study design was associated with size of treatment response . The bubble plot of fat content in diet also illustrated that the majority of studies reporting fat content in diet used either 40–45% or 60% kcal fat ( Figure 5C ) . 8/14 ( 57% ) drug classes were associated with a reduction in hepatocellular ballooning ( Figure 6A ) . Fibrates showed greater reduction in ballooning than other studied drug classes , however this could not be replicated at an individual drug level ( Figure 6—figure supplement 1 ) . Similar to previous analyses , difference in fasting glucose ( adj R217% , p=9 . 0×10−4 ) and weight ( adj R28% , p=0 . 01 ) were associated with the magnitude of treatment effect . Study design characteristics also influenced difference in ballooning , namely percentage of fat in diet ( Figure 6B , greater reduction in ballooning where a lower %kcal was used ) and percentage of fructose/glucose in diet ( Figure 6C ) ; however , there were only 12 studies contributing to this analysis . In addition , longer studies were associated with larger reductions in ballooning severity ( Figure 6D ) . The NAFLD activity score is a composite of steatosis , lobular inflammation , and ballooning scores . The results largely reflected those observed for the previous three meta-analyses ( Figure 7A ) . 10/14 ( 71% ) drug classes were associated with a significant reduction in NAS , with fibrates being the most beneficial drug class . Meta-regression found associations for difference in weight ( Figure 7B ) and glucose ( Figure 7C ) to account for 11% and 12% of heterogeneity in results , respectively . multiple-variable meta-regression models were able to account for more than 60% of variation in results ( in a small subset of cohorts ) using genetic background , fat in diet , age at start of intervention , weight and glucose difference , but without requiring drug or drug class ( Table 1 ) . Fibrosis stage is the histological feature that most strongly correlates with liver-related outcomes in humans with NAFLD ( Angulo et al . , 2015; Ekstedt et al . , 2015 ) , and was therefore pre-specified as the primary outcome measure for this study . However , it was reported in only 58/603 ( 9 . 6% ) of cohorts . Only FXR agonists and statins ( 2/5 , 40% drug classes ) were associated with a significant reduction in fibrosis stage ( Figure 8A ) , where the overall mean difference was −0 . 5 ( 95% CI −0 . 6 , −0 . 3 ) stages . Meta-regression replicated previous findings for other traits , showing that difference in weight was associated with reduction in fibrosis stage ( Figure 8B , adj R227% , p=0 . 004 ) . Funnel plots for steatosis grade , lobular inflammation , fibrosis stage , and NAS were asymmetric ( Figure 9 ) , supported by the results of Egger’s test for each analysis . Using the trim-and-fill method to account for these differences substantially altered the overall effect estimates: for steatosis grade , there was an 79% reduction in estimated effect size to −0 . 14 ( 95% −0 . 3 , + . 01 ) ; for lobular inflammation , a 70% reduction in effect size to −0 . 18 ( 95% −0 . 32 , −0 . 05 ) ; for fibrosis , 72% reduction to −0 . 12 ( 95% −0 . 33 , + . 08 ) ; and NAS , 55% reduction in effect size to −0 . 82 ( 95% −0 . 1 . 1 , −0 . 5 ) . We used a four-item scale to estimate study quality ( Figure 9—figure supplement 1 ) . We found that 497/603 ( 82% ) cohorts were at high risk of bias due to either absence of randomisation or absence of blinding . In addition , we used post-hoc power calculations to estimate the proportion of studies that were adequately powered . For analysis of hepatic TG , 39% ( 185/474 ) cohorts had a power of 80% or greater on post-hoc calculation . However , using the results from this meta-analysis , to achieve a power of 80% with significance set as p=0 . 05 , group size would need to be n = 16 . 4 . 2% ( 20/474 ) cohorts included 16 or more animals and would have met sufficient power to detect associations , based on these data . Similar results were obtained for histological steatosis grade: 70/174 ( 40% ) reported results consistent with >80% power but only 27/174 ( 16% ) had a group size large enough to be expected to reach 80% power . The majority of drug classes ( or individual drugs ) were found to show a significant reduction in severity of NAFLD . Fibrates ( for which most data were available for fenofibrate ) demonstrated the greatest improvement in several outcome measures ( Table 1 ) . Univariable meta-regression found that weight loss and lower fasting glucose were associated with a greater improvement in multiple outcomes ( Figure 10 ) . In addition , diet composition influenced the magnitude of treatment response for lobular inflammation , ballooning , and fibrosis . Heatmap illustrating the results of univariable meta-regression analyses using continuous variables . Beta-regression co-efficient was normalized within each outcome analysis ( e . g . steatosis grade ) to mean = 0 , standard deviation = 1 . Rows ( variables used as predictors in meta-regression ) and columns ( outcome measures for NAFLD ) are clustered for similarity .
Multiple drug classes improve NAFLD in rodents , however these results may be confounded by weight loss and alleviation of insulin resistance not observed in humans treated with the same drugs . Publication bias over-estimates these effect sizes by at least a third and a variety of other study design characteristics also influence treatment response . Therefore , standardisation of practices is needed in preclinical studies of metabolism to improve the translatability and reproducibility of findings .
The systematic review protocol was prospectively registered with SyRF ( Systematic Review Facility ) and is available from: https://drive . google . com/file/d/0B7Z0eAxKc8ApQ0p4OG5SblRlRTA/view . PubMed via MEDLINE and EMBASE was searched for published articles of experimental rodent models of fatty liver , NAFLD , or non-alcoholic steatohepatitis ( NASH ) . The following search term was used: ( ‘Non-alcoholic fatty liver disease’ OR ‘Nonalcoholic fatty liver disease’ OR ‘NAFLD’ OR ‘non-alcoholic steatohepatitis’ OR ‘nonalcoholic steatohepatitis’ OR ‘NASH’ OR ‘fatty liver’ OR ‘hepatic steatosis’ ) AND ( ‘mouse’ OR ‘animal’ OR ‘rat’ OR ‘murine’ OR ‘animal model’ OR ‘murine model’ OR ‘rodent model’ OR ‘experimental model’ ) NOT ( ‘Review’ ) . Both databases were searched using the ‘Animal’ filters ( de Vries et al . , 2014; Hooijmans et al . , 2010 ) , the results combined , and duplicates eliminated . The search was completed in January 2019 . Our inclusion criteria were as follows: primary research articles using mice or rats to model NAFLD ( to include hepatic steatosis , NASH , and NASH-fibrosis ) , use of pharmacological intervention with a control ( or placebo ) group , and that the pharmacological intervention class ( e . g . statins ) had been used in Phase 2 or three trials in humans for treatment of NAFLD/NASH . Studies were excluded if: not modelling NAFLD/NASH; studies in humans or any animal other than mice and rats; reviews , comments , letters , editorials , meta-analyses , ideas; articles not in English ( unless there was an available translation ) ; studies not reporting any relevant outcome metrics ( hepatic triglyceride content relative to hepatic protein ( e . g . mg/mg or µM/mg ) , NAFLD Activity Score [Brunt et al . , 2011; Kleiner et al . , 2005] or any of its components ) , portal inflammation grade [Brunt et al . , 2009] , or histological fibrosis stage ( 0–4 ) ; and studies using a pharmacological agent class that had not been used in Phase 2/3 studies in humans for NAFLD . Abstracts and titles were screened to identify relevant studies using Rayyan ( Ouzzani et al . , 2016 ) . Potentially relevant studies had their full-text extracted and were assessed against inclusion/exclusion criteria independently by two reviewers , with discrepancies settled by discussion with JPM . The variables extracted were as follows: phenotypic characteristics of animal model used ( sex , diet [including percentage of fat , glucose , fructose , sucrose , and cholesterol in diet] , rodent age , genetic alterations , background animal strain ) ; drug treatment ( dose , drug class , duration , age at intervention ) , hepatic triglyceride content and liver histology . Fructose/glucose concentration in diet was collected together as a single data point as they were frequently combined in diets . Liver histology results were extracted where the ( human ) NAFLD Activity Score ( NAS [0–8] ) and/or any of its components had been used ( steatosis grade [0–3] , lobular inflammation [0–3] , and ballooning severity [0–2]; portal inflammation severity [0–2] ) ; and/or histological fibrosis stage [0–4] . Studies frequently included multiple cohorts or interventional arms , which were defined as use of a different animal model of NAFLD , a different drug , or a different drug dose . Data were extracted for each cohort or interventional arm separately . Each paper was assessed in the following four areas: use of a protocol , reporting use of randomisation , reporting use of blinding , and a power calculation . ‘Use of a protocol’ assessed the article specifically referring to a protocol that was in place and prior to the start of the study . These were each given a score of 1 , and each paper was assigned an overall ‘quality score’ . A post-hoc power calculation was performed for each study using the means of each group and a common SD ( Cohen , 1988 ) using the pwr ( Champely , 2018 ) package in R . In addition , a ‘pre-test’ sample size calculation was performed using: the overall effect summary from meta-analysis , power = 80% , and p-value=0 . 05 . Multiple studies used a single placebo ( or control ) group for several experimental arms . Where possible , the experimental arms were combined into a single experimental cohort and compared to the control group ( Higgins and Green , 2011 ) . Where this was not appropriate ( e . g . interventions from different drug classes ) , the control group was divided evenly across interventional groups . Therefore , each control animal was included only once in analyses . Where possible , drugs were grouped into classes based upon their pharmacological mechanism of action . The majority were well-established classes of drugs: angiotensin receptor blockers , biguanides , dipeptidyl peptidase 4 ( DPP4 ) inhibitors , fibrates , glucagon-like peptide-1 ( GLP-1 ) agonists , statins etc . In some cases there was only a single drug represented in their class , for example: polyphenols – resveratrol , and cholesterol absorption inhibitors – ezetimibe . More novel agents fell into pharmacological classes based on mechanism that are less well established , for example: stearoyl–CoA desaturase-1 inhibitors , or PPARα/δ agonists . Other agents , particularly where the mechanism of action is unclear , were made a class of their own , for example , whilst eicosapentaenoic acid and docosahexaenoic acid are both omega-3 polyunsaturated fatty acids ( PUFA ) , their mechanism is not clear and therefore were classed individually , with other mixtures of PUFA being classed separately . Similarly , berberine and silymarin were classed individually . Where individual bacterial strains were used for probiotics they were classed accordingly ( e . g . Lactobacillus sp . ) , but where a mixture of strains were used a ‘Probiotic ( mix ) ’ category was allocated . For analyses by individual drugs , all agents were separated , though for some drugs ( e . g . berberine ) this was unchanged from their ‘drug class’ grouping . Prior to analysis , hepatic triglyceride content was normalized as a percentage of placebo ( or control ) for each cohort . Weight , fasting glucose , and fasting insulin of interventional groups were expressed as a proportion difference relative to placebo ( e . g . 20% lower fasting glucose in interventional group = 0 . 8 ) . All continuous variables were examined for normality using histograms and , where distributions were skewed , variables were logarithmically transformed prior to use in regression analyses . Primary outcome was the mean difference in histological fibrosis stage in the interventional group compared to control/placebo . Secondary outcomes were histological features: hepatic triglyceride ( TG ) content , steatosis grade , lobular inflammation , ballooning , and overall NAS . There was insufficient data to perform meta-analysis for portal inflammation severity . Random-effects meta-analysis using the Hartung-Knapp-Sidik-Jonkman method was used to calculate mean difference in the outcome measure . Each meta-analysis was run three times , once with subgrouping by drug class , then a sensitivity analysis using subgrouping by drug class after excluding outliers ( as described below ) , and then once using individual drugs . Drug classes , or individual drugs , were only included in meta-analyses where there was data from minimum three unique articles reporting that outcome . Drugs or drug classes were considered to have a significant effect on the outcome if their 95% CI did not cross zero . Drugs ( or drug classes ) were also assessed to have greater ( or smaller ) difference in the outcome measure if their 95% CI did not overlap with the 95% CI of the overall effect estimate . Additionally , for hepatic TG only , drugs were compared to a benchmark of 30% reduction in liver fat . This was based on data from MRI-PDFF in humans that suggests ≥30% reduction in liver fat is associated with a substantial histological response ( Jayakumar et al . , 2019; Loomba et al . , 2020; Stine et al . , 2020 ) . Heterogeneity within drug classes ( or individual drugs ) and across the whole dataset was reported using Cochran’s Q , Higgin’s and Thompson’s I2 , and 𝜏2 . Interpretation of I2 was performed according to the Cochrane Handbook where ‘considerable heterogeneity’ refers to PQ <0 . 05 and I2 = 75–100% ( Higgins and Green , 2011 ) . Potential outliers were identified using a Baujat plot ( Baujat et al . , 2002 ) and by assessment of standard deviation ( SD ) , where all studies with excess contribution to heterogeneity on visual inspection of the Baujat plot or SD >95th centile were excluded in a sensitivity analysis . Study distribution ( ‘publication’ ) bias was assessed using funnel plot with Egger’s test . Given evidence of study distribution bias , Duval and Tweedie’s trim-and-fill procedure ( Duval and Tweedie , 2000 ) was performed to estimate the impact of bias on the overall measure . Mixed-effects meta-regression was performed to assess which baseline variables were associated with heterogeneity in each outcome measure . Meta-regression was performed using both categorical variables ( e . g . drug class , sex , animal background , NAFLD model design ) and continuous variables ( e . g . percentage of components in diet , age at intervention , drug dose ) . For each regression analysis , variables were only included where three or more unique articles reported each variable . The number of cohorts included in each regression analysis is reported with their results . Univariable meta-regressions were considered significant where p-value<0 . 05 and were replicated in more than one outcome metric ( e . g . hepatic TG and steatosis grade ) . Univariable meta-regression was repeated for weight , glucose , and insulin difference after removal of models causing weight loss . These analyses of weight loss ( or gain ) with secondary changes in glycaemic control are most relevant to obese or insulin resistant animals . We hypothesised that trends would be strengthened after removal of models that did not recapitulate the metabolic syndrome . Models excluded were: methionine-choline deficient diet ( with or without added high-fat ) , orotic acid , choline deficient diet ( with or without added high-fat ) , and choline deficient L-amino-acid defined diet . Models were excluded irrespective of their genetic background , for example leptin receptor deficiency ( db/db ) plus methionine-choline deficient diet was excluded for this sensitivity analysis . For these three variables , due to replication of testing , statistical significance was set at p-value<0 . 025 . multiple-variable meta-regression was performed to assess what proportion of between-study heterogeneity could be accounted for by baseline characteristics ( using adjusted R2 ) . First variables were examined for multicollinearity and where two variables had Pearson correlation >0 . 6 , one was removed . Then , multimodel inference ( dmetar::multimodel . inference , RRID:SCR_019054 ) was used to obtain the model with the best fit for the data . Initially , drug ( or drug class ) was not included as an input variable as this greatly increased the number of variables and reduced the number of studies for inclusion . The optimum model ( defined by the lowest Akaike’s Information Criterion ) was then used in multiple-variable meta-regression ( known as ‘final model 1’ ) . The robustness of this model was tested using a permutation test ( metafor::permutest , RRID:SCR_003450 ) . This process was repeated to generate ‘final model 2’ , by additionally including individual drugs ( for TG ) or drug class ( for steatosis grade and NAS ) , as input variables in the multimodel inference stage . It was not possible to generate a 2nd multivariable meta-regression model including drug ( or drug class ) for lobular inflammation , ballooning , and fibrosis due to insufficient data . For multivariable meta-regression , individual variables were defined as ‘Top predictors’ if they had a predictor importance >0 . 8 on dmetar::multimodel . inference analysis . Individual variables were considered significant within each model where p-value<0 . 05 . Models were considered to significantly predict outcomes where p-value*<0 . 05 after use of metafor::permutest . Statistical analysis was performed using R 4 . 0 . 2 for Mac ( Harrer et al . , 2019; R Core Development team , 2019 ) with packages dmetar ( Harrer et al . , 2019 ) , meta ( RRID:SCR_019055 , [Schwarzer G , 2007] ) , and metafor ( Viechtbauer , 2010 ) . Graphs were also generated using GraphPad Prism ( RRID:SCR_002798 , v8 . 0 for Mac , GraphPad Software , La Jolla California , USA ) . | Obesity and diabetes are increasingly common diseases that can lead to other complications such as fatty liver disease . Fatty liver disease affects one in five people and is caused by a built-up of fat in the liver , which can result in scarring of the liver tissue and other serious complications . There is currently no cure for fatty liver disease . Drugs that have been effective in treating the condition in mice , lack efficacy in humans . To better understand why this is the case , Hunter , de Gracia Hahn , Duret , Im et al . conducted a review of over 5 , 000 published studies , analysing over 600 experiments . Hunter et al . asked which drugs improved fatty liver in mice the most and if they had the same effect in humans . They also tested whether the age of the mice affected the outcome of the experiments . The analyses revealed that the drugs that work best in mice are different to the ones that show some effect in humans . In mice , many of the drugs reduced their weight or lowered their blood sugar levels , which also improved the fatty liver condition . Moreover , drugs appeared to be less effective the older the mice were . However , most of these drugs do not cause weight loss or lower blood sugar levels in humans , suggesting that factors other than the intended action of these drug could affect the outcome of a mouse study . These findings will help shape future research into obesity , diabetes and fatty liver disease using mice . They highlight that results obtained from studies with mice so far do not predict if a drug will work in humans to treat fatty liver disease . Moreover , weight loss seems to be the most important factor linked to how efficiently a drug treats fatty liver disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"medicine"
] | 2020 | Weight loss, insulin resistance, and study design confound results in a meta-analysis of animal models of fatty liver |
To quantify gene regulation , a function is required that relates transcription factor binding to DNA ( input ) to the rate of mRNA synthesis from a target gene ( output ) . Such a ‘gene regulation function’ ( GRF ) generally cannot be measured because the experimental titration of inputs and simultaneous readout of outputs is difficult . Here we show that GRFs may instead be inferred from natural changes in cellular gene expression , as exemplified for the cell cycle in the yeast S . cerevisiae . We develop this inference approach based on a time series of mRNA synthesis rates from a synchronized population of cells observed over three cell cycles . We first estimate the functional form of how input transcription factors determine mRNA output and then derive GRFs for target genes in the CLB2 gene cluster that are expressed during G2/M phase . Systematic analysis of additional GRFs suggests a network architecture that rationalizes transcriptional cell cycle oscillations . We find that a transcription factor network alone can produce oscillations in mRNA expression , but that additional input from cyclin oscillations is required to arrive at the native behaviour of the cell cycle oscillator .
Much of the topology of the yeast transcriptional network is known from functional genomics approaches such as chromatin immunoprecipitation and mRNA expression data ( Lee et al . , 2002; Harbison et al . , 2004; Tsai et al . , 2005; Wu et al . , 2006; Hu et al . , 2007 ) . However , the nature of the nodes in such networks , where the input signals are integrated into a transcriptional response , remain elusive . The quantitative design of a network node is known as ‘gene regulation function’ ( GRF ) and is of central importance for understanding the regulatory dynamics of the network . In E . coli , the measurement of a GRF was demonstrated for a single node with two inputs ( Setty et al . , 2003; Kuhlman et al . , 2007 ) . However , the approach relies on specific inducers for the involved transcription factors and is applicable only to a very limited set of genes . GRFs must therefore be inferred from indirect experimental evidence . In multicellular organisms , GRFs for developmental genes can be inferred from the spatially varying profiles of morphogens ( Jaeger et al . , 2004; Segal et al . , 2008; Junker et al . , 2014 ) . For a single-cell organism such as yeast , the reconstruction of GRFs must instead rely on the temporal variation of input factors . The variation can either be intrinsic , as observed during the cell cycle ( Spellman et al . , 1998 ) , or it can be triggered by an external perturbation , as during DNA damage ( Workman et al . , 2006 ) or osmotic stress ( Miller et al . , 2011 ) . A major obstacle for inferring GRFs is that GRFs describe gene activity but expression data typically provide only total mRNA levels . This limitation is overcome by ‘Dynamic Transcriptome Analysis’ ( DTA ) , which uses nonperturbing metabolic RNA labeling to additionally obtain the amounts of newly synthesized mRNA , which can be used as a proxy for RNA synthesis rates and gene activity ( Miller et al . , 2011; Sun et al . , 2012 ) . Cell cycle progression in yeast is controlled by cyclin-dependent kinases ( CDKs ) that are activated by periodically expressed cyclins ( Evans et al . , 1983 ) . In addition , periodic transcription occurs in waves ( Breeden , 2000; Breeden , 2003 ) and plays a role in maintaining the cell cycle ( Wittenberg and Reed , 2005 ) . Several periodically expressed transcription factors control each other , forming a regulatory module ( Simon et al . , 2001 ) . The observation of periodic gene expression even in the absence of mitotic cyclins indicated that sequential transcription activation is sufficient for periodic expression of most cell cycle genes ( Orlando et al . , 2008; Haase and Reed , 1999 ) . However , coupling to cyclin-CDK activity serves as a pacemaker and increases robustness of the transcriptional oscillator ( Simmons Kovacs et al . , 2012 ) . Several models for a minimal transcriptional cell cycle oscillator have been proposed ( Simon et al . , 2001; Lee et al . , 2002; Orlando et al . , 2008; Sevim et al . , 2010; Simmons Kovacs et al . , 2012 ) . Parts of the transcriptional network were also reconstructed computationally from expression data ( Chen et al . , 2004; Wu et al . , 2006 ) . However , a comprehensive understanding of the transcriptional cell cycle network is still missing , and the degree of its dependence on the cyclin oscillator is debated ( Bristow et al . , 2014; Rahi et al . , 2016 ) . Here we develop a method to infer GRFs from DTA data and apply it to cell cycle genes in yeast . We use DTA data providing the mRNA synthesis rates and levels for synchronized S . cerevisiae cells over three cell cycles in two replicate experiments ( Eser et al . , 2014 ) . We consider target genes with significant regulatory inputs from one or more transcription factors . We restrict our analysis to cases where evidence for physical interaction between transcription factors and target genes exists or a genetic interaction is established . We apply our method to infer GRFs of cell cycle regulated transcription factors . We deduce possible models for a transcriptional cell cycle oscillator and test their capability to generate oscillations without cyclin-CDK activity . Our approach may be extended to quantitatively describe other gene regulatory systems , such as stress response mechanisms , apoptosis , or cell differentiation networks .
Our method to infer gene regulation functions ( GRFs ) from DTA data is illustrated in Figure 1 . After selecting a target gene of interest , we compile a list of known input factors and focus on those that display a significant fold-change in mRNA level over the time course of the experiment . We assume that their dynamics can rationalize the output dynamics ( Figure 1B ) via a smooth input-output relation , the GRF . This assumption is viable even for genes that belong to a larger regulatory network . We can treat each gene independently because the DTA data provide mRNA time traces m ( t ) for all input factors and output mRNA synthesis rates s ( t ) for most genes ( Figure 1A ) . The inputs may be transcription factors or cofactors ( Siggers et al . , 2011 ) , but for simplicity we refer to all inputs as transcription factors ( TFs ) . Here , we do not explicitly consider post-transcriptional regulation of TFs or potential inputs from regulatory RNAs . 10 . 7554/eLife . 12188 . 003Figure 1 . Reconstruction of regulation functions . ( A ) We select a target gene ( green ) and TFs ( blue ) , which are known from the literature to interact with the target gene . We consider only target genes and corresponding TFs , which show significant variation in our dataset . ( B ) The fitting task is to find a regulation model using the TF mRNA expression level ( upper plot ) , so that the time series of the target synthesis rate ( dots in lower plot ) is best described by the model synthesis rate sm ( solid curve in lower plot ) . ( C ) We describe the concentration of active TF protein p ( t ) by a simple model with translation rate νt and linear degradation rate λ ( see box quantitative model ) . For a correctly chosen λ the target gene synthesis rate plotted against the TF protein concentration collapses to a curve ( bottom plot ) . ( D ) Together with p ( t ) a regulation function ( target synthesis rate sm as a function of TF protein p ) is estimated . For a single TF the regulation function has four parameters ( b , α , K , n ) and two possible directions of regulation: activator and repressor ( see box mathematical model ) . To find the right regulatory direction we fit both and select the one , which yields the better score . ( E ) Two TFs can interact in multiple ways to generate a two dimensional combinatorial regulation function ( Buchler et al . , 2003 ) . For two TFs we estimate the protein models and six parameters for the regulation function . There are 10 non-trivial regulatory analog 'logic' operations to test , of which 3 examples are depicted ( see box quantitative model for the corresponding equations ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 003 We infer a GRF by constructing a parameterized model , which describes the measured target gene output s ( t ) via Hill-type functions of the input TF levels ( Figure 1 , 'quantitative model' ) . For the case of a single input , the GRF is parameterized by the basal activity b , the regulation amplitude α , the response threshold K , and the sensitivity n . The parameters are estimated by finding the global best fit of our model to the measured target gene activity using the simulated annealing technique , see ‘Materials and methods’ . The effect of the input TFs can be either activating or repressing . To discriminate between these possibilities , we fit models for both and select the one that yields the better score . Because protein levels are unknown , we infer a proxy p ( t ) for the TF concentration from its mRNA level ( see box 'quantitative model' in Figure 1 ) using a minimal model with a constant mRNA translation rate νt and a constant effective protein degradation rate λ . The resulting p ( t ) is therefore both delayed and smoothened with respect to the input mRNA concentration m ( t ) . The characteristic timescale for both the delay and the smoothening of the signal is the effective protein half-life τ1/2=ln ( 2 ) /λ . Figure 1C visualizes the effects of different protein degradation rates , and shows that information about the parameter λ for a TF is contained in the transcription rate trajectories of target genes: only for an appropriate choice of λ does the target gene activity s ( t ) plotted against p ( t ) collapse to a curve , which corresponds to the GRF of the target gene . This data collapse also serves as a visual consistency check for our approach . Note that besides the actual degradation of proteins , the timescale τ1/2 subsumes the effects of a number of molecular processes that are not yet characterized quantitatively . For instance , transport into and out of the nucleus , phosphorylation and dephosphorylation of transcription factors , and dilution of protein levels by cell growth all affect how the levels of activated transcription factors 'seen' by their target genes dynamically adapt after their mRNA level has changed . Effectively , these processes create a time delay , which is captured in our coarse-grained model by the single parameter . The loss of synchrony within the measured cell population does not significantly impact the ability of our model to infer GRFs . This is exemplified by the TF Swi4 and its target gene Rnr1 ( Figure 1 ) . Both Swi4 and Rnr1 are periodically expressed with a period of ∼60 min , corresponding to the cell cycle period of the used yeast strain under the conditions of the experiments . The loss of synchrony of cell cycle progression between different cells is observed as a dampening of the oscillations , due to the population average over cells that increasingly diverge in their relative cell cycle phase . However , this dampening occurs for inputs and outputs alike , and our nonlinear inference scheme is tolerant against a partial loss of synchrony: The limited cell-to-cell variation in TF levels at a given time point samples only a limited regime of the GRF input range , within which the nonlinear GRF appears locally linear and is therefore not significantly affected by the population average . In the following we focus our analysis mainly on cell cycle-dependent genes , since most genes with significant fold-change in our dataset are periodically expressed . Our method is applicable not only to genes with single input signals , but also to some cases of combinatorial regulation . We take the combinatorial interaction of multiple input factors into account by inferring the best fitting gene regulatory 'logic' ( Buchler et al . , 2003 ) along with the parameters that characterize the shape of the GRF . We combine activating and repressing Hill functions to model analog combinatorial 'logic' response functions ( Figure 1 ) . For example , two activating TFs can up-regulate a target gene either individually ( Figure 1E , 'OR logic' ) or cooperatively , if both are abundant ( Figure 1E , 'AND logic' ) . For two input TFs we fit ten different non-trivial GRFs and select the 'logic' function with the best score; see ‘Materials and methods’ for the complete list of the types of GRFs that we consider . The score is defined as the fraction of the variance in the data that is not explained by the model , see ‘Materials and methods’ . Our fits minimize the score as a function of the GRF parameters . To illustrate and test our method , we focused on the CLB2 cluster , which comprises a set of genes that are expressed during the G2/M phase of the cell cycle ( Spellman et al . , 1998 ) . We considered the input factors Fkh2 , Ndd1 , and Fkh1 . Fkh2 binds to a consensus regulatory DNA element as a complex with Mcm1 ( Spellman et al . , 1998; Zhu et al . , 2000 ) . The Fkh2-Mcm1-DNA complex recruits the coactivator Ndd1 ( Koranda et al . , 2000; Darieva et al . , 2003 ) . Fkh1 plays a complementary and partially redundant role ( Kumar et al . , 2000; Hollenhorst et al . , 2000; Hollenhorst et al . , 2001 ) . Because Mcm1 is not significantly periodically transcribed ( Figure 2—figure supplement 1 ) , we treat only Fkh2 , Ndd1 , and Fkh1 as significant regulatory inputs . Figure 2A and B show the results of our analysis for two target genes in the CLB2 cluster , Hof1 and Mob1 , which are not regulated by Fkh1 ( Harbison et al . , 2004; Tuch et al . , 2008; Zhu et al . , 2000 ) . In each case , the promoter structure is shown in the top right . TF binding sites and consensus motifs of the CLB2 cluster regulatory element are indicated as obtained by motif search in the upstream sequences of each target gene ( see ‘Materials and methods’ for details ) . The mRNA level time series for the input factors Fkh2 and Ndd1 are shown in the top left of each panel , with the inferred proxies p ( t ) for the corresponding protein levels shown below . Here , the p ( t ) time series for the same input TFs have been independently determined from the fits to the two target gene output time series . As a consequence , the resulting p ( t ) time series for the same TF can deviate from each other . This illustrates the uncertainty in the inference of the parameter λ from the output of an individual target gene . However , if several target genes of the same TF are known , then a mutually consistent estimate of λ can be determined . Since all candidate TFs for the proposed transcriptional cell cycle oscillator that we study below have multiple target genes , we describe a method for the unified inference of λ in the next section . 10 . 7554/eLife . 12188 . 004Figure 2 . Reconstructed regulation functions of two genes in the CLB2 cluster . For both target genes the measured TF expression level , the inferred protein level , the estimated regulation function with corresponding measured data points and the fit to the target synthesis rate are shown . Additionally predicted binding sites for relevant TFs within the target gene promoter regions are depicted . Both target genes , Hof1 and Mob1 , are regulated by Fkh2 and Ndd1 ( which is recruited to the complex Mcm1-Fkh2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 00410 . 7554/eLife . 12188 . 005Figure 2—figure supplement 1 . Non-periodic cell cycle regulators . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 005 The estimated GRFs in Figure 2 are depicted as surface plots , where the mesh represents the best-fit mathematical function within our set of GRF types and the discrete points show which parts of the GRF are sampled by the experimental data . The corresponding time series of the output promoter activity is shown in the bottom right of each panel , including both the model output ( continuous line ) and the experimental data ( points ) . As can be seen from these plots , the inferred GRFs capture the expression dynamics of the target genes relatively well ( best fit scores , i . e . the remaining fraction of the variance in the data not explained by the model , are indicated in the figure ) . In both cases , the estimation procedure yields an AND-like logic for the action of the input TFs Fkh2 and Ndd1 , such that full activation can be obtained only in the presence of both inputs . For the target gene Hof1 , however , the data points sample only a narrow region in the two-dimensional Ndd1-Fkh1 concentration space of the GRF , such that the shown two-dimensional GRF is an extrapolation that cannot be verified with this data set . This example illustrates a useful property of our GRF inference scheme: the sampling density in the input space of the GRF already indicates the range over which the GRF inference is supported by the data . The CLB2 cluster target gene Kip2 provides an example for a complex GRF inference task , since it has Fkh1 as regulatory input in addition to Fkh2 and Ndd1 ( Harbison et al . , 2004 ) . Given the combinatorial complexity of 3-input GRFs and the limited amount of expression data , we cannot simply extend our approach to simultaneously treat more than two inputs . Instead , it is necessary to limit our inference to the two most significant inputs . We used Kip2 as a test case for a method to identify the most significant inputs . As a reference , we first constructed a thermodynamic regulation model ( Bintu et al . , 2005 ) for Kip2 based on its promoter structure ( Figure 3A ) . This physico-chemical model is more detailed and requires more parameters than the Hill-type functions . It parameterizes the GRF by the TF affinities to their DNA binding sites , the interactions between TFs bound to proximal or overlapping sites , and the attraction of bound TF complexes towards RNA polymerase , see Figure 3A ( top right ) . We estimated the parameter values from the data with the same method as used above ( see ‘Materials and methods’ ) . The resulting 3-input GRF yielded a good description of the output ( Figure 3A , bottom right ) and predicted that the input from Fkh2 has only a minor effect . 10 . 7554/eLife . 12188 . 006Figure 3 . Comparison of the effective regulation function to a detailed promoter model . Target gene Kip2 within the CLB2 cluster is predominantly regulated by three input factors Fkh1 , Fkh2 and Ndd1 . ( A ) Results of fitting a detailed promoter model , using all three input factors . The regulation model skip2m is derived from the promoter structure which contains overlapping binding sites for Mcm1-Fkh2 and Fkh1 , using a thermodynamic approach as detailed in Bintu et al . ( 2005 ) . The displayed projected regulation function has been plotted with Fkh2 fixed at its average value which leads to only minor deviations from the full model . ( B ) Results for fitting an effective model to Kip2 . All three combinations of two out of three input factors have been tested with Fkh1/Ndd1 yielding the best score . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 00610 . 7554/eLife . 12188 . 007Figure 3—figure supplement 1 . Validation of our inferred GRFs with an independent dataset . All inferred GRFs presented in this article have been tested on the microarray dataset previously published in Pramila et al . ( 2006 ) . To match the scale of the data the GRFs have been inferred from , all TF data have been linearly transformed such that their minimum and maximum match those of the corresponding data in our dataset ( see ‘Materials and methods’ for details ) . Shown are the reversely transformed model output ( solid line ) with the target gene expression level of the test dataset ( stars ) . A comparison of the points in protein space which are used for the output in our dataset and the test dataset is shown in Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 00710 . 7554/eLife . 12188 . 008Figure 3—figure supplement 2 . Comparison of protein space coverage . Shown are the values of ( inferred ) protein concentration , which are used to generate the GRF outputs from our dataset ( black circles ) and from the test dataset ( red triangles ) . The points in protein space generated from the test dataset are used for the model output shown in Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 008 To test whether our general inference method would yield the same conclusion without constructing a physico-chemical promoter model , we applied our method for 2-input GRFs to each combination of two input factors out of Fkh1 , Fkh2 , and Ndd1 . Out of these combinations an AND-like GRF with Fkh1 and Ndd1 as the input obtained the best score ( Figure 3B ) , consistent with the prediction that Fkh2 has the least significant effect . Notably , the obtained fit to the time-dependent output is almost identical to the one obtained with the thermodynamic model ( bottom right panel of Figure 3A and B , respectively ) , despite the smaller number of fitted parameters in the general inference method ( 8 instead of 12 parameters ) . When the GRF of the thermodynamic model is plotted as a function of Fkh1 and Ndd1 with Fkh2 fixed at its average value ( Figure 3A , center ) the resulting projection is also very similar to the GRF of Figure 3B . Our analysis is therefore consistent with the interpretation of Ndd1 as the limiting factor in the formation of the Mcm1-Fkh2-Ndd1 DNA-bound complex in the Kip2 promoter during the cell cycle . The examples of Figures 2 and 3 illustrate that our method to infer GRFs not only recapitulates known roles of TFs on their target genes , but also provides additional quantitative insight into the individual or combinatorial effects of TFs . Our analysis of Kip2 regulation suggests that the two most significant regulatory inputs to a gene can be identified by applying the inference method for 2-input GRFs to each combination of two input factors and choosing the combination that yields the best score . In general , the relative significance of the different regulatory inputs to a gene will depend on the physiological conditions , and the method can only infer a relative significance for the conditions of the experiment . So far our analysis was limited to our calibration dataset ( Eser et al . , 2014 ) from which we infer our GRFs . To probe the consistency of our GRFs with independent data , and to test their ability to predict regulatory effects outside the regime of their calibration , we considered another published dataset that was measured under different , albeit similar , experimental conditions ( Pramila et al . , 2006 ) . We asked whether the same GRFs that we inferred from the data of Eser et al . ( 2014 ) would also be able to describe the mRNA dynamics of this dataset , which we had not used to calibrate our GRFs . Pramila et al . ( 2006 ) provide microarray data from α-Factor synchronized yeast cells , which follows transcript levels over two cell cycles at 5 min intervals . We used rescaled TF expression data from this test dataset ( see ‘Materials and methods’ ) as input for the GRFs inferred from our data . We then compared the output of our GRFs with the mRNA time series of the target genes in the test data . For this comparison , we used the GRFs and effective protein half-lives that we inferred for our analysis of transcriptional cell cycle oscillators presented in the next section . Hence , the output curves are predictions without fit parameters , and the good agreement achieved in the comparison with the test data ( Figure 3—figure supplement 1 ) suggests that our GRFs have predictive power beyond the regime of their calibration . Figure 3—figure supplement 2 shows that the regime of input TF concentrations that is effectively sampled by the test data is similar but not identical to the range in our calibration data set . It would also be desirable to be able to make in silico predictions of mutant behaviors with the inferred GRFs . For instance , a TF knockout could be emulated by setting the input from that TF to be zero for the GRFs of its target genes . However , other inputs of the target genes may also be affected by the knockout , leading to indirect effects on the expression pattern . Therefore , prediction of mutant behaviors will generally require a complete model of the genetic module in which that target gene is embedded . We discuss the inference of genetic modules in the next section , and return to the question of mutant behaviors further below . We next applied our method to multi-gene systems . As an illustration we wanted to test whether the data of Eser et al . ( 2014 ) is consistent with a proposed transcriptional cell cycle oscillator ( Haase and Reed , 1999; Orlando et al . , 2008; Simmons Kovacs et al . , 2012 ) . While the cell-cycle dependent oscillation in the expression of key cell-cycle TFs is clearly established , the underlying regulatory network is not comprehensively characterized . Furthermore , it is debated whether these TFs form a genuine transcriptional oscillator that can create transcriptional oscillations even without a functional protein-level cyclin oscillator ( Bristow et al . , 2014; Rahi et al . , 2016 ) . By applying our inference method to the dynamic transcriptome data of synchronized wild-type cells , we can test whether the oscillatory transcription profiles of a selected set of TFs form a consistent dynamic system of mutually regulating genes . To that end , we first pursued an autonomous transcriptional oscillator but included coupling to the cyclin-CDK oscillator where it turned out to be necessary for consistency with our data . Rather than basing our analysis on specific network architectures ( Simon et al . , 2001; Simmons Kovacs et al . , 2012 ) , we decided to systematically generate and rank all plausible networks according to their global consistency with the DTA data . The challenge of such an unbiased approach is the combinatorial explosion of the number of core networks that can be generated even from small numbers of TFs when each of them has several possible regulatory interactions . Due to this combinatorial explosion , any attempt to first construct all possible networks and then fit each of them to the dynamic transciptome profiles of all involved genes would necessarily fail . We therefore took a stepwise approach , in which we pre-fitted all possible combinations of at most two inputs to each node of the network , allowing us to then construct and evaluate all possible networks in a combinatorial scheme . As illustrated in Figure 4 , our complete approach involves four steps: ( A ) selection of candidate TFs , ( B ) construction of their interaction matrix , ( C ) inference of a unified protein proxy for each TF in the set using different target genes , and ( D ) inference of all possible GRFs and combination into candidate networks , which are ranked according to their compatibility with the DTA data . The final step overcomes the combinatorial complexity by exploiting the modularity of the problem . In the following , we briefly provide essential information for each of the steps; additional details are provided in ‘Materials and methods’ . 10 . 7554/eLife . 12188 . 009Figure 4 . Finding a regulatory model for the transcriptional cell cycle oscillator . ( A ) A set of potentially constituent TFs is selected . We require that these genes are ( i ) periodically expressed , ( ii ) regulate at least one other gene within the set and ( iii ) are regulated by at least one other gene within the set . ( B ) Interaction table of the resulting set of genes ( listed on the axis ) , as annotated in the literature . A green square indicates that the corresponding gene of that row regulates the respective gene on the column . ( C ) Before fitting regulation functions to target genes we pre-determine the effective protein degradation rate for each TF . To that end we first select a set of target genes , which are ( i ) periodically expressed , ( ii ) are well fitted by the TF and ( iii ) if there are other TFs regulating the target gene , they must be non-periodic so that they cannot contribute in explaining the expression pattern of the target gene . We then use MCMC to sample from the posterior distribution of the fitting parameters of each target gene and create histograms for the marginal distributions over the protein parameter . The consensus is fixed to the highest peak of the product histogram . ( D ) To each gene in the set of potentially constituent gene regulation models are fitted for all single inputs and all combinations of two inputs . To each fit we calculate a normalized score as the averaged squared residuals of the fit , divided by the data variance of the target synthesis rate . These 'regulatory nodes' are combined to networks , which we require to be strongly connected ( as illustrated on the bottom ) . The resulting set of possible network models are scored by the average normalized fitting score of the nodes and ranked . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 00910 . 7554/eLife . 12188 . 010Figure 4—source data 1 . Table of interactions between cell cycle TFs with references . If evidence exists that the gene heading a row transcriptionally regulates the gene heading a column , the corresponding references are listed in the respective field . A '0' indicates that no direct evidence for a transcriptional regulation exists . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 01010 . 7554/eLife . 12188 . 011Figure 4—source data 2 . Data for protein model inference . This table contains the inferred protein model parameter λ for each TF in the cell cycle oscillator model and both replicated . Additionally listed are the sets of target genes on which the inferences were performed . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 01110 . 7554/eLife . 12188 . 012Figure 4—source data 3 . List of inferred regulation directions of genes in the trancriptional cell cycle oscillatorDirection of regulation ( activating or repressing ) and , if available , references for each best fitted regulation function in the transcriptional cell cycle oscillator model . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 01210 . 7554/eLife . 12188 . 013Figure 4—figure supplement 1 . Fitted GRF for Hcm1 . Depicted for both replicates are transcript levels of the TFs , the inferred protein levels , the inferred GRF and a comparison of model output and data for the target gene activity . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 01310 . 7554/eLife . 12188 . 014Figure 4—figure supplement 2 . Fitted GRF for Yhp1 . Depicted for both replicates are transcript levels of the TFs , the inferred protein levels , the inferred GRF and a comparison of model output and data for the target gene activity . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 01410 . 7554/eLife . 12188 . 015Figure 4—figure supplement 3 . Fitted GRF for Swi4 . Depicted for both replicates are transcript levels of the TFs , the inferred protein levels , the inferred GRF and a comparison of model output and data for the target gene activity . This GRF will be revised below . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 01510 . 7554/eLife . 12188 . 016Figure 4—figure supplement 4 . Fitted GRF for Fkh2 . Depicted for both replicates are transcript levels of the TFs , the inferred protein levels , the inferred GRF and a comparison of model output and data for the target gene activity . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 01610 . 7554/eLife . 12188 . 017Figure 4—figure supplement 5 . Fitted GRF for Fkh1 . Depicted for both replicates are transcript levels of the TFs , the inferred protein levels , the inferred GRF and a comparison of model output and data for the target gene activity . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 01710 . 7554/eLife . 12188 . 018Figure 4—figure supplement 6 . Fitted GRF for Yox1 . Depicted for both replicates are transcript levels of the TFs , the inferred protein levels , the inferred GRF and a comparison of model output and data for the target gene activity . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 01810 . 7554/eLife . 12188 . 019Figure 4—figure supplement 7 . Fitted GRF for Fhl1 . Depicted for both replicates are transcript levels of the TFs , the inferred protein levels , the inferred GRF and a comparison of model output and data for the target gene activity . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 01910 . 7554/eLife . 12188 . 020Figure 4—figure supplement 8 . Fitted GRF for Ndd1 . Depicted for both replicates are transcript levels of the TFs , the inferred protein levels , the inferred GRF and a comparison of model output and data for the target gene activity . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 020 For the selection of candidate TFs , we imposed the following criteria ( Figure 4A ) : ( i ) distinct periodic expression in the DTA data , ( ii ) control by at least one other TF within the set , and ( iii ) at least one regulatory target within the set . The criteria ( ii ) and ( iii ) are necessary to obtain closed ( strongly connected ) networks . We focused on closed networks in order to identify possible core motifs for an autonomous transcriptional oscillator . However , we will see below that a consistent interpretation of the data requires additional input from the cyclin-CDK oscillator . To quantify the degree of periodicity of gene expression profiles , we relied on a previously established method ( Eser et al . , 2014 ) . We considered the top 500 periodic genes , out of which 20 were TFs . We then used the direct interactions between these genes documented in the YEASTRACT database ( Teixeira et al . , 2006 ) to apply our criteria ( ii ) and ( iii ) . This resulted in the following set of 11 TFs ( see ‘Materials and methods’ for details ) : Hcm1 , Swi5 , Ace2 , Yhp1 , Swi4 , Fkh1 , Fkh2 , Ash1 , Yox1 , Fhl1 , and Ndd1 . The factors Swi5 and Ace2 are known to be homologous ( Doolin et al . , 2001 ) and display synchronous expression in the DTA data set . We therefore considered them as a single network node for the purpose of our study . Since Swi5 already has all the inputs that Ace2 has within our set , the combined node is simply referred to as Swi5 in the following . Figure 4B shows the interaction matrix for the remaining set of 10 candidate network nodes . This matrix includes the direct interactions from the YEASTRACT database , which are experimentally confirmed TF-TF interactions ( see Figure 4—source data 1 for the original reference for each interaction ) . Additionally , we included regulation of the Swi4 gene by Yox1 and Yhp1 , which can bind to its ECB promoter element ( Mai and Miles , 2002; Pramila et al . , 2002; Darieva et al . , 2010 ) . We next estimated the effective protein degradation rate λ for all TFs within our set . We exploited the fact that each of these TFs also has multiple target genes outside of the set . Since the parameter λ is a property of the TF and not of its targets , the value of λ should be consistent between all targets . We therefore devised a method that infers a unified proxy for λ from a set of target genes . The method is illustrated in Figure 4C and detailed in ‘Materials and methods’ . Briefly , for each TF in the set we selected target genes , which have a significant regulatory input from the TF over the cell cycle . We then used Markov Chain Monte Carlo sampling to estimate posterior distributions for the parameter λ at each target gene , and combined the results into a unified estimate ( Figure 4—source data 2 ) . The corresponding effective protein half-lives τ1/2=ln ( 2 ) /λ range from 5 min ( Ash1 ) to 50 min ( Yox1 ) and provide the basis for our unified proxy p ( t ) for the protein dynamics of each TF . In the crucial final step , we constructed multiple candidate GRFs for each gene in our set and combined them into a large number of candidate network models ( Figure 4D ) . For each gene , we fitted GRFs with all possible regulations involving either one or two inputs chosen from the interaction matrix of Figure 4B . Each fit has an associated score normalized such that it measures the fraction of data variance not described by the fitted model ( see ‘Materials and methods’ ) . This assures a meaningful relative weighting of the individual nodes in the combined network scores used below . Two examples for GRFs of best-scoring regulations are shown in Figure 5 , the regulation of Ash1 by Swi5 and the regulation of Swi5 by the two inputs Fkh1 and Ndd1; see Figure 4—figure supplements 1–8 for the remaining best-scoring GRFs . Importantly , all best-scoring GRFs correctly predicted whether a TF is activating or repressing , wherever experimental evidence exists ( see Figure 4—source data 3; Chua et al . , 2006; Di Talia et al . , 2009 ) . Furthermore , the shape of the GRFs obtained from the two biological replicates in the DTA data set were reasonably consistent , as also illustrated by the examples in Figure 5 . 10 . 7554/eLife . 12188 . 021Figure 5 . Results of node fitting in the cell cycle network . ( A ) Regulation diagram of our best ranking model for a transcriptional cell cycle oscillator . Arrows represent positive ( pointing clockwise ) and negative ( pointing counter-clockwise ) interactions and are color-coded with respect to their corresponding TF . Only the regulatory interactions indicated by solid lines resulted from our network inference method , while the additional interactions indicated by dashed lines were introduced in the subsequent manual analysis ( see main text ) . ( B ) Example of a regulation model with two input factors: Swi5 is regulated by Fkh1 and Ndd1 by an analog XOR-like function . ( C ) Example of a one input factor regulation model: Ash1 is activated by Swi5 . Depicted in ( B ) and ( C ) are , for both replicates respectively , data of the TF ( s ) total mRNA expression level , the inferred TF protein time series , the fitted regulation function and the resulting target synthesis rate . Data points of the target synthesis rate are displayed as stars or colored spheres . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 02110 . 7554/eLife . 12188 . 022Figure 5—figure supplement 1 . Network scores and best ranking networks . ( A ) Histograms of network scores in replicate datasets 1 and 2 , respectively . ( B–F ) The five best ranking networks ( rank 1–5 ) . ( B ) best ranking network . ( C ) Rank 2 . In comparison to ( B ) Swi5 is regulated only by Fkh1 . ( C ) Rank 3 . In comparison to ( B ) Ndd1 is regulated only by Hcm1 . ( E ) Rank 4 . In comparison to ( B ) Fkh2 is part of the network , regulating Ndd1 together with Hcm1 . ( F ) Rank 5 . In comparison to ( B ) Yox1 is regulated only by Fhl1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 02210 . 7554/eLife . 12188 . 023Figure 5—figure supplement 2 . The GRF for Swi4 is an outlier . Histograms for the scores of all possible GRFs for each TF selected for the cell cycle oscillator . Marked in red are the scores of the respective best ranking GRF for each TF . The outlier in both replicates ( marked with star ) represents the GRF of Swi4 , scoring considerably worse than the other best ranking GRFs . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 02310 . 7554/eLife . 12188 . 024Figure 5—figure supplement 3 . Current biological model and the effective model for regulation of Swi4 . Also depicted are references with experimental evidence for the corresponding interaction . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 02410 . 7554/eLife . 12188 . 025Figure 5—figure supplement 4 . GRF for Swi4 with cyclin input . Aside from repressors Yhp1 and Yox1 , Swi4 is also auto-activated by SBF ( Swi4-Swi6 ) , which in turn is ( indirectly ) activated by Cln1 and Cln2 . We model this regulation by additional input from Swi4 and Cln2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 02510 . 7554/eLife . 12188 . 026Figure 5—figure supplement 5 . Concerted regulation by both Swi4 and Ash1 are necessary to model the expression pattern of Yhp1 . ( A ) shows a fit of the GRF for Yhp1 with Swi4 as only input factor , while ( B ) shows a fit with Ash1 as only input factor . Both explain less than 25% of the data variance in both replicates . ( C ) Combined regulation of Yhp1 by Swi4 and Ash1 yields a satisfactory fit . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 02610 . 7554/eLife . 12188 . 027Figure 5—figure supplement 6 . Concerted regulation by both Hcm1 and Ndd1 are necessary to model the expression pattern of Fkh1 . ( A ) shows a fit of the GRF for Fkh1 with Hcm1 as only input factor , while ( B ) shows a fit with Ndd1 as only input factor . ( C ) Combined regulation of Fkh1 by Hcm1 and Ndd1 yields superior fits to single inputs . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 027 In principle , 116 , 872 , 448 different gene networks can be combinatorially constructed from our 10 nodes using the different regulations that we allow for each node . However , out of these only 952 , 100 satisfy our criterion of being strongly connected . Their score distribution is shown in Figure 5—figure supplement 1A . For the construction of candidate networks , we do not only allow the best-scoring regulations , but also the suboptimal ones , since due to the constraints the globally optimal network does not necessarily consist only of optimal nodes . The globally optimal genetic network depicted in Figure 5A has a strong overall similarity to transcriptional oscillator networks that were previously proposed . It contains all of our candidate nodes except Fkh2 . All nodes with the exception of Ash1 have two regulatory inputs . To confirm that multiple inputs are indeed necessary to explain the expression pattern of these genes , we compared the best fits with inputs from only a single TF with the respective combinatorial regulation . Figure 5—figure supplement 5 illustrates this for the case of Yhp1 regulation by Swi4 and Ash1 , while Figure 5—figuer supplement 6 shows analogous plots for Fkh1 regulation by Hcm1 and Ndd1 . 10 . 7554/eLife . 12188 . 028Figure 6 . Simulation results of the reconstructed transcriptional cell cycle oscillator . Simulated mRNA expression levels for all genes in the network are shown in arbitrary units and rescaled to equal sizes . For the first 205 min ( shaded grey ) Swi4 receives periodic input from Cln2 using measured expression data . After 205 min the input from Cln2 is set to a constant average value to simulate loss of cyclin activity . ( A ) The network inferred from replicate dataset 1 recovers sustained oscillations . ( B ) Oscillations in the network inferred from replicate dataset 2 dampen over time without periodic input . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 02810 . 7554/eLife . 12188 . 029Figure 6—figure supplement 1 . Test of predictions for mutant strains by model simulations . ( A ) ( Bean et al . , 2005 ) reported that in a Swi4 deletion mutant the Swi4 target gene Yox1 shows heavily reduced expression and a weak oscillation with a period shorter than the cell cycle . This oscillation might results from secondary feedback loops in the transcriptional cell cycle network . We reproduced this behavior within our ODE model of the transcriptional cell cycle oscillator by setting Swi4 expression to zero . The resulting output for Yox1 shows the expected fast oscillations . ( B ) It has been shown that in a Δyhp1Δyox1 double mutant Swi4 expression exhibits a delayed and longer peak , reaching into S and G2 phase ( Pramila et al . , 2002 ) . We could partially recapture this qualitative behavior with our ODE model of the transcriptional cell cycle oscillator by setting Yox1 and Yhp1 expression to zero . The resulting model output shows a prolonged peak duration , albeit a delayed onset that is not observed in the data . ( C ) In the same mutant strain as in ( A ) Rnr1 , a target gene of Swi4 and Mbp1 , was reported to have delayed peak time with an increases amplitude ( Bean et al . , 2005 ) . This behavior cannot be reproduced within our model , because Mbp1 shows no periodic transcriptional activity in our data ( see Figure 2—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 02910 . 7554/eLife . 12188 . 030Figure 6—figure supplement 2 . Relative change of model parameter after re-fitting the full ODE model compared to the individual fits . To minimize propagation of errors at the individual nodes , the parameters of the GRFs optimized again for the network score of the ODE model , starting from their original value ( see ‘Materials and methods’ ) . The relative change of the majority of parameters is well below 10% . DOI: http://dx . doi . org/10 . 7554/eLife . 12188 . 030 Figure 5—figure supplement 1 compares our five best-scoring networks . The networks on rank 2 , 3 , and 5 have the same nodes as our best-ranking network . They differ only by the absence of one regulatory input at a single gene , supporting the best-ranking network as a consensus . The network on rank 4 features Fkh2 as additional node , and differs in several regulatory interactions , making it incompatible with our consensus network . Taken together , these results illustrate that our inference method extracts useful regulatory information about genetic modules from DTA data . However , we will see in the next section that further manual analysis of the output generated by the inference method can considerably increase the biological insight . By design the gene regulatory network obtained in the previous section does not include regulation by the 'external' cyclin-CDK oscillator . However , further analysis suggests that a completely autonomous network is incompatible with our data . Close inspection of the fitting results reveals that the expression dynamics of the Swi4 gene is not adequately explained even by the best-scoring GRF . Indeed , the Swi4 node already stands out in the score statistics ( Figure 5—figure supplement 2 ) – whereas the best-scoring GRFs for the other nodes leave only between 14% and 38% of the variation in the data unexplained , 70% and 52% unexplained variance remain for Swi4 in replicates 1 and 2 , respectively . The best-scoring regulation for Swi4 is by Yhp1 and Yox1 , consistent with the known Swi4 regulation by Yhp1 and Yox1 during the G1 phase ( McInerny et al . , 1997; Pramila et al . , 2002 ) . However , the clear difference between the shape of the output profile of Swi4 and its best fit by the Yhp1 and Yox1 inputs ( Figure 4—figure supplement 3 ) suggests that additional regulatory inputs are significantly affecting the transcription rate of the Swi4 gene . The molecular role of Swi4 in cell cycle regulation is well studied . Swi4 is the DNA binding component of the SBF complex , which activates Cln1/Cln2 ( Koch et al . , 1996 ) and late G1 genes , thereby promoting progression from G1 to S phase ( Nasmyth and Dirick , 1991 ) . Further , Swi4 activates its own expression via SBF ( Iyer et al . , 2001 ) . SBF activity is inhibited in early G1 by the repressor Whi5 ( de Bruin et al . , 2004; Costanzo et al . , 2004 ) , which in turn is exported from the nucleus at cell cycle START , promoted by G1 cyclins Cln1-3 and Cdc28 ( Wijnen et al . , 2002; de Bruin et al . , 2004; Costanzo et al . , 2004 ) . Commitment to START is primarily achieved by a positive feedback loop between SBF and Cln1/Cln2 to rapidly exclude the SBF repressor Whi5 from the nucleus ( Bean et al . , 2006; Skotheim et al . , 2008 ) . For the purpose of our study , we modeled this complex set of interactions by a reduced regulation scheme , as illustrated in Figure 5—figure supplement 3 . Following Skotheim et al . ( 2008 ) , we considered Cln1/2 to be the main control of Whi5 nuclear export . This led us to an effective scheme where Swi4 has Cln1/2 and Swi4 itself as additional inputs ( indicated by the dashed links in Figure 5A ) . With these added inputs and a matching GRF ( see ‘Materials and methods’ ) , the Swi4 output profile was equally well described as the other nodes of the network ( Figure 5—figure supplement 4 ) . So far we considered the dynamics at each node separately from the other nodes , and it remains to be shown that our entire network model as a dynamical system generates oscillations compatible with the dynamic transcriptome data . We therefore used the inferred GRFs and effective protein half-lives to construct a complete set of differential equations for the mRNA expression levels and protein concentrations in our network , see ‘Materials and methods’ . We integrated these equations to simulate the dynamics of the full network starting from the measured mRNA levels and inferred protein concentrations at t=0 as initial values . In addition , we fed the measured dynamics of Cln2 as an external input into our network acting on the Swi4 node . Figure 6 shows the obtained model trajectories for the mRNA expression levels of all nodes , together with the corresponding experimental data . For both replicates , the model adequately captures the behavior of the dynamic transcriptome data . This agreement prompted us to test if our simulation model could also qualitatively predict the behavior of mutant strains . Towards this end , we considered expression data from two studies with TF knockout strains ( Bean et al . , 2005; Pramila et al . , 2002 ) and compared it to the behavior of our simulation model with the expression levels of the mutated genes set to zero . Bean et al . ( 2005 ) measured the effects of deleting Swi4 using the cdc20 block-release protocol for cell-cycle synchronization . The wild-type mRNA expression time series plotted in Figure 1 , of Bean et al . ( 2005 ) qualitatively resembles the corresponding data of Eser et al . ( 2014 ) after taking into account an apparent time-shift , which is likely due to the different protocol for cell-cycle synchronization . In their Swi4 deletion strain , Bean et al . ( 2005 ) found a strongly reduced expression of Yox1 with a weak rapid oscillation ( period ∼45 min ) . This phenotype is recapitulated by our model prediction , see Figure 6—figure supplement 1A . In our simulation , the rapid oscillations are caused by indirect effects via Fhl1 , the second input of Yox1 . A second case is shown in panel B of the same figure , where the expression of the Swi4 gene is considered in a Δyox1Δyhp1 double mutant background , as studied experimentally by Pramila et al . ( 2002 ) . The microarray data exhibits a delayed and prolonged peak expression , which reaches into S and G2 phase ( Pramila et al . , 2002 ) . Our model qualitatively reproduces this behavior when Yox1 and Yhp1 expression is set to zero . It should be noted , however , that the effect is significantly more pronounced in the model than in the data , suggesting that our model for S . cerevisiae transcriptional cell cycle oscillations is missing a mechanism that buffers against the effects of the Δyox1Δyhp1 double mutation . While the qualitative agreement obtained in Figure 6—figure supplement 1A and B suggests that our inferred model indeed captures important aspects of the transcriptional regulation of cell cycle genes , it is clear that it will fail to predict the effect of deletions that unmask post-transcriptional effects . This is illustrated in Figure 6—figure supplement 1C for the case of Rnr1 , a target gene of the transcription factors Swi4 and Mbp1 . While the periodic expression of Swi4 can explain the cell cycle dependent activity of the SBF complex ( Swi4-Swi6 ) , Mbp1 is not periodically transcribed , and the activity of the MBF complex ( Mbp1-Swi6 ) is likely cell cycle dependently modulated by cyclin-dependent posttranscriptional regulation ( de Bruin et al . , 2008 ) . Accordingly , our method infers a GRF based almost entirely on regulatory input from Swi4 , and predicts that the transcription rate of Rnr1 is essentially constant in a Swi4 mutant strain , as shown in Figure 6—figure supplement 1C . In contrast , the Swi4 deletion strain of Bean et al . ( 2005 ) exhibits an oscillatory expression of Rnr1 , with a delayed peak time and an increased amplitude . As shown by Bean et al . ( 2005 ) , Rnr1 is in fact redundantly regulated by SBF and MBF , such that only the swi4-mbp1 double mutant displays an Rnr1 expression that is essentially cell-cycle-independent . Finally , we returned to the global behavior of the transcriptional oscillator . Given that Cln2 already provides an oscillatory input signal to our simulation model , we can conclude only that the network can be driven towards transcriptional oscillations with the correct amplitudes , shapes , and relative phases in all of its nodes . Can it also oscillate autonomously , without external driving ? To test this , we continued to run the network dynamics simulation beyond the last measured timepoint ( t=205 min ) with the Cln2 input set to zero , mimicking the loss of cell cycle dependent cyclin activity . We found that the network has the intrinsic ability to oscillate . After an initial disturbance due to the abrupt disappearance of the Cln2 input , regular oscillations were recovered in most genes for replicate 1 ( Figure 6A ) , while dampened oscillations were obtained for replicate 2 ( Figure 6B ) . Hence , depending on the parameter values , which were separately adjusted to describe each data set ( see ‘Materials and methods’ ) , the network model is able to produce either regular or dampened oscillations . However , one node of the network appears to require the Cln2 input for oscillation: for both replicates Ash1 oscillations cease almost immediately after the Cln2 input is taken away . Interestingly , Ash1 has also been reported to be non-oscillating in some cyclin mutant strains ( Orlando et al . , 2008 ) , suggesting that its ability to oscillate is particularly sensitive with respect to perturbations of the cyclin-CDK oscillator . While most genes still oscillate , the oscillation period is shortened by ∼20% in the absence of the Cln2 input , underlining the importance of cyclin-CDK activity for the timing of gene expression . Despite this change in the absolute period , the relative ordering of the genes in terms of the phase of their oscillations remains the same as with the Cln2 input . Taken together , we arrived at a model for a transcriptional cell cycle oscillator that ( i ) is quantitatively consistent with the dynamic transcriptome data for synchronized wild-type cells , and ( ii ) can produce ( dampened ) oscillations without cyclin input , albeit not with the native properties , supporting the notion that cyclin action is required for normal oscillatory behavior ( Simmons Kovacs et al . , 2012; Rahi et al . , 2016 ) .
We established a method to infer gene regulation functions ( GRFs ) from the intrinsic cellular dynamics of the transcriptome . GRFs are key quantities for the modeling of transcription regulation and provide a basis for the quantitative analysis of functional genetic modules . Inference of GRFs is necessary , since their direct measurement is notoriously difficult while indirect measurements are limited to specific TFs ( Setty et al . , 2003 ) and confounded by extraneous factors ( Kuhlman et al . , 2007 ) . We demonstrated , for the first time , the inference of individual GRFs as well as a functional genetic module from dynamic transcriptome data , which in our example was obtained from wild-type synchronized yeast cells . Our inferred GRFs agreed well between biological replicates , were able to capture the expression dynamics of the target genes also in an independent test dataset , and correctly predicted whether the effect of a TF on its target is activating or repressing wherever experimental evidence was available . The method identifies the two inputs of a gene that are most significant for the description of the experimental data set . We illustrated the consistency of this reduced description with a more detailed physico-chemical model that takes all known inputs into account . The amount of data needed to infer GRFs rises strongly ( exponentially ) with the number of regulatory inputs , due to the combinatorial complexity of the task . Given the data that is currently available , our inference of GRFs with two inputs is at the limit of what is possible in a systematic and unbiased way . Clearly , GRFs could be extracted more directly and reliably if the time-dependent protein levels of all input TFs were also measured , as well as their activity and nuclear localization . However , we showed that one can obtain surprisingly consistent GRFs from high-accuracy dynamic transcriptome data alone , which simultaneously provides the transcript levels of the inputs and the mRNA synthesis rates of their targets . These GRFs are not based on detailed mechanistic models of the underlying molecular processes , but correspond to effective regulation models which subsume many processes , including the transport of the TFs to and from the nucleus and possibly activation of the TFs , e . g . by phosphorylation . The ability of our method to infer GRFs from dynamic transcriptome data depends on two general conditions: First , prior knowledge about which TFs potentially regulate which targets is required , since the information contained in the time series does not suffice to discriminate the correct regulatory interactions from all possible wrong associations . For the examples that we considered , the prior information provided by the YEASTRACT database proved sufficient . This database lists more input TFs than are relevant under the conditions that we study . However , since the number of irrelevant interactions is small ( compared to the number of all possible inputs ) , our method is able to identify the most relevant ones . Second , the information contained in the data must be sufficiently redundant to permit self-consistency conditions to be imposed during the inference procedure . We made use of redundant information derived from the length of the time series ( the transcriptome data covers three cell cycles ) and the pleiotropy of the cell cycle transcription factors ( which must produce consistent effects in multiple targets ) . Self-consistency conditions are essential for our method to circumvent the need for protein data . One of the most promising applications of GRF inference is the quantitative analysis of functional genetic modules . Modules composed of interacting molecules and genes are widely considered central for the understanding of cellular functions ( Hartwell et al . , 1999 ) . While the topologies of possible genetic modules can be generated from known regulatory interactions , any attempt to quantitatively test the compatibility of such modules with expression data will have to infer the GRFs of the modules nodes . We introduced and illustrated an unbiased approach to this task . Instead of testing a small set of preselected candidate network modules against the dynamic transcriptome data , we systematically assessed close to a million candidates for the core module of the yeast transcriptional cell cycle network . Our best ranking network has several features in common with previously proposed transcriptional cell cycle networks ( Simon et al . , 2001; Lee et al . , 2002; Orlando et al . , 2008; Simmons Kovacs et al . , 2012 ) . Compared to the latest proposal , which was constructed by hand from genes that remained periodic in a cdc28 mutant ( Simmons Kovacs et al . , 2012 ) , our best ranking network contains the same genes , except that it adds the nodes Yox1 and Fkh1 but does not explicitly consider Mcm1 as significant regulatory input ( although Mcm1 certainly plays an important mechanistic role ) . Our overall network topology is also similar , with sequential forward activation and backward inhibition as dominant motifs , consistent with previous studies ( Simon et al . , 2001; Orlando et al . , 2008; Simmons Kovacs et al . , 2012 ) . Beyond recapitulating prior results , our unbiased inference method has led us to a mathematical model that is consistent with the DTA data and capable of generating oscillations , while also predicting a coupling to the cyclin-CDK oscillator . The mathematical models obtained with our inference approach are valuable for further quantitative analysis . As an illustration , we tested to what extent our core module could generate oscillations without input from the cyclin-CDK oscillator . While its dynamics was sensitive to loss of the cyclin-CDK input , it was still able to display either regular or dampened oscillations . This behavior is reminiscent of the diverse dynamical behavior generated by synthetic biochemical oscillators for different initial conditions or parameters ( Weitz et al . , 2014 ) . Our results support a picture whereby the transcriptional oscillations do not strictly depend on the cyclin input , but are intimately coupled with the post-translational oscillations to produce the observed wild-type behavior ( Simmons Kovacs et al . , 2012 ) . While our model cannot resolve the detailed mechanism of the coupling , e . g . because it cannot take interactions with cell cycle checkpoints into account ( Bristow et al . , 2014 ) , it does point towards an interesting possibility to reconcile the seemingly contradictory experimental results of Simmons Kovacs et al . ( 2012 ) and Rahi et al . ( 2016 ) : The model results of Figure 6 indicate that the transcriptional oscillations can change from continuous to dampened within a narrow parameter range . Simmons Kovacs et al . ( 2012 ) measured periodic transcription immediately after disabling a temperature sensitive allele of cdc28 by a shift to the restrictive temperature . In contrast , Rahi et al . ( 2016 ) applied a ‘cyclin depletion protocol’ to clear all cyclins from the cells before testing for transcriptional oscillations . Dampened oscillations might indeed account for both observations , a periodic transcription with a decaying amplitude immediately after loss of G1 cyclin activity , and cessation of all periodic events after an extended ‘cyclin depletion protocol’ . The present study provides a proof of principle for our approach for extracting quantitative information about GRFs from gene expression time series . Other studies have already demonstrated the rich information that dynamic expression data provide about regulatory interactions ( Dunlop et al . , 2008; Lipinski-Kruszka et al . , 2015 ) and regulatory mechanisms ( Westermark and Herzel , 2013 ) . In the future , we expect that inference and quantitative analysis of functional modules will develop into an extremely powerful approach as dynamic transcriptome data is systematically collected for genetic variants and under different physiological conditions . With suitable data , the approach could be extended to include inputs from regulatory RNAs and regulation on the post-transcriptional level .
The dataset used in this work ( cDTA for the yeast cell cycle ) has been previously published ( Eser et al . , 2014 ) . The complete dataset is available at ArrayExpress ( RRID:SCR_002964 ) under accession code E-MTAB-1908 . To fit GRFs to the time series of input and output signals , we use a score function that quantifies the fraction of the variance in the output signal that is not explained by the candidate GRF with the given input signals . For a given target gene i , the score Si is defined as a ratio , where the numerator corresponds to the mean squared deviation between the genes measured synthesis rate sid ( tn ) and the corresponding model synthesis rate sim ( tn ) , averaged over all 42 time points in the dataset ( tn=0 , 5 , 10 , … , 205 min ) . The denominator corresponds to the variance of the time series sid ( tn ) , such thatSi=1N∑n=142 ( sid ( tn ) −sim ( tn ) ) 2/1N∑n=142 ( sid ( tn ) −⟨sid⟩ ) 2 where ⟨sid⟩ denotes the time-averaged measured synthesis rate of gene i . The score Si is 0 for a perfect fit and 1 if the model is just a constant value corresponding to the average of the expression data . The fitting task is to minimize the score with respect to the model parameters . The model consists of two parts , the model for the protein dynamics and the parameterization of the different types of GRFs . The GRFs are functions of the protein levels of the input TFs . We model the dynamics pi ( t ) of each TF protein level bydpi ( t ) dt=νimi ( t ) −λipi ( t ) , with a translation rate νi and an effective degradation rate λi , as discussed in the main text . This generates a time series for the TF protein level , pi ( tn ) =pi ( 0 ) e−λitn+e−λitn∫0tnνieλiτmi ( τ ) dτ , where the mRNA time series mi ( tn ) is interpolated by cubic splines and the integral is evaluated numerically . The initial values pi ( 0 ) are fixed by first solving the equation until the initial condition has essentially decayed and then extrapolating backwards to obtain a smooth expression pattern . The ratio νi/λi sets the absolute level of protein relative to the mRNA level , while the dynamics of the relative protein level is governed by the parameter λi alone . The absolute protein levels are not relevant for our analysis , since all GRFs respond to ratios of protein levels to effective binding constants that are on an arbitrary scale here . Thus , we can set the ratio νi/λi to one , such that dpi/dt=λi ( mi-pi ) . This leaves λi as the only protein model parameter to be inferred for each TF . Our GRFs are parameterized via Hill functions as shown in Figure 1 and listed below . We consider two different forms of one-input GRFs , corresponding to activation and repression , and ten different forms of two-input GRFs , corresponding to analog versions of the standard Boolean logic functions . For the one-input GRFs , the parameters are the basal transcription rate b , the maximal fold change α , the sensitivity of the response n , and the TF concentration K at which the target gene is half activated or repressed , respectively . Two-input GRFs involve a K and n parameter for each input and a global b and α parameter . We consider the following types of two-input GRFs:p1ANDp2:sm=b+α[p1n1K1n1+p1n1×p2n2K2n2+p2n2]p1ORp2:sm=b+α[p1n1K1n1+p1n1+p2n2K2n2+p2n2]p1ANDNOTp2:sm=b+α[p1n1K1n1+p1n1×K2n2K2n2+p2n2]NOTp1ANDp2:sm=b+α[K1n1K1n1+p1n1×p2n2K2n2+p2n2]p1ORNOTp2:sm=b+α[p1n1K1n1+p1n1+K2n2K2n2+p2n2]NOTp1ORp2:sm=b+α[K1n1K1n1+p1n1+p2n2K2n2+p2n2]p1NORp2:sm=b+α[K1n1K1n1+p1n1×K2n2K2n2+p2n2]p1NANDp2:sm=b+α[K1n1K1n1+p1n1+K2n2K2n2+p2n2]p1XORp2:sm=b+α[ ( p1n1K1n1+p1n1×K2n2K2n2+p2n2 ) + ( K1n1K1n1+p1n1×p2n2K2n2+p2n2 ) ]p1EQp2:sm=b+α[ ( p1n1K1n1+p1n1×p2n2K2n2+p2n2 ) + ( K1n1K1n1+p1n1×K2n2K2n2+p2n2 ) ] To minimize the score and identify the best-fit GRF , we use simulated annealing with a self-adapting cooling schedule ( Lam and Delosme , 1988 ) to find the global minimum of the score as a function of the protein and GRF parameters . We perform this separately for each of the GRF types that we consider and select the one that yields the best score . To predict binding sites in the promoter region of target genes we retrieved 700 bp sequences upstream of the consensus TSS from YEASTRACT ( RRID:SCR_006076 , Teixeira et al . , 2006 ) and matched binding motifs of relevant transcription factors in the JASPAR database ( RRID:SCR_003030 , Sandelin et al . , 2004 ) with a relative score cutoff of 0 . 8 . Additionally , we matched the published consensus regulatory motif of the CLB2 cluster ( Spellman et al . , 1998 ) , using MAST ( RRID:SCR_001783 , Bailey et al . , 1998 ) . Microarray data from Pramila et al . ( 2006 ) were obtained from the GEO database ( RRID:SCR_005012 ) . Because the data are logarithmic , and each gene is individually normalized , we exponentiated the data and performed a linear transformation to make its range comparable to our dataset . The linear transformation , m~it ( t ) =ai+bimit ( t ) with mit ( t ) denoting the expression time course of gene i in the test dataset , was performed such that the minimum and maximum expression of each gene in the test data is equivalent to the minimum and maximum in our dataset , i . e . , the constants were chosen as bi= ( max[mi]-min[mi] ) / ( max[mit]-min[mit] ) and ai=min[mi]-bimin[mit] . The obtained expression time series for the TFs , m~it ( t ) , was then used together with the GRFs inferred from our data to generate predictions for the target gene expression to be compared with the corresponding values in the Pramila et al . ( 2006 ) dataset . Documented regulatory interactions between TFs and target genes have been downloaded from the YEASTRACT database ( http://yeastract . com/download/ , Teixeira et al . , 2006 ) and matched to the set of genes measured in the DTA experiments . Based on this information , TFs satisfying conditions ( ii ) and ( iii ) in Figure 4A were selected . To estimate a unified protein proxy for a TF we first select a set of suitable target genes by the following requirements: ( i ) the target genes must be periodically expressed ( analogously to requirement ( i ) in Figure 4A ) ; ( ii ) for each target gene there must be a GRF ( single input or combinatorial ) with the TF as input and a score lower than 0 . 5; ( iii ) each target gene is fitted with a GRF in which the regulatory sign of the TF ( activator or repressor ) corresponds to the literature reference ( c . f . Figure 4—source data 3 ) . On each target gene of the resulting set a MCMC algorithm is performed to sample the score function of the corresponding GRF and a histogram of the sampled posterior distribution over the parameter λ is generated ( Hastings , 1970; Andrieu et al . , 2003 ) . Finally the posterior distributions of all target genes are combined by histogram multiplication and the mode of the product histogram is determined to be the estimated parameter of the unified protein proxy of the TF . We model coupling to the primary cyclin-CDK oscillator by placing Swi4 under the control of Cln2 and itself ( via SBF ) . We include regulation by Cln2 and Swi4 as additional ( additive ) activators:sswi4m=b+α[ ( Kyhp1n1Kyhp1n1+pyhp1n1×Kyox1n2Kyox1n2+pyox1n2 ) + ( pcln2m1Kcln2m1+pcln2m1×pswi4m2Kswi4m2+pswi4m2 ) ] . In a network with N interacting transcription factors , we describe mRNA and protein expression level of each gene i by the following equations: ( 1 ) dmi ( t ) dt=sim ( pj ( t ) , … ) −δimi ( t ) , dpi ( t ) dt=λi ( mi ( t ) −pi ( t ) ) . The mRNA synthesis rate sim is given by the regulation model found by the reconstruction method and coupled , correspondingly , to the protein expression level of other TF’s in the model . In this coarse-grained description of a gene regulatory network we assume the mRNA degradation rate δi to be constant ( as the protein degradation rate λi ) . We adjust the parameters found by the reconstruction method by minimizing the squared residues between modeled and measured mRNA expression level , normalized for each gene by the data variance ( as for the normalized node score defined in Figure 3D ) . Since all nodes had been fitted individually , errors at each node propagate to downstream nodes in a simulation , distorting their behavior . Therefore , we re-fitted the global output of the ODE model to the expression data , using gradient descent and the parameters from the previous fits as starting values . The relative change in the re-fitted parameters with respect to their original values is shown in Figure 6—figure supplement 2 . The majority of the parameters change less than 10% . To simulate a TF knockout , we set the expression level of the respective gene to a constant zero throughout the simulation . Since the TF is embedded in an extended network , indirect effects in the expression of target genes can also be captured by this method . | Living cells rely on networks of genes to control their behavior , including how they grow , develop and respond to stress . Genes encode instructions needed to make proteins and other molecules , and much of the control is exerted at the first stage of protein production , known as transcription . During this process , a gene is copied to make molecules known as transcripts that may later be used as templates to make proteins . Many genes encode proteins that act to regulate transcription . Therefore , an individual gene may receive inputs from other genes , and these inputs affect how much transcript the gene produces , which can be considered as the gene’s output . While these inputs and outputs can often be wired together to form a network , it is less clear exactly how all the different inputs at a gene interact to determine its output . These interactions are known as “gene regulation functions” , and knowing them would be an important step towards understanding gene networks , which would help us to predict how cells will behave in different situations . Gene regulation functions are difficult to measure directly , so researchers would like to find other ways to assess them indirectly . A recently developed experimental technique called “dynamic transcriptome analysis” seemed promising as it measures both the inputs and outputs of all genes in a cell over time . Hillenbrand et al . used this technique to infer gene regulation functions with one or two inputs in yeast cells . Comparing these estimates with experimental data from previous studies showed that these inferred gene regulation functions could successfully predict the output of a gene based on its inputs . Hillenbrand et al . then used these estimates to search and model a well-known genetic network that is thought to be part of the molecular clockwork that controls the timing of events that cause a cell to divide . Currently , the approach used by Hillenbrand et al . treats gene regulation functions like “black boxes” . This means that , while an output can be predicted if the inputs are known , it cannot reveal all of the detailed mechanisms behind it . Gaining insights into the inner workings of these black boxes will require taking more data into account , such as how abundant the proteins that regulate transcription are , where they are located within cells or whether they are active or not . Therefore , the next challenge is to incorporate these kinds of data to gain a fuller picture of how gene networks operate within cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology"
] | 2016 | Inference of gene regulation functions from dynamic transcriptome data |
Eusociality is a distinct form of biological organization . A key characteristic of advanced eusociality is the presence of non-reproductive workers . Why evolution should produce organisms that sacrifice their own reproductive potential in order to aid others is an important question in evolutionary biology . Here , we provide a detailed analysis of the selective forces that determine the emergence and stability of non-reproductive workers . We study the effects , in situations where the queen of the colony has mated once or several times , of recessive and dominant sterility alleles acting in her offspring . Contrary to widespread belief based on heuristic arguments of genetic relatedness , non-reproductive workers can easily evolve in polyandrous species . The crucial quantity is the functional relationship between a colony’s reproductive rate and the fraction of non-reproductive workers present in that colony . We derive precise conditions for natural selection to favor the evolution of non-reproductive workers .
Eusociality is a form of social organization where some individuals reduce their own lifetime reproductive potential to raise the offspring of others ( Wilson , 1971; Crespi and Yanega , 1995; Gadagkar , 2001; Hunt , 2007; Nowak et al . , 2010 ) . Primary examples are ants , bees , social wasps , termites , and naked mole rats . There have been ~10–20 origins of eusociality , about half of them in haplodiploid species and the other half in diploid ones ( Andersson , 1984 ) . A crucial step in the origin of eusociality is cancellation of dispersal behavior ( Abouheif and Wray , 2002; Nowak et al . , 2010 , 2011; Hunt , 2012; Tarnita et al . , 2013 ) . Individuals who stay at the nest begin to work at tasks such as care of the young , defense of the nest , and foraging behavior . Eusociality can evolve if the strong reproductive advantages of the queen , including reduced mortality and increased rate of oviposition , arise already for small colony sizes ( Nowak et al . , 2010 ) . Haplodiploidy is the sex-determination system of ants , bees , and wasps . With haplodiploidy , females arise from fertilized eggs and are diploid . They have two homologous sets of chromosomes , one padumnal ( paternally inherited ) and one madumnal ( maternally inherited ) . Males arise from unfertilized eggs and are haploid . They have one set of chromosomes , which is madumnal . With haplodiploid genetics , mated females can become queens and lay female and male eggs . In many cases , unmated females become workers , which can lay male eggs but not female eggs . Therefore , in a haplodiploid colony , male eggs could come from the queen or from the workers . This creates intracolony competition between the fertilized queen and the unfertilized workers over the production of males . In this paper , we investigate genetic mutations that affect the phenotype of the workers and make them non-reproductive ( or sterile ) . We calculate conditions both for the evolutionary invasion and evolutionary stability of such mutations . The typical theoretical approach for investigating the evolution of non-reproductive workers makes use of coefficients of relatedness . Relatedness of one individual to another is the probability that a random allele in the former is also in the latter due to recent common ancestry . In particular , there are three coefficients of relatedness that are of primary interest ( Hamilton , 1964; Trivers and Hare , 1976 ) . First , consider the relatedness of a female to one of her sons , Rson . In parthenogenetically producing a son , a diploid female transmits a haploid genome to him . Under Mendelian segregation , each allele in her genome has probability 1/2 of inclusion in this transmitted haploid complement , and so Rson = 1/2 . Next , consider the relatedness of a female to one of her brothers , Rbrother . The female and her brother share a mother , and so the probability that an allele inherited maternally by the female is the same allele as inherited maternally by her brother is 1/2 . On the other hand , the male has no father , and so there is no chance that an allele in the female is equivalent to one in her brother through paternal inheritance . A random allele in the diploid female has equal chance of being padumnal or madumnal , and so Rbrother = ( 1/2 ) ( 0 ) + ( 1/2 ) ( 1/2 ) = 1/4 . Finally , consider the relatedness of a female to one of her sisters , Rsister . This coefficient of relatedness depends on the number , n , of different males that the queen mates with before laying eggs . For a padumnal allele in a female to be identical , by paternal descent , to that in a sister requires them only to share the same father ( probability 1/n ) , since that father is haploid and therefore always transmits the same allele . Sisters share a mother , and so the probability that an allele inherited maternally by one female is the same allele as inherited maternally by her sister is 1/2 . Therefore , Rsister = ( 1/2 ) ( 1/n ) + ( 1/2 ) ( 1/2 ) = ( 2 + n ) / ( 4n ) . The traditional investigation of evolution of non-reproductive workers uses the following relatedness-based heuristic . The relatedness of a female to her male offspring is Rson = 1/2 . If the queen mates with only a single male ( n = 1 ) , then the relatedness of a female worker to one of her random sisters is Rsister = 3/4 > Rson . The naive conclusion is that worker altruism should readily evolve , because a worker can more efficiently spread her genes by raising her sisters . This conjecture is known as the ‘haplodiploidy hypothesis’ ( Hamilton , 1964 , 1972 ) . If the queen mates with more than two males ( n>2 ) , then Rsister < Rson . Now the preference of an unfertilized worker is reversed , because she has a higher relatedness to her own male offspring than to one of her random sisters . These old arguments suggest that queen monogamy and haplodiploid genetics synergistically act as a driving force for the evolution of worker altruism ( Hamilton , 1964 , 1972 ) . But worker-laid eggs do not compete only with queen-laid female eggs . The high relatedness of a female to her sisters in Hymenopteran colonies is cancelled by the low relatedness of the same female to her brothers ( Trivers and Hare , 1976 ) . In a colony with a singly mated queen , the relatedness of a worker to a sister is Rsister = 3/4 . But the relatedness of a worker to a brother is only Rbrother = 1/4 , regardless of the number of times the queen mates . So , when a worker female helps her queen reproduce , she aids in the production both of sisters ( to whom she is highly related ) and brothers ( to whom she is not ) . In the relatedness-based argument , these effects exactly cancel each other out , and so the unusually high relatedness of sisters in eusocial colonies cannot be the simple solution to the puzzle of worker altruism that it was once thought to be ( Trivers and Hare , 1976 ) . This is true even when the population sex ratio is female-biased , because when more reproductive females are produced than males , the average reproductive success of a female is lower than that of the average male exactly in proportion to their relative abundances ( Craig , 1979 ) . More recently , it was proposed that each eusocial lineage must have passed through a ‘monogamy window’—a period of evolutionary history in which queens were singly mated ( Boomsma , 2009 ) . This argument assumes that worker-laid eggs compete equally with queen-laid female and male eggs . If the queen is singly mated ( n = 1 ) , and if the colony’s sex ratio is 1/2 , then a worker has an average relatedness to siblings ( sisters and brothers ) of ( Rsister + Rbrother ) /2 = ( ( 3/4 ) + ( 1/4 ) ) /2 = 1/2 . Relatedness of a worker to her son is also Rson = 1/2 . In this case , assuming that worker-laid eggs substitute equally for queen-laid female and male eggs , the argument suggests that any infinitesimal benefit of non-reproductive workers to colony productivity should lead to evolution of worker altruism . If queens mate more than once ( n≥2 ) , then the average relatedness of a worker to a random sibling falls below 1/2 , and evolution of worker altruism is supposed to be strongly disfavored ( Boomsma , 2009 ) . There also exist hypothetical scenarios in which the relatedness values of a female to a random sister and a random brother would not cancel ( Trivers and Hare , 1976 ) : for example , the sex ratio could vary from colony to colony , while the average sex ratio in the population remains at 1/2 . Evolution of helping might then be expected in the female-biased colonies ( Trivers and Hare , 1976 ) . Some recent papers ( Gardner et al . , 2012; Alpedrinha et al . , 2013 ) examine this case of split sex ratios and also question the importance of haplodiploidy for the evolution of helping . Based on their analysis , the authors conclude that haplodiploidy can have either a positive or negative influence on the evolution of helping depending on colony variables , and they determine the effect of haplodiploidy on the evolution of helping to generally be small . But they claim , in agreement with Boomsma ( 2009 ) , that monandry is a key requirement for the evolution of a worker caste . They argue that , in the case of lifetime monogamy , “any small efficiency benefit from rearing siblings ( b/c > 1 ) would lead to helping being favored by natural selection . ” ( Gardner et al . , 2012 ) In this paper , we investigate the situation where worker-laid male eggs compete directly with queen-laid male eggs . In other words , there are two types of reproduction events to consider: A worker can lay a male egg , or , instead , the queen can lay a male egg while the worker helps to raise the queen’s male egg . In either case , a male is produced . Thus , the sex ratio of the colony is unaffected by which of these two strategies is realized . The reproductive competition between the queen and the workers is only over the male offspring . ( We do not assume that the sex ratio is equal to 1/2; we only assume that it is independent of the fraction of worker-produced males . ) This possibility represents the simplest scenario for studying the evolution of non-reproductive workers and is biologically plausible ( Winston , 1987; Sundstrom , 1994; Sundstrom et al . , 1996; Queller and Strassmann , 1998; Hammond et al . , 2002 ) . Once this case is understood , subsequent analysis can consider the situation where both the fraction of worker-produced male offspring and the colony’s sex ratio vary at the same time . The relatedness of a worker to her male offspring , Rson = 1/2 , is always larger than the relatedness of a worker to a brother , Rbrother = 1/4 . Thus , if worker-laid male eggs compete primarily with queen-laid male eggs , then relatedness-based arguments might predict that worker altruism should not evolve with any number of matings of the queen unless non-reproductive workers provide some benefit to the colony . We study a model of competition between worker-laid and queen-laid male eggs that incorporates both haplodiploid genetics and a variable number of matings of the queen . Specifically , we analyze the selection pressure acting on the emergence and stability of non-reproductive workers in situations where the queen has mated once or several times . Our model assumes that the evolution of sterile workers has a negligible effect on the sex ratio of a population , which allows us to isolate and identify the specific selective forces . We derive exact conditions for the invasion and stability of non-reproductive workers .
We study the evolution of a non-reproductive worker caste in the context of haplodiploid genetics , where females are diploid and males are haploid . Virgin queens can mate with one or several males . The parameter n denotes the number of matings of the queen . Unfertilized workers help raise the offspring of the queen , but they can also lay male eggs . We analyze the conditions under which a wild-type allele , A , can be invaded by a mutant allele , a , which causes workers to be non-reproductive . Since we consider a loss of function event ( the loss of the tendency to produce eggs ) , it is more likely that the mutation is recessive rather than dominant . Therefore in the main text we present the conditions for a recessive mutant allele . In the Methods , we give derivations and results for both recessive and dominant alleles . If the mutant allele is recessive , then aa workers are sterile , while AA and Aa workers still lay male eggs . For n matings , there are 3 ( n + 1 ) types of mated queens ( Figure 1A ) . We use the notation AAm , Aam , and aam to denote the genotype of the queen and the number , m , of her matings that were with mutant males , a . The parameter m can assume values 0 , 1 , . . . , n . For example , for triple mating ( n = 3 ) , an AA2 queen has mated with one A male and two a males . 10 . 7554/eLife . 08918 . 003Figure 1 . Haplodiploid genetics and multiple matings . The wild-type allele is A . The mutant allele inducing worker sterility is a . ( A ) There are three types of virgin queens: AA , Aa , and aa . Each queen mates n times . Of those matings , n − m are with wild-type males ( type A ) and m are with mutant males ( type a ) . Hence , there are 3 ( n + 1 ) types of fertilized queens ( colonies ) . ( B ) Relative proportions of offspring for each colony type if the mutant allele , a , for worker sterility is recessive . For example , if the queen’s genotype is Aa , then half of her sons are A and the other half are a . We denote this by A + a . If the queen’s genotype is aa and she has mated with both types of males , 0< m < n , then she has both Aa and aa workers ( in proportion n − m and m , respectively ) ; her Aa workers produce male eggs , which have an equal proportion of A and a genotypes . ( C ) Relative proportions of offspring for each colony type if the mutant allele , a , for worker sterility is dominant . DOI: http://dx . doi . org/10 . 7554/eLife . 08918 . 003 The genotype of the colony is determined by the genotype of the queen and the sperm she has stored . There are 3 ( n + 1 ) types of colonies that need to be considered to formulate the full dynamics . The different colony types , and corresponding offspring with a recessive sterility allele , a , are shown in Figure 1B . For each colony , there are three types of offspring: queen-laid females , queen-laid males , and worker-laid males . Figure 1B can be understood by considering how the queen and the workers produce their offspring . Consider the offspring of type AAm colonies . The queen makes a female by randomly selecting one of the two alleles from her own genotype and pairing it with an allele selected randomly from the sperm of one of her mates . In type AAm colonies , the type AA queen mates with ( n − m ) type A males and m type a males . From her own genotype , the queen always selects an A allele . From her mates’ sperm , the queen selects an A allele with probability ( n − m ) /n or an a allele with probability m/n . Notice that in Figure 1B , C , for simplicity of presentation , we omit the overall normalization of each entry . So , for example , for type AAm colonies , we simply write that the queen produces n − m type AA females for every m type Aa females ( first row , second column of Figure 1B ) . The correct normalizations are included in the calculations of the Materials and methods section . The queen makes a male by randomly selecting one of the two alleles from her own genotype . Because the queen in a type AAm colony only carries the A allele , she can only produce type A drones ( first row , third column of Figure 1B ) . For a type AAm colony , in the 'Workers’ Sons' column , we must consider the rates of production of type AA and type Aa females by the queen , and we must consider the offspring of the type AA and type Aa females that the queen produces . The fraction of queen-produced females of type AA is , as described above , ( n − m ) /n , and each type AA female produces only type A males . The fraction of queen-produced females of type Aa is , as described above , m/n , and each type Aa female produces type A and type a males with equal probability . The total fraction of worker-laid males that are of type A is ( n − m ) /n + ( 1/2 ) ( m/n ) = ( 2n − m ) /2n . The total fraction of worker-laid males that are of type a is ( 1/2 ) ( m/n ) = m/2n . Therefore , the workers of type AAm colonies produce 2n − m type A males for every m type a males ( first row , fourth column of Figure 1B ) . The logic behind the offspring of type Aam and type aam colonies is the same . The only other point is that type aa workers are non-reproductive . To see how worker sterility enters into Figure 1B , consider the worker-produced males of type aam colonies . The queen of type aam colonies produces n − m type Aa females for every m type aa females ( third row , second column of Figure 1B ) . Type Aa workers produce equal numbers of type A and type a males . Type aa workers are non-reproductive; they do not contribute to the colony’s production of worker-produced males . So , in type aam colonies , all worker-produced males come from type Aa workers , and type A and type a males are therefore produced by workers in equal amounts ( third row , fourth column of Figure 1B ) . The entries in Figure 1C originate from the same reasoning . The only difference is that , if the sterility allele , a , is dominant , then type Aa and type aa workers do not contribute to a colony’s production of worker-produced males . The entries in Figure 1C are described in detail in the Materials and methods . In our analysis , we neglect stochastic effects . This is reasonable if we assume that the number of individuals produced by a colony is very large . In this case , the fractions of colony offspring of different genotypes in a generation do not differ significantly from the entries in Figure 1B , C . A crucial quantity is the functional relationship between the fraction of males produced by the queen , p , and the fraction of non-reproductive workers , z , that are present in a colony . The parameter z can vary between 0 and 1 . If z = 0 , then there are no non-reproductive workers in the colony . If z = 1 , then all workers in the colony are non-reproductive . We denote by pz the fraction of males that come from the queen if the fraction of non-reproductive workers is z . The quantity p0 denotes the fraction of males that come from the queen if there are no non-reproductive workers in the colony . We expect p0 to be less than 1 . The quantity p1 denotes the fraction of males that come from the queen if all workers are non-reproductive . Clearly , p1 = 1 . It is natural to assume that pz is an increasing function of z , but various functional forms are possible . Perhaps the simplest possibility is that pz is a linearly increasing function of z . Intuitively , this means that the fraction , 1 − pz , of male eggs that originate from workers is simply proportional to the fraction , 1 − z , of workers that are reproducing . But there are nonlinear intracolony effects that modulate worker production of male eggs . For example , the queen might efficiently suppress worker reproduction via aggression or removal of worker-laid eggs ( Free and Butler , 1959; Michener , 1974; Oster and Wilson , 1978; Fletcher and Ross , 1985 ) , if only a small number of workers attempt to reproduce . If too many workers reproduce , then the queen could be overwhelmed , and her effect on removing worker-laid eggs is diminished . In this equally plausible scenario , the fraction , pz , of male eggs that originate from the queen would be expected to increase sublinearly with the fraction , z , of workers that are sterile . Several sample forms of the function pz are shown in Figure 2A . 10 . 7554/eLife . 08918 . 004Figure 2 . For understanding the evolution of non-reproductive workers , the following two functions are crucial . ( A ) The function pz denotes the fraction of male offspring that come from the queen if a fraction , z , of the workers are non-reproductive . Therefore , 1 − pz is the fraction of male offspring that come from the workers . Clearly , pz should be an increasing function . More workers that are sterile means a larger fraction of males that come from the queen . If all workers are non-reproductive , then all males come from the queen , p1 = 1 . ( B ) The function rz denotes the reproductive rate ( or efficiency ) of the colony if a fraction , z , of the workers are non-reproductive . Without loss of generality , we normalize such that r0 = 1 . If worker sterility has an advantage , then it should increase colony efficiency for some values of z , but the function rz need not be monotonically increasing . It is possible that maximum colony efficiency is obtained for an intermediate value of z . Several possibilities for the colony efficiency function , rz , are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08918 . 004 The mutant allele can be favored by natural selection if non-reproductive workers provide a benefit to the colony , which is of course a natural assumption for the evolution of worker altruism . Division of labor has the potential to improve efficiency ( Cole , 1986; Naeger et al . , 2013 ) . Another key component in our analysis is the functional relationship between the rate , r , at which the colony produces reproductive units ( virgin queens and males ) and the fraction of sterile workers , z . We use the notation rz to describe the reproductive rate of a colony where a fraction , z , of workers are non-reproductive . The quantity r0 denotes the reproductive rate of the colony if none of the workers are non-reproductive . Non-reproductive workers have a chance to be favored by natural selection if rz > r0 for some z . But the function rz need not be monotonically increasing . It is possible that there is an optimum fraction of non-reproductive workers , which maximizes the overall reproductive rate of the colony . We will study various functional forms of rz . Several sample forms of the function rz are shown in Figure 2B .
If the mutant allele is recessive , then AA and Aa workers lay male eggs , while aa workers are non-reproductive . For single mating , n = 1 , we find that the a allele can invade an all-A resident population provided ( 1 ) r1/2r0>6+2p05+3p0 What is the intuition behind this condition ? There are five colony types , AA0 , AA1 , Aa0 , Aa1 , and aa0 , which are relevant for determining if the mutant allele can invade . Four of those colony types do not produce sterile workers ( z = 0 ) , so the parameters p0 and r0 enter into Equation 1 . In colonies of type Aa1 , half of the workers are sterile ( z = 1/2 ) ; thus the parameter r1/2 enters into Equation 1 . Moreover , both the queen and the workers in Aa1 colonies each produce 50% type A males and 50% type a males; therefore the parameter p1/2 is irrelevant for the invasion and absent from Equation 1 . If all males are initially produced by the workers ( p0 = 0 ) , then the ratio of the efficiency of type Aa1 colonies to type AA0 colonies , r1/2/r0 , must be greater than 6/5 for non-reproductive workers to appear . Notice that the critical value of r1/2/r0 is a decreasing function of p0 . Intuitively , this means that if worker sterility has a smaller phenotypic effect on a colony ( such that p0 is closer to p1 = 1 ) , then the efficiency gain from sterile workers does not need to be as high to facilitate the invasion of sterility . If p0 is small , then we get efficiency thresholds for r1/2/r0 of ~1 . 1-1 . 2 . If p0 is large , then we get efficiency thresholds that are close to 1 . As long as p0 is not infinitesimally close to 1 , the ratio r1/2/r0 must always be greater than 1 by a finite amount . Sterility cannot invade if sterile workers do not appreciably improve colony efficiency . We note that other studies report the evolution of a worker caste with infinitesimal efficiency benefits in singly mated colonies ( Boomsma , 2007 , 2009; Gardner et al . , 2012 ) . But these papers consider competition between worker offspring and queen-laid female eggs , which induces sex-ratio effects that complicate the analysis . Another recent study argues that eusociality can evolve even if sterile workers are relatively inefficient at raising siblings ( Avila and Fromhage , 2015 ) . But this work focuses centrally on the evolution of nest formation , where nest-site limitation and dispersal mortality impose ecological constraints on independent breeding . In our study , we analyze the scenario where nests have already formed , and non-reproductive workers emerge as a subsequent step in the path to advanced eusociality . For double mating , n = 2 , we find that the a allele can invade an all-A resident population provided ( 2 ) r1/4r0>6+3p05+3p0+p1/4 Now there are six colony types , AA0 , AA1 , AA2 , Aa0 , Aa1 , and aa0 , which are relevant for determining if the mutant allele can invade . Colony types AA0 , AA1 , AA2 , Aa0 , and aa0 do not produce sterile workers ( z = 0 ) , so the parameters p0 and r0 appear in Equation 2 . Colonies of type Aa1 produce a fraction 1/4 of sterile workers ( z = 1/4 ) . The Aa1 queen uses the sperm from the type A male that she has mated with to produce AA and Aa workers in equal proportion . Or the Aa1 queen uses the sperm from the type a male to produce Aa and aa workers with equal proportion . Thus , 1/4 of the workers are of type aa and are non-reproductive . Correspondingly , the parameters p1/4 and r1/4 appear in Equation 2 . The maximum critical value of r1/4/r0 for evolution of non-reproductive workers is 6/5 , and the minimum critical value is 1 . The threshold of r1/4/r0 is large ( ~1 . 1-1 . 2 ) if p0 is small . The threshold of r1/4/r0 is close to 1 if p0 is large . Provided that p0 is not infinitesimally close to 1 , the ratio r1/4/r0 must always be greater than 1 by a finite amount for sterile workers to be able to invade . It is not clear , a priori , that Equation 2 would be easier or harder to satisfy than Equation 1 . Empirical knowledge of the parameters p0 , p1/4 , r0 , r1/4 , and r1/2 is needed to determine whether sterility invades more easily for single mating than for double mating . An illustration of the parameter space and whether single or double mating is more conducive to development of sterility is shown in Figure 3A . It is clear from Equation 1 that , holding all other parameters constant , an increase in r1/2 favors the invasion of the sterility allele for n = 1 . This is easy to see in Figure 3A: The upper panels ( higher r1/2 ) involve invasion of the sterility allele for n = 1 , while the lower panels do not . Similarly , from Equation 2 , it is clear that , holding all other parameters constant , an increase in r1/4 favors the invasion of the sterility allele for n = 2 . Again , this is illustrated in Figure 3A: the right panels ( higher r1/4 ) are associated with invasion of the sterility allele for n = 2 , while the left panels are not . 10 . 7554/eLife . 08918 . 005Figure 3 . Regions of the parameter space for the evolution of non-reproductive workers for single and for double mating . ( A ) For single mating , n = 1 , the invasion of a recessive worker sterility allele depends on the parameters p0 and r1/2; for double mating , n = 2 , it depends on the parameters p0 , p1/4 , and r1/4 . ( B ) The evolutionary stability of a recessive worker sterility allele depends on the parameters p0 , r1/2 , and r1 for single mating , and on the parameters p1/2 , r1/2 , and r1 for double mating . We set r0 = 1 as baseline . DOI: http://dx . doi . org/10 . 7554/eLife . 08918 . 005 The region of parameter space for which sterility invades for double mating but not for single mating is arbitrarily large . The region of parameter space for which sterility invades both for double mating and for single mating is also arbitrarily large . These features apply generally for different values of p0 and p1/4 . For many possible combinations of those parameters , worker sterility invades for double mating but not for single mating . For example , if p0 = 0 . 8 and p1/4 = 0 . 9 , then for single mating the invasion condition is r1/2 > 1 . 027 while for double mating the invasion condition is r1/4 > 1 . 012 . ( Here , without loss of generality , we set r0 = 1 . ) The latter condition could be easier to satisfy—even if rz increases linearly with z . We note that colony reproductive efficiency , rz , would not necessarily be expected to increase monotonically with the fraction of sterile workers , z . The law of diminishing returns may apply to the addition of non-reproductive workers to a colony . Non-reproductive workers contribute positively to the colony’s total reproductive output by performing colony maintenance and helping to raise other individuals’ offspring . But by not laying any eggs , non-reproductive workers are also negatively affecting the colony’s total reproductive output . Consequently , colony reproductive efficiency may be maximized if some workers reproduce while other workers focus their efforts on colony maintenance . In our model , this would correspond to rz reaching a maximum for some 0 < z < 1 . Assuming that p0 and p1/4 are small , we find that a fairly substantial benefit to colony reproductive rate ( around 10% to 20% ) must be provided by a non-reproductive worker caste . The large thresholds predicted by our model might help to explain the rarity of the evolution of non-reproductive worker castes in social insects . Additional work is needed to connect the parameters of our model with biological measurements of colony dynamics . Numerical simulations of the evolutionary dynamics for different parameter values are shown in Figure 4 . 10 . 7554/eLife . 08918 . 006Figure 4 . Numerical simulations of the evolutionary dynamics nicely illustrate the conditions specified by Equations 1 and 2 . The sterility allele is recessive . For numerically probing invasion , we use the initial condition XAA , 0=1-10-2 and XAA , 1=10-2 . We set r0 = 1 . A: Single mating , n = 1 . Parameters p0 = 0 . 1 and r1 = 1 . 29 . B: Double mating , n = 2 . Parameters p0 = 0 . 2 , p1/4 = 0 . 4 , p1/2 = 0 . 6 , r1/2 = 1 . 24 and r1 = 1 . 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 08918 . 006 We have also calculated the condition for the evolutionary stability of non-reproductive workers . For single mating , n = 1 , we find that the a allele is stable against invasion of A in an all-a resident population provided ( 3 ) r1r0-1-p022r1r1/2-1>1 Three colony types , aa1 , aa0 , and Aa1 , are relevant for determining if the a allele for sterility is evolutionarily stable to invasion by the A allele . Type aa1 colonies produce only sterile workers ( z = 1 ) , hence the appearance of r1 in Equation 3 . Type aa0 colonies produce no sterile workers ( z = 0 ) , hence the appearance of p0 and r0 in Equation 3 . Type Aa1 colonies produce 50% sterile workers ( z = 1/2 ) , hence the appearance of r1/2 in Equation 3 . The parameter p1/2 is irrelevant because the queen and the reproductive workers in type Aa1 colonies each produce 50% type A males and 50% type a males . For double mating , n = 2 , we find that the a allele is evolutionarily stable provided ( 4 ) r1r1/2- ( 1-p1/2 ) 2r1r1/2-1>1 Three colony types , aa2 , aa1 , and Aa2 , are relevant for determining if the a allele for sterility is evolutionarily stable . Type aa2 colonies produce only sterile workers ( z = 1 ) , hence the appearance of r1 in Equation 4 . Type aa1 and type Aa2 colonies each produce 50% sterile workers ( z = 1/2 ) , hence the appearance of p1/2 and r1/2 in Equation 4 . The conditions for invasion and stability with more than two matings are given in the Materials and methods . Empirical knowledge of the parameters p0 , p1/2 , r0 , r1/2 , and r1 is needed to determine if worker sterility is more stable for single mating than for double mating . For many possible combinations of those parameters , worker sterility is evolutionarily stable for double mating but not for single mating . For example , if p0 = 0 . 6 , p1/2 = 0 . 9 , r0 = 1 , and r1/2 = 1 . 05 , then for single mating the stability condition is r1 > 1 . 105 while for double mating the stability condition is only r1 > 1 . 087 . The latter condition is less stringent . The parameter space for evolutionary stability for specific values of pz is shown in Figure 3B . Equations 1–4 tell us how non-reproductive workers evolve in a population of otherwise reproductive workers . The simplest case of singly mated queens already shows rich behavior . In Figure 5A , the four possibilities are shown: Sterility may not invade and be unstable ( lower left ) , invade but be unstable ( lower right ) , not invade but be stable ( upper left ) , or invade and be stable ( upper right ) . For example , notice that if r1/2 = 0 . 6 and r1 = 0 . 9 , then worker sterility does not invade but is evolutionarily stable , even though both efficiency parameters are less than 1 . As another example , notice that as long as r1/2 exceeds about 1 . 077 , the quantity r1 can be arbitrarily small and worker sterility will still invade . It is also interesting that , for a fixed value of r1 , increasing the value of r1/2 does not necessarily promote the stability of worker sterility , and doing so can actually render non-reproductive workers evolutionarily unstable . Complexities such as these are not readily accounted for by heuristic relatedness-based arguments . If the value of p0 is very close to 1 , then arbitrarily small changes in colony efficiency can positively or negatively influence the evolutionary invasion or stability of worker sterility ( Figure 5B ) . Numerical simulations of the evolutionary dynamics demonstrating the four possible behaviors are shown in Figure 6 . 10 . 7554/eLife . 08918 . 007Figure 5 . Evolution of non-reproductive workers for single mating ( n = 1 ) . We consider a recessive sterility allele , a . There are four possible scenarios: The mutant allele cannot invade but is evolutionarily stable ( bistability ) ; the mutant allele can invade and is evolutionarily stable; the mutant allele can invade but is unstable ( coexistence ) ; the mutant allele cannot invade and is unstable . Only three parameters matter: p0 , r1/2 , and r1; p0 denotes the fraction of male offspring that come from the queen if there are no sterile workers in the colony ( z = 0 ) ; r1/2 and r1 denote respectively the reproductive rate ( efficiency ) of the colony if z = 1/2 and z = 1 of all workers are sterile . The baseline value is r0 = 1 . ( A ) Phase diagram for p0 = 0 . 5 . ( B ) Phase diagram for p0 = 0 . 9 . As p0 gets closer to 1 , the intersection of the critical curves approaches the point ( r1/2 , r1 ) = ( 1 , 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08918 . 00710 . 7554/eLife . 08918 . 008Figure 6 . Numerical simulations of the evolutionary dynamics that show the four behaviors in Figure 5A . The sterility allele is recessive , and we consider single mating ( n = 1 ) . For each of the four panels , we use the initial conditions: A: XAA , 0=1-10-3 and XAA , 1=10-3; B: Xaa , 1=1-10-3 and Xaa , 0=10-3; C: XAA , 0=0 . 27 and Xaa , 0=0 . 73 ( lower curve ) , and XAA , 0=0 . 26 and Xaa , 0=0 . 74 ( upper curve ) ; D: XAA , 0=1-10-2 and XAA , 1=10-2 ( lower curve ) , and Xaa , 1=1-10-2 and Xaa , 0=10-2 ( upper curve ) . We set r0 = 1 . A: Parameters p0 = 0 . 5 , r1/2 = 1 . 0869 , and r1 = 1 . 1521 . B: Parameters p0 = 0 . 5 , r1/2 = 1 . 0669 , and r1 = 1 . 1321 . C: Parameters p0 = 0 . 5 , r1/2 = 1 . 0669 , and r1 = 1 . 1521 . D: Parameters p0 = 0 . 5 , r1/2 = 1 . 0869 , and r1 = 1 . 1321 . DOI: http://dx . doi . org/10 . 7554/eLife . 08918 . 008 Figure 7 shows some examples . In Figure 7A , worker sterility invades for double mating but not for single mating . Here , rz increases sublinearly in z . In Figure 7B , the value of p1/4 is only slightly increased compared with its value in Figure 7A . In Figure 7B , the efficiency function , rz , is linearly increasing , and worker sterility invades for double mating but not for single mating . For the parameter values in Figure 7C , worker sterility is stable for double mating but not for single mating . Here , rz increases somewhat faster than linearly in z . In Figure 7D , the value of p1/2 is only slightly increased compared with its value in Figure 7C . In Figure 7D , the efficiency function , rz , is linearly increasing , and worker sterility is stable for double mating but not for single mating . 10 . 7554/eLife . 08918 . 009Figure 7 . Comparing the effect of single mating ( n = 1 ) and double mating ( n = 2 ) on the evolution of worker sterility . Whether or not single or double mating favors the evolution of worker sterility depends on the functions pz and rz . The function pz specifies the fraction of male offspring that come from the queen if a fraction , z , of all workers in the colony is non-reproductive . The function rz specifies the reproductive rate ( or efficiency ) of the colony if a fraction , z , of all workers in the colony is non-reproductive . We consider a recessive mutant allele , a , for worker sterility . ( A , B ) For these parameter choices , the mutant allele causing worker sterility can invade for double mating but not for single mating . ( C , D ) For these parameter choices , the mutant allele causing worker sterility is evolutionarily stable for double mating but not for single mating . DOI: http://dx . doi . org/10 . 7554/eLife . 08918 . 009 For a dominant sterility allele , there is typically a large region of parameter space for which sterility evolves for double mating but not for single mating . There is also an arbitrarily large region of parameter space for which sterility evolves for double mating and for single mating . For a recessive allele , it is possible that more than two matings are necessary for the emergence of worker sterility . These additional examples are presented in the Materials and methods .
Single mating of queens has often been claimed to be a key factor in the evolution of eusociality . This paradigm derives from heuristic relatedness-based arguments . For example , Boomsma’s ‘monogamy window hypothesis’ ( Boomsma , 2007 , 2009 ) holds that persistent monandry is crucial in the evolution of a worker caste . His argument is that , under monandry , because a worker is equally related to her own offspring as to her mother’s offspring ( 1/2 ) , her reduced reproduction will be selected for if it increases production of the latter more than it decreases production of the former . By this argument , under monandry but not multiple mating , very small colony-level efficiency gains from workers should lead to their evolution . On the empirical side , Hughes et al . ( 2008 ) study a phylogeny of the Hymenoptera , and , employing an ancestral state reconstruction analysis , infer that each of the eight independent transitions to eusociality in the Hymenoptera occurred in a monandrous ancestral species . They take this correlation to be evidence for the causal claim that monandry is key to the evolution of eusociality . But , if most ancestral species in the clade were monandrous [as appears to be the case: Hughes et al . ( 2008 ) , Fig S1; Nonacs ( 2011 ) ] , then the fact that the ancestors of eusocial species in the clade were monandrous would not be surprising ( Nonacs , 2011 ) . As an extreme example , if Hughes et al . were to repeat their study for the trait of haplodiploidy , they would also find that each of the eight independent transitions to eusociality occurred in a haplodiploid ancestral species ( since all Hymenoptera are haplodiploid ) . It would be absurd to consider this to be evidence for the theoretical claim that haplodiploidy is key to the evolution of eusociality . An important aspect of the evolution of advanced eusociality is the cancellation of all worker reproduction—in particular , the production of males . Here , we have performed a rigorous mathematical analysis of the conditions under which worker non-reproduction evolves . Our analysis has revealed that monandry does not play a crucial role in the evolution of non-reproductive worker castes . Indeed , in some cases , non-reproduction evolves when queens are multiply mated , but not when they are singly mated . Our results therefore show that the dominant paradigm , that monandry is crucial for the evolution of eusociality because it maximizes relatedness among siblings , needs to be revised . It may still turn out that monandry is important in the evolution of non-reproductive castes , but this would have to be for other reasons ( Nowak et al . , 2010; Nonacs , 2011; Hunt , 2012; Wilson and Nowak , 2014 ) . These insights have been achieved because of a more general treatment of the colony-level effects of non-reproductive workers than is allowed for by simple relatedness-based arguments . In our model , we have explicitly accounted for two key parameters that are often neglected in other studies: rz , the reproductive rate of the colony , and pz , the proportion of male offspring that come from the queen , if a fraction z of the colony’s workers are non-reproductive . The conditions under which selection favors non-reproductive workers have been shown to depend crucially , and in interesting ways , on these two parameters . Empirical measurement of these parameters is therefore required to understand the selective forces underlying the evolution of non-reproductive castes in social insects . This suggests a line of future research . It is important to distinguish between worker non-reproduction ( workers produce no offspring , but may still retain the ability to do so ) , and the more specific phenomenon of worker sterility ( workers do not have the ability to reproduce ) . An embedded distinction is that , in very many eusocial insect species , worker females have lost the ability to lay fertilized ( female ) eggs—e . g . , through loss or degradation of the spermatheca—but retain the ability to lay unfertilized ( male ) eggs—i . e . , they have functional ovaries ( Bourke , 1988; Hölldobler and Wilson , 1990 ) . In comparison , complete sterility ( sexual and asexual ) is rare in the eusocial insects; for example , only 9 of the roughly 300 genera of ants are known to have evolved complete worker sterility ( Bourke , 1995 ) . In our model , we have assumed that workers can lay only unfertilized male eggs . Our model therefore best applies to those transitions from species whose workers lay only male eggs to species where the workers are non-reproductive . This is probably the most common and important route to a non-reproductive worker caste in the social insects ( Bourke , 1988 ) . It is important to realize that our model applies significantly more generally than to just the ( comparatively few ) transitions to complete worker sterility . We have focused here on the case where a single ( recessive or dominant ) allele turns off worker production of males . However , mathematically , this assumption can easily be relaxed by supposing that the allele in question merely alters the frequency of worker male production by some amount . Thus , our approach is flexible enough to handle a variety of molecular mechanisms for worker reproductive restraint , which are still being elucidated for particular species ( Abouheif and Wray , 2002; Dearden , 2006; Khila and Abouheif , 2008; Moczek et al . , 2011; Cameron et al . , 2013; Sadd et al . , 2015; Kapheim et al . , 2015 ) . One potentially important factor in the evolution of worker non-reproduction is worker policing . Here , if a worker lays a ( male ) egg , there is some chance that another worker will destroy the egg . This reduces the incentive for workers to lay male eggs in the first place , therefore selecting for decreased worker reproduction ( Bourke , 1999; Wenseleers et al . , 2004; Ratnieks et al . , 2006 ) . A slightly modified version of our model can cover the case of worker policing ( with appropriate interpretations of the parameters rz and pz ) . A detailed investigation of this situation is desirable , and is in progress . Moreover , our analysis demonstrates that worker policing , though perhaps conducive to the evolution of worker non-reproduction , is not necessary for it . Our analysis makes no use of inclusive fitness theory , which is an unnecessary construct ( Nowak et al . , 2010; Allen et al . , 2013 ) . Indeed , our analysis shows that the evolution of non-reproductive workers depends on precise functional relationships between worker reproduction , queen reproduction , and colony efficiency , which inclusive fitness heuristics cannot account for . We note that inclusive fitness theory , which has dominated this area for decades , has not produced a mathematical analysis of even the most basic factors leading to the evolution of non-reproductive workers . A clear understanding of how natural selection acts on the evolution of any social behavior is possible once the field has recognized the limitations of inclusive fitness and has moved beyond them .
Consider a large population of insects . There are many colonies in the population , and each colony produces many offspring over its lifetime . The particular species under investigation has a haplodiploid genetic system . Females carry homologous pairs of maternal and paternal chromosomes , while males carry a single set of chromosomes . Queens can produce diploid female workers and gynes ( future queens ) from her own genotype combined with the genotype of each of the male drones that she has mated with . Queens can also produce drones using her own genotype . Female workers can produce drones as well . Thus there can be competition over whether the queen or the workers produce most of the males in a colony . To investigate the selective forces behind non-reproductive workers—i . e . , workers that do not parthenogenetically produce haploid drones—we propose two alleles , A and a . The phenotype corresponding to the A allele is such that workers produce drones . The phenotype corresponding to the a allele is such that workers do not produce drones . An important parameter is the number , n , of males with which the colony’s queen has mated . A schematic of the mating events is shown in Figure 1A . There are several possibilities: A type AA gyne mates with n − m type A males and m type a males . A type Aa gyne mates with n − m type A males and m type a males . A type aa gyne mates with n − m type A males and m type a males . The mating events are random . A virgin queen mates with n randomly chosen males in the population . Notice that , for mating , the gynes and drones are considered well-mixed: A gyne from one colony can mate with n drones , each chosen randomly from among the colonies in the population . For n = 1 , we obtain single mating ( monandry ) . For n = 2 , we obtain double mating . The following system of ordinary differential equations describes the selection dynamics in continuous time: ( 5 ) X˙AA , m=dXAA , mdt=nmxAAyAn-myam-ϕXAA , mX˙Aa , m=dXAa , mdt=nmxAayAn-myam-ϕXAa , mX˙aa , m=dXaa , mdt=nmxaayAn-myam-ϕXaa , m The overdot denotes the time derivative , d/dt . We use the overdot notation for any time derivative . We understand Equation 5 as follows . We represent the genotype of a colony by the genotype of its queen and the sperm she has stored from her matings . Each queen carries homologous pairs of maternal and paternal chromosomes , so each queen has one of three possible combinations of the A and a alleles in her own genotype: AA , Aa , or aa . A particular queen has mated with n − m type A males and m type a males . The number of colonies that are headed by type AA queens who have mated with n − m type A males and m type a males is denoted by XAA , m . The number of colonies that are headed by type Aa queens who have mated with n − m type A males and m type a males is denoted by XAa , m . The number of colonies that are headed by type aa queens who have mated with n − m type A males and m type a males is denoted by Xaa , m . The variables xAA , xAa , and xaa denote the numbers of gynes of the three possible genotypes in the population . The variables yA and ya denote the numbers of drones of the two possible genotypes in the population . A gyne randomly mates with n drones to become a queen . The binomial coefficient accounts for all possible sequences in which a female can mate with m males carrying an a allele out of n total matings . We require that the total number of colonies sums to a constant value , c , at all times: ( 6 ) ∑m=0n ( XAA , m+XAa , m+Xaa , m ) =c Colonies compete for resources which are limited . Notice that ϕ in Equation 5 represents a density-dependent death rate . We use ϕ to model the effect of environmental constraints in limiting the total number of colonies . To enforce the density constraint , Equation 6 , on the colony variables , we set ( 7 ) ϕ=c-1 ( xAA+xAa+xaa ) ( yA+ya ) n Our choice to analyze the evolutionary dynamics of sterile workers in continuous time is a matter of preference . Working in continuous time usually simplifies the analysis . For example , when we derive conditions for the invasion of a recessive allele or the stability of a dominant allele , the perturbative expansion of the colony variables must be performed to second order , and the calculations become quite messy . There are a couple of key biological parameters in our model . The emergence of sterile workers can affect the fraction of male eggs in a colony that originate from the queen . If a fraction z of workers in a colony are non-reproductive , then the fraction of male offspring that originate from the queen is denoted by pz . The queen and the unfertilized females may compete for production of male eggs . The function pz for 0 ≤ z < 1 likely varies for different species . It is reasonable to expect that pz is an increasing function of z; an increase in the proportion of workers that are non-reproductive results in a larger proportion of queen-produced males . If all workers are non-reproductive , then z = 1 and p1 = 1 . The other key function in our model is the efficiency , rz , of a colony in which a fraction z of workers are non-reproductive . An appropriate biological intuition is that the parameter rz is the total number of offspring produced by a colony when a fraction , z , of workers in the colony are non-reproductive . As we shall see , the ratios of colony efficiency values , rz , for colonies with different genotypes—i . e . , the relative reproductive efficiencies of colonies with different genotypes—are important quantities for understanding the evolutionary dynamics of a mutation that causes workers to be non-reproductive . As baseline , we set r0 = 1 . Non-reproductive workers forego their own reproductive potential in order to help raise their nestmates’ offspring . If this division of labor has some advantage for the colony , then we expect rz > 1 for some values of z . It is not necessary , however , that rz is a monotonically increasing function . Since we are focused on the evolutionary dynamics of the colony variables , XAA , m , XAa , m , and Xaa , m for 0 ≤ m ≤ n , we rewrite the first term on the right-hand side of Equation 5 in terms of the colony variables . We express each of the gyne and drone numbers , xAA , xAa , xaa , yA , and ya , as a linear combination of the colony variables , XAA , m , XAa , m , and Xaa , m . The coefficients in these linear relationships depend on whether the allele , a , that acts in a worker to induce that worker’s sterility is dominant or recessive . For additional insight , we perform random sampling of the parameter space to obtain some intuition whether evolution of non-reproductive workers is more or less likely for single or double mating . We will also evaluate the likelihood of selection favoring invasion or evolutionary stability of alleles ( mutations ) that induce non-reproductive workers . Thus , we do random sampling of the parameter regions shown in Figures 3A , 5A , and 8 . In each case , the outcome depends on two efficiency values , which we call rz1 and rz2 with z1 < z2 . For Figure 3A , those values are r1/4 and r1/2 . For Figure 5A and for Figure 8 , those values are r1/2 and r1 . The outcome of this numerical experiment depends on how we choose to randomize the colony efficiency values , rz1 and rz2 . There are many ways to do this . Here , we consider two possibilities: P ( rz1 , rz2 ) =12πσ2exp -[ ( rz1-μ ) 2+ ( rz2-μ ) 2]2σ2 There is no correlation between rz1 and rz2 . The average is μ = 1 . We choose σ = 0 . 1 for Figure 3A . We choose σ = 0 . 2 for Figures 5A and 8 . P ( rz1 , rz2 ) =12πσ21-ρ2exp -[ ( rz1-μ ) 2+ ( rz2-μ ) 2-2ρ ( rz1-μ ) ( rz2-μ ) ]2σ2 ( 1-ρ2 ) We set ρ = 0 . 8 . Now , there is positive correlation between rz1 and rz2 . We choose μ and σ as for Procedure 1 . Table 1 shows the outcome of this numerical experiment for the parameter values used in Figures 3A and 8 . Table 2 shows the outcome of this numerical experiment for the parameter values used in Figure 5A . For example , consider the first row of Table 1 . We set p0 = 0 . 2 and p1/4 = 0 . 4 with a recessive sterility allele , as this corresponds with Figure 3A . Procedure 1 is used for selecting values of r1/4 and r1/2 . For a randomly chosen pair of efficiency values r1/4 and r1/2 , the probabilities that the sterility allele does not invade , invades only for n = 1 , invades only for n = 2 , and invades for n = 1 and n = 2 are 0 . 7769 , 0 . 0644 , 0 . 1465 , and 0 . 0122 , respectively . For the second row of Table 1 , Procedure 2 is used for selecting values of r1/4 and r1/2 . For a randomly chosen pair of efficiency values r1/4 and r1/2 , the probabilities that the sterility allele does not invade , invades only for n = 1 , invades only for n = 2 , and invades for n = 1 and n = 2 are 0 . 8237 , 0 . 0177 , 0 . 0997 , and 0 . 0589 , respectively . The third and fourth rows of Table 1 and the rows of Table 2 are understood in the same way . 10 . 7554/eLife . 08918 . 013Table 1 . Numerical experiments . We randomly select the two relevant colony efficiency values from a bivariate normal distribution . For Procedure 1 , the two efficiency values are uncorrelated . For Procedure 2 , they are correlated ( with correlation 0 . 8 ) . The results of the numerical experiment for Figures 3A and 8 are shown . For Figure 3A , which describes a recessive allele inducing non-reproductive workers , we randomly generate values for r1/4 and r1/2 . For Figure 8 , which describes a dominant allele inducing non-reproductive workers , we randomly generate values for r1/2 and r1 . The table shows the likelihood of the four possible outcomes: non-reproductive workers ( i ) do not invade , ( ii ) invade for single mating but not for double mating , ( iii ) invade for double mating but not for single mating , and ( iv ) invade for both single and double mating . For this particular randomization experiment , double mating is more favorable than single mating for the invasion of non-reproductive workers . All p values are exactly as in the corresponding Figures . DOI: http://dx . doi . org/10 . 7554/eLife . 08918 . 013Does notInvades for n = 1Invades for n = 2Invades for bothinvadebut not n = 2but not n = 1n = 1 and n = 2Figure 3A , Proc . 1 , recessive0 . 77690 . 06440 . 14650 . 0122Figure 3A , Proc . 2 , recessive0 . 82370 . 01770 . 09970 . 0589Figure 8 , Proc . 1 , dominant0 . 79440 . 01290 . 08300 . 1097Figure 8 , Proc . 2 , dominant0 . 79270 . 01460 . 02600 . 166710 . 7554/eLife . 08918 . 014Table 2 . Numerical experiments . With the equivalent Procedures , we explore the likelihood of the four scenarios regarding invasion and/or stability for single mating . Results of the numerical experiment for Figure 5A , describing a recessive allele , are shown . We randomly generate values for r1/2 and r1 . The value p0 = 0 . 5 is exactly as in Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 08918 . 014Does not invadeDoes not invadeInvadesInvadesand is unstablebut is stablebut is unstableand is stableFigure 5A , Proc . 1 , recessive0 . 34840 . 30140 . 30070 . 0495Figure 5A , Proc . 2 , recessive0 . 52950 . 12030 . 23790 . 1123 For both Procedures , we find that the invasion of non-reproductive workers is more likely favored for double mating , n = 2 , than for single mating , n = 1 . | Certain wasps , bees and ants live in highly organized social groups in which one member of a colony ( the queen ) produces all or almost all of the offspring . This form of social organization – called eusociality – raises an important question for evolutionary biology: why do individuals that forego the chance to reproduce and instead raise the offspring of others evolve ? One factor linked to the evolution of eusociality in insects is a system that determines the gender of offspring known as haplodiploidy . In this system , female offspring develop from fertilized eggs , while male offspring develop from unfertilized eggs . The queen mates with male insects and so she can produce both male and female offspring . On the other hand , the workers – which are also female – do not mate and therefore can only produce male offspring . So , should these workers produce their own male eggs , or should all male offspring come from the queen ? The answer to this question could depend on whether the queen has mated with a single male ( monandry ) or with multiple males ( polyandry ) because this affects how closely related the other insects in the colony are to each other . It is a widespread belief that monandry is important for the evolution of non-reproductive workers . Here , Olejarz et al . develop a mathematical model that explores the conditions under which natural selection favors the evolution of non-reproductive workers . Contrary to the widespread belief , it turns out that non-reproductive workers can easily evolve in polyandrous species . The crucial quantity is the relationship between the overall reproductive rate of the colony and the fraction of non-reproductive workers present in that colony . Olejarz et al . challenge the view that single mating is crucial for the evolution of non-reproductive workers . The study demonstrates the need for precise mathematical models of population dynamics and natural selection instead of informal arguments that are only based on considerations of genetic relatedness . | [
"Abstract",
"Introduction",
"Model",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"evolutionary",
"biology"
] | 2015 | The evolution of non-reproductive workers in insect colonies with haplodiploid genetics |
Genome instability in yeast and mammals is caused by RNA–DNA hybrids that form as a result of defects in different aspects of RNA biogenesis . We report that in yeast mutants defective for transcription repression and RNA degradation , hybrid formation requires Rad51p and Rad52p . These proteins normally promote DNA–DNA strand exchange in homologous recombination . We suggest they also directly promote the DNA–RNA strand exchange necessary for hybrid formation since we observed accumulation of Rad51p at a model hybrid-forming locus . Furthermore , we provide evidence that Rad51p mediates hybridization of transcripts to homologous chromosomal loci distinct from their site of synthesis . This hybrid formation in trans amplifies the genome-destabilizing potential of RNA and broadens the exclusive co-transcriptional models that pervade the field . The deleterious hybrid-forming activity of Rad51p is counteracted by Srs2p , a known Rad51p antagonist . Thus Srs2p serves as a novel anti-hybrid mechanism in vivo .
Genome instability can lead to a range of alterations in both the sequence and structure of chromosomes . While such changes may help drive evolution , more often they are associated with decreased organism fitness and increased susceptibility to disease ( Aguilera and Gómez-González , 2008 ) . Historically , most genome instability was thought to occur as the result of errors during replication , or the failure of DNA repair pathways . However , work in Saccharomyces cerevisiae and mammalian cells has demonstrated that genome instability also arises from lesions generated from the formation of RNA–DNA hybrids ( Huertas and Aguilera , 2003; Li and Manley , 2005; Kim and Jinks-Robertson , 2009; Paulsen et al . , 2009; Wahba et al . , 2011; Stirling et al . , 2012 ) . Many important aspects of hybrid-mediated genome instability remain to be elucidated . Genome-wide screens in budding yeast and human cells have revealed that levels of RNA–DNA hybrids increase when RNA biogenesis is disturbed at sites of transcription initiation or repression , elongation , splicing , degradation , and export ( Huertas and Aguilera , 2003; Li and Manley , 2005; Paulsen et al . , 2009; Wahba et al . , 2011; Stirling et al . , 2012 ) . The co-transcriptional binding of many RNA processing and transcription factors suggests that they prevent hybrid formation by restricting the access of nascent RNA molecules to the DNA template at the site of transcription ( Aguilera and García-Muse , 2012 ) . Recent studies suggest that these RNA biogenesis factors are not sufficient to prevent transient hybrid formation at some loci in wild-type budding yeast; rather , hybrids form but are removed rapidly by hybrid removal factors , including two endogenous RNase H enzymes , and Sen1p , an RNA–DNA helicase ( Mischo et al . , 2011; Wahba et al . , 2011 ) . In RNA biogenesis mutants the elevated levels of hybrid formation overwhelm the capacity of these hybrid removal factors , allowing the accumulation of hybrids and genome instability . While a number of factors that prevent hybrid formation or persistence have been identified , little is known about factors that promote the formation of RNA–DNA hybrids in vivo . One potential factor is RNA polymerase , which generates negative supercoiling behind the elongating polymerase . The negative supercoiling facilitates DNA unwinding , and may allow RNA access to the DNA template ( Roy et al . , 2010 ) . A connection between negative supercoiling and hybrid formation is supported by in vivo work on topoisomerase mutants in bacteria , yeast , and human cells ( Drolet et al . , 1995; Tuduri et al . , 2009; El Hage et al . , 2010 ) . Another potential promoter of hybrid formation is RecA , the bacterial strand exchange protein that normally promotes the invasion of single-stranded DNA into duplex DNA to repair DNA damage . Studies a decade ago showed that RecA promotes RNA–DNA hybrid formation in vitro ( Kasahara et al . , 2000; Zaitsev and Kowalczykowski , 2000 ) . This observation supported a model of RecA-dependent hybrid formation that had been postulated as an alternative mechanism to initiate DNA replication ( Cao and Kogoma , 1993; Hong et al . , 1995 ) . The intriguing possibility that RecA or its eukaryotic ortholog , Rad51p , might play a role in vivo to promote hybrid formation has not been pursued further . The in vitro studies on RecA-dependent hybrid formation also challenged ideas of when hybrid formation occurs . In these studies , RecA-mediated R-loops formed when RecA was mixed with RNA and its homologous DNA template in the absence of active transcription . This observation showed that hybrids could , at least in vitro , form post-transcriptionally ( i . e . , in trans ) ( Kasahara et al . , 2000; Zaitsev and Kowalczykowski , 2000 ) . However , to date , models postulated from in vivo studies suggest that RNA–DNA hybrids occur co-transcriptionally ( in cis ) from the invasion of the duplex DNA at the site of transcription by a nascent RNA transcript ( Huertas and Aguilera , 2003; Aguilera and García-Muse , 2012 ) . The cis mechanism is supported by the co-transcriptional nature of many of the RNA biogenesis steps implicated in preventing hybrid formation . However , the cis mechanism does not fully explain how mutants in post-transcriptional processes , such as RNA degradation and export , would cause hybrid-mediated instability ( Wahba et al . , 2011 ) . Therefore , the formation of RNA–DNA hybrids may occur in trans as well as in cis . The intriguing possibility that hybrids may occur in trans and contribute to genomic instability has not been assessed . In this work we used S . cerevisiae as the model system to test in vivo the role of Rad51p in hybrid formation . We report that the formation of RNA–DNA hybrids and associated genome instability in at least four RNA biogenesis mutants requires Rad51p and its activator , Rad52p . Furthermore , the deleterious hybrid-forming activity of Rad51p is suppressed in wild-type cells by Srs2p , a Rad51p inhibitor . Additionally , we developed a model locus system that allows us to monitor hybrid-mediated genome instability as a result of transcription . We manipulate this system to provide compelling evidence that hybrids and ensuing genome instability can occur via a trans mechanism that is dependent on Rad51p .
The conditions that drive the initial formation of RNA–DNA hybrids in vivo are not well understood . With the bacterial in vitro experiments in mind , we wondered whether hybrid formation was simply a strand exchange reaction , similar to that mediated by Rad51p during DNA repair and homologous recombination . To test this possibility , we examined the effect of deleting RAD51 on hybrid formation and the associated genome instability in RNA biogenesis mutants of budding yeast . We chose a representative set of mutants defective in elongation ( leo1Δ ) , repression ( med12Δ and sin3Δ ) , and degradation ( kem1Δ and rrp6Δ ) . We assayed directly for the presence of RNA–DNA hybrids in wild-type cells and these mutants by staining chromosomes in spread nuclei with S9 . 6 antibody ( see ‘Materials and methods’ ) . Previously , we demonstrated the specificity of the S9 . 6 antibody for hybrids by two approaches . First , S9 . 6 staining in spreads of RNA biogenesis mutants is reduced to that seen in wild-type cells by post treatment of chromosome spreads with RNase H ( Wahba et al . , 2011 ) . Similarly , spreads of an RNA biogenesis mutant over-expressing RNase H no longer stained with S9 . 6 . As reported previously , less than 5% of wild-type nuclei stain with this antibody ( Figure 1A , Figure 1—figure supplement 1 ) . In contrast , 80–85% of nuclei in our representative set of RNA biogenesis mutants showed robust staining , indicating the formation of stable hybrids at many loci in most cells ( Figure 1A , Figure 1—figure supplement 1 ) . The deletion of RAD51 ( rad51Δ ) in these mutants diminished S9 . 6 staining in nearly all nuclei from the RNA biogenesis mutants threefold to fourfold to near background levels ( Figure 1A , Figure 1—figure supplement 2 ) . To corroborate our cytological method , we isolated total nucleic acids from wild-type , sin3Δ ( a representative RNA biogenesis mutant ) , and sin3Δ rad51Δ cells , transferred them to a solid matrix and monitored binding of S9 . 6 . S9 . 6 binding to sin3Δ nucleic acids was elevated approximately tenfold relative to sin3Δ rad51Δ ( Figure 1—figure supplement 3 ) . These results strongly suggest that hybrid formation in these mutants is highly dependent upon Rad51p . 10 . 7554/eLife . 00505 . 003Figure 1 . Deletion of RAD51 suppresses RNA–DNA hybrids and YAC instability . ( A ) Left panel: Representative images of chromatin spreads stained with S9 . 6 antibody , showing reduced RNA–DNA hybrid staining in mutants with a deletion of RAD51 ( rad51Δ ) . Right panel: The percent of total nuclei scored that stain positively for RNA–DNA hybrid in chromatin spreads is quantified . A total of 50–100 nuclei from two independent experiments were scored for each genotype . ( B ) Rate of yeast artificial chromosome ( YAC ) instability in mutants is also reduced when RAD51 is deleted . Error bars represent standard deviation calculated from at least four independent colonies . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 00310 . 7554/eLife . 00505 . 004Figure 1—figure supplement 1 . Larger panels of chromatin spreads showing multiple nuclei of single mutants stained with S9 . 6 antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 00410 . 7554/eLife . 00505 . 005Figure 1—figure supplement 2 . Larger panels of chromatin spreads showing multiple nuclei of double mutants stained with S9 . 6 antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 00510 . 7554/eLife . 00505 . 006Figure 1—figure supplement 3 . Dot blotting with S9 . 6 antibody . Roughly 1 μg of DNA from indicated genotypes was spotted onto the membrane and stained with the S9 . 6 antibody . As a reference known amounts of pre-formed RNA–DNA hybrids were also spotted . Pre-formed RNA–DNA hybrids were made by performing a first strand synthesis reaction on total RNA . Amounts were quantified using Quant-iT Picogreen ( Invitrogen , Carlsbad , CA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 006 One prediction from the cytological results is that the suppression of hybrid formation by rad51Δ should also lead to the suppression of hybrid-mediated chromosome instability . To measure hybrid-mediated genome instability , we exploited an assay we developed previously using a yeast artificial chromosome ( YAC ) ( Wahba et al . , 2011; see ‘Materials and methods’ ) . The total rate of YAC instability ( the sum of chromosome loss and terminal deletions ) in wild-type cells was 6 × 10−4 per division . Notably , rad51Δ alone caused no increase in YAC instability . In our subset of RNA biogenesis mutants , YAC instability increased fivefold to tenfold ( Figure 1B ) . The introduction of rad51Δ into the RNA biogenesis mutants completely suppressed the elevated YAC instability , both chromosome loss and terminal deletions , in leo1Δ , kem1Δ , rrp6Δ , and sin3Δ mutants . In med12Δ , YAC instability was mostly but not entirely suppressed despite the near complete suppression of hybrid formation as monitored by spreads , indicating that in the med12Δ rad51Δ strain a subset of the YAC instability was hybrid independent . Overall , the suppression of hybrid-mediated chromosome instability by rad51Δ corroborates its elimination of RNA–DNA hybrids and associated destabilizing lesions . To further validate the occurrence of Rad51-dependent hybrids , we sought to develop a model locus that can be used to induce hybrid formation and hybrid-mediated instability at a known region . From our previous study on RNA biogenesis mutants that induce hybrids , we noted that many of these mutants allow cryptic transcription , and likely the production of aberrant transcripts ( Wyers et al . , 2005; Cheung et al . , 2008; Wahba et al . , 2011 ) . Based on this observation , we introduced a portion of the GAL1-10 promoter into the YAC ( henceforth referred to as YAC-GALpr ) , such that the addition of galactose to the media would induce GALpr-dependent transcription of neighboring non-yeast sequences ( Figure 2A ) . We analyzed transcription of the human and vector sequences flanking GALpr by qRT-PCR . This analysis revealed approximately one hundredfold induction of RNA at least 1 kb on both sides of the GAL promoter ( Figure 2B ) . 10 . 7554/eLife . 00505 . 007Figure 2 . Hybrid-mediated YAC instability is induced in wild-type when high rates of transcription are induced on the YAC using the GAL1-10 promoter ( GALpr ) . ( A ) Schematic of the YAC-GALpr construct . Total yeast artificial chromosome ( YAC ) length is 350 kb , of which 324 kb come from human chromosome VII . The GALpr was integrated 10 kb from the telomere , on the arm with the URA3 marker . ( B ) Quantitative RT-PCR monitoring changes in RNA levels on the YAC 5 hr after induction with galactose . YAC RNA is normalized to actin RNA , and represented as fold change , as compared to RNA levels detected in uninduced cells . Above the table is a schematic representation of the YAC region from which RNA is measured , with the qRT-PCR fragments used in quantification indicated with black dashes . The region in gray represents the GAL1-10 promoter and selectable marker integrated in the YAC-GALpr strain . ( C ) DIP analysis to monitor RNA–DNA hybrid formation in the YAC-GALpr strain in the absence of galactose , and 2 hr after induction with galactose . Error bars represent standard deviation calculated from two independent DIP experiments . ( D ) Rates of YAC instability in strains with YAC ( black bars ) or YAC-GALpr ( gray bars ) 5 hr after addition of galactose to the media . Strains carried either an RNase H over-expressing plasmid or an empty control vector . ( E ) Induced YAC instability is suppressed when RAD51 is deleted . Error bars represent standard deviation calculated from at least three independent colonies . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 00710 . 7554/eLife . 00505 . 008Figure 2—figure supplement 1 . ( A ) DIP analysis of YAC strain prior to and 2 hr after addition of galactose to the media . ( B ) Monitoring of DIP signal in the YAC-GALpr strain at a distal region , showing low levels of hybrid signal upon induction with galactose as compared to . ( C ) DIP signals are reduced around the YAC-GALpr module upon return to repressive conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 00810 . 7554/eLife . 00505 . 009Figure 2—figure supplement 2 . The percent of terminal deletions and chromosome loss events recovered after 5 hr of growth in galactose-containing media is comparable for YAC and YAC-GALpr strains . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 009 Using the model locus , we monitored the presence of transcription-induced hybrids specifically proximal to GALpr . Total nucleic acids were isolated from strains containing either the YAC or YAC-GALpr in the presence or absence of galactose . These samples were subjected to DNA immunoprecipitation ( DIP ) analysis with the S9 . 6 antibody that should only precipitate DNA in RNA–DNA hybrids ( Mischo et al . , 2011; see ‘Materials and methods’ ) . Using primers specific to the YAC region proximal to the GALpr , low DIP signals were observed in YAC-GALpr cultures in the absence of galactose , as well as in cultures with the YAC , with and without the addition of galactose ( Figure 2C , Figure 2—figure supplement 1A ) . Thus , hybrids form rarely in the YAC sequences proximal to the GALpr in the absence of their transcription . In contrast , we observed a dramatic increase in the DIP signal for hybrids on the YAC sequences proximal to YAC-GALpr , 2 hr after induction by galactose ( Figure 2C ) . The specificity of this increased DIP signal was evident by the fact that no elevation in hybrid signal was detected in two regions of the YAC-GALpr distal to the GALpr ( Figure 2—figure supplement 1B ) . Additionally , lower DIP signals coincide with the transcriptional start site of the GALpr , where there is little transcript detectable upon addition of galactose ( Figure 2C ) . Furthermore , the DIP signal in the YAC-GALpr strains was suppressed when transcription was repressed by the addition of glucose ( Figure 2—figure supplement 1C ) . Finally , hybrid formation at YAC-GALpr was dependent upon RAD51 ( see ‘Rad51p-dependent hybrid formation can occur in trans’ ) . These data provide molecular evidence for the formation of RAD51-dependent hybrids at the YAC sequences transcribed by induction of GALpr . To determine whether the Rad51p-dependent hybrids induced by YAC-GALpr led to genome instability , we monitored the instability of YAC-GALpr upon galactose treatment . Indeed , its instability was elevated 25-fold with a distribution of chromosome loss and terminal deletions similar to that seen in wild-type cells and RNA biogenesis mutants ( Figure 2 , Figure 2—figure supplement 2 ) . Furthermore , this transcription-induced YAC instability was suppressed by over-expression of RNase H or deletion of RAD51 ( Figure 2D–E ) . Thus both by DIP and YAC instability , hybrids induced by transcription at the model YAC-GALpr locus , like those induced by RNA biogenesis mutants , required Rad51p for their formation . A second prediction concerning Rad51p-mediated hybrid formation is that Rad51p should be detectable near sites of hybrid formation . To test this prediction we used our YAC-GALpr model locus to assay for the presence of Rad51p binding around the site of hybrid formation . We generated cultures of strains containing the YAC or YAC-GALpr that had been grown in the presence or absence of galactose . These cultures were fixed and assayed for Rad51p binding to the YAC sequences by chromatin immunoprecipitation ( ChIP ) ( see ‘Materials and methods’ ) . ChIP was performed using two independent antibodies , anti-HA against a C-terminal haemagglutinin ( HA ) tagged Rad51p and a polyclonal rabbit anti-Rad51p . No Rad51p binding was detected either on YAC-GALpr in the absence of galactose or on the YAC in the presence or absence of galactose ( Figure 3A ) . Thus the level of Rad51p binding to the YAC or vector sequences in the absence of transcription was very low if any . In contrast , using either antibody for ChIP , significant Rad51p binding was detected around the GAL promoter on YAC-GALpr upon the addition of galactose and induction of transcription ( Figure 3A , Figure 3—figure supplement 1 ) . Notably , Rad51p binding appears to extend further than the region of hybrid formation detected by DIP ( Figure 2C and Figure 3A ) . Rad51p is known to spread from regions of ssDNA into dsDNA ( Zaitsev and Kowalczykowski , 2000 ) , and it is possible that in our model locus Rad51p is spreading from the ssDNA or RNA–DNA hybrid into the neighboring dsDNA . To test further the correlation of transcription and Rad51p binding , we added dextrose to the galactose-treated YAC-GALpr cultures to repress galactose-induced YAC-GALpr transcription ( see ‘Materials and methods’ ) . In these cultures , Rad51p binding disappeared ( Figure 3—figure supplement 2 ) . Taking these findings together , we observe Rad51p binding to the region of the hybrid-forming locus on the YAC-GALpr only when transcripts from this region are induced . 10 . 7554/eLife . 00505 . 010Figure 3 . Rad51p binding is detectable around the YAC-GALpr module upon induction of transcription . Cells growing exponentially in YEP-lactic acid were split , and galactose added to one half . The other half was collected immediately for the –Gal sample and fixed for chromatin immunoprecipitation ( ChIP; see ‘Materials and methods’ ) . After 120 min , the +Gal sample was similarly fixed for ChIP . Input DNA and DNA coimmunprecipitated with α-HA or -γ-H2a . X ( IP ) antibody were amplified using primer sets along the yeast artificial chromosome ( YAC ) as annotated with black dashes on the YAC-GALpr or YAC schematic above each graph . ( A ) ChIP of Rad51-HA in the YAC-GALpr strain shows an increased signal in Rad51-HA binding 2 hr after induction of transcription by addition of galactose to the media ( top panel ) . No change in the RAD51-HA signal is observed in the YAC strain ( bottom panel ) . ( B ) ChIP of γ-H2a . X in YAC-GALpr reveals no significant change in signal within 2 hr of galactose induction . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 01010 . 7554/eLife . 00505 . 011Figure 3—figure supplement 1 . Rad51p binding is detectable around the YAC-GALpr module upon induction of transcription . Chromatin immunoprecipitation ( ChIP ) of untagged Rad51 using a polyclonal antibody , showing an increased signal in Rad51 binding 2 hr after induction of transcription with galactose . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 01110 . 7554/eLife . 00505 . 012Figure 3—figure supplement 2 . Rad51p binding is reduced around the YAC-GALpr module upon return to repressive conditions . After 2 hr of growth in galactose-containing media , dextrose was then added to the media , and cells were allowed to grow , maintained in exponential phase , for 3 hr . Cells were then fixed and used for chromatin immunoprecipitation ( ChIP ) of Rad51-HA . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 01210 . 7554/eLife . 00505 . 013Figure 3—figure supplement 3 . Rad51 and γ-H2a . X binding at an inducible break site on chromosome III . ( A ) Chromatin immunoprecipitation ( ChIP ) of Rad51-V5 around a double-strand break induced by a site-specific HO endonuclease under control of a galactose inducible promoter . The graph shows levels of Rad51-V5 binding prior to , and 2 hr after galactose was added to the media . ( B ) ChIP of γ-H2a . X around the break site prior to , and 2 hr after galactose was added to the media . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 01310 . 7554/eLife . 00505 . 014Figure 3—figure supplement 4 . Wild-type levels of YAC instability are observed after 2 hr of transcription induction . YAC-GALpr strains were grown for 2 hr in galactose-containing media , followed by addition of dextrose and growth for 3 more hours . Error bars represent standard deviation calculated from two independent colonies . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 014 We propose that the binding of Rad51p observed at the model locus is due to its role in hybrid formation . However , hybrids are thought to induce double-strand breaks ( DSBs ) , and Rad51p binds at DSBs to initiate DNA repair through homologous recombination ( Sugawara et al . , 2003; Figure 3—figure supplement 3A ) . Therefore , the presence of Rad51p at the hybrid-forming locus might be due to its function in repair rather than in hybrid formation . To address this alternative explanation for Rad51p binding , we performed molecular and functional tests for the formation of DSBs 2 hr after the induction of transcription . As a molecular assay , we monitored a 20 kb region surrounding the GAL promoter for the accumulation of phosphorylated histone H2AX ( γ-H2AX ) by ChIP . This modification is one of the most dramatic and earliest markers of DSB formation , arising within minutes and spanning large regions of chromatin adjacent to the break ( Shroff et al . , 2004; Figure 3—figure supplement 3B ) . However , we did not detect a ChIP signal for γ-H2AX above background level in the YAC-GALpr strain even under conditions that induced Rad51p binding ( Figure 3B ) . Thus by this molecular assay Rad51p binding occurs at the site of hybrid formation prior to hybrid-induced DNA damage . As a functional test , we took advantage of the fact that adding dextrose after 2 hr suppressed transcription and Rad51p binding at the model hybrid locus . We reasoned that if Rad51p binding during the 2 hr prior to the addition of dextrose reflected Rad51p association with hybrid-induced DNA damage , then this damage would manifest as increased YAC instability . However , no increase in YAC instability was observed ( Figure 3—figure supplement 4 ) , indicating that binding of Rad51p to this locus during the first 2 hr was unlikely to result from DNA damage . Thus neither our molecular nor functional test supports the binding of Rad51p to the model locus prior to hybrid-induced DNA damage , pointing to a direct role of Rad51p in hybrid formation . Does the formation of all hybrids require Rad51p ? Studies from our laboratory and the Aguilera laboratory suggest that hybrids not only form in RNA biogenesis mutants but also transiently in wild-type cells ( Mischo et al . , 2011; Wahba et al . , 2011 ) . The latter fail to persist because of their rapid removal by RNases H and Sen1 ( Mischo et al . , 2011; Wahba et al . , 2011 ) . To test whether these naturally occurring hybrids are also dependent on Rad51p , we monitored hybrid staining and YAC instability in rnh1Δrnh201Δ in the absence of RAD51 . Neither hybrid staining nor YAC instability was suppressed ( Figure 4A , B ) , indicating that the transient hybrids in wild-type cells are not Rad51p dependent . Thus both Rad51p-dependent and -independent mechanisms for hybrid formation exist . 10 . 7554/eLife . 00505 . 015Figure 4 . Deletions of RAD51 and RAD52 do not affect RNA–DNA hybrid formation in rnh1Δrnh201Δ . ( A ) Representative images of chromatin spreads stained with S9 . 6 antibody . ( B ) Rate of yeast artificial chromosome ( YAC ) instability is similar in rnh1Δrnh201Δ and strains lacking RAD51 ( rad51Δ ) or RAD52 ( rad52Δ ) . Error bars represent standard deviation calculated from at least six independent colonies . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 015 In the in vitro bacterial studies , RecA promoted hybrid formation in the absence of active transcription , suggesting that RNA–DNA hybrids can form post-transcriptionally , or in trans . To test whether in vivo hybrids could form in trans , we constructed a strain , LW7003 , in which chromosome III contained 3 . 5 kb of vector and human sequences surrounding the galactose promoter of the YAC-GALpr ( henceforth referred to as the YAC-GALpr module ) . This strain also contained the original unmodified YAC , allowing us to investigate whether transcription of the YAC-GALpr module on chromosome III could induce both hybrid formation on the YAC and YAC instability ( Figure 5A ) . 10 . 7554/eLife . 00505 . 016Figure 5 . Transcription of YAC sequences from chromosome III causes RNA–DNA hybrid formation in trans on the YAC . ( A ) Schematic representation of the trans assay is depicted . The GALpr , selectable marker ( CLONAT ) , and 1 . 1 kb of yeast artificial chromosome ( YAC ) DNA was integrated on chromosome III . ( B ) Schematic representation of where the primer sets used to monitor hybrid formation in trans are depicted . Hybrid formation is monitored by DIP 2 hr after induction with galactose in RAD51 and rad51Δ strains . Error bars represent standard deviation from two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 016 To test directly whether hybrids can form in trans , DIP was performed on cultures of our LW7003 strain after growth in the presence or absence of galactose . One primer set that monitored hybrids from both the YAC and YAC-GALpr module generated a strong DIP signal only in the presence of galactose ( Figure 5B , primer 1 ) . This combined hybrid signal was eliminated when the rad51Δ was introduced in this strain ( Figure 5B , primer 1 ) . These results minimally corroborate our previous demonstration of hybrids forming in cis and show that hybrid formation is dependent upon Rad51p . Two other primer sets that monitored hybrids only from the YAC also revealed a strong DIP signal only in the presence of galactose ( Figure 5B , primers 2 and 3 ) . These results demonstrated transcription-dependent hybrid formation in trans . This trans-specific hybrid signal was eliminated when rad51Δ was introduced into our strain . The RAD51-dependent DIP results strongly support the formation of Rad51p-dependent hybrids in trans . We also tested for hybrid formation on the YAC in trans by monitoring YAC instability in LW7003 . As expected , no increase in YAC instability was observed in this strain in the absence of galactose ( Figure 6A ) . However , YAC instability increased tenfold upon galactose-induced transcription of the YAC-GALpr module on chromosome III ( Figure 6A , black bars ) . The transcription-induced YAC instability was dependent on the homology between the YAC and the transcribed YAC sequences from the YAC-GALpr module on chromosome III , as deletion of the corresponding 1 kb of homology from the YAC completely suppressed the transcription-induced YAC instability ( Figure 6—figure supplement 1 ) . The elevated YAC instability was blocked by RNase H over-expression , indicating the YAC instability was hybrid dependent ( Figure 6A , gray bars ) . YAC instability was also blocked after introduction of the rad51Δ in LW7003 ( Figure 6B ) . Thus transcription from the YAC-GALpr module on chromosome III acted in trans to cause the YAC to rearrange through a hybrid- and Rad51p-dependent mechanism . 10 . 7554/eLife . 00505 . 017Figure 6 . Transcription of YAC sequences in trans causes hybrid-mediated YAC instability . ( A ) Rates of yeast artificial chromosome ( YAC ) instability in strains carrying an empty control vector ( black bars ) or RNase H over-expressing vector ( gray bars ) showing an increased rate of instability upon induction of transcription that is reduced when RNase H is over-expressed . Error bars represent standard deviation calculated from at least three independent colonies . ( B ) Rate of YAC instability is suppressed when RAD51 is knocked out . ( C ) Pulse-field gel and Southern analysis with HIS3 probe of FOAresistant , His+ colonies , showing that 9/10 colonies analyzed have YACs rearranged to a smaller size . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 01710 . 7554/eLife . 00505 . 018Figure 6—figure supplement 1 . Levels of YAC instability in the trans assay with and without a region of homology on the YAC . The 1 kb region of the yeast artificial chromosome ( YAC ) inserted on chromosome III ( YAC-GALpr module ) was replaced on the YAC with a LEU2 cassette . Wild-type levels of YAC instability are observed in that strain after 2 hr of induction with galactose . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 01810 . 7554/eLife . 00505 . 019Figure 6—figure supplement 2 . Schematic representation of an alternative for how HIS+ URA− colonies may arise in the trans assay . Upon induction of transcription , breaks may occur in cis on chromosome III . Repair via break-induced replication using homologous yeast artificial chromosome ( YAC ) sequences as a substrate can lead to a chromosome III;YAC translocation . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 019 If the YAC instability induced by the YAC-GALpr module on chromosome III is mediated by hybrids formed in trans on the YAC , then these hybrids should lead to a similar distribution of YAC loss and terminal deletion as hybrids induced in cis . Indeed hybrids induced in trans and in cis both lead to a similar distribution of YAC instability events; on average 85% are HIS− URA− ( chromosome loss ) and 15% are HIS+ URA− ( putative terminal deletions ) . However , the total rate of YAC instability increased only 10-fold by hybrids formed in trans ( from the YAC-GALpr module on chromosome III ) compared to 25-fold by hybrids formed in cis . Thus , hybrid formation in trans may be less efficient than in cis . While we have assumed that HIS+ URA− clones of LW7003 reflect terminal deletions of the YAC , these clones may have had rearrangements that occurred by an indirect mechanism as a result of hybrid-induced double strand breaks in cis . In this model hybrids would form in cis at the module on chromosome III and cause DSBs there . These DSBs in cis would induce recombination between the YAC sequences on the broken chromosome III and the YAC , resulting in a chromosome III;YAC translocation that has the same genetic phenotype ( HIS+ URA− ) as YAC terminal deletions ( Figure 6—figure supplement 2 ) . To determine what fraction of rearrangements may have occurred by this indirect mechanism , we performed pulse-field gel and Southern analysis on DNA isolated from 10 independent HIS+ URA− colonies of LW7003 . Amongst the 10 YAC rearrangements analyzed , nine were shorter than the existing YAC , consistent with the formation of YAC terminal deletions ( Figure 6C ) . Only one rearrangement was the size expected if a chromosome III;YAC translocation had occurred . Thus the structure of most rearranged YACs in LW7003 is consistent with the formation of terminal deletions through the formation of hybrids in trans . These results further support our hypothesis that hybrids can form in trans by a Rad51p mechanism , causing chromosome instability at sites distinct from the site of hybrid RNA transcription . During homologous recombination , the activity of Rad51p is regulated by a number of factors that modulate Rad51p binding to ssDNA and dsDNA ( Krejci et al . , 2003; Sugawara et al . , 2003 ) . Because of the importance of such accessory factors for Rad51p function , we wondered whether they might also help regulate Rad51 in hybrid formation . To test this , we deleted positive and negative regulators of Rad51–DNA filament formation . Rad52p is required for the binding of Rad51p to ssDNA ( Figure 7A; Song and Sung , 2000 ) . Deletion of RAD52 ( rad52Δ ) in our panel of transcriptional mutants completely suppressed hybrid staining , as assayed by chromosome spreads ( Figure 7B , Figure 7—figure supplement 1 ) . Note that we were unable to test suppression of YAC instability in the double mutants because the rad52Δ alone caused substantial hybrid-independent YAC instability , an expected result given its central role in many repair pathways . Nonetheless , the suppression of hybrid staining by rad52Δ suggests that hybrid formation is not simply a consequence of rogue activity by Rad51p but rather occurs as part of the canonical Rad51p repair pathway . 10 . 7554/eLife . 00505 . 020Figure 7 . Deletion of RAD52 suppresses RNA–DNA hybrids . ( A ) Schematic showing the major proteins canonically involved in regulating Rad51p binding in DNA repair . Following resection , replication protein A ( RPA ) polymerizes onto ssDNA . Rad52 then interacts with RPA and catalyzes its exchange for Rad51p . The Rad51–ssDNA filament promotes the pairing and strand exchange reaction with a homologous region in duplex DNA . Srs2p , Rad54p , and Rdh54p all regulate the Rad51 filament by dismantling Rad51 from ssDNA and dsDNA , respectively . ( B ) Representative images of chromatin spreads stained with S9 . 6 antibody and quantification of nuclei , showing reduced RNA–DNA hybrid staining in mutants with RAD52 knocked out . A total of 50–100 nuclei from two independent experiments were scored for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 02010 . 7554/eLife . 00505 . 021Figure 7—figure supplement 1 . Larger panels of chromatin spreads showing multiple nuclei of rad52Δ mutants stained with S9 . 6 antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 021 A number of inhibitors of Rad51p have been identified . SRS2 is a helicase involved in removing Rad51p filaments formed on ssDNA ( Krejci et al . , 2003 ) , and Rad54p and Rdh54p are two translocases that promote the removal of Rad51p from double-stranded DNA ( Shah et al . , 2010 ) . We wondered whether these inhibitors might help suppress the rogue hybrid-forming activity of the Rad51p pathway in wild-type cells . To test this we deleted SRS2 , RAD54 , and RDH54 from cells and measured hybrid formation and YAC instability . Neither single nor double deletions of RAD54 and RDH54 significantly increased hybrid formation or YAC instability ( Figure 8A , B , Figure 8—figure supplement 1 ) . In contrast , deletion of SRS2 increased both hybrid staining and YAC instability . Both of these phenotypes of the srs2Δ were suppressed in the srs2Δ rad51Δ mutant ( Figure 8C ) . Thus , Srs2p antagonizes the hybrid-forming activity of the Rad51p pathway and represents another mechanism by which cells protect their genome against hybrid formation . 10 . 7554/eLife . 00505 . 022Figure 8 . Deletion of SRS2 , but not RAD54 and RDH54 increases genome instability and hybrid formation . ( A ) Rate of yeast artificial chromosome ( YAC ) instability is increased in srs2Δ but not in rad54Δ and rdh54Δ single and double mutants . Error bars represent standard deviation calculated from at least six independent colonies . ( B ) Representative images of chromatin spreads stained with S9 . 6 antibody , showing increased RNA–DNA hybrid staining in srs2Δ . ( C ) Left panel: Rate of YAC instability in srs2Δ is suppressed when RAD51 is knocked out . Right panel: Hybrid staining is also reduced in the srs2Δ rad51Δ double mutant . ( D ) srs2Δ mutants with URA3 integrated at the rDNA were assayed for loss of the URA3 marker , showing increased instability . Error bars represent standard deviation calculated from at least six independent colonies . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 02210 . 7554/eLife . 00505 . 023Figure 8—figure supplement 1 . Larger panels of chromatin spreads showing multiple nuclei of srs2Δ and rdh54Δrad54Δ mutants stained with S9 . 6 antibody . Quantification of number of nuclei that stained positively for RNA–DNA hybrids in srs2Δ and rdh54Δrad54Δ . Total number of nuclei scored is 50–100 per genotype , from two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 023 The hybrid staining pattern in srs2Δ nuclei was reminiscent of the pattern observed in sin3Δ cells , exhibiting an apparent enrichment of RNA–DNA hybrids at the RDN locus on chromosome XII , the site of 150 tandem rDNA copies ( Wahba et al . , 2011 ) . We measured rDNA instability by monitoring the rate of excision of a URA3 marker inserted at the RDN locus ( Heidinger-Pauli et al . , 2010 ) . In srs2Δ cells , the rate of rDNA instability is twenty threefold greater compared to wild-type cells , a marked increase in instability as compared to the fourfold increase in YAC instability ( Figure 8C , D ) . Together these results suggest that Srs2p has a particularly important role in protecting the highly transcribed rDNA locus against Rad51p-dependent hybrid formation and repeat instability .
In this study we describe a compelling series of observations that demonstrate an in vivo role for Rad51p in promoting formation of RNA–DNA hybrids . First , cytological detection of RNA–DNA hybrids in RNA biogenesis mutants is dramatically suppressed when RAD51 is deleted . Second , Rad51p is required for the associated hybrid-mediated instability of a YAC in the RNA biogenesis mutants . Third , Rad51p is required for the hybrid formation and YAC instability that results from galactose-induced transcription of specific YAC sequences in our hybrid model locus . Fourth , cofactors canonically known to regulate Rad51p binding and function also modulate hybrid formation . Removal of Rad52p , a positive regulator of Rad51p , blocks hybrid formation . Conversely , the removal of Srs2p , a negative regulator of Rad51 filament formation , increases both hybrid formation and genome instability in a Rad51p-dependent manner . Finally , Rad51p binds to the YAC at the site of hybrid formation proximal to the galactose promoter , in a transcription-dependent manner , prior to any evidence that DSBs have formed . Taken together these results strongly suggest that Rad51p plays a direct role in RNA–DNA hybrid formation . Establishing a role for Rad51p in the formation of RNA–DNA hybrids marks the first direct in vivo evidence of a factor that facilitates hybrid formation . Indications that a strand-exchange mechanism is important for hybrid formation were suggested by in vitro work on RecA , where RecA catalyzes assimilation of complementary RNA into a homologous region in duplex DNA ( Kasahara et al . , 2000; Zaitsev and Kowalczykowski , 2000 ) . Interestingly , previous in vivo studies showed that over-expressing Rad51p can compromise genome integrity ( Richardson et al . , 2004; Shah et al . , 2010 ) . We suggest that this instability may result in part from the ability of Rad51p to promote RNA–DNA hybrid formation . This deleterious activity of Rad51p is also intriguing because a number of studies have shown that RAD51 expression is up-regulated in tumor cells ( Klein , 2008 ) . The increased expression is part of a coordinated up-regulation of DNA repair proteins in response to increased damage in cancerous cells ( Zhou and Elledge , 2000 ) . These high levels of Rad51p have been interpreted as evidence for Rad51p acting as a tumor suppressor by ensuring non-faulty repair of DNA damage . However , our results are consistent with a different interpretation: high levels of Rad51p may be oncogenic , driving hybrid-mediated genomic instability that promotes carcinogenesis . How can Rad51p promote hybrid formation ? In the forward reaction , the conventional mechanism for Rad51p action , it forms a filament on ssDNA , finds a homologous region of dsDNA , and then catalyzes a strand exchange ( Sung , 1994 ) . By analogy , Rad51p may form a filament on RNA and promote its invasion of dsDNA . However , the bacterial studies of RecA suggest an alternative mechanism in which Rad51 catalyzes RNA–DNA hybrid formation through an inverse strand exchange reaction . In this case , RecA first binds to ssDNA in a gap , forms a filament on adjacent dsDNA , and then promotes pairing and exchange with complementary RNA . Each of these mechanisms has strengths and weaknesses to explain our current findings . For example , the forward reaction but not the inverse strand exchange predicts the association of Rad51p with the hybrid locus should depend upon the induction of the hybrid forming RNA , as we observe . Conversely , the DNA-based inverse strand reaction more easily explains the role of DNA-dependent Rad51p cofactors like Rad52p . Furthermore , in vitro RecA is unable to catalyze the forward reaction with RNA . Clearly an exciting future direction will be to determine the biochemical nature of Rad51p-mediated hybrid formation by either of these mechanisms or an alternative mechanism like stabilization of the R-loop by binding to the RNA–DNA hybrid or the extruded DNA strand ( Figure 9A–C ) . 10 . 7554/eLife . 00505 . 024Figure 9 . Three models for how Rad51p may mediate RNA–DNA hybrid formation . ( A ) In the forward reaction , Rad51p polymerizes onto RNA , and mediates strand exchange with homologous DNA , forming an RNA–DNA hybrid . ( B ) In the inverse reaction , Rad51p forms a filament on dsDNA and promotes strand exchange with homologous RNA . ( C ) A third alternative is that Rad51 forms a filament on the extruded ssDNA , stabilizing an open D-loop that allows RNA to bind to homologous sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 00505 . 024 Here we show that Rad51p allows transcripts of YAC sequences generated on chromosome III to act in trans to form hybrids and cause hybrid-mediated instability of the YAC . Thus RNA molecules , with the aid of Rad51p can invade duplex DNA to form RNA–DNA hybrids at sites distinct from the site of hybrid RNA transcription . The ability of hybrids to form in trans forces a broadening of previous in vivo models for hybrid formation that only considered co-transcriptional mechanisms . The lower level of instability observed in trans compared to in cis implies that hybrid formation occurs less efficiently in trans . A lower efficiency might be expected since the RNA concentration at a cis site of hybrid formation will invariably be higher than at a trans site of hybrid formation . Indeed , an effect for RNA concentration on hybrid formation has been documented in the in vitro reactions with RecA ( Zaitsev and Kowalczykowski , 2000 ) . While we show that Rad51p is clearly required to mediate hybrid formation in trans , it is likely also important for hybrid formation in cis . However , the fact that rad51Δ has no effect on hybrid formation in the rnh1Δrnh201Δ mutant implies that alternative mechanisms for hybrid formation , presumably in cis , exist . Notably , hybrid formation in trans might be a more potent promoter of genome instability than hybrid formation in cis , particularly for hybrid RNA generated from a highly repetitive element , which in trans can cause instability at a plethora of targets . The formation of hybrids in trans has potential applications beyond genome instability . Intriguingly , recent studies in mammalian cells and budding yeast have developed a system for generating targeted DNA breaks using the CRISPR system , where target specificity is determined by a guide RNA complementary to the region of interest ( Jinek et al . , 2012; DiCarlo et al . , 2013 ) . It is unknown how the guide RNA finds and hybridizes to its homologous DNA region . From our work , it is clear that these two steps could be mediated through Rad51p . Additionally , positive roles for hybrid formation in modulating transcription state have been found in both human cells and fission yeast ( Ginno et al . , 2012; Nakama et al . , 2012 ) . In fission yeast , there is evidence that RNA–DNA hybrids may provide a platform for RNAi-mediated heterochromatin formation , driving transcriptional silencing ( Nakama et al . , 2012 ) . If transcripts from one locus can induce hybrid formation and modulate heterochromatin formation at a homologous locus , this could provide a rapid mechanism for the silencing of repetitive elements—including transposable elements—in a genome . Cells have developed a number of mechanisms to keep RNA–DNA hybrids in check . We provide evidence for a novel Srs2p-dependent pathway that limits the formation of Rad51p-dependent hybrids . We show that deletion of SRS2 causes Rad51-dependent hybrid formation and YAC instability . Srs2p interacts directly with Rad51p , dismantling inappropriate Rad51p filaments from DNA ( Krejci et al . , 2003; Burgess et al . , 2009 ) . By removing Rad51p from DNA , Srs2p could potentially inhibit hybrid formation by a mechanism like the inverse strand exchange . Alternatively , Srs2p might be playing a direct role in dismantling RNA–DNA hybrids . Its bacterial homolog UvrD has been shown to catalyze the unwinding of RNA–DNA hybrids in vitro ( Matson , 1989 ) . The work reported here , coupled with previous studies , has revealed that cells possess at least four lines of defense against RNA–DNA hybrids: ( 1 ) suppressing the deleterious hybrid-forming activity of Rad51p by Srs2p; ( 2 ) suppressing the amount of RNA in the nucleus with hybridization potential through proper RNA biogenesis; ( 3 ) unwinding RNA–DNA hybrids by helicases such as Sen1p; and ( 4 ) degrading RNA in RNA–DNA hybrids by RNases H . The fact that hybrids form when any one of these anti-hybrid pathways is abrogated clearly indicates that they are not completely functionally redundant . The increased propensity for hybrids to form and cause instability at the rDNA locus in srs2Δ cells is particularly intriguing , as it may be that the abundance of rRNA , as well as homologous rDNA loci , makes this region a particularly good substrate for Rad51p-mediated hybrid formation . When Rad51p activity is no longer limiting because of inactivation of Srs2p , hybrid formation may overwhelm the anti-hybrid activities of RNases H or Sen1p . Alternatively , the RNases H or Sen1 may be occluded from the nucleolus , making the rDNA locus more susceptible to hybrid formation by elevated Rad51p activity . It will be exciting to map where hybrids form in RNase H , sen1-1 and srs2Δ mutants to elucidate whether these different anti-hybrid systems are dedicated to protect different regions of the genome .
Full genotypes for the strains used in this study are listed in Supplementary file 1A . Strain LW6811 , the YAC-GALpr strain , was made by integrating the GAL1-10 promoter along with the selectable marker CLONAT at site 323 , 280 kb on the YAC . The trans YAC module in LW7003 encompasses 1 kb of the YAC , along with the GALpr and CLONAT marker integrated on chromosome III in place of the BUD5 ( YCR038C ) open reading frame . All integrations were done using standard one-step PCR techniques . The 70mers used for integration are listed in Supplementary file 1B . The empty control and RNase H plasmids used are 2μ plasmids , previously described in Wahba et al . , 2011 . Yeast strains were grown in YEP or minimal media supplemented with 2% glucose . 5-Fluoroorotic ( 5-FOA ) was purchased from BioVectra ( Charlottetown , PE ) . Cells were dilution streaked out on SC-URA plates to select for the YAC terminal marker ( URA3 ) . Single colonies were then picked and resuspended in 0 . 5 ml of water , diluted , and 105 cells were plated onto 5-FOA and –HIS 5-FOA plates . Plating efficiency was monitored by plating 200 cells onto rich media plates . Plates were incubated at 30°C for 3 d after which the number of colonies formed on each plate was counted . The number of colonies that grow on 5-FOA , normalized for plating efficiencies , is a measure of the rate of events . Chromosome spreads were performed as previously described ( Wahba et al . , 2011 ) . Slides were incubated with the mouse monoclonal antibody S9 . 6 directed to RNA–DNA hybrids , and available in the hybridoma cell line HB-8730 . The primary antibody was diluted 1:2000 in blocking buffer ( 5% BSA , 0 . 2% milk , 1× PBS ) for a final concentration 0 . 25 μg/ml . The secondary Cy3-conjugated goat anti-mouse antibody ( No . 115-165-003 ) was obtained from Jackson ImmunoResearch ( West Grove , PA ) and diluted 1:2000 in blocking buffer . Indirect immunofluorescence ( IF ) was observed using an Olympus IX-70 microscope with a 100×/NA 1 . 4 objective , and Orca II camera ( Hamamatsu , Bridgewater , NJ ) . Cells were picked from SC-URA plates , resuspended in SC-URA media , and grown to saturation . Fresh YEP or -URA media with 2% lactic acid , 3% glycerol was inoculated to an optical density ( OD ) of ∼0 . 3 , and allowed to double to an OD of ∼1 . 0 . Galactose was then added to a final concentration of 2% . Cells were shaken at 30°C and then plated onto 5-FOA 0 , 2 and 5 hr after induction with galactose . Plating efficiency was monitored by plating 200 cells onto rich media plates . Genomic DNA was isolated using the Qiagen Genomic DNA kit ( Qiagen , Hilden , Germany ) . Roughly 1 µg of DNA was resuspended to a final volume of 50 µl in nuclease-free water , and spotted directly onto a nylon GeneScreen Plus membrane ( NEF988; PerkinElmer , Waltham , MA ) using a Bio-Dot Microfiltration Apparatus ( Bio-Rad , Hercules , CA ) . The membrane was UV-crosslinked and blocked with 5% 1× PBS/0 . 1% Tween-20 prior to incubation with primary and secondary antibodies . A 5 µg aliquot of S9 . 6 antibody was used for the primary , and a 25 , 000× dilution of goat anti-mouse HRP ( Bio-Rad ) was used as the secondary . The HRP signal was developed with Clarity Western ECL Substrate ( Bio-Rad ) and exposed to autoradiography film . Total RNA was isolated using an RNeasy Mini Kit ( Qiagen ) . Reverse transcriptase was carried out with specified primer pairs using the OneStep RT-PCR Kit ( Qiagen ) and quantified using SYBR Green ( Invitrogen , Carlsbad , CA ) and the DNA Engine Opticon Continuous Fluorescence Detection System ( CMJ Research ) . DIP analysis was performed as previously described ( Mischo et al . , 2011; Alzu et al . , 2012 ) . Briefly , 150–200 µg of genomic DNA isolated using the Qiagen Genomic DNA kit was sonicated , precipitated , and resuspended in 50 µl of nuclease-free water . Then 350 µl of FA buffer ( 1% Triton X-100 , 0 . 1% sodium deoxycholate , 0 . 1% SDS , 50 mM HEPES , 150 mM NaCl , 1 mM EDTA ) was added to the DNA , and incubated for 90 min with 5 µg of S9 . 6 antibody prebound to magnetic protein A beads . Beads were then washed and the DNA eluted according to standard ChIP protocols . % RNA–DNA hybrid amounts were quantified using quantitative PCRs on DNA samples from DIP and total DNA with the DyNAmo HS SYBR Green qPCR kit ( Thermo Scientific , Waltham , MA ) . Cells used for ChIP experiments were grown in YEP media with 2% lactic acid , 3% glycerol and collected either before galactose was added ( −Gal ) or 2 hr after the addition of galactose at a final 2% concentration ( +Gal ) . Standard ChIP was performed as described previously ( Unal et al . , 2004 ) . Briefly , 5 × 108 cells were crosslinked in 1% formaldehyde for 30 min at room temperature . Chromatin was sheared 20 times for 45 s each ( settings at duty cycle: 20% , intensity: 10 , cycles/burst: 200 ) with 30 s of rest in between using a Covaris S2 . Immunoprecipitation of Rad51-HA or untagged Rad51p was done with anti-HA antibody ( Roche , Mannheim , Germany ) or anti-Rad51p polyclonal antibody ( Santa Cruz , Dallas , TX ) . Immunoprecipitation of γ-H2a . X was done with anti-γ-H2AX ( Abcam , Cambridge , UK ) . A no primary antibody control is also run to ensure specificity . Appropriate dilutions of input and immunoprecipitated DNA samples were used for PCR analysis to ensure linearity of the PCR signal . PCR and data analysis was carried out as described previously ( Unal et al . , 2004 ) . With the exception of the experiment shown in Figure 3—figure supplement 1 which was carried out once , all experiments were done at least twice and a representative data set is shown . ChIP primers are listed in Supplementary file 1B . Yeast genomic DNA was prepared in 1% pulse-field grade agarose plugs ( SeaPlaque 50100 ) and resolved as previously described ( Schwartz and Cantor , 1984 ) with a Bio-Rad CHEF-DR III system . The following parameters were used: 6 V/cm , 120° angle , 20–50 s switch times , 17 hr at 14°C . For Southern analysis , gels were transferred onto a GeneScreen Plus membrane ( PerkinElmer NEF988 ) and probed with a 0 . 5 kb fragment containing HIS3 sequence . Cells were dilution streaked out on SC-URA . The rate of rDNA instability was calculated from 5-FOA plates as described above for YAC instability . | Cells with an unusually large number of mutations—either in the form of changes to the DNA sequence or changes in the number or structure of chromosomes—are said to show genome instability . Although these mutations sometimes boost an organism's chances of survival and reproduction , they more often have detrimental effects , which can include cancer . Genome instability can arise as a result of mistakes occurring during the repair of damaged DNA , or due to inappropriate hybridization of RNA to its DNA template . These RNA–DNA hybrids had been thought to occur strictly during the transcription of DNA into RNA . During this process , the two strands of the DNA molecule separate behind the moving RNA polymerase , and this provides an opportunity for the newly formed RNA to hybridize back to its DNA template . When these RNA–DNA hybrids persist , they give rise to DNA damage that leads to genome instability . Although much is known about the factors that prevent the formation of hybrids , or promote their removal , little is known about how hybrids form in the first place . Now , Wahba et al . have identified one such mechanism in the model yeast , Saccharomyces cerevisiae . It involves a protein called Rad51p , which helps to join stretches of nucleic acids together to repair breaks in DNA . However , Wahba et al . showed that if Rad51p is not properly regulated , it can also trigger the formation of RNA–DNA hybrids; yeast cells that lack the gene for Rad51p showed significantly reduced levels of hybrid formation . Moreover , dysfunctional Rad51p causes RNA sequences to anneal to DNA throughout the genome , rather than just at the site in which the RNA was originally produced . This means that RNA sequences produced during transcription are much more of a threat to genomic stability than previously thought . The work of Wahba et al . presents a paradox in which a protein that is normally involved in repairing DNA can itself cause damage if it is not carefully regulated . It also raises the possibility that the elevated levels of Rad51p expression observed in cancer cells could be a cause , rather than a consequence , of mutations . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2013 | The homologous recombination machinery modulates the formation of RNA–DNA hybrids and associated chromosome instability |
SLAMF6 is a homotypic receptor of the Ig-superfamily whose exact role in immune modulation has remained elusive . Its constitutive expression on resting and activated T cells precludes it from being a bona fide exhaustion marker . By breeding Pmel-1 mice with SLAMF6 -/- mice , we generated donors for T cells lacking SLAMF6 and expressing a transgenic TCR for gp100-melanoma antigen . Activated Pmel-1xSLAMF6 -/- CD8+ T cells displayed improved polyfunctionality and strong tumor cytolysis . T-bet was the dominant transcription factor in Pmel-1 x SLAMF6 -/- cells , and upon activation , they acquired an effector-memory phenotype . Adoptive transfer of Pmel-1 x SLAMF6 -/- T cells to melanoma-bearing mice resulted in lasting tumor regression in contrast to temporary responses achieved with Pmel-1 T cells . LAG-3 expression was elevated in the SLAMF6 -/- cells , and the addition of the LAG-3-blocking antibody to the adoptive transfer protocol improved the SLAMF6 -/- T cells and expedited the antitumor response even further . The results from this study support the notion that SLAMF6 is an inhibitory immune receptor whose absence enables powerful CD8+ T cells to eradicate tumors .
The SLAM family of receptors ( SFRs ) is a set of six receptors expressed on hematopoietic cells ( Wu and Veillette , 2016; Cannons et al . , 2011; Veillette , 2010; Calpe et al . , 2008 ) . All SFRs , except 2B4 , are homotypic binders , that is they engage the same ectodomain sequence , either in cis ( same cell ) or in trans ( adjacent cell ) configuration . Most hematopoietic cell types express 3–5 members of the SLAM family . SFRs generate signals via a bi-phasic mechanism of recruitment to tyrosines in the immunoreceptor tyrosine-based switch motifs ( ITSMs ) in their cytoplasmic domain . SLAM associated protein ( SAP ) , a small protein containing the Src homology 2 ( SH2 ) -domain , was shown to be the default adaptor of the SFRs , interchanging with protein tyrosine phosphatases , mainly SHP-1 , but also SHP-2 , inositol phosphatase SHIP-1 and protein tyrosine kinase Csk ( Wu and Veillette , 2016; Cannons et al . , 2011; Veillette , 2010; Calpe et al . , 2008 ) . SLAMF6 , also known as CD352 , LY108 , or NTB-A , is a homotypic SFR expressed on T cells , NK cells , B cells , and dendritic cells ( Bottino et al . , 2001; Zhong and Veillette , 2008 ) . Kageyama et al . linked SLAMF6 to the anchoring of T cells to their target cells , and subsequent cytolysis of the target ( Kageyama et al . , 2012 ) . According to these authors , functional SAP is critical for SLAMF6 activity . In mice lacking SAP , SLAMF6 was shown to inhibit T cell function ( Kageyama et al . , 2012; Zhao et al . , 2012; Bottino et al . , 2001 ) . The role of SLAMF6 in healthy T cells expressing normal SAP levels was generally inferred from contextual data and is not yet clear . There are indications that SLAMF6 plays an activating role in double-positive thymocytes ( Dutta et al . , 2013 ) along with evidence that it plays an inhibitory role in iNKT cells and CD8+ T cells ( Lu et al . , 2019; Eisenberg et al . , 2018 ) . Gene expression profiles of T cell subsets link SLAMF6 to the progenitor-exhausted state ( Miller et al . , 2019 ) and to the tuning of the critical number of T cells required for proper differentiation ( Polonsky et al . , 2018 ) . To elucidate the net function of SLAMF6 , we generated a transgenic mouse with the Pmel-1 melanoma-specific T-cell receptor ( TCR ) expressed in CD8+ T cells , in which the SLAMF6 gene was knocked out . In this report , we show for the first time that SLAMF6 -/- CD8+ T cells display improved anti-melanoma activity and prevent melanoma growth more effectively than CD8+ T cells with intact and functional SLAMF6 . Since SLAMF6 is constitutively expressed on T cells , it acts as an inhibitory checkpoint receptor whose absence allows the eradication of established tumors by CD8+ T cells .
SLAMF6 is an immune receptor constitutively expressed on non-activated and activated T cells ( Eisenberg et al . , 2018 ) . The level of SLAMF6 transcription and receptor expression , however , is dynamic , changing with time and activation states . To record SLAMF6 expression in a longitudinal manner , human tumor-infiltrating lymphocytes ( TILs ) were activated for 5 days , and SLAMF6 transcript and protein expression were measured ( Figure 1A–C ) . After 1 day of activation , there was an initial decrease in the SLAMF6 transcript that switched to over-expression ( Figure 1C ) . From 3 days after activation onward , SLAMF6 receptor expression consistently increased ( Figure 1A and B ) . Interestingly , the increased expression was most pronounced in T cells activated in the absence of IL-2 ( Figure 1D ) . A similar pattern was observed for the expression of the murine SLAMF6 receptor on Pmel-1 CD8+ T cells ( Figure 1E ) . To identify other immune-related genes that may cluster with SLAMF6 , longitudinal RNA sequencing data were generated from CD4 T cells from two healthy human donors . Five groups of genes ( clusters A-E ) were identified ( Figure 1F ) . Cluster A represents genes highly expressed in non-activated cells , and downregulated upon activation , such as BCL2 and JAK1 . Cluster B represents fast-rising genes undergoing transcription as early as 3–6 hr following activation , but down-regulated after that; genes in this cluster include the activation marker CD69 . Clusters C and D include intermediate genes , upregulated after 6 hr ( C ) or between 12 and 24 hr ( D ) , and downregulated later . Lastly , cluster E includes late-rising genes , such as LAG3 . The SLAMF6 transcript appears in cluster C , rising at 6 hr of activation and staying high after that ( Figure 1G ) . Other genes in cluster C are CD44 , encoding a glycoprotein that takes part in T cell activation ( Huet et al . , 1989 ) , and CD28 and ICOS , which encode co-stimulatory immune receptors ( Turka et al . , 1990; Dong et al . , 2001 ) . The increase in the transcription of receptors that are stably expressed at all times may hint at enhanced recruitment and degradation of these receptors during activation , possibly in the immune synapse ( Onnis and Baldari , 2019 ) . Because SLAMF6 is constantly present on T cells , it is difficult to decipher its effect when it acts as a ligand , introduced in trans . To solve this problem , we generated a cell line derived from B16-F10/mhgp100 melanoma , over-expressing SLAMF6 ( Figure 2A ) . Compared to the wild-type cell line , the SLAMF6-expressing melanoma cells co-cultured with Pmel-1 CD8+ T cells led to decreased IFN-γ secretion by the lymphocytes ( Figure 2B and C ) . To evaluate the effect of SLAMF6 expression on melanoma rejection in vivo , SLAMF6-expressing melanoma was compared to the parental B16-F10/mhgp100 line , in an adoptive T cell transfer regimen , using activated Pmel-1 CD8+ T cells ( Figure 2D ) . The B16-F10/mhgp100/SLAMF6+ tumors grew more aggressively: on day 23 , the mean tumor volume was 431 mm3 in the SLAMF6-expressing melanomas , compared to 137 mm3 in the non-modified tumors ( p=0 . 04 , Student’s t-test ) ( Figure 2E ) . These experiments show that trans-activation of SLAMF6 on lymphocytes , which in this system was achieved with the SLAMF6-expressing melanoma , inhibits the melanoma-specific CD8+ T cell response and allows rapid tumor growth . To evaluate the role of SLAMF6 in melanoma-cognate T cells , we generated a new mouse strain by breeding Pmel-1 mice with SLAMF6 -/- mice . The offspring of this cross represented a new strain , which could serve as a source of CD8+ T cells lacking SLAMF6 and expressing the transgenic TCR against the H-2Db gp100: 25–33 peptide ( Figure 3A ) . Evaluation of the lymphocyte subsets in these mice showed a lower percentage ( 15% ) of CD8+ cells in Pmel-1 x SLAMF6 -/- mouse spleens compared to Pmel-1 splenocytes ( 24% ) ( Figure 3B and Figure 3—figure supplement 1A ) . Despite the smaller percent of CD8+ cells in the spleens of the SLAMF6-deficient mice , the ratio of CD8+ subpopulations ( naive , effector , effector memory , and central memory ) was similar in both mouse strains ( Figure 3C ) . In the initial in vitro activation assays , it was already clear that the Pmel-1 x SLAMF6 -/- T cells have improved functional capacity . Their proliferative response to peptide stimulation was preserved , which was mandatory to produce ample numbers of CD8+ lymphocytes for the adoptive T cell transfer regimen ( Figure 3D ) . CFSE dilution curves were identical in the two mouse strains , as were the activation-induced cell death ( AICD ) rates ( Figure 3—figure supplement 1B , C ) . After 3 days of activation , the -/- mice had higher expression levels of CD25 and CD137 ( 4-1BB ) activation markers ( Figure 3E ) . In parallel , higher PD-1 expression was detected on day 7 ( Figure 3F ) , which in this experimental context was initially taken as an indicator of activation , but was later attributed to SAP deficiency in the SLAMF6 -/- lymphocytes ( see below in Results and Figure 5A and B ) . PD-1 overexpression was also noted by Lu et al . in iNKT cells from SFR -/- mice ( Lu et al . , 2019 ) . The expression of other SLAM family members and ligands ( CD48 , LY9 , CD244 , CD8+4 , CD319 ) on T cells during activation was similar in Pmel-1 and Pmel-1 x SLAMF6 -/- cells ( Figure 3—figure supplement 1D ) . Lastly , we phenotyped the T cells following 3-day and 7-day activation to assess subset ratios based on CD44 and CD62L differentiation markers . While initially , all Pmel-1 T cells were naive , only a negligible number of the activated Pmel-1 x SLAMF6 -/- cells remained in the naïve CD62Lhigh/CD44low state compared to 10% of the Pmel-1 cells . The complete shift of the activated Pmel-1 x SLAMF6 -/- lymphocytes towards effector and effector memory phenotypes indicates the strength of their response to activation ( Figure 3G and H ) . Pmel-1 x SLAMF6 -/- lymphocytes , generated to evaluate the effect of SLAMF6 on antigen-specific activated T cells , showed stronger activation and global acquisition of an effector phenotype . We attribute these results , in contrast to those previously obtained with lymphocytes from SLAMF6 -/- mice , to the role of SLAMF6 in the immune synapse , as the results could only be obtained if a synapse formation had been initiated via cognate TCRs . In the previous experiments , the effect of SLAMF6 deletion was evaluated in the activation and proliferation phases . To test if this superior activation also affects anti-tumor immunity , we characterized Pmel-1 x SLAMF6 -/- melanoma specific T cells in the effector phase . Comparing IFN-γ secretion in response to melanoma cells by Pmel-1 versus Pmel-1 x SLAMF6 -/- lymphocytes showed that cytokine production by the Pmel-1 x SLAMF6 -/- lymphocytes was significantly higher at all effector-to-target ratios ( p=0 . 05 , Figure 4A and B and Figure 4—figure supplement 1A and B ) . In addition to IFN-γ , higher secretion of GM-CSF and lower levels of IL-10 and IL-13 were measured in the SLAMF6 -/- T cells ( Figure 4C and D ) . Since GM-CSF is a strong recruiter of innate immune cells , while IL-10 and IL-13 drive suppressor traits , this secretion profile supports autocrine and paracrine immune activation . mRNA data validated the secretion assays ( Figure 4—figure supplement 1C ) . Importantly , Pmel-1 x SLAMF6 -/- T cells produced higher levels of granzyme B in response to melanoma compared to Pmel-1 T cells , which are already strong killers due to their TCR design ( Figure 4E and F ) . These results indicate that in the absence of the SLAMF6 modulatory effect , even strong cytolysis can be further enhanced . To evaluate the antitumor activity of Pmel-1 x SLAMF6 -/- cells , we assessed adoptive cell transfer ( ACT ) of 7 day pre-activated gp100:25–33-specific , Pmel-1 or Pmel-1 x SLAMF6 -/- CD8+ T cells , transferred into mice bearing palpable B16-F10/mhgp100 melanoma in their back skin , followed by a 2-day course of intraperitoneal IL-2 ( Figure 4G–J ) . The spider curves comparing melanoma growth following Pmel-1 versus Pmel-1 x SLAMF6 -/- T cell transfer revealed that in the first 4 weeks post-transfer , tumor growth was inhibited in both groups . However , on week 4 , tumors treated with Pmel-1 ACT escaped control and grew again in six of the seven animals , whereas mice receiving Pmel-1 x SLAMF6 -/- ACT survived longer , and three of the seven treated mice remained tumor-free for over 80 days ( Figure 4H ) . Vitiligo was noted in all mice attaining complete response , showing up earlier in the -/- group , at the 6th week , indicating the strength of the response ( Figure 4—figure supplement 1D ) . In a similar experiment , peptide-activated Pmel-1 CD8+ T cells or Pmel-1 x SLAMF6 -/- CD8+ T cells were transferred into mice bearing 7-day-old tumors . A week later , mice were sacrificed , and spleens , tumors , and tumor-draining lymph nodes were extracted and evaluated for the presence of transferred T cells , using flow cytometry to detect the gp10025-33 tetramer . A higher proportion of gp10025-33 tetramer+ cells was found in the draining lymph nodes of mice that had received Pmel-1 x SLAMF6 -/- cells ( Figure 4K ) compared to those who had received Pmel-1 cells . Tumors from both groups had a similar density of infiltrating lymphocytes ( Figure 4—figure supplement 1E ) . In summary , melanoma-specific T cells lacking SLAMF6 showed improved functional capacity both in vitro and in vivo and induced longer lasting tumor remission with longer tumor control compared to their wild-type counterparts . SLAMF6 homotypic interactions may involve receptors on a same-cell population , in cis , or interactions in trans , between effector T cells and SLAMF6-expressing antigen-presenting targets . In Figure 2 , we showed that aberrant presentation of SLAMF6 in trans on a melanoma target diminished T cell antitumor activity . Using the Pmel-1 x SLAMF6-/- CD8+ T cells , we then established matrices to determine the regulatory effect of SLAMF6 on the effector T cells and their antigen-presenting targets . SLAMF6-/- mice served as a source of APC devoid of SLAMF6 , and peptide-loaded EL4 cells transduced to express SLAMF6 served as another SLAMF6-expressing APC ( Figure 5B ) . Figure 5A shows that knocking-out SLAMF6 in peptide-pulsed APCs enhances the response of Pmel-1 T cells , attesting to the inhibitory effect of the trans positioning of SLAMF6 . A modest increase in function was observed in the SLAMF6 -/- T cells compared to the WT T cells when cultured with WT APCs , attesting to the improvement in the absence of cis positioned SLAMF6 . However , when SLAMF6-/- T cells were activated by SLAMF6-/- APCs , their cytokine secretion increased even more ( p=0 . 001 , Anova test ) . This finding suggests that not only the absence of SLAMF6 on the T cells improves their secretion capacity , but also SLAMF6-deficient APCs are inherently better . This observation was confirmed with the EL4 APCs ( Figure 5C ) , which inhibited the WT Pmel-1 T cells , as expected , but surprisingly also inhibited the Pmel-1 x SLAMF6-/- CD8+ T cells , which were expected to be resistant to this effect . These results show that the inhibitory effect of SLAMF6 derives from both in cis and in trans interactions . However , the role of SLAMF6 on APCs warrants further investigation . The goal of the next series of experiments was to identify mechanisms underlying the improved effector function of Pmel-1 x SLAMF6 -/- lymphocytes . We initially evaluated the level of SAP , the primary adaptor required for SLAMF6 signaling , encoded by the Sh2d1a gene , which was intact in the SLAMF6 -/- mice ( Figure 6A ) . SAP is a critical adaptor that recruits Fyn kinase to SLAMF6 . However , while SAP transcript was found at similar levels in WT and SLAMF6 -/- lymphocytes , SAP protein was not detectable in SLAMF6 -/- cells ( Figure 6B ) . The discrepancy between the transcript and the protein could be due to the rapid degradation of cytoplasmic SAP in its unbound form . SAP protein deficiency also implies that SLAMF6 is its major anchor in non-activated CD8+ T cells , even though SAP also mediates the inhibitory activity of PD-1 ( Peled et al . , 2018 ) . As shown in Figure 3F , PD-1 expression is more than two-fold increased in 7 day activated Pmel-1 x SLAMF6 -/- lymphocytes compared to Pmel-1 , but , as we have shown , this gap did not affect functionality . Thus , PD-1 overexpression can represent the failure of this receptor to generate a negative feedback loop in over-activated T cells . Next , we measured by flow cytometry the level of phosphorylated ribosomal protein S6 ( rpS6 ) , an integrator of important signaling pathways , including PI3K/AKT/mTOR and RAS-ERK ( Figure 6C ) . No difference in phosphorylated rpS6 was found in the Pmel-1 x SLAMF6 -/- T cells . Therefore , we proceeded to identify transcription regulators whose activity differed between the Pmel-1 cells and the Pmel-1 x SLAMF6 -/- cells . The most prominent regulator found was T-bet , which increased more than twofold in activated Pmel-1 x SLAMF6 -/- splenocytes , followed by Eomes ( Figure 6D and Figure 6—figure supplement 1 ) . T-bet was originally described as the key transcription factor defining type 1 T helper ( Th ) cells; it has since been found to play a major role in the acquisition of effector functions by CD8+ T cells ( Szabo et al . , 2000 ) . Type one inflammatory signals induce T-bet expression ( Joshi et al . , 2007 ) , which is in line with the intensive IFN-γ secretion we observed by the Pmel-1 x SLAMF6 -/- lymphocytes ( Figure 4A ) . Lastly , the level of immune receptors that mediate exhaustion was recorded during prolonged activation . Five days after the end of a 7-day activation course , Pmel-1 x SLAMF6 -/- T cells displayed similar levels of PD-1 , CD244 ( SLAMF4 ) , and TIM-3 to those in the Pmel-1 T cells , but higher expression of LAG-3 ( Figure 6E ) . Hypothesizing that LAG-3 represents a compensatory mechanism for the enhanced activation of Pmel-1 x SLAMF6 -/- T cells , we used a blocking antibody against LAG-3 in the gp100 activation assay . We then measured IFN-γ secretion by activated T cells in response to B16-F10/mhgp100 melanoma cells . As expected , blocking LAG-3 on Pmel-1 x SLAMF6 -/- lymphocytes increased their cytokine secretion significantly , whereas it did not affect Pmel-1 cells . Overall , knocking out SLAMF6 together with LAG-3 blockade resulted in a three-fold increase in IFN-γ production ( Figure 6F ) . To evaluate the combination of SLAMF6 -/- cells and LAG-3 blocking antibody , we conducted an ACT experiment in melanoma-bearing mice , using Pmel-1 x SLAMF6 -/- CD8+ T cells activated in vitro in the presence of anti-LAG-3 and sustained by intraperitoneal IL-2 ( days 8 and 9 ) and anti-LAG-3 ( days 8 , 10 , 15 , 18 , 21 ) . The control arm consisted of Pmel-1 x SLAMF6 -/- CD8+ T cells stimulated and sustained with an isotype antibody ( Figure 6G–I ) . Mice treated by the combination of SLAMF6-deficient T cells and anti-LAG-3 antibody showed faster reduction and disappearance of their tumors ( Figure 6H ) , apparent already on day 16 ( p=0 . 04 ) ( Figure 6I ) . These results demonstrate that blocking the compensatory rise of LAG-3 on SLAMF6-/- T cells improved their anti-tumor effect even further .
The aim of this study was to characterize the role of SLAMF6 in CD8+ T cells , in the context of an antitumor response . The data obtained identify SLAMF6 as a receptor whose absence significantly improves CD8+-mediated tumor regression , suggesting that it is an inhibitory checkpoint . Historically , SFRs were studied for their part in X-linked lymphoproliferative disease ( XLP ) , a complex genetic immune dysfunction caused by a SAP mutation . XLP is characterized by a compromised immune response to Epstein-Barr virus ( EBV ) but also by unrestrained T lymphoblast proliferation , which is not necessarily EBV-induced . Thus , it is unclear whether loss of SAP converts all SFRs into ‘super-inhibitory’ receptors or whether , on the contrary , loss of SAP unleashes lymphocytes to proliferate , free from re-stimulation-induced apoptosis ( Katz et al . , 2014; Kageyama et al . , 2012; Zhao et al . , 2012; Bottino et al . , 2001 ) . Since SAP is an adaptor common to all SLAM family receptors , the role of each individual receptor was obscured by the shared defect . In this situation , SLAMF6 was considered a receptor with a dual function , depending on the interplay between SAP and SHP-1 and SHP-2 , protein phosphatases that bind to tyrosines on the cytoplasmic tail of the receptor ( Veillette , 2010; Cannons et al . , 2011; Detre et al . , 2010 ) . SLAMF6 duality was echoed in data from Veillette that showed differing effects of SLAMF6 on NK cells , enhancing function in the priming phase while suppressing cells in the effector-phase ( Wu et al . , 2016 ) . Also , mice lacking individual SFRs exhibit minor immune deviations ( Wu and Veillette , 2016; Cannons et al . , 2011; Veillette , 2010; Calpe et al . , 2008 ) . In the past , we described that targeting SLAMF6 with its soluble ectodomain yielded CD8+ T cells that do not need IL-2 supplementation , either in vitro or in vivo , to eradicate established melanoma ( Eisenberg et al . , 2018 ) . The beneficial effect of the soluble ectodomain of SLAMF6 prompted us to generate melanoma-specific SLAMF6 -/- T cells , to characterize the role of the receptor in a solid tumor model . A key finding using the new Pmel-1 x SLAMF6 -/- mice described in this manuscript is the absence , in fact , of a dichotomy in SLAMF6 action in effector T cells . On the contrary , knocking-out SLAMF6 in murine antigen-specific CD8+ T cells disclosed an unequivocal inhibitory role for the receptor . In its absence , TCR triggering of anti-melanoma CD8+ T cells yielded a strong effector phenotype , higher IFN-γ secretion , improved cytolysis , and better outcomes in the adoptive transfer of SLAMF6 -/- anti melanoma CD8+ T cells to treat established melanoma . This study identifies SLAMF6 as a powerful inhibitor of antitumor immune response . The absence of viable SAP in SLAMF6 -/- lymphocytes hints that this adaptor takes a major part in the inhibitory effect of SLAMF6 . To explore the role of SLAMF6 in T cells without the confounding effects of its function in other cell types , we generated a system in which effector T cells interact with their tumor target based on specific epitope recognition and subsequently generate an immunological synapse . The synapse is a subcellular structure involved in the effect of SLAMF6 and is crucial for its study ( Zhao et al . , 2012 ) . However , although we revealed the inhibitory effect of SLAMF6 in the Pmel-1 x SLAMF6 -/- mice , the source and configuration of SLAMF6/SLAMF6 homotypic binding in the wild-type situation were still difficult to characterize . We had to generate a SLAMF6-positive B16-F10/mhgp100 melanoma line to measure the effect , or more exactly , the degree of suppression , that SLAMF6 trans-activation has on the capacity of melanoma-cognate CD8+ T cells to eradicate tumors . As shown ( Figure 2E ) , the SLAMF6-expressing melanoma suppressed T cell efficacy and consequently grew faster . This observation received further support from similar data generated with peptide-pulsed thymoma cells transduced to express SLAMF6 ( Figure 5C ) . However , the improved IFNγ secretion of SLAMF6-/- T cells , when co-cultured with SLAMF6-lacking APCs compared to WT APCs , implies that an inherent mechanism , most likely lack of cis-inhibition in the antigen presenting cells , is also responsible for the effect ( Figure 5A ) . The molecular mechanisms underlying the increased functional capacity of Pmel-1 T cells lacking SLAMF6 have common features with XLP , as the absence of SAP implies . But while XLP is a global defect of all cell types of the immune system , and therefore yields mixed derangements , the absence of SLAMF6 is remarkable for the enhanced functionality of CD8+ T cells , in which it is the dominating SFR . The transcriptional landscape of SLAMF6 -/- T cells was governed by the higher expression of T-bet . T-bet is a transcription factor that contributes to Th1 and Th17 phenotypes in CD4 T cells . T-bet is prevalent in cytolytic innate lymphocytes residing in tissues and B cells of mouse strains prone to autoimmunity ( Plank et al . , 2017; Nixon and Li , 2018 ) . The increased activity of T-bet in SLAMF6 -/- CD8+ T cells implies that T-bet-regulated pathways may operate in CD8+ T cells in the absence of functioning SLAMF6 , generating ‘type 1’ inflammatory traits and high cytotoxicity . The improved production of IFN-γ and GM-CSF , in parallel with reduced IL-10 and IL-13 , is also typical for type one phenotypes . SLAMF6 should be distinguished from typical exhaustion markers because it is expressed on CD8+ T cells , regardless of their state of activation . Yigit et al . suggested that blocking SLAMF6 using an antibody can correct the exhaustion phase of T cells ( Yigit et al . , 2019 ) , but we favor the notion that SLAMF6 hampers T cells at any stage , as reflected from the functional superiority of short-term activated Pmel-1 T cells . Depleting SLAMF6 improved CD8+ T cells in the short and long-term , as was most evident when the WT Pmel-1 cells induced the regression of melanoma only for a limited period while the Pmel-1 x SLAMF6 -/- cells led to lasting responses in mice ( Figure 4H ) . While searching for new immunotherapeutic targets , the field of immunotherapy is moving to combination therapies , and to biomarker-based treatment choices , to target the escape mechanisms used by tumors . From the results presented here , we conclude that SLAMF6 is an important checkpoint receptor with a significant inhibitory effect on T cells . The balance between SLAMF6 and LAG-3 , and the enhancing effect of LAG3 blocking suggests that targeting both may have a valuable combinatorial , and perhaps even a synergistic , effect ( Figure 6G–I ) . In summary , we have shown that SLAMF6 is a constitutive inhibitory immune receptor; in its absence , CD8+ T cells acquire stronger reactivity against tumor cells . The strong effector trait is attributed to a series of T-bet-mediated transcriptional events that drive CD8+ T cells to exert strong cytotoxicity and achieve long-lasting tumor control . SLAMF6 is an attractive target for the development of checkpoint inhibitors for systemic treatment of cancer and for the improvement of antitumor cellular therapies .
pCMV3-mSLAMF6 and pCMV3-negative control vectors were purchased from SINO Biological Inc , Eschborn , Germany . For flow cytometry , cells were labeled with the following reagents: anti-CD16/32 ( 93 ) , anti-SLAMF6 ( 330-AJ ) , anti-TNFa ( MP6-XT22 ) , anti-CD19 ( 6D5 ) , anti-CD44 ( IM7 ) , anti-TIM3 ( RMT3-23 ) , and anti-LAG-3 ( C9B7W ) ( all from Biolegend , San Diego , CA ) . Anti-IFN-ɣ ( XMG1 . 2 ) , anti-CD8+ ( 53–6 . 7 ) , anti-GZMB ( NGZN ) , anti-CD4 ( GK1 . 5 ) , and anti-CD25 ( PC61 . 5 ) were from Biogems . Anti-CD62L ( MEL-14 ) , anti-Vb13 ( MR12-3 ) , anti-CD69 ( H1 . 2F3 ) , anti-CD137 ( 17B5 ) , anti-PD1 ( J43 ) , and anti-CD244 ( eBio244F4 ) were from eBioscience . Anti-human SLAMF6 ( REA ) was purchased from Miltenyi Biotec , Bergisch Gladbach , Germany . Anti-pS6 ( D57 . 2 . 2E ) was from Cell Signaling Technology , Danvers , MA . Anti-LAG-3 ( C9B7W ) and the corresponding isotype were from InVivoMab , BioXcell , NH , USA . The Mart-126-35 iTag MHC tetramer was from MBL , Woburn , MA . For Immunobloting: anti-Ly108 ( Rat , 3E11 , Merck , Kenilworth , NJ ) , anti-SAP ( Rat , 1A9 , Biolegend ) , anti-β actin ( Mouse , sc-47778 , Santa Cruz Biotechnology , TX ) , anti-SHP1 ( Rabbit , generated in A . V . ’s laboratory ) . C57BL/6 mice were purchased from Harlan laboratories . Pmel-1 ( a kind gift from M . Baniyash ) and SLAMF6 -/- mice ( a kind gift from I . Shachar ) were self-bred . The Pmel-1 mice carry a rearranged TCR specific for a 9-mer epitope ( 25-32 ) from murine Pmel 17 , overexpressed on transformed melanocytes and homologous to the human melanoma-associated antigen gp100 . All experiments were performed with 8- to 12-week-old female mice . Generation of Pmel-1 x SLAMF6 -/- mice . Pmel-1 and SLAMF6-/- mice were bred to generate Pmel-1 X SLAMF6-/- mice according to the ethics requirements ( Authority for biological and biomedical models , Hebrew University , Jerusalem , Israel ) . When the mice reached 3 weeks of age , 2 mm of the mouse tail were cut , 200 µl 50 mM NaOH 0 . 2 Mm EDTA were added , and the tails were incubated at 95°C for 20 min for DNA purification . 200 µl 80 mM TRIS-HCL , pH5 , were added to stop the reaction . The DNA purified from the tails was used in PCR reactions for genotyping of mice in the SLAMF6 locus on chromosome 1 ( primers adapted from the Jackson laboratories website ) and in the Pmel-1 locus on chromosome 2 ( Ji et al . , 2014 ) . The identification of the genomic insertion site of the Pmel-1 TCR α and β transgenes was performed by next-generation sequencing . Pmel-1 or Pmel-1xSLAMF6-/- mouse splenocytes ( 2 × 106/ml ) were activated with 1 µg/ml of mouse gp10025-33 peptide for 6 days with IL-2 30 IU/ml . Fresh medium containing IL-2 was added every other day . RNA isolation and qPCR . RNA was isolated from cells using the GenElute Mammalian Total RNA kit ( Sigma Aldrich , MA ) according to the manufacturer’s protocol . RNA was then transcribed to cDNA using qScript cDNA Synthesis kit ( Quantabio , Beverly , MA ) according to the manufacturer’s instructions , and RT-PCT or qRT-PCR was performed using the following primers: RT-PCR for Sh2d1a was performed in the SensQuest lab cycler machine ( Danyel Biotech ) ; the products were then run on 1 . 5% agarose gel . Adoptive cell transfer experiments: Animal studies were approved by the Institutional Review Board - Authority for biological and biomedical models , Hebrew University , Jerusalem , Israel ( MD-14602–5 and MD-15421–5 ) . Statistics . Statistical significance was determined by unpaired t-test ( two-tailed with equal SD ) using Prism software ( GraphPad ) . A p-value<0 . 05 was considered statistically significant . Analysis of more than two groups was performed using the one-way ANOVA test . * , p≤0 . 05; ** , p≤0 . 01; *** , p≤0 . 001 . For each experiment , the number of replicates performed and the statistical test used are stated in the corresponding figure legend . | The immune system helps to protect our bodies from illnesses and infections . Immunotherapies are medicines designed to treat diseases , such as cancer , by boosting the immune system against the condition . This is a powerful approach but so far immunotherapies have only had partial success and there is a need for further improvements . One protein called SLAMF6 is found on cells from the immune system that attack and kill cancer cells . Immunotherapies that suppress SLAMF6 on immune cells called killer T cells could increase immune system activity helping to treat cancers , particularly melanoma skin cancers . So far the potential for SLAMF6 as a target for immunotherapy has not been fully explored . Hajaj et al . created mice with killer T cells that recognized skin cancer cells and lacked SLAMF6 . These modified cells were better at fighting cancer , producing more anti-cancer chemicals called cytokines and killing more cancer cells . The modified cells had a lasting effect on tumors and helped the mice to live longer . The effects could be further boosted by treating the mice in combination with other immunotherapies . SLAMF6 is a possible new target for skin cancer immunotherapy that could help more people to live longer following cancer diagnosis . The next step is to create a drug to target SLAMF6 in humans and to test it in clinical trials . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"immunology",
"and",
"inflammation",
"cancer",
"biology"
] | 2020 | SLAMF6 deficiency augments tumor killing and skews toward an effector phenotype revealing it as a novel T cell checkpoint |
The ancestral condition from which humans evolved is critical for understanding the adaptive origin of bipedal locomotion . The 4 . 4 million-year-old hominin partial skeleton attributed to Ardipithecus ramidus preserves a foot that purportedly shares morphometric affinities with monkeys , but this interpretation remains controversial . Here I show that the foot of Ar . ramidus is most similar to living chimpanzee and gorilla species among a large sample of anthropoid primates . The foot morphology of Ar . ramidus suggests that the evolutionary precursor of hominin bipedalism was African ape-like terrestrial quadrupedalism and climbing . The elongation of the midfoot and phalangeal reduction in Ar . ramidus relative to the African apes is consistent with hypotheses of increased propulsive capabilities associated with an early form of bipedalism . This study provides evidence that the modern human foot was derived from an ancestral form adapted to terrestrial plantigrade quadrupedalism .
Terrestrial bipedalism is widely regarded as a shared-derived characteristic of the hominin clade and understanding its evolution is one of the central foci of biological anthropology ( Darwin , 1871; Wasburn , 1967; Fleagle et al . , 1981; Richmond et al . , 2001; Gebo , 1996; Begun , 2004; Lovejoy et al . , 2009a; White et al . , 2015 ) . There are numerous adaptive explanations for the origin of bipedalism ( Darwin , 1871; Hewes , 1961; Lovejoy , 1981; Rose , 1991; Washburn , 1960; Hunt , 1996 ) that are inherently difficult to test directly ( Smith and Wood , 2017 ) , but each of them depends on alternative hypothetical models for the morphology and locomotor behavior of the human-chimpanzee last common ancestor ( LCA; Richmond et al . , 2001 ) . Hypotheses for the locomotor behavior of the LCA include vertical climbing ( Stern , 1975; Prost , 1980; Fleagle et al . , 1981 ) , terrestrial knuckle-walking ( Wasburn , 1967; Gebo , 1992; Gebo , 1996; Pilbeam , 1996; Richmond and Strait , 2000; Richmond et al . , 2001; Begun , 2004; Inouye and Shea , 2004 ) , below-branch suspension ( Keith , 1923; Tuttle , 1969; Young et al . , 2015 ) , arboreal bipedality ( Thorpe et al . , 2007 ) , and more generalized quadrupedalism with slow , deliberate climbing ( Lovejoy et al . , 2009a; White et al . , 2009; White et al . , 2015 ) . These behavioral hypotheses make different assumptions about whether the LCA was adapted to arboreality or terrestriality ( Wasburn , 1967; Gebo , 1996; Gebo , 1992; Schmitt , 2003; Crompton et al . , 2010 ) . No matter the specific elements of the proposed behavioral hypothesis for the Homo-Pan LCA , hominin bipedalism is either the result of an initial evolutionary shift towards terrestriality from an arboreal ancestor ( Schmitt , 2003; Lovejoy et al . , 2009a; Lovejoy et al . , 2009b; Lovejoy et al . , 2009c; White et al . , 2009; Crompton et al . , 2010; White et al . , 2015 ) or , alternatively , a secondary shift from a semi-terrestrial quadrupedal ancestor ( Figure 1; Wasburn , 1967; Gebo , 1992; Gebo , 1996; Pilbeam , 1996; Richmond and Strait , 2000; Richmond et al . , 2001; Inouye and Shea , 2004 ) . The partial skeleton of the early hominin Ardipithecus ramidus purportedly lacks postcranial specializations associated with hominoid-like orthogrady , vertical climbing , and suspension ( Lovejoy et al . , 2009a; Lovejoy et al . , 2009b; Lovejoy et al . , 2009c; White et al . , 2009; White et al . , 2015 ) , which challenges conventional understandings of ape and human ancestry ( Darwin , 1871; Keith , 1923; Morton , 1922; Wasburn , 1967; Gebo , 1992; Gebo , 1996; Williams , 2012 ) . In particular , the foot of Ar . ramidus was argued to possess monkey-like midfoot stabilizing and propulsive morphologies that were inferred to be primitive for great apes ( Lovejoy et al . , 2009a ) . Lovejoy and colleagues ( Lovejoy et al . , 2009a; White et al . , 2015 ) suggested that the Ar . ramidus foot was consistent with ‘above branch plantigrady’ and a form of locomotion termed ‘arboreal multigrady’ ( White et al . , 2015 ) . The African apes are viewed as ‘adaptive cul-de-sacs’ ( Lovejoy et al . , 2009b: pg . 104 ) that independently evolved adaptations for vertical climbing given the purported lack of ‘the peculiar substrate-conforming , hand-like grasping foot of living African apes’ ( White et al . , 2015: pg . 4883 ) in Ar . ramidus and in White and colleagues’ reconstruction of the Homo-Pan LCA . This interpretation of the Ar . ramidus foot is consistent with hominin bipedalism emerging from a more generalized , exclusively arboreal , quadrupedal ancestor ( Straus , 1949; Schmitt , 2003; Lovejoy et al . , 2009a; Lovejoy et al . , 2009b; Crompton et al . , 2010 ) . Foot proportions ( e . g . , tarsal , metatarsal , and phalangeal lengths ) are hypothesized to reflect variation in locomotor behavior among anthropoid primates ( Midlo , 1934; Schultz , 1963a; Schultz , 1963b; Jolly , 1967; Strasser , 1992; Strasser , 1994 ) . Furthermore , the modern human foot has been highly modified in response to the biomechanical constraints of terrestrial bipedalism ( Morton , 1922; Gebo , 1992; Harcourt-Smith and Aiello , 2004 ) . Foot proportions may therefore provide insight into long-standing debates about alternative models for the Homo-Pan LCA ( Keith , 1923; Wasburn , 1967; Stern , 1975; Prost , 1980; Fleagle et al . , 1981; Gebo , 1992; Gebo , 1996; Richmond et al . , 2001; Schmitt , 2003; Lovejoy et al . , 2009b; Crompton et al . , 2010; White et al . , 2015 ) . The locomotor behavior of the earliest hominins significantly alters inferences about the paleobiology of the Homo-Pan LCA and our understanding of how bipedalism evolved . This study utilizes recent methodological advances to ( 1 ) test alternative hypotheses about the relationship between foot proportions and locomotor behavior among extant anthropoid primates , ( 2 ) characterize the morphometric affinities of the Ar . ramidus foot ( ARA-VP-6/500 ) on the basis of foot proportions , and ( 3 ) estimate the foot proportions of the Homo-Pan LCA . Specifically , this study uses a combination of evolutionary modeling ( Hansen , 1997; Butler and King , 2004; Ingram and Mahler , 2013 ) and ancestral state estimation ( Elliot and Mooers , 2014 ) methods to make inferences about the evolutionary history of foot proportions in the anthropoid clade and its implications for the adaptive origin of hominin bipedal locomotion .
The morphometric affinities of the Ar . ramidus foot were evaluated by constructing a morphospace based on six geometric mean-standardized variables that are preserved in the ARA-VP-6/500 foot skeleton using Principal Components Analysis ( PCA , Figure 2 ) . The first three principal components account for 96% of the total variance in the sample and clearly separate taxonomic groups along previously hypothesized axes of morphological variation ( Schultz , 1963a; Schultz , 1963b; Jolly , 1967; Strasser , 1992; Strasser , 1994 ) . The first principal component accounts for 63% of the variance and is positively loaded by the lengths of the fifth metatarsal and fourth proximal phalanx . The second principal component accounts for 18% of the variance and is positively loaded by the lengths of the first metatarsal and fourth proximal phalanx , and negatively loaded by the length of the fifth metatarsal . The third principal component , which represents 15% of the variance , is positively loaded by the length of the first metatarsal ( Table 1 ) . The distribution of anthropoid taxa in the PCA is consistent with predictions based on locomotor behavior . For example , the more terrestrial taxa fall at the negative end ( Homo , Pan , Gorilla , Theropithecus , Papio , and Erythrocebus ) with shorter metatarsals and phalanges , whereas the most arboreal , suspensory , taxa fall at the positive end of PC1 ( e . g . , Pongo and Ateles ) . The terrestrial taxa are sort into those that are heel-strike plantigrade ( Homo , Pan , and Gorilla ) and those that are digitigrade ( Theropithecus , Papio , Erythrocebus ) . Hylobatids , atelids , and Pongo are distinguished from other arboreal taxa along the same axis that separates terrestrial heel-strike plantigrade taxa from terrestrial digitigrade taxa . A UPGMA cluster analysis ( cophenetic correlation coefficient = 0 . 82 ) shows that of the 44 extant taxa presented here Ar . ramidus is most similar to Pan and Gorilla ( Figure 2—figure supplement 1 ) . Univariate comparisons show that Ar . ramidus possesses a cuboid that is only slightly elongated relative to African apes ( Figure 2—figure supplement 2A ) , a relatively short fourth proximal phalanx ( Figure 2—figure supplement 2B ) , and an intrinsically elongated first metatarsal like African apes and atelids ( Figure 2—figure supplement 2C ) . The scaling of various tarsals , metatarsals , and phalanges with body mass was investigated using phylogenetic generalized least squares regression ( pGLS ) to account for the statistical non-independence of the data due to phylogenetic relationships . The parameters for each of the pGLS models include the intercepts and slopes of the variables regressed on log body mass , their standard error ( s . e . ) , T , and the p-value ( Supplementary file 1 ) . All variables scale with slight negative allometry in that larger species tend to have relatively shorter metatarsals , phalanges , and tarsals . The only exception is the length of the talar trochlea , which scales isometrically with body mass . Pagel’s lambda ( λ ) is a parameter commonly estimated in pGLS regression analyses as a measure of phylogenetic signal that can be used to transform the branches of the phylogenetic tree to improve model fit ( Pagel , 1999; Revell , 2010 ) . A λ value of 1 would be consistent with expectations under a Brownian motion evolutionary model . The tarsal measurements show departure from Brownian motion ( λ = 0 . 574 or less ) , whereas the higher λ values of the metatarsal and phalangeal variables are consistent with a Brownian motion model . Several of the λ values are not significantly different from either 0 or 1 , which suggests the true λ for these models is uncertain . The PC scores used in the evolutionary analyses are not correlated with body mass ( pGLS p = 0 . 08 or higher ) , which suggests that body mass is not responsible for driving the differences in intrinsic foot proportions among anthropoid groups . The adaptive implications of an African ape-like morphology in Ar . ramidus ( Figure 2 ) were evaluated using evolutionary modeling . The input data include the first three principal components from the PCA described above . This approach was chosen to reduce the dimensionality of the dataset for evolutionary modeling and ancestral state estimation . Alternative a priori evolutionary hypotheses were constructed and include a Brownian motion , single-optimum Ornstein-Uhlenbeck ( OU ) , and several multi-optima OU models ( Butler and King , 2004 ) ( Figure 3—figure supplement 1 ) . Alternative multi-optima OU models were constructed using different selective regimes associated with locomotion in extant taxa . The evolutionary models differ in increasing complexity where each model includes additional phenotypic optima in a hierarchical manner . Model comparisons using multiple criteria ( AIC , AICc , SIC ) show that the multi-optima OU models are a better fit to the data than a Brownian or single-optimum OU model ( Supplementary file 2 ) , which suggests there are multiple adaptive peaks associated with foot proportions among anthropoid primates . An additional evolutionary model was constructed without identifying selective regimes a priori and resulted in a similar pattern of selective regimes compared to the best fitting a priori hypothesis ( Figure 3 ) . The evolutionary hypothesis with the most favorable AICc value is the most complex and includes selective regimes associated with bipedalism , terrestrial plantigrady , terrestrial semiplantigrady , arboreal quadrupedalism , arboreal quadrupedalism with increased frequency of hindlimb-assisted suspension , and arboreal quadrupedalism with increased frequency of climbing . Simulations show that there is adequate power to distinguish between alternative models and provides support for the a priori model selection results obtained using AICc ( Figure 3—figure supplements 1–2 ) . A molecular consensus phylogeny with branch lengths proportional to elapsed time was superimposed on the multivariate data and ancestral values were estimated using a Markov chain Monte Carlo ( MCMC ) method . This approach relaxes assumptions of neutrality and gradualism and therefore minimizes the effect of exceptional lineage divergences on the estimation of ancestral values . The estimated value for the Homo-Pan LCA is nearest to the African apes and highly distinct from all other taxa ( Figure 4 ) . The 95% credibility intervals for the PC scores of the node representing the Homo-Pan LCA are relatively narrow and include the mean values for Pan paniscus and Gorilla gorilla . The estimated ancestral values for both hominids and hominoids are nearest to Alouatta and Lagothrix , which is consistent with prior suggestions based on the comparative morphology of the foot in extant and fossil taxa ( Gebo , 1996; Sarmiento , 1983; Langdon , 1985; Harrison , 1986 ) .
The multivariate evolutionary modeling analyses presented here confirms that among anthropoid primates intrinsic foot proportions are linked to locomotor behavior ( Midlo , 1934; Schultz , 1963a; Schultz , 1963b; Jolly , 1967; Strasser , 1992; Strasser , 1994 ) . The combination of evolutionary modeling with ancestral estimations provides evidence for homoplasy in the evolution of anthropoid foot proportions , which strengthens hypotheses about the link between morphology and behavior . For example , Erythrocebus probably evolved a terrestrially adapted , digitigrade foot characterized by short phalanges , a short hallux , and long non-hallucal metatarsals independently of Theropithecus and Papio . These terrestrial monkeys also tend to have a longer midfoot , which increases the pedal load arm and enhances propulsive capabilities . Although most anthropoids sampled here are highly arboreal , the foot of Alouatta , Lagothrix , and hylobatids converge on similar foot proportions associated with arboreal climbing ( though the upper limb of hylobatids is highly autapomorphic ) , such as a long hallux , relatively shorter metatarsals , and longer phalanges derived from a more generalized anthropoid ancestral condition . The more suspensory anthropoids Ateles and Pongo may have independently evolved towards the part of the morphospace associated with pedal elongation . African apes and Ar . ramidus occupy a distinct phenotypic optimum characterized by short phalanges , short metatarsals , moderately elongated tarsals , and a long hallux . Although the more terrestrial hominines ( Homo , Ardipithecus , Pan , Gorilla ) and cercopithecines ( Papio , Theropithecus , Erythrocebus ) probably evolved short phalanges in parallel ( Figures 1 and 2 ) , the hominines retain an intrinsically elongated hallux and short non-hallucal metatarsals while the terrestrial cercopithecines display the opposite configuration ( Schultz , 1963a; Schultz , 1963b ) . The difference in foot proportions between terrestrially adapted apes and monkeys are similar to the differences between plantigrade and digitigrade carnivorans ( Taylor , 1976 ) . The long metatarsals of digitigrade quadrupeds increase the range of plantarflexion at the talocrural joint at the expense of decreasing the effective mechanical advantage of the ankle plantar-flexors ( Biewener , 1989 ) . African apes and Ar . ramidus are separated from terrestrial monkeys along the same axis that distinguishes generalized arboreal anthropoids ( e . g . , Cercopithecus ) from hallux-elongated taxa associated with behaviors variously described as ‘climbing’ ( Hylobatids , Alouatta , Lagothrix ) . The unique combination of these traits in the hominine foot supports the hypothesis that such feet are adapted to both terrestrial heel-strike plantigrade ( rather than digitigrade ) quadrupedalism and vertical climbing , which is consistent with previous suggestions based on comparative anatomy and behavioral observations ( Fleagle et al . , 1981; Gebo , 1996; Prost , 1980; Gebo , 1992; DeSilva , 2009; Prang , 2015 ) . If Ar . ramidus and the Homo-Pan LCA were adapted to more generalized quadrupedalism and climbing as originally suggested ( Lovejoy et al . , 2009a; White et al . , 2009 ) or ‘arboreal multigrady’ as later revised ( White et al . , 2015 ) ; see also Fernández et al . , 2018 ) , then the Ar . ramidus foot skeleton should not have been placed in the same selective regime as the African apes ( Figure 3 ) . The ancestral estimations provide support for the hypothesis that modern humans evolved from an ancestor with African ape-like foot proportions . Modern humans have the longest cuboids of any of the anthropoid primates sampled here ( Figure 2 ) , which reflects a biomechanical strategy for lengthening the foot’s load arm to enhance aspects of propulsion ( i . e . , range of motion ) while simultaneously restricting metatarsal length to minimize bending moments associated with plantigrady ( Lovejoy et al . , 2009a ) . Although the cuboid elongation of Ar . ramidus from the estimated LCA is more modest than previously reported ( Lovejoy et al . , 2009a ) , it is more parsimoniously interpreted to be derived in the direction of modern humans from an African ape-like ancestor with a short midfoot , rather than an ancestral retention from a more monkey-like great ape ancestor ( Lovejoy et al . , 2009a; Lovejoy et al . , 2009b; White et al . , 2015; McNutt et al . , 2018 ) . Midfoot elongation is consistent with the functional hypothesis of increased propulsive capabilities associated with the form of bipedalism practiced by Ar . ramidus ( Lovejoy et al . , 2009a; White et al . , 2015; Suwa et al . , 2009; Kimbel et al . , 2014; Kozma et al . , 2018; Fernández et al . , 2018 ) . These analyses show that the origin of bipedalism cannot be explained as an initial shift toward terrestriality from a more exclusively arboreal ancestor ( Figure 3 ) . Instead , early hominins , including Ar . ramidus , evolved from an ancestor with an African ape-like foot adapted to terrestrial plantigrade quadrupedalism ( Morton , 1922; Wasburn , 1967; Gebo , 1992; Gebo , 1996 ) . Evidence for terrestrial plantigrade quadrupedalism in the Homo-Pan LCA raises the question of whether or not purported knuckle-walking traits in hominines might be homologies ( Richmond et al . , 2001; Gebo , 1996; Begun , 2004; Inouye and Shea , 2004 ) rather than homoplasies ( Dainton and Macho , 1999; Dainton , 2001; White et al . , 2015; Lovejoy et al . , 2009b; Kivell and Schmitt , 2009 ) . The knuckle-walking hand posture of African apes is hypothesized to be a secondary adaptation to terrestriality in taxa that retain long hands in association with below-branch forelimb suspension ( Tuttle , 1969 ) . However , there is no special relationship between plantigrade foot postures and knuckle-walking hand postures ( e . g . , plantigrade non-primate mammals do not have a knuckle-walking hand posture ) . Whether or not modern humans evolved from a knuckle-walking ancestor relies on the analysis of the hand and wrist ( Tuttle , 1969; Gebo , 1996; Dainton , 2001; Begun , 2004; Inouye and Shea , 2004; Dainton and Macho , 1999; Kivell and Schmitt , 2009; Almécija et al . , 2015 ) . The hand of Ar . ramidus was argued not to display traits associated with forelimb suspension in extant taxa or show evidence of a knuckle-walking ancestry ( White et al . , 2015; Lovejoy et al . , 2009b; Almécija et al . , 2015 ) , but many of those observations have not yet been independently validated . The terrestrial specializations of the hominine foot are likely to be homologous because they are present in Ar . ramidus and they are consistent with model-based ancestral estimations . Critically , they carry a similar set of implications for the origin of bipedalism regardless of the terrestrial hand posture of the Homo-Pan LCA . This study provides evidence that modern humans evolved from an ancestor with an African ape-like foot associated with terrestrial plantigrady and vertical climbing . Hominin upright walking therefore likely emerged in the context of semi-terrestrial quadrupedalism . Explaining the adaptive origin of hominin bipedalism ( i . e . , ‘why’ bipedalism evolved ) will continue to be a challenging endeavor ( Smith and Wood , 2017 ) . However , the comparative and fossil material provides evidence for patterns of evolution ( i . e . , ‘how’ bipedalism evolved ) and strongly suggests that hypotheses of a non-African ape-like morphology for the foot of the Homo-Pan LCA ( Lovejoy et al . , 2009a; White et al . , 2015; Lovejoy et al . , 2009b ) are inconsistent with the results from this study . The Ar . ramidus ARA-VP-6/500 partial skeleton is remarkable for its preservation of multiple areas of anatomy ( Lovejoy et al . , 2009a; White et al . , 2009; White et al . , 2015 ) . This study provides evidence that intrinsic foot proportions reflect locomotor diversity among anthropoid primates , but it will be important to consider other regions in future comparative studies . The hypothesis that hominins evolved from a semi-terrestrial quadrupedal ancestor could be tested with detailed quantitative analyses of other aspects of Ar . ramidus postcranial morphology .
The extant sample is composed 385 individuals representing 45 taxa: Homo sapiens , Ardipithecus ramidus , Pan troglodytes , Pan paniscus , Gorilla beringei beringei , Gorilla beringei graueri , Gorilla gorilla , Pongo pygmaeus , Pongo abelii , Hylobates lar , Hylobates muelleri , Hylobates klossii , Hylobates agilis , Hoolock hoolock , Symphalangus syndactylus , Macaca fascicularis , Macaca nemestrina , Papio anubis , Papio hamadryas , Theropithecus gelada , Lophocebus albigena , Mandrillus sphinx , Mandrillus leucophaeus , Cercocebus spp . , Erythrocebus patas , Chlorocebus aethiops , Cercopithecus mitis , Miopithecus talapoin , Nasalis larvatus , Colobus polykomos , Trachypithecus cristatus , Semnopithecus entellus , Ateles geoffroyi , Ateles fusciceps , Ateles paniscus , Ateles belzebuth , Alouatta palliata , Alouatta seniculus , Lagothrix lagotricha , Cebus spp . , Saimiri spp . , Aotus spp . , Pithecia spp . , Chiropotes spp . , and Callicebus donicophilus . These specimens are housed at the following collections: American Museum of Natural History ( AMNH ) , Cleveland Museum of Natural History ( CMNH ) , Harvard Museum of Comparative Zoology ( MCZ ) , United States National Museum of National History , Smithsonian Institution ( USNM ) , Field Museum ( FM ) , Berkeley Museum of Vertebrate Zoology ( MVZ ) , Human Evolution Research Center ( HERC ) at the University of California , Berkeley , and the Royal Museum for Central Africa ( RMCA ) . The modern human sample is composed of recent modern individuals of European and African ancestry housed at the Hamann-Todd Collection at the CMNH as well as a Native American population from California housed at the Phoebe A . Hearst Museum of Anthropology ( PAHMA ) at the University of California , Berkeley . Measurements of Ardipithecus ramidus ( ARA-VP-6/500 ) were initially taken on casts at the University of California , Berkeley . Observations were then made on the original fossils at the National Museum of Ethiopia ( NME ) and measurements confirmed by T . White . Six measurements were taken on the foot of each individual using Mitutoyo digital calipers: maximum talar articular length , talar trochlea length , cuboid length , first metatarsal ( MT1 ) length , fifth metatarsal length ( MT5 ) , and fourth proximal phalanx ( PP4 ) length . Talar neck length was derived by subtracting the talar trochlea length from maximum talar articular length . Maximum talar articular length is defined as the maximum proximodistal distance between the most proximal margin of the talar trochlea and the most distal point on the talar head . Talar trochlea length is defined as the maximum proximodistal distance between the most proximal margin of the talar trochlea and the most distal point of the talar trochlea . Cuboid length is defined as the proximodistal distance between the dorsal margin of the calcaneal facet and the most distal point of the tarsometatarsal joint , taken in dorsal view in approximate anatomical orientation . The purpose of measuring cuboid length in this manner is to explicitly avoid the cuboid beak or calcaneal process since it varies extensively among great apes ( Lewis , 1983 ) and because it confounds the cuboid length measurement as a representation of midfoot length since it is articular and housed within a corresponding concavity on the cuboid facet of the calcaneus . MT1 length is defined as the maximum proximodistal distance between the most proximal points on the metatarsal base ( with calipers held flush ) and the most distal point of the metatarsal head . MT5 length is defined as the proximodistal distance between the most proximal point of the cuboid-MT4 articular margin and the most distal point on the metatarsal head . PP4 length is defined as maximum proximodistal distance between the most proximal point of the phalangeal base and the most distal point of the trochlea . The MT5 was chosen to represent non-hallucal metatarsal length because it is preserved in the ARA-VP-6/500 foot . The fossil is missing most of its metatarsal head and its length was estimated by Lovejoy et al . ( 2009a ) using a combination of anatomical and statistical estimation . There is a nearly complete third metatarsal of Ar . ramidus , but it derives from a different locality and is therefore associated with a different individual . There is a partially preserved second metatarsal of Ar . ramidus also from a different individual ( Lovejoy et al . , 2009a ) . Therefore , for this study , the metrics are based on the preserved elements of the ARA-VP-6/500 foot of Ar . ramidus . The individual elements of the bony foot skeleton contribute to the production of three movements used in various forms of primate locomotor behavior that are hypothesized to be reflected in intrinsic foot proportions: hallucal adduction and flexion , non-hallucal digital flexion , and plantarflexion at the talocrural joint . Increasing the length of the first metatarsal should increase the moment arm of the intrinsic adductor musculature such as the m . adductor hallucis across a range of hallucal abduction angles , and therefore should increase the hallucal adduction force during grasping in non-human primates ( Cartmill , 1979 ) . Increasing hallucal metatarsal and non-hallucal phalangeal lengths also contributes to increasing the span of the pedal grasp in taxa with a mobile hallux , which also helps to maintain a friction grip in pedal grasping ( Cartmill , 1979 ) . Previous studies have modeled the foot skeleton as a second-class lever , where the fulcrum is at the metatarsophalangeal joints , the load passes through the talocrural joint at the rearfoot , and the force is produced by the plantarflexor muscles . The load arm is the distance between the fulcrum and the load , whereas the effort arm ( or ‘power arm’ ) is the distance between the insertion of the ankle plantarflexors on the calcaneal tuberosity and the talocrural joint ( Schultz , 1963a; Schultz , 1963b; Strasser , 1992 ) . Increasing the effort arm of the foot relative to the load arm increases the mechanical advantage of the foot skeleton as a lever ( Schultz , 1963a; Schultz , 1963b; Strasser , 1992 ) . However , increasing the load arm increases the range of motion for a given amount of plantarflexor contraction ( Schultz , 1963a; Schultz , 1963b ) . The length of the effort arm of the Ar . ramidus foot is unknown because its calcaneus is not well preserved ( Lovejoy et al . , 2009a ) . There are multiple anatomical strategies for increasing the length of the foot skeleton’s load arm . The length of the load arm can be increased by lengthening the metatarsals , tarsals ( e . g . , the talus and/or cuboid ) , or any combination of these elements . Increasing the length of the metatarsals subjects them to greater bending moments during stance phase , and therefore increases the possibility of injury , so humans achieve a longer load arm by instead increasing the length of the tarsals . As such , in a bipedal heel-strike plantigrade foot , the load arm is increased by lengthening the cuboid and other midtarsal elements . In contrast , in quadrupedal semiplantigrade or digitigrade primates , and indeed other terrestrial cursorial mammals ( Taylor , 1976 ) , the load arm is lengthened by increasing the length of the metatarsals . One implication of these differing anatomical arrangements ( e . g . , increasing metatarsal versus tarsal lengths ) , is that foot proportions may also be a correlate of foot postures ( plantigrady and digitigrady or semiplantigrady ) , which is one of the hypotheses tested by this study using evolutionary modeling . To correct for differences in scale among species , each measurement was divided by the geometric mean of all six measurements per individual ( Jungers et al . , 1995 ) . Morphometric affinities were evaluated using an unweighted pair-group with arithmetic mean ( UPGMA ) cluster analysis on Euclidean distances . The cophenetic correlation coefficient was used to assess the degree to which the resulting dendrogram represented the true pairwise distances between taxa ( Sokal and Rohlf , 1962 ) . Principal Components Analysis ( PCA ) on all six geometric mean-standardized variables was used to reduce , ordinate , and visualize the multivariate data . All evolutionary modeling and ancestral state estimation analyses use the first three principal components ( PCs ) derived from the PCA . The PC scores were used instead of the original data to avoid analytical problems surrounding correlations among variables ( Clavel et al . , 2015 ) and to maximize statistical power ( Boettiger et al . , 2012 ) . Ardipithecus ramidus was added to a molecular phylogenetic tree from 10 k trees ( Arnold et al . , 2010 ) as a stem hominin ( Strait and Grine , 2004; White et al . , 2009; Dembo et al . , 2015 ) with a branch length of 1 . 4 million years in accordance with first and last appearance data for the genus ( 5 . 8–4 . 4 Ma ( Haile-Selassie , 2001; WoldeGabriel et al . , 2001 ) ) using Mesquite software ( Maddison and Maddison , 2017 ) . The branch length for Homo sapiens was reduced in order to improve estimation of phenotypic optima in evolutionary analyses ( Butler and King , 2004; Ingram and Mahler , 2013 ) . The scaling of individual foot elements with body mass among extant taxa ( Smith and Jungers , 1997 ) was conducted using phylogenetic generalized least squares regression ( Grafen , 1989 ) with the ‘caper’ package ( Orme , 2013 ) in R ( R Core Team , 2017 ) . To increase the fit of the evolutionary model to the data , branch lengths were transformed using Pagel’s lambda ( Pagel , 1999 ) , which was estimated with maximum likelihood as a measure of phylogenetic signal in the residual error of each pGLS model ( Revell , 2010 ) . A phylomorphospace was constructed by superimposing the phylogenetic tree on the average principal component ( PC ) scores for each taxon and ancestral values were estimated using a Markov chain Monte Carlo method ( MCMC ) that relaxes assumptions of neutrality and gradualism ( Elliot and Mooers , 2014 ) . Ancestral values estimated under constant rate Brownian motion are affected by taxa that are exceptionally phenotypically derived compared to their close relatives because it is assumed that all branches have evolved at the same rate ( Elliot and Mooers , 2014 ) . The assumption of a constant evolutionary rate therefore results in an ‘averaging effect’ of ancestral values ( Schluter et al . , 1997; Elliot and Mooers , 2014 ) . Therefore , estimates of ancestral values for continuous traits that assume a constant evolutionary rate are potentially biased in the direction of more derived branches characterized by higher phenotypic evolutionary rates . Since this study is focused on estimating ancestral values for humans , great apes , and hominoids , the stable model ( Elliot and Mooers , 2014 ) was specifically chosen in light of evidence for molecular and morphological evolutionary rate differences in hominoids relative to other anthropoids ( Steiper et al . , 2004 ) and in Homo relative to Pan ( Weaver and Stringer , 2015 ) . Ancestral states ( PC scores ) were estimated under a constant rate Brownian motion model and a stable model using StableTraits software version 1 . 5 ( Elliot and Mooers , 2014 ) . Two independent Markov chains were run with 2 , 000 , 000 iterations at a thinning rate of 200 , resulting in 10 , 000 samples each . Priors on evolutionary rate were set to the default settings as implemented in StableTraits and which prevent rates from approaching zero . The two chains converged after 500 , 000 iterations as evidenced by a proportional scale reduction factor ( PSRF ) value approaching 1 ( Brooks and Gelman , 1998 ) . Therefore , the first 600 , 000 iterations for each of the two chains were discarded as burn in . The stable model returns a list of median ancestral values for internal nodes along with their 95% credibility interval ( Elliot and Mooers , 2014 ) . The constant rate Brownian motion model was compared to the stable model using the Bayesian predictive information criterion ( BPIC ) , which is analogous to Akaike’s Information Criterion ( AIC ) in a Maximum Likelihood framework ( Ando and Tsay , 2010 ) . A vector of median ancestral values of the PC scores was supplied for internal nodes and a phylomorphospace was constructed given the topology of the molecular phylogenetic tree using the ‘phytools’ package ( Revell , 2012 ) in R ( R Core Team , 2017 ) . A model-based approach was used to evaluate alternative evolutionary hypotheses: Brownian motion , single-optimum Ornstein-Uhlenbeck ( OU ) , and multi-optimum OU . Prior to evolutionary model comparison , a multivariate extension ( Adams , 2014 ) of Blomberg’s K statistic ( Blomberg et al . , 2003 ) was computed to estimate the phylogenetic signal in the first three principal components using the ‘geomorph’ package ( Adams et al . , 2017 ) in R ( R Core Team , 2017 ) . The Brownian motion model , which is commonly used in phylogenetic comparative analyses , is defined by the stochastic differential equation ( SDE ) : dX ( t ) = σdB ( t ) where X is the trait value , t is time , dB ( t ) is random white noise , and σ is the magnitude of random fluctuations in the evolutionary process . Under Brownian motion all trait changes are independent of previous ones , as well as those on other branches , and trait variance is proportional to time ( i . e . , branch lengths ) . Alternatively , the OU process was quantitatively formalized by Hansen ( 1997 ) to model stabilizing selection as the stochastic differential equation ( SDE ) : dX ( t ) = α ( θ – X ( t ) ) dt + σdB ( t ) where θ is the trait optimum and α is the strength of the ‘restraining force’ acting on a trait around an optimum . Brownian motion is therefore a special case of OU when α = 0 . Hansen ( 1997 ) views phenotypic optima ( θ ) as peaks in an adaptive landscape , which are a compromise among the many possibly conflicting selective demands acting on a trait at any given time . Multi-optima Hansen models reflect adaptive hypotheses based on observations of extant primate posture and locomotion culled from the literature and quantitatively formalized as alternative arrangements of hypothetical phenotypic optima ( θ ) following the methodology of Butler and King ( 2004 ) . Species means represent local optima surrounding a global optimum ( θ ) which correspond to a selective regime ( Hansen , 1997 ) . As such , individual species may differ significantly from one another while simultaneously occupying the same global phenotypic optimum , possibly due to other factors such as drift or pleiotropy ( Hansen , 1997 ) . Therefore , in evolutionary model comparison , the focus is on the number and arrangement of global phenotypic optima ( θ ) and their surrounding local optima ( i . e . , species means ) rather than on individuals within species . Several recent studies have used modeling methods to test evolutionary hypotheses in paleoanthropology ( Almécija et al . , 2015; Grabowski and Jungers , 2017; Fernández et al . , 2018 ) . The evolutionary hypotheses for the link between foot proportions and behavior are informed by observations of locomotor behavior in the wild reported in the literature . The first evolutionary hypothesis is a Brownian motion model ( H1 ) . The second evolutionary hypothesis is the first Hansen model and it reflects a single global phenotypic optimum ( H2 ) . Support for these hypotheses would suggest that there are no major adaptive differences in foot proportions between anthropoid primate groups . Subsequent hypotheses represent multi-optima OU models of increasing complexity ( i . e . , number of phenotypic optima ) . The second Hansen model has three selective regimes associated with advanced bipedalism ( Homo ) , mostly terrestrial quadrupedal locomotion ( Pan , Gorilla , Papio , Theropithecus , Erythrocebus , Chlorocebus aethiops ) and mostly arboreal quadrupedal locomotion in all other taxa ( H3 ) . African apes are competent climbers and possess foot adaptations related to this form of locomotion ( DeSilva , 2009 ) . However , observations from the wild show them to be highly terrestrial . Papio , Theropithecus , Erythrocebus , and Chlorocebus aethiops are the most terrestrial among the cercopithecoids . Previous studies suggest there may be effects of arboreality and terrestriality on foot proportions ( Schultz , 1963a; Jolly , 1967 ) . The third Hansen model is an elaboration of the previous one that splits the terrestrial regime into two: terrestrial heel-strike plantigrady in Pan , and Gorilla , and terrestrial semiplantigrady in Papio , Theropithecus , Erythrocebus , and Chlorocebus aethiops ( H4 ) . Semiplantigrady is defined as any habitual foot posture in which some , but not all , tarsals are in contact with the substrate during stance phase . Heel-strike plantigrady is defined as a foot posture in which the tarsals , principally the calcaneus and its proximal tuberosity , contact the substrate at the beginning of stance phase . Several studies have suggested that African ape tarsal morphology reflects their heel-strike plantigrade foot posture ( Gebo , 1992; Gebo , 1996 ) . Previous work on mammalian foot proportions implies that foot proportions may reflect foot posture ( Taylor , 1976 ) . The fourth Hansen model separates the arboreal taxa from the previous model into mostly arboreal taxa that engage in little climbing and hindlimb-assisted suspension versus mostly arboreal taxa that engage in climbing and hindlimb-assisted suspension more frequently ( i . e . , Pongo abelii , Pongo pygmaeus , Ateles , Alouatta , Lagothrix , H5 ) . Numerous studies have shown that suspensory locomotion is correlated with limb and joint morpology ( Fleagle et al . , 1981; Gebo , 1996 ) . The majority of these studies have been focused on the upper limb and shoulder ( Fleagle et al . , 1981; Gebo , 1996; Hunt , 1996; Young et al . , 2015 ) . The fifth Hansen model further separates mostly arboreal taxa into those that frequently engage in active climbing ( i . e . , Hylobates , Symphalangus , Alouatta ) from the more hindlimb-assisted suspensory taxa such as Pongo and Ateles ( H6 ) . This model attempts to distinguish between possible effects of hindlimb-assisted suspension versus active climbing ( Sarmiento , 1983; Langdon , 1985; Harrison , 1986 ) . Standard model selection criteria ( AICc ) were used to evaluate alternatives and to choose the model that best fit the data . Alternative adaptive hypotheses were evaluated using the ‘OUCH’ package ( Butler and King , 2004 ) in R ( R Core Team , 2017 ) . An additional method was used to identify phenotypic optima without a priori information using the SURFACE method , which stands for SURFACE Uses Regime Fitting with Akaike Information Criterion to model Convergent Evolution ( Ingram and Mahler , 2013 ) . The purpose of SURFACE is to estimate the macroevolutionary adaptive landscape ( i . e . , the number and arrangement of phenotypic optima ) using only a data set and a phylogeny . SURFACE uses a stepwise AIC algorithm to fit a series of Hansen models in two phases: a forward phase in which selective regimes are added , and a backward phase , in which selective regimes are collapsed . The original intent of the SURFACE method was to test for convergence in a clade while minimizing potential biases in the identification of hypothetically convergent ecomorphs a priori ( Ingram and Mahler , 2013 ) . These analyses were carried out using the ‘SURFACE’ package ( Ingram and Mahler , 2013 ) in R ( R Core Team , 2017 ) . Comparison of the model identified by SURFACE ( H7 ) with the a priori models was conducted using the ‘OUCH’ package ( Butler and King , 2004 ) in R ( R Core Team , 2017 ) . Therefore , a total of seven evolutionary hypotheses describing the evolution of the anthropoid foot were tested . Model comparison offers a powerful method for testing evolutionary hypotheses , but several researchers have noted the importance of conducting simulations to evaluate statistical power in model selection ( Boettiger et al . , 2012; Cooper et al . , 2016 ) . Simulations were conducted using a Monte Carlo method in order to evaluate statistical power in model selection closely following the approach outlined in Boettiger and colleagues ( Boettiger et al . , 2012 ) . The purpose of this approach is to determine whether alternative evolutionary models can be distinguished from one another given the data set and phylogeny , and if so , which of the models is best ( Boettiger et al . , 2012; Lst and Ané , 2014 ) . It is therefore an alternative to other model selection criteria such as Akaike’s Information Criterion ( AIC ) . First , parameters ( e . g . , log-likelihood ) were estimated from the data ( i . e . , the first three principal components ) under models A and B ( e . g , Brownian motion versus single optimum Ornstein-Uhlenbeck ) . Second , 1000 data sets were simulated under the estimated parameters for each of the two models . Third , models A and B were both re-fit to each of the two 1000 simulated data sets , producing four sets of 1000 log-likelihoods . Finally , the likelihood ratio statistic , which is defined as δ = −2 ( logL0 – logL1 ) , where logL0 is the log-likelihood of model A and logL1 is the log-likelihood of model B , was calculated , resulting in two distributions of 1000 values for the likelihood ratio statistic under both models . The difference between the distributions reflects statistical power and the proximity of the empirical likelihood ratio statistic to the distributions indicates which of the two models is best . The following simulations were conducted to compare models of increasing complexity: Brownian motion vs . OU1 , OU2 vs . OU3 , OU3 vs . OU4 , and OU4 vs . OU5 . These simulations were conducted using R ( R Core Team , 2017 ) . | Walking on two legs is considered to be one of the first steps towards becoming human . While some animals are also able to walk on two legs , such as kangaroos , birds , and some rodents , the way they move is nevertheless quite distinct to the way humans walk . How animals evolve traits is influenced by the characteristics of their ancestors . But what exactly was the common ancestor of humans and chimpanzees like ? Most primates are suited for a life in the trees . But some also have skeletal characteristics associated with living on the ground . For example , the feet of chimpanzees and gorillas show adaptations that suit life on the ground , such as walking on the sole of the foot with a heel first foot posture . So far , it was unclear whether the ancestor of humans and chimpanzees was primarily adapted to living on the ground or in the trees . To investigate this further , Prang studied the oldest-known fossil foot ( 4 . 4 million years ) attributed to the hominin Ardipithecus ramidus . This involved using evolutionary models to evaluate the relationship between foot bone proportions and the locomotory behaviour of monkeys and apes . The results revealed that humans evolved from an ancestor that had a foot similar to living chimpanzees and gorillas . The African ape foot is uniquely suited to life on the ground , including shorter toe bones , but also shows some adaptations to life in the trees , such as an elongated , grasping big toe . Therefore , the locomotion of our common ancestor probably bore a strong resemblance to these two ape species . Moreover , if the last common ancestor already had ground-living characteristics , the first step of the evolution of human bipedalism did not involve descending from the trees to the ground , as our ancestors had already achieved this milestone in some form and frequency . This is an important discovery . If this ancestor already had adaptations for life on the ground , why did only humans evolve to walk upright despite the retention of climbing capabilities in the earliest human relatives ? A next step could be to investigate what selective pressures favored upright walking in a partly ground-living African ape . This may provide us with more insight into our own evolutionary story as well as the ways in which living primates evolve adaptations in an ecological context . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"evolutionary",
"biology"
] | 2019 | The African ape-like foot of Ardipithecus ramidus and its implications for the origin of bipedalism |
Hybrid sterility is one of the reproductive isolation mechanisms leading to speciation . Prdm9 , the only known vertebrate hybrid-sterility gene , causes failure of meiotic chromosome synapsis and infertility in male hybrids that are the offspring of two mouse subspecies . Within species , Prdm9 determines the sites of programmed DNA double-strand breaks ( DSBs ) and meiotic recombination hotspots . To investigate the relation between Prdm9-controlled meiotic arrest and asynapsis , we inserted random stretches of consubspecific homology on several autosomal pairs in sterile hybrids , and analyzed their ability to form synaptonemal complexes and to rescue male fertility . Twenty-seven or more megabases of consubspecific ( belonging to the same subspecies ) homology fully restored synapsis in a given autosomal pair , and we predicted that two or more DSBs within symmetric hotspots per chromosome are necessary for successful meiosis . We hypothesize that impaired recombination between evolutionarily diverged chromosomes could function as one of the mechanisms of hybrid sterility occurring in various sexually reproducing species .
Hybrid sterility ( HS ) is a postzygotic reproductive isolation mechanism that enforces speciation by restricting gene flow between related taxa . HS is a universal phenomenon observed in many eukaryotic inter-species hybrids , including examples in yeast , plants , insects , birds , and mammals ( Coyne and Orr , 2004; Maheshwari and Barbash , 2011 ) . In the early days of genetics , HS was difficult to accommodate in Darwin's theory of evolution by natural selection . In time , however , the Bateson–Dobzhansky–Muller incompatibility ( BDMI ) hypothesis ( Muller and Pontecorvo , 1942; Dobzhansky , 1951; Orr , 1996 ) explicated HS , and more generally any hybrid incompatibility , as a consequence of the independent divergence of mutually interacting genes resulting in aberrant interaction of the new alleles that have not been tested by natural selection . HS has several common features across various sexually reproducing eukaryotic species . Haldane's rule posits that if one sex of the F1 offspring of two different animal races is absent , rare , or sterile , it is the heterogametic sex ( XY or ZW ) ( Haldane , 1922 ) . Another common feature refers to the disproportionately large role of Chr X compared to that of autosomes in reproductive isolation ( Presgraves , 2008 ) . More recently , interaction between selfish genomic elements causing meiotic drive and their suppressors has been implicated in some instances of reproductive isolation ( Orr , 2005; Zhang et al . , 2015 ) . The molecular mechanisms underlying HS remain an unresolved question . Historically , genic and chromosomal mechanisms of HS had been hypothesized , but the latter were soon dismissed as unlikely on the grounds that large chromosomal rearrangements do not segregate with HS genetic factors ( Dobzhansky , 1951 ) . Other possible forms of non-genic chromosomal HS were not considered because of the limited knowledge of the carrier of genetic information at the time . Thus , for the past 80 years or so , the focus on the genic control of HS prevailed ( Dobzhansky , 1951; Orr , 1996; Forsdyke , 2017 ) . In studies mapping HS genes , the Drosophila group of species has been the model of choice , yet only five Drosophila HS genes , namely OdsH , JYAlpha , Ovd , agt , and Taf1 , have been identified to date , none of which has a known interacting partner predicted by the BDMI hypothesis ( Ting et al . , 1998; Masly et al . , 2006; Phadnis and Orr , 2009 ) . The low success rate of the positional cloning of HS genes was explained by the oligogenic or polygenic nature of HS phenotypes and by the inherent difficulty in genetically dissecting the phenotype that prevents its own transfer to progeny . Over 40 years ago , we introduced the house mouse ( Mus musculus ) as a mammalian model for the genetic analysis of HS . The first mouse HS locus Hst1 was genetically mapped in crosses of laboratory inbred strains ( predominantly of Mus musculus domesticus ( Mmd ) origin ) with wild Mus musculus musculus ( Mmm ) mice ( Forejt and Iványi , 1974 ) . Later , we developed the PWD/Ph and PWK/Ph inbred strains purely from the wild Mmm mice of Central Bohemia ( Gregorová and Forejt , 2000 ) and used them in the positional cloning of Hst1 by high-resolution genetic crosses and physical mapping ( Gregorová et al . , 1996; Trachtulec et al . , 1997 ) . Finally , we identified the Hst1 locus with the PR-domain-containing nine gene ( Prdm9 ) ( Mihola et al . , 2009 ) , which codes for histone H3 lysine 4/lysine 36 methyltransferase ( Powers et al . , 2016 ) the first and still the only HS gene known in vertebrates . Most of the tested laboratory inbred strains share either the Prdm9Dom2 or the Prdm9Dom3 allele ( Parvanov et al . , 2010; Brunschwig et al . , 2012 ) . The former allele was found in inbred strains producing sterile male hybrids when crossed with PWD females , whereas the Prdm9Dom3 was observed in the strains that yielded quasi-fertile males in the same type of inter-subspecific crosses ( Forejt et al . , 2012 ) . The male sterility of ( PWD x C57BL/6 ) F1 ( henceforth PB6F1 ) hybrids depends on the interaction of the heterozygous allelic combination Prdm9Msc/Prdm9Dom2 with the PWD allelic form of the X-linked Hybrid sterility X Chromosome 2 locus , Hstx2Msc ( Dzur-Gejdosova et al . , 2012; Bhattacharyya et al . , 2014 ) . For the sake of clarity and to stress the origin of the alleles , we will use the names Prdm9PWD , Prdm9B6 and Hstx2PWD in the rest of this paper . Any other tested allelic combination of these two major HS genes yields fully fertile or subfertile male hybrids ( Dzur-Gejdosova et al . , 2012; Flachs et al . , 2012 ) . The proper allelic combination of Prdm9 and Hstx2 genes is necessary but not sufficient to govern HS completely because less than 10% instead of the expected 25% of ( PWD x B6 ) x B6 male backcross progeny replicated the infertility of male PB6F1 hybrids ( Dzur-Gejdosova et al . , 2012 ) . Initially , we explained this 'missing heritability' by assuming the genic interaction of three or more additional HS genes with a small effect that had escaped the genetic screen ( Dzur-Gejdosova et al . , 2012 ) . However , an alternative , non-genic explanation emerged from the analysis of meiotic phenotypes of sterile hybrids . We observed multiple unsynapsed autosomal pairs decorated by phosphorylated histone γH2AX as a mark of persisting unrepaired DNA double-strand breaks ( DSBs ) in approximately 90% of primary spermatocytes of infertile PB6F1 hybrids . The asynapsis was accompanied by disturbed transcriptional inactivation of sex chromosomes at the first meiotic prophase ( Bhattacharyya et al . , 2013 , 2014 ) . The failure of intersubspecific homologs to synapse was clearly dependent on interhomolog interactions , and we suggested that their fast-evolving nongenic DNA divergence could be the causal factor . Because meiotic asynapses of different origin are known to compromise the normal progression of the first meiotic division ( Forejt , 1984 , 1996; Mahadevaiah et al . , 2008; Burgoyne et al . , 2009 ) , we proposed that Prdm9 and Hstx2-directed asynapsis per se could be the ultimate cause of the sterility of male hybrids . Recently , the role of PRDM9 zinc-finger-domain binding sites within noncoding genomic DNA has been demonstrated in PB6F1 male HS . Replacement of the mouse sequence encoding the PRDM9 zinc-finger array with the orthologous human sequence reversed sterility in ( PWD x B6-Prdm9Hu ) F1 hybrid males ( Davies et al . , 2016 ) . In PB6F1 hybrids , roughly 70% of the Prdm9-directed DSBs hotspots identified by the DMC1 ChIP-seq method were enriched on the 'nonself' homologous chromosome , as the DSB hotspots determined by the B6 allele of Prdm9 were found predominantly on PWD chromosomes , and vice versa . Such hotspots were designated as asymmetric DSB hotspots . Chromosome-specific quantification of asymmetry correlated well with the asynapsis rate across five arbitrarily chosen chromosomes of PB6F1 hybrids ( Davies et al . , 2016; Smagulova et al . , 2016 ) . Another , non-exclusive interpretation of DMC1 ChIP-seq data pointed to significant enrichment of PRDM9-independent hotspots in the PB6F1 hybrid testis , which occurs in promoters and other regulatory motifs and which is characteristic of spermatogenic arrest in Prdm9 knockout males ( Smagulova et al . , 2016 ) . Recently , one third of PRDM9-dependent DSBs were reported within sequences that have at least some repetitive character , indicating that inappropriately high DSB levels in transposons and other repetitive elements may contribute to the infertility seen in some mouse hybrids ( Yamada et al . , 2017 ) . In this work , we studied the relationship between meiotic chromosome asynapsis , intersubspecific heterozygosity and male HS in a series of PB6F1 hybrids carrying recombinant chromosomes with Mmm/Mmm consubpecific ( belonging to the same subspecies ) PWD/PWD homozygous intervals on Mmm/Mmd intersubpecific ( belonging to different subspecies ) PWD/B6 heterozygous background . We report the restoration of synapsis of intersubspecific chromosome pairs in the presence of 27 Mb or more of consubspecific sequence , and the reversal of HS by targeted suppression of asynapsis in the four most asynapsis-sensitive chromosomes . Our findings point to the chromosomal basis of Prdm9-directed hybrid male infertility as a ( nonexclusive ) alternative to a widely accepted concept of hybrid sterility driven by multiple genic incompatibilities .
First , we ascertained the frequency of meiotic asynapsis separately for each chromosome pair of PB6F1 hybrid males by combining the use of fluorescence in-situ hybridization ( FISH ) to decorate chromatin from individual chromosomes with immunostaining of synaptonemal complex protein 3 ( SYCP3 ) ( a major component of axial/lateral elements ) , to visualize synaptonemal complexes , and HORMA domain-containing protein-2 ( HORMAD2 ) ( Wojtasz et al . , 2012 ) , to identify the axial elements of unsynapsed chromosomes ( Figure 1A ) . Altogether , 4168 pachynemas from 40 PB6F1 hybrid males were analyzed . All autosomes of hybrid males displayed a certain degree of asynapsis , classified as complete , partial , or intermingled ( more than two tangled univalents within labeled chromatin cloud ) , with frequencies ranging from 2 . 6% ( Chr 1 ) to 42 . 2% ( Chr 19 ) ( Figure 2—source data 1 ) . A strong bias was evident towards higher asynapsis rates in the five smallest autosomes ( p=5 . 2×10−14 , comparison of Generalized Linear Mixed Models [GLMM] , Figure 2A ) . Recently , SPO11 oligos released during the processing of DSBs were sequenced , mapped and quantified at chromosome-wide scale in male mice of the B6 laboratory inbred strain ( Lange et al . , 2016 ) . This information , together with the estimated frequency of asymmetric DSB hotspots in PB6F1 hybrids ( Davies et al . , 2016; Smagulova et al . , 2016 ) , enabled us to calculate the possible correlation between the number of DSBs within symmetric hotspots ( hereafter symmetric DSBs ) per chromosome per cell and synapsis between intersubspecific homologs . The calculation is based on and limited by the following premises: ( i ) the overall densities of DSBs on individual chromosomes of B6 and PB6F1 hybrid males are similar; ( ii ) approximately 250 DSBs occur per leptotene/zygotene cell ( Kauppi et al . , 2013 ) ; and ( iii ) the 0 . 28 proportion of symmetric DSB hotspots in ( PWD x B6 ) F1 hybrid males ( Davies et al . , 2016 ) is constant in all autosomes . Under these conditions , a strong negative correlation ( Spearman’s ρ=−0 . 760 , p=0 . 0003 ) of asynapsis rate with predicted symmetric DSBs ( Lange et al . , 2016 ) can be seen ( Figure 2—source data 2 ) . This correlation is stronger than the correlation of the asynapsis rate with the chromosomal physical length ( Spearman’s ρ=−0 . 681 , p=0 . 0013 ) . Even though the chromosomal length and the expected number of symmetric DSB hotspots strongly correlate ( Spearman’s ρ=0 . 916 , p=1 . 1 × 10−7 ) , we observed that it is the symmetric DSB hotspots that affect the asynapsis rate . The chromosomal length does not add any additional explanation of the asynapsis rate to that provided by symmetric DSBs ( p=0 . 709 , comparison of GLMM models ) . On the contrary , the symmetric DSBs add an additional explanation of the asynapsis rate to that provided by the chromosomal length ( p=0 . 046 , comparison of GLMM models ) . Thus , our findings suggest that synapsis of a pair of homologous chromosomes depends on the presence of a certain minimum number of symmetric DSBs , as we elaborate further using a simulation described in the 'Discussion' . Further , we examined the asynapsed chromosomes of PB6F1 hybrids for localization of active chromatin using confocal fluorescence microscopy after Cot-1 RNA FISH ( Hall et al . , 2014 ) and HORMAD2 immunolabeling . Fluorescence signal quantification revealed that subnuclear regions of asynapsed chromosomes composed of sex chromosomes and/or autosomal univalents were lacking active euchromatin in contrast to other regions of the pachytene nuclei ( Figure 1—video 1 ) . We propose that the absence of active euchromatin is a consequence of the meiotic synapsis failure of intersubspecific chromosomes , known as meiotic silencing of unsynapsed chromatin ( MSUC [Burgoyne et al . , 2009] ) , which can act as an epigenetic component contributing to the meiotic phenotypes of sterile hybrids ( Larson et al . , 2016 ) . We have shown previously that meiotic asynapsis affects intersubspecific ( PWD/B6 ) but not consubspecific ( PWD/PWD ) pairs of homologous chromosomes in sterile male hybrids from crosses of PWD females and B6 . PWD-Chr # consomic males ( Gregorová et al . , 2008; Bhattacharyya et al . , 2013 , 2014 ) . Here , we searched for the minimum length of the PWD/PWD consubspecific sequence that still could secure synapsis of a chromosome and potentially restore fertility in the hybrids . Instead of replacing the whole B6 chromosome with its PWD homolog , we generated recombinant PWD/B6 and B6/PWD ( centromere/telomere ) chromosomes . To do that , we crossed the male hybrids between two B6 . PWD-Chr # consomic strains with a PWD female and estimated the minimum size and location of consubspecific PWD/PWD stretches needed for synapsis rescue , as shown in Figure 3A . In three such generated 'two-chromosome crosses' ( hereafter referred to as 2-chr crosses ) we investigated the effect of the PWD/PWD consubspecific intervals on the asynapsis rate in six different chromosomes — two in a given experiment , namely Chr 5 and Chr 12 ( Figure 3—source datas 1 and 2 ) , Chr 7 and Chr 15 ( Figure 3—source datas 3 and 4 ) and Chr 17 and Chr 18 ( Figure 3—source datas 5 and 6 ) . Altogether , 122 chromosomes from over 12 , 000 pachynemas were examined . All male progeny of the 2-chr crosses were fully sterile , with low testis weight and the absence of sperm in the epididymis . The analysis of data from six recombinant chromosomes revealed the common features described below . Introduction by recombination of 27 Mb or more of a consubspecific ( PWD/PWD ) interval into a pair of intersubspecific ( PWD/B6 ) homologs effectively suppressed the asynapsis rate below the baseline of 5% in all six studied autosomes ( Figure 3B ) . The efficiency of synapsis rescue was gradual with an apparent change point ( Figure 3B ) . To describe the pattern in the data , a segmented regression model was used ( see 'Materials and methods' ) . The model based on the data pooled from all 2-chr crosses was selected as the best model with an estimated change point at 27 . 14 Mb ( 19 . 36; 34 . 91 ) ( 95% CI ) ( see Figure 3—source data 7 ) . The slope of the decrease of asynapsis in the region of consubspecific intervals shorter than 27 . 14 Mb differed for different chromosomes ( p=3 × 10−11 , F-test ) . For each chromosome , the asynapsis rate decreased from the maximal value measured for non-recombinant PWD/B6 ( with 0 Mb of PWD/PWD ) down below 5% estimated for the change-point value of 27 . 14 Mb of PWD/PWD interval . In spite of the known role of subtelomeric ( bouquet ) association in chromosome pairing ( Ishiguro et al . , 2014; Scherthan et al . , 2014 ) , the location of the consubspecific sequence at the telomeric end was not essential for synapsis ( p=0 . 9573 , F-test ) . The PWD/PWD intervals of sufficient size rescued synapsis whether located at the centromeric ( proximal , n = 14 cases ) , interstitial ( n = 3 ) , or telomeric ( distal , n = 14 ) position ( Figure 3—source data 1–6 ) . The experiments described above have shown that a randomly located consubspecific PWD/PWD interval of 27 or more Mb on otherwise intersubspecific PWD/B6 background is sufficient to restore the pachytene synapsis of a given autosomal pair . To check the causal relationship between meiotic chromosome asynapsis and HS , we attempted to reverse HS by reducing the asynapsis in the four most asynapsis-prone chromosomes . Provided that hybrid male sterility is directly dependent on chromosome synapsis , we predicted ( by multiplying the probabilities of the synapsis of individual chromosomes obtained in F1 hybrids ) that complete elimination of asynapsis of four of the shortest autosomes ( excluding Chr 17 to avoid Prdm9PWD/PWD interference ) could increase the proportion of primary spermatocytes that have the full set of synapsed autosomes up to 26 . 7% and could potentially abolish the apoptosis of these cells to yield around 5 million sperm cells in the epididymis of the hybrid males . To evaluate this prediction experimentally , random intervals of consomic Chrs 15PWD , 16PWD , 18PWD , and 19PWD were transferred onto the genetic background of B6 mice in a four-generation cross as shown in Figure 4A . Eleven G3 males selected for maximal extent of PWD sequence on these chromosomes were crossed to PWD females ( Figure 4—source data 1 ) . The resulting G4 hybrid male progeny ( hereafter referred to as a 4-chr cross ) displayed various degrees of PWD homozygosity in the studied consomic autosomes on an otherwise intersubspecific PWD/B6 genetic background . As predicted , a significant fraction of hybrid males did indeed show partial rescue of spermatogenesis . In the PB6F1 cross , 100% of hybrid males displayed no sperm in the epididymis , whereas in the 4-chr cross , only 51 . 7% of 87 G4 males were aspermic , 19 . 5% had a 0 . 01–0 . 74 × 106 sperm count , and 28 . 7% had 1 . 0–13 . 7 × 106 sperm cells ( Figure 4—figure supplement 1 , Figure 4—source data 2 ) . Next , we asked whether the reversal of meiotic arrest correlates with the recovery of meiotic synapsis of recombined chromosomes and with the size of PWD/PWD consubspecific stretches in the four manipulated chromosomes . Eighteen G4 males were deliberately selected according to their fertility parameters , 13 with HS partial rescue , displaying sperm cells in the epididymis ( 0 . 1 × 10–6 . 9 × 106 ) , and five aspermic controls . The meiotic analysis of over 6500 pachynemas from the genotyped males confirmed the prediction based on the results of 2-chr crosses . The nonrecombinant PWD/PWD consubspecific bivalents were always fully synapsed , whereas all nonrecombinant PWD/B6 intersubspecific pairs revealed the highest frequencies of asynapsis . All recombinant chromosomes with consubspecific intervals of sufficient length ( Figure 4—source data 3; see Figure 3—source data 7 for change point estimates ) effectively restored synapsis . Moreover , the presence of sperm cells corresponded with the rescue of synapsis of consomic chromosomes . As a rule of thumb , the hybrids had sperm when asynapsis was suppressed in at least three of four segregating chromosomes and when the probability of all four consomic chromosomes being synapsed was >0 . 7 ( p=0 . 0014 , Mann-Whitney test ) . Chrs 16 , 18 , and 19 contributed the strongest effect ( Figure 4—source data 3 ) . Provided that the probability of failure of the synapsis of each chromosome was completely independent of the rest of the hybrid genome , then the asynapsis rate of a particular nonrecombinant intersubspecific chromosome pair would be the same in F1 hybrids , 2-chr crosses , and the 4-chr cross . Moreover , the frequency of pachynemas with all chromosomes synapsed could be predicted by multiplication of the observed frequencies of the synapsis of individual chromosomes ( see Figure 2—source data 2 ) . Such predicted values would be close to the values directly read from the meiotic spreads and would lie along the diagonal in Figure 5 . As shown below , both types of analysis clearly revealed that the asynapsis rate of a particular chromosome depends on the synapsis status of other chromosomes . First , in PB6F1 hybrids , the observed 13 . 1% ( 11 . 4–14 . 9% ) ( 95% CI ) of fully synapsed pachynemas was double ( p=0 . 023 , Mann-Whitney test ) the 6 . 6% ( 5 . 3–8 . 1% ) rate expected by the multiplication of the observed synapsis rates of individual chromosomes ( Figure 5—source data 1 ) , indicating a trans effect of synapsed autosomes on the probability of the asynapsis of other PWD/B6 chromosome pairs . The trans effect was more pronounced in 2-chr cross and 4-chr cross experiments . Second , at the level of individual chromosomes , the most straightforward comparison was between the nonrecombinant PWD/B6 chromosomes , where the asynapsis rate was dramatically reduced in 2-chr crosses or the 4-chr cross ( odds ratio [OR]=0 . 687 , p=0 . 0002 , GLMM ) compared to F1 hybrid rates . The trans effect was analyzed further for Chromosomes 15 , 16 , 18 and 19 by comparing the asynapsis rate of a given non-recombinant PWD/B6 pair with the other three analyzed chromosomes in the 4-chr cross and in F1 hybrids . The Supplement 1 to Figure 5 shows a negative correlation from r=−0 . 45 for Chr 16 to r =−0 . 88 for Chr 15 . On average , if the predicted synapsis rate of three chromosomes is increased by ten percent , we can expect a 4 . 18% 2 . 72–5 . 34% ) ( 95% CI ) decrease of asynapsis rate of the fourth chromosome ( p=0 . 0266 , log-log regression ) . However , for the chromosomes with at least 34 . 9 Mb of PWD/PWD segment ( right bound of 95% CI of change point estimate ) , for which an additional length of PWD/PWD was not shown to affect asynapsis rate anymore , the trans effect could not be detected ( p=0 . 186 , comparison of GLMM models ) . To conclude , the trans effect is the second non-genic effect modifying the asynapsis rate primarily caused by the cis-acting inter-homolog incompatibility in PB6F1 primary spermatocytes . The significance and the magnitude of the trans effect depends on the cis-acting inter-homolog incompatibility .
Davies et al . ( 2016 ) found that the DNA-binding zinc-finger domain of the PRDM9 molecule is responsible for sterility in PB6F1 hybrids . Further , they found that in the sterile hybrids , most PRDM9PWD-specific hotspots reside on B6 chromosomes and , vice versa , that most of the PRDM9B6-binding sites are activated on PWD chromosomes . This asymmetry could be explained in part by erosion of the PRDM9-binding sites due to preferential transmission to progeny of the altered hotspots motifs ( Boulton et al . , 1997; Myers et al . , 2010 ) . In a parallel study , Smagulova et al . ( 2016 ) identified a novel class of strong hotspots in PB6F1 hybrids that are absent in PWD and B6 parents and that are apparently related to asymmetric hotspots described by Davies et al . ( 2016 ) . Moreover , Prdm9-independent 'default' hotspots were particularly enriched in Chr X , and we noticed that the percentage of these ‘default’ hotspots in autosomes correlates with the present data on asynapsis rate in F1 hybrids ( Spearman’s ρ=0 . 69 , p=0 . 0012 ) . These Prdm9-independent hotspots may represent the late-forming DSBs on unsynapsed chromosomes and , as such , they may be a consequence rather than the cause of meiotic asynapsis ( see Kauppi et al . [2013] ) . We found that meiotic asynapsis affects each autosomal pair in PB6F1 intersubspecific hybrids at distinctively unequal rates , with shorter chromosomes affected more often than longer ones . A similar pattern of higher sensitivity of smaller autosomes to the synapsis failure was observed in mice with lowered dosages of SPO11 ( Kauppi et al . , 2013 ) and in the consequent two-fold DSB reduction . The fact that the asynapsis rate of sterile F1 hybrids correlates better with SPO11-oligo-derived DSB density ( inferred from B6 mouse strain data [Lange et al . , 2016] ) than with the chromosome length bringsexperimental support for the idea ( Davies et al . , 2016 ) linking the asynapsis in sterile PB6F1 hybrids to an insufficient number of symmetric DSB hotspots . Provided that a shortage of symmetric hotspots ( Davies et al . , 2016 ) is the ultimate cause of the failure of meiotic synapsis of intersubspecific homologs , then the full synapsis could be restored by exchanging the asymmetric hotspots for the symmetric ones . To test this prediction experimentally , we constructed pairs of PWD/B6 intersubspecific homologs carrying stretches of PWD/PWD consubspecific intervals , which by definition cannot carry asymmetric hotspots . We found that chromosomes with 27 Mb or longer stretches of consubspecific sequence always rescued full synapsis in hybrid males . The position of the consubspecific interval along the chromosome was not critical for synapsis rescue , in accordance with the finding that synaptonemal complexes nucleate at multiple recombination sites in each chromosome ( Zickler and Kleckner , 2015; Finsterbusch et al . , 2016 ) . We assume that the presence of symmetric DSB in the PWD/PWD homozygous stretches exceeded the threshold of a minimum number of timely repaired DSBs , thus rescuing normal meiotic synapsis . Allowing for the assumptions enumerated in the 'Results' section , the number of DSBs necessary for proper synapsis of a given chromosome can be estimated on the basis of the expected distribution of symmetric DSB hotspots on all autosomes and their asynapsis ratios in sterile F1 hybrids ( Figure 2—source data 2 ) . We aimed to model how the induction and repair of DSBs influence proper meiotic synapsis , and tried to estimate the minimum number of symmetric DSBs per chromosome sufficient for full meiotic synapsis . Our model predicts that in approximately 25% of cases , a chromosome is asynapsed because there are only asymmetric DSBs and no symmetric DSBs ( slope of the regression of P[asynapsis] on P[0 symmetric DSBs]=4 . 22 ) . Assuming a critical threshold of the required DSBs , the remaining 75% of asynapsis could occur on chromosomes with one symmetric DSB and with the other DSBs being asymmetric . Thus , it is consistent with our data that two symmetric DSBs per chromosome could be sufficient for full development of the synaptonemal complex , as shown in Figure 6 ( slope of the regression of P[asynapsis] on P[0 or one symmetric DSBs]=1 . 00 ) . The same conclusion also holds true for 2-chr cross and 4-chr cross experiments ( Figure 6—figure supplements 1 and 2 ) . The deviations from the diagonal in Figure 6 depicting the 4-chr cross can be explained by the trans-effect described in the 'Results' section . The trans effect as reported in this paper refers to the enhanced probability of synapsis of a pair of intersubspecific homologs depending on successful pairing of other chromosomes in males with the Prdm9PWD/Prdm9B6 , Hstx2PWD 'hybrid sterility' genotype . The mechanism of the trans effect is unknown . Kauppi et al . ( 2013 ) discussed the chain reactions of asynapsis , in which asynapsis of one or more chromosomes observed in male mice with lowered dosage of SPO11 increased the risk of asynapsis of other chromosomes by engaging them in nonhomologous synapsis among themselves or with the non-PAR region of X chromosome . An alternative explanation of the trans effect could involve an unspecified rate- or time-limiting step involving the DSB repair machinery . The engagement of the X chromosome in autosomal asynapsis could also be related to the fertility of female hybrids ( Forejt , 1996; Kauppi et al . , 2013 ) . PB6F1 hybrid females are fertile , obeying Haldane's rule ( Haldane , 1922 ) but their oogenesis is impaired , having 45% of pachynemas with one or more asynapsed autosomes . Nevertheless , the effect of Prdm9 on asynapsis in female PB6F1 meiosis seems weak or absent ( Bhattacharyya et al . , 2013 , 2014 ) . The vast majority of literature on the genetic mechanism of infertility of inter ( sub ) specific hybrids focuses on the genetic mapping of hybrid sterility genes , their possible epistatic incompatibilities and evolutionarily diverged structure or expression pattern ( Maheshwari and Barbash , 2011; Presgraves , 2010; Civetta , 2016; Mack and Nachman , 2017 ) . Likewise , our early studies considered univalents in PB6F1 primary spermatocytes as a secondary consequence of meiotic arrest caused by genic incompatibilities ( Forejt and Iványi , 1974; Forejt , 1996 ) . However , quantitative meiotic analyses revealed that 90% of primary spermatocytes carry one or more pairs of asynapsed homologs and , more importantly , that asynapsis is chromosome-autonomous , depending on inter-homolog ( cis- ) interactions ( Bhattacharyya et al . , 2013 , 2014 ) . The findings described in this paper provide the first direct link between Prdm9-controlled asynapsis and meiotic arrest in PB6F1 male hybrids . We showed that by deliberately manipulating the synapsis of homologous chromosomes , we could modify the extent of meiotic arrest in intersubspecific PB6F1 hybrids in a predictable way . Admittedly , the exact molecular basis of meiotic asynapsis and subsequent spermatogenic arrest in PB6F1 males is still unclear and the lack of symmetric DSBs is not necessarily the only explanation for the sterility of thePB6F1hybrid . For instance , multimerization of PRDM9 mediated by PRDM9's zinc fingers ( Baker et al . , 2015; Altemose et al . , 2017 ) could alter its DNA-binding properties and enable the default PRDM9-independent hotspots ( Smagulova et al . , 2016 ) to appear and to generate the Prdm9 null-like phenotype ( Hayashi et al . , 2005 ) . Recent identification of DSB hotspots within repetitive sequences ( Yamada et al . , 2017 ) indicate another potential threat to homologous synapsis in intersubspecific hybrids , which could be caused by illegitimate interactions with nonhomologous chromosomes or by the absence of an allelic PRDM9 binding site in the genome of the other subspecies . Models of the molecular mechanism of hybrid sterility will need to take into consideration that all intersubspecific pairs of homologs synapse properly in the majority of pachynemas . A particular chromosome fails to synapse in as few as 2% and maximally in 45% of PB6F1 pachynemas . Oddly , the B6 allele of the X-linked Hstx2 locus dramatically increases the pairing efficiency but causes only a small change , if any alteration , in the profile of asymmetric hotspots ( Davies et al . , 2016; Smagulova et al . , 2016 ) . The asymmetric DSBs could affect meiotic pairing by hindering repair when searching for the allelic site of a homologous chromosome as a template . The homologous sequence may be inaccessible because of its inappropriate chromatin conformation , such as lack of trimethylation of Lysine four and Lysine 36 of histone H3 or because a critical alteration of the PRDM9-binding motif may provoke the antirecombination activity of the mismatch repair machinery ( Chakraborty and Alani , 2016 ) to prevent the repair . It is also probable that some 'difficult' DSBs could be repaired using sister chromatid as a DNA template during the delayed phase of repair when the nonhomologous compensatory synapsis can occur and when the non PAR X chromosome DSBs are most probably repaired ( Kauppi et al . , 2013 ) . However , such inter-sister recombination cannot contribute to the homolog's synapsis . Hybrid sterility , as well as pairing of homologous chromosomes and meiotic recombination , are universal biological phenomena common to the majority of sexually reproducing organisms . We hypothesize that meiotic pairing and hybrid sterility controlled by Prdm9 could represent a special case of a more universal reproductive isolation mechanism that is based on meiotic recombination . It is tempting to speculate that the mechanisms that safeguard recombination between homologous allelic sequences could function as checkpoints that disable recombination after homologous sequences have diverged sufficiently during the isolation of closely related taxa . Originally , such an inter-species barrier was proposed by Radman and colleagues ( Rayssiguier et al . , 1989; Stambuk and Radman , 1998 ) to prevent homeologous recombination between Escherichia coli and Salmonella typhiimurium . Among eukaryotes , the role of the mismatch repair system in reproductive isolation has been reported in Saccharomyces species ( Hunter et al . , 1996; Greig et al . , 2003; Liti et al . , 2006 ) . An exciting possibility , which is experimentally testable , posits an antirecombination machinery as a means to gradually restrict gene flow between related taxa , a 'cause in fact' of speciation .
The mice were maintained at the Institute of Molecular Genetics in Prague and Vestec , Czech Republic . The project was approved by the Animal Care and Use Committee of the Institute of Molecular Genetics AS CR , protocol No 141/2012 . The principles of laboratory animal care , Czech Act No . 246/1992 Sb . , compatible with EU Council Directive 86/609/EEC and Appendix of the Council of Europe Convention ETS , were observed . Simple sequence length polymorphisms ( SSLP ) markers used for genotyping consomic chromosomes in 2-chr crosses and 4-chr cross are listed in Figure 3—source data 8 . The PWD/Ph inbred strain originated from a single pair of wild mice of the Mus musculus musculus subspecies trapped in 1972 in Central Bohemia , Czech Republic ( Gregorová and Forejt , 2000 ) . The C57BL/6J ( B6 ) inbred strain was purchased from The Jackson Laboratory . The panel of 27 chromosome substitution strains C57BL/6J-Chr #PWD ( abbreviated B6 . PWD-Chr # ) was prepared in our laboratory ( Gregorová et al . , 2008 ) and is maintained by the Institute of Molecular Genetics AS CR , Division BIOCEV , Vestec , Czech Republic , and by The Jackson Laboratory , Bar Harbor , Maine , USA . All mice were maintained in a 12 hr light/12 hr dark cycle in a specific pathogen-free barrier facility . The mice had unlimited access to a standard rodent diet ( ST-1 , 3 . 4% fat; VELAZ ) and water . All males were killed at age 60–70 d . For immunocytochemistry , the spread nuclei were prepared as described ( Anderson et al . , 1999 ) with modifications . Briefly , a single-cell suspension of spermatogenic cells in 0 . 1M sucrose with protease inhibitors ( Roche ) was dropped on 1% paraformaldehyde-treated slides and allowed to settle for 3 hr in a humidified box at 4°C . After brief H2O and PBS washing and blocking with 5% goat sera in PBS ( vol/vol ) , the cells were immunolabeled using a standard protocol with the following antibodies: anti-HORMAD2 ( 1:700 , rabbit polyclonal antibody , a gift from Attila Toth ) and SYCP3 ( 1:50 , mouse monoclonal antibody , Santa Cruz , #74569 ) . Secondary antibodies were used at 1:500 dilutions and incubated at room temperature for 60 min: goat anti-Rabbit IgG-AlexaFluor568 ( MolecularProbes , A-11036 ) and goat anti-Mouse IgG-AlexaFluor647 ( MolecularProbes , A-21235 ) . The images were acquired and examined using a Nikon Eclipse 400 microscope with a motorized stage control using a Plan Fluor objective , 60x ( MRH00601; Nikon ) and captured using a DS-QiMc monochrome CCD camera ( Nikon ) and the NIS-Elements program ( Nikon ) . The images were processed using the Adobe Photoshop CS software ( Adobe Systems ) . XMP XCyting Mouse Chromosome N Whole Painting Probes ( Metasystems ) were used for the DNA FISH analysis of asynapsis of all autosomes , one at a time , as described ( Turner et al . , 2005 ) , with slight modifications . Testes from 8-week-old mice were dissected and spread meiocyte nuclei were prepared as described previously ( Mahadevaiah et al . , 2009 ) with a modification , which relies on a reversed sequence of RNA FISH and immunofluorescence staining . Briefly , after cell fixation and permeabilization , the immunofluorescent labeling was performed for 90 min at 20°C with primary anti-HORMAD2 and anti-SYCP3 antibodies . Secondary antibodies were selected as above and incubated at room temperature for 60 min . After washing and postfixation steps , the immunostained nuclei were processed with RNA fluorescence in situ hybridization . The Cot-1 DNA biotin-labeled probe was incubated overnight at 37°C , and then the hybridized biotinylated Cot-1 probe was labelled with a FITC–avidin conjugate and the fluorescent signal was amplified as described previously ( Chaumeil et al . , 2008 ) . The images of the immunofluorescence stained and Cot-1 RNA FISH-labeled spread spermatocytes were examined and photographed using confocal microscope DMI6000CEL – Leica TCS SP8 . To model the dependence between the asynapsis rate and the number of symmetric DSBs , we determined the probabilistic distribution of the number of symmetric DSBs . The distribution was determined by simulation and with parameters based on previous studies . ( i ) The number of DSBs per cell ( Bhattacharyya et al . , 2013 ) was modeled as an observation from the normal distribution N ( 250 , sd = 20 ) . ( ii ) We assumed a number of DSBs proportional to SPO11 oligos ( Lange et al . , 2016 ) in each autosome . ( iii ) The positions of DSBs in the particular autosome were simulated from the uniform distribution , U ( 0 , Autosome_length ) . ( iv ) For the intersubspecific part of the autosomal pair , the number of symmetric DSBs was simulated from the binomial distribution Bi ( n = N_DSBs_in_het_part , p=0 . 28 ) . For the consubspecific part of the autosome , all DSBs were taken to be symmetric . The total number of symmetric DSBs in the autosome was taken as the sum of symmetric DSBs in the respective parts . Steps ( i ) to ( iv ) were performed in N = 100000 simulations to obtain a probabilistic distribution ( Source Code 1 ) . The effects of the number of Spo11 oligos and the chromosomal length on the asynapsis rate were investigated using a GLMM model . In all of the GLMM models used in this work , the asynapsis was modeled as a binary response to the fixed effects under investigation and a random intercept for each animal . In F1 hybrids , 95% confidence intervals for the the observed rates of asynapsed pachynemas and for the expected rated of asynapsed pachynemas were calculated by bootstrap . The estimates of mean asynapsis rate in respective chromosomes , their standard errors and their 95% confidence intervals were based on the GLMM model . In 2-chr crosses and the 4-chr cross , 95% confidence intervals of asynapsis rate were constructed on the basis of the likelihood ratio to also capture the uncertainty in the cases when the zero asynapsis rate per mouse and per chromosome was observed . On the basis of the nature of the dependence between the asynapsis rate and the length of the consubspecific PWD/PWD region on chromosomes of 2-chr crosses and the 4-chr cross , we fitted the data to segmented two-part continuous regression models ( Muggeo , 2003 ) . We fitted the models for all the chromosomes separately ( see Figure 3—source data 7 ) , being aware of the limitations caused by the lack of animals having specific lengths of the consubspecific PWD/PWD region in the respective chromosomes . As the best model describing the dependence of asynapsis rate on the lengths of PWD/PWD intervals , we selected piecewise linear models fitting 1 ) pooled data from 2-chr crosses and 2 ) pooled data from both 2-chr crosses and the 4-chr cross . These models are not severely affected by the lack of observations with specific lengths of the PWD/PWD segment neither by outliers . All calculations were performed in R 3 . 2 . 2 ( RRID:SCR_001905 ) ; the change point models and the GLMM models were fitted using the packages segmented and lme4 , respectively ( Muggeo and Ferrara , 2008; Bates et al . , 2015 ) . | It has been known for centuries that hybrids between closely related species are often infertile . This hybrid sterility was an enigma for Charles Darwin , who understood that it influenced how new species formed but could not fit it with his theory of evolution by natural selection . Sex cells – in mammals , the egg or sperm cells – form by a process called meiosis . During meiosis , chromosomes formed of DNA inherited from the mother pair up with the equivalent chromosomes inherited from the father and exchange sections of their DNA . This process is called synapsis and homologous recombination . A gene called Prdm9 determines where the DNA will break and recombine . Prdm9 plays a major role in determining whether the male hybrid offspring of two laboratory strains of mice ( which come from different subspecies ) are sterile . In sterile hybrids , the two versions of Prdm9 interact in ways that disturb the DNA repair process . However , these interactions are not enough on their own to cause hybrid males to be sterile . The currently prevailing view is that interactions between a large number of other – currently unidentified – genes also contribute to sterility . But could there be other processes involved that do not involve gene interactions ? To investigate , Gregorova , Gergelits et al . utilized strains of hybrid mice where a pair of chromosomes of one subspecies was substituted by the corresponding pair from the other subspecies . This generated hybrids with stretches of DNA that came entirely from a single subspecies . Having such a stretch that lasted for 27 million or more DNA base pairs fully restored synapsis in a given pair of chromosomes during meiosis . Hybrid sterility was reversed when synapsis was restored in the four chromosomes that were most strongly affected by synapsis not occurring . The results presented by Gregorova , Gergelits et al . provide a direct link between Prdm9-controlled chromosome synapsis and the interruption of meiosis . Hybrid sterility occurs in all sexually reproducing organisms , as does chromosome pairing during meiosis . Thus Prdm9 could control a particular case of a more universal mechanism that enables new species to form . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"genetics",
"and",
"genomics"
] | 2018 | Modulation of Prdm9-controlled meiotic chromosome asynapsis overrides hybrid sterility in mice |
Transposons are mobile genetic elements that are found in nearly all organisms , including humans . Mobilization of DNA transposons by transposase enzymes can cause genomic rearrangements , but our knowledge of human genes derived from transposases is limited . In this study , we find that the protein encoded by human PGBD5 , the most evolutionarily conserved transposable element-derived gene in vertebrates , can induce stereotypical cut-and-paste DNA transposition in human cells . Genomic integration activity of PGBD5 requires distinct aspartic acid residues in its transposase domain , and specific DNA sequences containing inverted terminal repeats with similarity to piggyBac transposons . DNA transposition catalyzed by PGBD5 in human cells occurs genome-wide , with precise transposon excision and preference for insertion at TTAA sites . The apparent conservation of DNA transposition activity by PGBD5 suggests that genomic remodeling contributes to its biological function .
Transposons are genetic elements that are found in nearly all living organisms ( Feschotte and Pritham , 2007 ) . They can contribute to the developmental and adaptive regulation of gene expression and are a major source of genetic variation that drives genome evolution ( Cordaux and Batzer , 2009 ) . In humans and other mammals , they comprise about half of the nuclear genome ( Smit , 1999 ) . The majority of primate-specific sequences that regulate gene expression are derived from transposons ( Jacques et al . , 2013 ) , and transposons are a major source of structural genetic variation in human populations ( Stewart et al . , 2011 ) . While the majority of genes that encode transposase enzymes tend to become catalytically inactive and their transposon substrates tend to become immobile in the course of organismal evolution , some can maintain their transposition activities ( Liu et al . , 2007; Munoz-Lopez and Garcia-Perez , 2010 ) . In humans , at least one hundred L1 long interspersed repeated sequences ( LINEs ) actively transpose in human genomes and induce structural variation ( Kazazian , 2004 ) , including somatic rearrangements in neurons that may contribute to neuronal plasticity ( Erwin et al . , 2014 ) . The human Transib-like transposase RAG1 catalyzes somatic recombination of the V ( D ) J receptor genes in lymphocytes and is essential for adaptive immunity ( Hiom et al . , 1998 ) . The Mariner-derived transposase SETMAR functions in single-stranded DNA resection during DNA repair and replication in human cells and can catalyze DNA transposition in vitro ( Liu et al . , 2007; Shaheen et al . , 2010 ) . Among transposase enzymes that can catalyze excision and insertion of transposon sequences , DNA transposases are distinct in their dependence only on the availability of competent genomic substrates and cellular repair enzymes that ligate and repair excision sites , as compared to retrotransposons , which require transcription of the mobilized sequences ( Berg and Howe , 1989 ) . Most DNA transposases utilize an RNase H-like domain with three aspartate or glutamate residues ( so-called DDD or DDE motif ) that catalyze magnesium-dependent hydrolysis of phosphodiester bonds and strand exchange ( Keith et al . , 2008; Mitra et al . , 2008; Dyda et al . , 2012 ) . The IS4 transposase family , which includes piggyBac transposases , is additionally distinguished by precise excisions without modifications of the transposon flanking sequences ( De Palmenaer et al . , 2008 ) . The piggyBac transposase and its transposon were originally identified as an insertion in lepidopteran Trichoplusia ni cells ( Fraser et al . , 1985 ) . The piggyBac transposon consists of 13-bp inverted terminal repeats ( ITRs ) and 19-bp subterminal inverted repeats located 3 and 31 base pairs from the 5′ and 3′ ITRs , respectively ( Elick et al . , 1997 ) . PiggyBac transposase can mobilize a variety of ITR-flanked sequences and has a preference for integration at TTAA target sites in the host genome ( Fraser et al . , 1983; Beames and Summers , 1990; Wang and Fraser , 1993; Fraser et al . , 1995; Elick et al . , 1997; Handler et al . , 1998; Mitra et al . , 2008 ) . Members of the piggyBac superfamily of transposons have colonized a wide range of organisms ( Sarkar et al . , 2003 ) , including a recent and likely ongoing invasion of the bat Myotis lucifugus ( Mitra et al . , 2013 ) . The human genome contains five paralogous genes derived from piggyBac transposases , PGBD1-5 ( Smit and Riggs , 1996; Sarkar et al . , 2003 ) . PGBD1 and PGBD2 invaded the common mammalian ancestor , and PGBD3 and PGBD4 are restricted to primates , but are all contained as single coding exons , fused in frame with endogenous host genes , such as the Cockayne syndrome B gene ( CSB ) -PGBD3 fusion ( Sarkar et al . , 2003; Newman et al . , 2008 ) . Thus far , only the function of PGBD3 has been investigated . CSB-PGBD3 is capable of binding DNA , including endogenous piggyBac-like transposons in the human genome , but has no known catalytic activity , though biochemical and genetic evidence indicates that it may participate in DNA damage response ( Bailey et al . , 2012; Gray et al . , 2012 ) . PGBD5 is distinct from other human piggyBac-derived genes by having been domesticated much earlier in vertebrate evolution approximately 500 million years ( My ) ago , in the common ancestor of cephalochordates and vertebrates ( Sarkar et al . , 2003; Pavelitz et al . , 2013 ) . PGBD5 is transcribed as a multi-intronic but non-chimeric transcript predicted to encode a full-length transposase ( Pavelitz et al . , 2013 ) . Furthermore , PGBD5 expression in both human and mouse appears largely restricted to the early embryo and certain areas of the embryonic and adult brain ( Sarkar et al . , 2003; Pavelitz et al . , 2013 ) . These intriguing features prompted us to investigate whether human PGBD5 has retained the enzymatic capability of mobilizing DNA .
Human PGBD5 contains a C-terminal RNase H-like domain that has approximately 20% sequence identity and 45% similarity to the active lepidopteran piggyBac , ciliate piggyMac , and mammalian piggyBat transposases ( Figure 1 ) ( Sarkar et al . , 2003; Baudry et al . , 2009; Mitra et al . , 2013 ) . In addition , the human genome contains 2358 sequence elements with similarity to the piggyBac transposable elements ( Table 1 and Figure 2A ) . Specifically , MER75 ( MER75 , MER75A , MER75B ) and MER85 elements show considerably similar ITR sequences as compared to lepidopteran piggyBac transposons ( Table 2 and Figure 2B ) . A total of 328 piggyBac-like elements in the human genome have intact ITRs and exhibit duplications of their presumed TTAA target sites ( Table 1 and Figure 2C ) . We reasoned that even though the ancestral transposon substrates of PGBD5 cannot be predicted due to its very ancient evolutionary origin ( ∼500 My ) , preservation of its transposase activity should confer residual ability to mobilize distantly related piggyBac-like transposons . To test this hypothesis , we used a synthetic transposon reporter PB-EF1-NEO comprised of a neomycin resistance gene flanked by T . ni piggyBac ITRs ( Figure 3B ) ( Cary et al . , 1989; Fraser et al . , 1995 ) . We transiently transfected human embryonic kidney ( HEK ) 293 cells , which lack endogenous PGBD5 expression with the PB-EF1-NEO transposon reporter plasmid in the presence of a plasmid expressing PGBD5 , and assessed genomic integration of the reporter using clonogenic assays in the presence of G418 to select cells with genomic integration conferring neomycin resistance ( Figure 3C , Figure 3—figure supplement 1 ) . Given the absence of suitable antibodies to monitor PGBD5 expression , we expressed PGBD5 as an N-terminal fusion with the green fluorescent protein ( GFP ) . We observed significant rates of neomycin resistance of cells conferred by the transposon reporter with GFP-PGBD5 , but not in cells expressing control GFP or mutant GFP-PGBD5 lacking the transposase domain ( Figure 3C ) , despite equal expression of all transgenes ( Figure 3—figure supplement 2 ) . The efficiency of neomycin resistance conferred by the transposon reporter with GFP-PGBD5 was approximately 4 . 5-fold less than that of the T . ni piggyBac-derived transposase ( Figure 3D ) , consistent with their evolutionary divergence . These results suggest that human PGBD5 can promote genomic integration of a piggyBac-like transposon . 10 . 7554/eLife . 10565 . 003Figure 1 . Human PGBD5 is a distinct piggyBac-like transposase . Sequence alignment of piggyBac-like transposases frog Uribo 2 , bat piggyBat , lepidopteran piggyBac , and human PGBD5 . Catalytic residues conserved among piggyBac transposases are highlighted in red . Human PGBD5 D168 , D194 , and D386 residues , identified in our study ( Figure 7 ) , are marked in yellow . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 00310 . 7554/eLife . 10565 . 004Table 1 . Summary of annotated human piggyBac-like elementsDOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 004Total piggyBac-like elementsIntact elements*MER75475144MER75A9362MER75B11427MER8590595UCON292400Looper5310*Denotes elements with intact ITR sequences that align with the consensus without gaps and contain a TTAA target site duplication . ITR , inverted terminal repeat . 10 . 7554/eLife . 10565 . 005Figure 2 . Human piggyBac-like transposable elements have intact inverted terminal repeat sequences similar to the T . ni piggyBac transposon . ( A ) Chromosome ideogram representing the distribution of annotated piggyBac-like elements in the human genome ( version hg19 ) . ( B ) Multiple sequence alignment of the piggybac inverted terminal repeat ( ITR ) sequence with the consensus ITR sequences of the MER75 and MER85 families of piggyBac-like elements . ( C ) Chromosome ideogram representing the distribution of piggyBac-like elements annotated in the human genome ( version hg19 ) with TTAA target site duplication as well as ITR sequences aligning with the consensus ( intact ITRs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 00510 . 7554/eLife . 10565 . 006Table 2 . Sequence identity matrix of the piggyBac inverted terminal repeat sequences and consensus sequences of the MER75 and MER85 human piggyBac-like elementsDOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 006piggyBacMER75MER85piggyBac100%––MER7553%100%–MER8556%63%100%10 . 7554/eLife . 10565 . 007Figure 3 . PGBD5 induces genomic integration of synthetic piggyBac transposons in human cells . ( A ) Schematic of the human PGBD5 protein with its C-terminal transposase homology domain , as indicated . ( B ) Schematic of synthetic transposon substrates used for DNA transposition assays , including transposons with mutant ITR marked by triangles in red , and transposons lacking ITRs marked in blue . ( C ) Representative photographs of crystal violet-stained colonies obtained after G418 selection of HEK293 cells co-transfected with the transposon reporter plasmid along with transposase cDNA expression vectors . ( D ) Quantification of G418-selection clonogenic assays , demonstrating the integration activities of GFP-PGBD5 , PGBD5 N-terminus , T . ni . piggyBac , and green fluorescent protein ( GFP ) control ( GFP-PGBD5 vs GFP; p = 0 . 00031 ) . ( E ) Quantification of genomic transposon integration using quantitative PCR of GFP-PGBD5 and GFP expressing cells using intact ( black ) , mutant ( red ) , and deleted ( blue ) ITR-containing transposon reporters ( intact vs mutant ITR; p = 0 . 00011 ) . Error bars represent standard errors of the mean of 3 biologic replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 00710 . 7554/eLife . 10565 . 008Figure 3—figure supplement 1 . Assay for genomic integration of transposon reporters . Schematic showing the procedure to assay for genomic integration of transposon reporters using G418 selection to clone genomically integrated neomycin resistant cells . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 00810 . 7554/eLife . 10565 . 009Figure 3—figure supplement 2 . GFP-PGBD5 , PGBD5 N-terminus , and T . ni . piggyBac are equally expressed upon transfection in HEK293 cells . Quantitative RT-PCR specific to GFP-PGBD5 , PGBD5 N-terminus , and T . ni . piggyBac shows equal mRNA expression of all three transposases ( PGBD5 N-terminus p = 0 . 17 , T . ni . piggyBac p = 0 . 092 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 00910 . 7554/eLife . 10565 . 010Figure 3—figure supplement 3 . Sanger sequencing traces of the ITR of the synthetic transposon reporter plasmids . Top to bottom: intact transposon , deleted transposon , and mutant transposon . Black line indicates location of GGG to ATA mutations in mutant transposon ITR . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 01010 . 7554/eLife . 10565 . 011Figure 3—figure supplement 4 . Quantitative assay of genomic integration of transposon reporters . Cells were transfected as described in ‘Materials and methods’ . Next , cells were expanded and genomic DNA was isolated . Quantitative real-time PCR was performed with primers specific to the transposon sequence as well as to the TK1 reference gene . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 01110 . 7554/eLife . 10565 . 012Figure 3—figure supplement 5 . Quantitative genomic PCR standard curve for transposon specific primers . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 012 If neomycin resistance conferred by the PGBD5 and the transposon reporter is due to genomic integration and DNA transposition , then this should require specific activity on the transposon ITRs . To test this hypothesis , we generated transposon reporters with mutant ITRs and assayed them for genomic integration ( Figure 3B , Figure 3—figure supplement 3 ) . DNA transposition by the piggyBac family transposases involves hairpin intermediates with a conserved 5′-GGGTTAACCC-3′ sequence that is required for target site phosphodiester hydrolysis ( Mitra et al . , 2008 ) . Thus , we generated reporter plasmids lacking ITRs entirely or containing complete ITRs with 5′-ATATTAACCC-3′ mutations predicted to disrupt the formation of productive hairpin intermediates ( Mitra et al . , 2008 ) . To enable precise quantitation of mobilization activity , we developed a quantitative genomic PCR assay using primers specific for the transposon reporter and the endogenous human TK1 gene for normalization ( Figure 3—figure supplements 4 , 5 ) . In agreement with the results of the clonogenic neomycin resistance assays , we observed efficient genomic integration of the donor transposons in cells transfected by GFP-PGBD5 as compared to the minimal signal observed in cells expressing GFP control ( Figure 3E ) . Deletion of transposon ITRs from the reporter reduced genomic integration to background levels ( Figure 3E ) . Consistent with the specific function of piggyBac family ITRs in genomic transposition , mutation of the terminal GGG sequence in the ITR significantly reduced the integration efficiency ( Figure 3E ) . These results indicate that specific transposon ITR sequences are required for PGBD5-mediated DNA transposition . DNA transposition by piggyBac superfamily transposases is distinguished from most other DNA transposon superfamilies by the precise excision of the transposon from the donor site and preference for insertion in TTAA sites ( Cary et al . , 1989; Fraser et al . , 1995 ) . To determine the structure of the donor sites of transposon reporters mobilized by PGBD5 , we isolated plasmid DNA from cells 2 days after transfection , amplified the transposon reporter using PCR , and determined its sequence using capillary Sanger sequencing ( Figure 4—figure supplement 1 ) . Similar to the hyperactive T . ni piggyBac , cells expressing GFP-PGBD5 , but not those expressing GFP control vector , exhibited robust excision of ITR-flanked transposon with apparently precise repair of the donor plasmid ( Figure 4A , B and Figure 4—figure supplement 1 ) . These results suggest that PGBD5 is an active cut-and-paste DNA transposase . 10 . 7554/eLife . 10565 . 013Figure 4 . PGBD5 precisely excises piggyBac transposons . ( A ) Representative agarose electrophoresis analysis of PCR-amplified PB-EF1-NEO transposon reporter plasmid from transposase-expressing cells , demonstrating efficient excision of the ITR-containing transposon by PGBD5 , but not GFP or PGBD5 N-terminus mutant lacking the transposase domain . T . ni piggyBac serves as positive control . ( B ) Representative Sanger sequencing fluorogram of the excised transposon , demonstrating precise excision of the ITR and associated duplicated TTAA sequence , marked in red , demonstrating integrations of transposons ( green ) into human genome ( blue ) with TTAA insertion sites and genomic coordinates , as marked . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 01310 . 7554/eLife . 10565 . 014Figure 4—figure supplement 1 . Schematic of transposon excision assay . Cells were transfected as described in ‘Materials and methods’ . Next , DNA was isolated . PCR was performed with primers specific to sequences flanking the transposon . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 014 To validate chromosomal integration and determine the location and precise structure of the insertion of the reporter transposons in the human genome , we isolated genomic DNA from G418-resistant HEK293 cells following transfection with PGBD5 and PB-EF1-NEO and amplified the genomic sites of transposon insertions using flanking sequence exponential anchored ( FLEA ) PCR , a technique originally developed for high-efficiency analysis of retroviral integrations ( Pule et al . , 2008 ) . We adapted FLEA-PCR for the analysis of genomic DNA transposition by using unique reporter sequence to prime polymerase extension upstream of the transposon ITR into the flanking human genome , followed by reverse linear extension using degenerate primers , and exponential amplification using specific nested primers to generate chimeric amplicons suitable for massively parallel single-molecule Illumina DNA sequencing ( Figure 5 ) ( Henssen et al . , 2015a ) . This method enabled us to isolate specific portions of the human genome flanking transposon insertions , as evidenced by the reduced yield of amplicons isolated from control cells lacking transposase vectors or expressing GFP ( Figure 6—figure supplement 1 ) . To identify the sequences of the transposon genomic insertions at single-base pair resolution , we aligned reads obtained from FLEA-PCR Illumina sequencing to the human hg19 reference genome and synthetic transposon reporter , and identified split reads that specifically span both ( Figure 5 ) . These data have been deposited to the Sequence Read Archive ( http://www . ncbi . nlm . nih . gov/sra/ , accession number SRP061649 , Henssen et al . , 2015c ) , with the processed and annotated data available from the Dryad Digital Repository ( Henssen et al . , 2015b ) . 10 . 7554/eLife . 10565 . 015Figure 5 . Schematic of transposon-specific flanking sequence exponential anchored–polymerase chain reaction amplification ( FLEA-PCR ) and massively parallel single molecule sequencing assay for mapping and sequencing transposon insertions . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 015 To infer the mechanism of genomic integration of transposon reporters , we analyzed the sequences of the insertion loci to determine integration preferences at base-pair resolution and identify potential sequence preferences . We found that transposon amplicons isolated from cells expressing GFP-PGBD5 , but not those isolated from GFP control cells , were significantly enriched for TTAA sequences , as determined by sequence entropy analysis ( Crooks et al . , 2004 ) ( Figure 6A ) . To discriminate between potential DNA transposition at TTAA target sites and alternative mechanisms of chromosomal integration , we classified genomic insertions based on target sites containing TTAA and those containing other sequence motifs ( Table 3 ) . Consistent with the DNA transposition activity of PGBD5 , we observed significant induction of TTAA-containing insertions in cells expressing GFP-PGBD5 and transposons with intact ITRs , as compared to control cells expressing GFP , or to cells transfected with GFP-PGBD5 and mutant ITR transposons ( Table 3 ) . Sequence analysis of split reads containing transposon-human junction at TTAA sites revealed that , in almost every case examined ( n = 65 out of 66 ) , joining between TTAA host and transposon DNA occurred precisely at the GGG/CCC terminal motif of the donor transposon ITR ( Figure 6C ) , in agreement with its requirement for efficient DNA transposition ( Figure 3E ) . Consistent with the genome-wide transposition induced by PGBD5 , we identified transposition events in all human chromosomes , including both genic and intergenic loci ( Figure 6B ) . Thus , PGBD5 can mediate canonical cut-and-paste DNA transposition of piggyBac transposons in human cells . 10 . 7554/eLife . 10565 . 016Figure 6 . PGBD5 induces DNA transposition in human cells . ( A ) Analysis of the transposon integration sequences , demonstrating TTAA preferences in integrations in cells expressing GFP-PGBD5 , but not GFP control . X-axis denotes nucleotide sequence logo position , and y-axis denotes information content in bits . ( B ) Circos plot of the genomic locations PGBD5-mobilized transposons plotted as a function of chromosome number and transposition into genes ( red ) and intergenic regions ( gray ) . ( C ) Alignment of representative DNA sequences of identified genomic integration sites . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 01610 . 7554/eLife . 10565 . 017Figure 6—figure supplement 1 . Representative agarose gel image of amplicons from flanking sequence exponential anchored–polymerase chain reaction amplification ( FLEA-PCR ) . Arrow indicated expected size of degenerate anchor primer amplicons . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 01710 . 7554/eLife . 10565 . 018Table 3 . Analysis of transposon integration sequences in human genomes induced by PGBD5DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 018Intact transposonMutant transposonTTAA ITRNon-ITRTTAA ITRNon-ITRTransposase GFP-PGBD582% ( 65 ) †18% ( 14 ) 11% ( 4 ) ‡89% ( 33 ) GFP Control17% ( 2 ) 83% ( 10 ) 40% ( 27 ) 60% ( 40 ) Cells expressing GFP-PGBD5 and intact transposons exhibit significantly higher frequency of genomic integration as compared to either GFP control , or GFP-PGBD5 with mutant transposons , with 82% ( 65 out of 79 ) of sequences demonstrating DNA transposition of ITR transposons into TTAA sites ( †p = 1 . 8 × 10-5 ) . Mutation of the transposon ITR significantly reduces ITR-mediated integration , with only 11% ( 4 out of 37 ) of sequences ( ‡p = 0 . 0016 ) . Numbers in parentheses denote absolute numbers of identified insertion sites . GFP , green fluorescent protein; ITR , inverted terminal repeat . Requirement for transposon substrates with specific ITRs , their precise excision , and preferential insertion into TTAA-containing genomic locations is all consistent with the preservation of PGBD5's DNA transposase activity in human cells . Like other cut-and-paste transposases , piggyBac superfamily transposases are thought to utilize a triad of aspartate or glutamate residues to catalyze phosphodiester bond hydrolysis , but the catalytic triad of aspartates previously proposed for T . ni piggyBac is apparently not conserved in the primary sequence of PGBD5 ( Figure 1 ) ( Sarkar et al . , 2003; Keith et al . , 2008; Mitra et al . , 2008; Nesmelova and Hackett , 2010 ) . Thus , we hypothesized that distinct aspartic or glutamic acid residues may be required for DNA transposition mediated by PGBD5 . To test this hypothesis , we used alanine-scanning mutagenesis and assessed transposition activity of GFP-PGBD5 mutants using quantitative genomic PCR ( Figure 7 and Figure 7—figure supplements 1–3 ) . This analysis indicated that simultaneous alanine mutations of D168 , D194 , and D386 reduced apparent transposition activity to background levels , similar to that of GFP control ( Figure 7A ) . We confirmed that the mutant GFP-PGBD5 proteins have equivalent stability and expression as the wild-type protein in cells by immunoblotting against GFP ( Figure 7B ) . Phylogenetic analysis of vertebrate piggyBac homologs from Danio rerio , Python bivittatus , Xenopus tropicalis , Gallus gallus , Mus musculus , and Homo sapiens showed that PGBD5 proteins are divergent from other piggyBac-like proteins ( Figure 8 ) and include conservation of functionally important D168 , D194 , and D386 residues that distinguish them from lepidopteran piggyBac and human PGBD1-4 ( Figure 8—figure supplement 1 ) . These results suggest that PGBD5 represents a distinct member of the evolutionarily ancient piggyBac-like family of DNA transposases . 10 . 7554/eLife . 10565 . 019Figure 7 . Structure-function analysis of PGBD5-induced DNA transposition using alanine scanning mutagenesis . ( A ) Quantitative PCR analysis of genomic integration activity of alanine point mutants of GFP-PGBD5 , as compared to wild-type and GFP control-expressing cells . D168A , D194A , and D386A mutants ( red ) exhibit significant reduction in apparent activity ( Asterisks denote statistical significance: p = 0 . 00011 , p = 0 . 000021 , p = 0 . 000013 vs GFP-PGBD5 , respectively ) . Dotted line marks threshold at which less than 1 transposon copy was detected per haploid human genome . Error bars represent standard errors of the mean of 3 biological replicates . ( B ) Western immunoblot showing equal expression of GFP-PGBD5 mutants , as compared to wild-type GFP-PGBD5 ( green ) . β-actin ( red ) serves as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 01910 . 7554/eLife . 10565 . 020Figure 7—figure supplement 1 . Sanger sequencing trances of pRecLV103-GFP-PGBD5 D>A and E>A mutants ( D168A , D175A , E188A , D192A , D194A , E203A , E205A , E236A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 02010 . 7554/eLife . 10565 . 021Figure 7—figure supplement 2 . Sanger sequencing trances of pRecLV103-GFP-PGBD5 D>A and E>A mutants ( D241A , D244A , E284A , E285A , E287A , D303A , E365A , E373A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 02110 . 7554/eLife . 10565 . 022Figure 7—figure supplement 3 . Sanger sequencing trances of pRecLV103-GFP-PGBD5 D>A and E>A mutants ( D386A , D387A , D425A , E439A , E444A , E449A , D450A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 02210 . 7554/eLife . 10565 . 023Figure 8 . PGBD5 homologs are divergent from other piggyBac genes in vertebrates . Phylogenetic reconstruction of the evolutionary relationships among piggyBac transposase-derived genes in vertebrates , demonstrating the PGBD5 homologs represent a distinct subfamily of piggyBac-like derived genes . Scale bar represents Grishin distance . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 02310 . 7554/eLife . 10565 . 024Figure 8—figure supplement 1 . PGBD5 glutamic acid resitues D168 , D194 , and D386 are conserved across species . Clustal O alignment of PGBD5 proteins across species ( Danio rerio , Xenopus tropicalis , Homo sapiens , Mus musculus , Gallus gallus , Python bivittatus ) . D168 , D194 , and D386 residues are highlighted in yellow . DOI: http://dx . doi . org/10 . 7554/eLife . 10565 . 024
Our current findings indicate that human PGBD5 is an active piggyBac transposase that can catalyze DNA transposition in human cells . DNA transposition by PGBD5 requires its C-terminal transposase domain and depends on specific ITRs derived from the lepidopteran piggyBac transposons ( Figure 3 ) . DNA transposition involves transesterification reactions mediated by DNA hairpin intermediates ( Mitra et al . , 2008 ) . Consistent with the requirement of intact termini of the piggyBac , Tn10 , and Mu transposons ( Elick et al . , 1997 ) , elimination or mutation of the terminal GGG nucleotides from the transposon substrates also abolishes the transposition activity of PGBD5 ( Figure 3 ) . PGBD5-induced DNA transposition is precise with preference for insertions at TTAA genomic sites ( Figure 4 ) . Since our analysis was limited to ectopically expressed PGBD5 fused to GFP and episomal substrates derived from lepidopteran piggyBac transposons , it is possible that endogenous PGBD5 may exhibit different activities on chromatinized substrates in the human genome . Current structure-function analysis indicates that PGBD5 requires three aspartate residues to mediate DNA transposition ( Figure 7 ) , but its transposase domain appears to be distinct from other piggyBac transposase enzymes with respect to its primary sequence ( Figure 1 and Figure 8 ) ( Keith et al . , 2008 ) . Thus , the three aspartate residues required for efficient DNA transposition by PGBD5 may form a catalytic triad that functions in phosphodiester bond hydrolysis , similar to the DDD motif in other piggyBac family transposases , or alternatively may contribute to other steps in the transposition reaction , such as synaptic complex formation , hairpin opening , or strand exchange ( Elick et al . , 1997; Keith et al . , 2008; Mitra et al . , 2008 ) . In addition , we find that alanine mutations of the three required aspartate residues in the PGBD5 transposase domain significantly reduce but do not completely eliminate genomic integration of the transposon reporters ( Figure 7 ) . This could reflect residual catalytic activity despite these mutations , or that PGBD5 expression may affect other mechanisms of DNA integration in human cells . The evolutionary conservation of the transposition activity of PGBD5 suggests that it may have hitherto unknown biologic functions among vertebrate organisms . DNA transposition is a major source of genetic variation that drives genome evolution , with some DNA transposases becoming extinct and others domesticated to evolve exapted functions . The evolution of transposons' activities can be highly variable , with some organisms such as Zea mays undergoing continuous genome remodeling and recent twofold expansion through endogenous retrotransposition , Drosophila and Saccharomyces owing over half of their known spontaneous mutations to transposons , and primate species including humans exhibiting relative extinction of transposons ( Feschotte and Pritham , 2007 ) . Indeed , transposase-derived genes domesticated in humans have evolved to have endogenous functions other than genomic transposition per se . For example , human RAG1 is a domesticated Transib transposase that has retained its active transposase domain , and can transpose ITR-containing transposons in vitro , but catalyzes somatic recombination of immunoglobulin and T-cell receptor genes in lymphocytes across signal sequences that might be derived from related transposons ( Landree et al . , 1999; Fugmann et al . , 2000 ) . Human SETMAR is a Mariner-derived transposase with a divergent DDN transposase domain that has retained its endonuclease activity and functions in double-strand DNA repair by non-homologous end joining ( Liu et al . , 2007 ) . The human genome encodes over 40 other genes derived from DNA transposases ( Smit , 1999; Feschotte and Pritham , 2007 ) , including THAP9 that was recently found to mobilize transposons in human cells with as of yet unknown function ( Majumdar et al . , 2013 ) . RAG1 , THAP9 , and PGBD5 are , to our knowledge , the only human proteins with demonstrated transposase activity in human cells . The distinct biochemical and structural features of PGBD5 indicated by our findings are consistent with its unique evolution and function among human piggyBac-derived transposase genes ( Sarkar et al . , 2003; Pavelitz et al . , 2013 ) . PGBD5 exhibits deep evolutionary conservation predating the origin of vertebrates , including a preservation of genomic synteny across lancelet , lamprey , teleosts , and amniotes ( Pavelitz et al . , 2013 ) . This suggests that while PGBD5 likely derived from an autonomous mobile element , this ancestral copy was immobilized early in evolution and PGBD5 can probably no longer mobilize its own genomic locus , at least in germ line cells . The human genome contains several thousands of miniature inverted repeat transposable elements ( MITEs ) with similarity to piggyBac transposons ( Figure 2 and Table 1 ) ( Sarkar et al . , 2003; Feschotte and Pritham , 2007 ) . CSB-PGBD3 can bind to the piggyBac-derived MER85 elements in the human genome ( Bailey et al . , 2012; Gray et al . , 2012 ) . Similarly , it is possible that PGBD5 can act in trans to recognize and mobilize one or several related MITEs in the human genome . Recently , single-molecular maps of the human genome have predicted thousands of mobile element insertions , and the activity of PGBD5 or other endogenous transposases may explain some of these novel variants ( Chaisson et al . , 2015; Pendleton et al . , 2015 ) . Given that both human RAG1 and ciliate piggyMac domesticated transposases catalyze the elimination of specific genomic DNA sequences ( Hiom et al . , 1998; Baudry et al . , 2009 ) , it is reasonable to hypothesize that PGBD5's biological function may similarly involve the excision of as of yet unknown ITR-flanked sequences in the human genome or another form of DNA recombination . Since DNA transposition by piggyBac family transposases requires substrate chromatin accessibility and DNA repair , we anticipate that additional cellular factors are required for and regulate PGBD5 functions in cells . Likewise , just as RAG1-mediated DNA recombination of immunoglobulin loci is restricted to B lymphocytes , and rearrangements of T-cell receptor genes to T lymphocytes , potential DNA rearrangements mediated by PGBD5 may be restricted to specific cell types and developmental periods . PGBD5 localizes to the cell nucleus and is expressed during embryogenesis and neurogenesis , but its physiological function is not yet known ( Pavelitz et al . , 2013 ) . Generation of molecular diversity through DNA recombination during nervous system development has been a long-standing hypothesis ( Dreyer et al . , 1967; Wu and Maniatis , 1999 ) . The recent discovery of somatic retrotransposition in human neurons ( Coufal et al . , 2009; Evrony et al . , 2012; Upton et al . , 2015 ) , combined with our finding of DNA transposition activity by human PGBD5 , which is highly expressed in neurons , suggests that additional mechanisms of somatic genomic diversification may contribute to vertebrate nervous system development . Because DNA transposition is inherently topological and orientation of transposons can affect the arrangements of reaction products ( Claeys Bouuaert et al . , 2011 ) , potential activities of PGBD5 can depend on the arrangements of accessible genomic substrates , leading to both conservative DNA transposition involving excision and insertion of transposon elements , as well as irreversible reactions such as DNA elimination and chromosomal breakage-fusion-bridge cycles , as originally described by McClintock ( 1942 ) . Finally , given the potentially mutagenic activity of active DNA transposases , we anticipate that unlicensed activity of PGBD5 and other domesticated transposases can be pathogenic in specific disease states , particularly in cases of aberrant chromatin accessibility , such as cancer .
All reagents were obtained from Sigma–Aldrich ( St . Louis , MO , United States ) if not otherwise specified . Synthetic oligonucleotides were obtained from Eurofins MWG Operon ( Huntsville , AL , United States ) and purified by HPLC . HEK293 and HEK293T were obtained from the American Type Culture Collection ( ATCC , Manassas , Virginia , United States ) . The identity of all cell lines was verified by STR analysis and lack of Mycoplasma contamination was confirmed by Genetica DNA Laboratories ( Burlington , NC , United States ) . Cell lines were cultured in DMEM supplemented with 10% fetal bovine serum and 100 U/ml penicillin and 100 µg/ml streptomycin in a humidified atmosphere at 37°C and 5% CO2 . Human PGBD5 cDNA ( Refseq ID: NM_024554 . 3 ) was cloned as a GFP fusion into the lentiviral vector pReceiver-Lv103-E3156 ( GeneCopoeia , Rockville , MD , United States ) . piggyBac ITRs ( 5′-TTAACCCTAGAAAGATAATCATATTGTGACGTACGTTAAAGATAATCATGTGTAAAATTGACGCATG-3′ and 5′-CATGCGTCAATTTTACGCAGACTATCTTTCTAGGGTTAA-3′ ) , as originally cloned by Malcolm Fraser et al . ( Elick et al . , 1997; Handler et al . , 1998 ) , were cloned into PB-EF1-NEO to flank IRES-driven neomycin resistance gene , as obtained from System Biosciences ( Mountain View , CA , United States ) . Plasmid encoding the hyperactive T . ni piggyBac transposase , as originally generated by Nancy Craig et al . ( Li et al . , 2013 ) , was obtained from System Biosciences . Site-directed PCR mutagenesis was used to generate mutants of PGBD5 and piggyBac , according to manufacturer's instructions ( Agilent , Santa Clara , CA , United States ) . Plasmids were verified by restriction endonuclease mapping and Sanger sequencing and deposited in Addgene . Lentivirus packaging vectors psPAX2 and pMD2 . G were obtained from Addgene ( Cudre-Mauroux et al . , 2003 ) . HEK293 cells were seeded at a density of 100 , 000 cells per well in a 6-well plate and transfected with 2 μg of total plasmid DNA , containing 1 μg of transposon reporter ( PB-EF1-NEO or mutants ) and 1 μg of transposase cDNA ( pRecLV103-GFP-PGBD5 or mutants ) using Lipofectamine 2000 , according to manufacturer's instructions ( Life Technologies , CA , United States ) . After 24 hr , transfected cells were trypsinized and re-plated for functional assays . Upon transfection , cells were cultured for 48 hr and total RNA was isolated using the RNeasy Mini Kit , according to manufacturer's instructions ( Qiagen , Venlo , Netherlands ) . cDNA was synthesized using the SuperScript III First-Strand Synthesis System ( Invitrogen , Waltham , MA , United States ) . Quantitative real-time PCR was performed using the KAPA SYBR FAST PCR polymerase with 20 ng template and 200 nM primers , according to the manufacturer's instructions ( Kapa Biosystems , Wilmington , MA , United States ) . PCR primers are listed in Supplementary file 1 . Ct values were calculated using ROX normalization using the ViiA 7 software ( Applied Biosystems ) . Upon transfection , cells were seeded at a density of 1000 cells per 10-cm dish and selected with G418 sulfate ( 2 mg/ml ) for 2 weeks . Resultant colonies were fixed with methanol and stained with crystal violet . Upon transfection , cells were cultured for 48 hr and DNA was isolated using the PureLink Genomic DNA Mini Kit , according to manufacturer's instructions ( Life Technologies ) . Reporter plasmid sequences flanking the neomycin resistance cassette transposons were amplified using hot start PCR with an annealing temperature of 57°C and extension time of 2 min , according to the manufacturer's instructions ( New England Biolabs , Beverly , MA , United States ) using the Mastercycler Pro thermocycler ( Eppendorf , Hamburg , Germany ) . PCR primers are listed in Supplementary file 1 . The PCR products were resolved using agarose gel electrophoresis and visualized by ethidium bromide staining . Identified gel bands were extracted using the PureLink Quick Gel Extraction Kit ( Invitrogen ) and Sanger sequenced to identify excision products . Upon transfection , cells were selected with puromycin ( 5 μg/ml ) for 2 days to eliminate non-transfected cells . After selection , cells were expanded for 10 days without selection and genomic DNA isolated using PureLink Genomic DNA Mini Kit ( Life Technologies ) . Quantitative real-time PCR was performed using the KAPA SYBR FAST PCR polymerase with 20 ng template and 200 nM primers , according to the manufacturer's instructions ( Kapa Biosystems ) . PCR primers are listed in Supplementary file 1 . Ct values were calculated using ROX normalization using the ViiA 7 software ( Applied Biosystems ) . We determined the quantitative accuracy of this assay using analysis of serial dilution PB-E1-NEO plasmid as reference ( Figure 3—figure supplement 5 ) . To amplify genomic transposon integration sites , we modified FLEA-PCR ( Pule et al . , 2008 ) , as described in Figure 5 ( Henssen et al . , 2015a ) . First , linear extension PCR was performed using 2 μg of genomic DNA and 100 nM biotinylated linear primer using the Platinum HiFidelity PCR mix , according to manufacturer's instructions ( Invitrogen Corp . ) . Linear extension parameters for PCR were: 95°C ( 45 s ) , 62°C ( 45 s ) , 72°C ( 3 min ) for 30 cycles . Reaction products were purified by diluting the samples in a total volume of 200 μl of nuclease-free water and centrifugation using the Amicon Ultra 0 . 5 ml 100 K at 12 , 000×g for 10 min at room temperature ( EMD Millipore , Billerica , MA , United States ) purification . Retentate was bound to streptavidin ferromagnetic beads on a shaker at room temperature overnight ( Dynal , Oslo , Norway ) . Beads were washed with 40 μl of washing buffer ( Kilobase binder kit; Dynal ) , then water , then 0 . 1 N NaOH , and finally with water again . To anneal the anchor primer , washed beads were resuspended in a total volume of 20 μl containing 5 μM anchor primer , 500 nM dNTP , and T7 DNA polymerase buffer ( New England Biolabs ) . Samples were placed in a heating block pre-heated to 85°C and allowed to passively cool to 37°C . Once annealed , 10 units of T7 DNA polymerase ( New England Biolabs ) was added and the mixtures were incubated for 1 hr at 37°C . Next , the beads were washed five times in water . To exponentially amplify the purified products , beads were resuspended in a total volume of 50 μl containing 500 nM of exponential and Transposon1 primers and the Platinum HiFidelity PCR mix . PCR was performed with the following parameters: 95°C for 5 min , followed by 35 cycles of 95°C for 45 s , 62°C for 45 s , and 72°C for 3 min . PCR products were purified using the Invitrogen PCR purification kit ( Invitrogen Corp . ) . Second nested PCR was performed using 1/50th of the first exponential PCR product as template using the Platinum HiFidelity PCR with 500 nM of exponential and Transposon2 primers . PCR was performed with the following parameters: 35 cycles of 95°C for 45 s , 62°C for 45 s , and 72°C for 3 min . Final PCR products were purified using the Invitrogen PCR purification kit , according to the manufacturer's instructions ( Invitrogen Corp . ) . Equimolar amounts of purified FLEA-PCR amplicons were pooled , as measured using fluorometry with the Qubit instrument ( Invitrogen ) and sized on a 2100 BioAanalyzer instrument ( Agilent Technologies ) . The sequencing library construction was performed using the KAPA Hyper Prep Kit ( Kapa Biosystems ) and 12 indexed Illumina adaptors from IDT ( Coralville , IO , United States ) , according to the manufacturer's instructions . After quantification and sizing , libraries were pooled for sequencing on a MiSeq ( pooled library input at 10 pM ) on a 300/300 paired end run ( Illumina , San Diego , CA , United States ) . An average of 575 , 565 paired reads were generated per sample . The duplication rate varied between 56 and 87% . Because of the use of FLEA-PCR amplicons for DNA sequencing , preparation of Illumina sequencing libraries is associated with the formation of adapter dimers ( ILLUMINA , 2015 ) . We used cutadapt to first trim reads to retain bases with quality score >20 , then identify reads containing adapter dimers and exclude them from further analyses ( parameters -q 20 -b P7 = <P7_index> -B P5 = <P5_index> -discard; where <P7_index> is the P7 primer adapter with the specific barcode for each library , and <P5_index> is the generic P5 adapter sequence: GATCGGAAGAGCGTCGTGTAGGGAAAGAGTGTAGATCTCGGTGGTCGCCGTATCATT ) ( Lindgreen , 2012 ) . Anchor primer sequences were then trimmed from the reads retained using cutadapt ( -g^GTGGCACGGACTGATCNNNNNN ) . Filtered and trimmed reads were mapped to a hybrid reference genome consisting of the hg19 full chromosome sequences and the PB-EF1-NEO plasmid sequence using bwa-mem using standard parameters ( Li and Durbin , 2010 ) . Mapped reads were then analyzed with LUMPY using split read signatures ( Layer et al . , 2014 ) , and insertion loci were identified using the called variants flagged as interchromosomal translocations ( BND ) between the plasmid sequence and the human genome . Breakpoints were resolved to base-pair accuracy using split read signatures when possible . Insertion loci were taken with 10 flanking base pairs and aligned with MUSCLE to establish consensus sequence ( Layer et al . , 2014 ) . Genomic distribution of insertion loci was plotted using ChromoViz ( https://github . com/elzbth/ChromoViz ) . All analysis scripts are available from Zenodo ( Henaff et al . , 2015a ) . Lentivirus production was carried out as described in Kentsis et al . ( 2012 ) . Briefly , HEK293T cells were transfected using TransIT with 2:1:1 ratio of the pRecLV103 lentiviral vector , and psPAX2 and pMD2 . G packaging plasmids , according to manufacturer's instructions ( TransIT-LT1 , Mirus , Madison , WI , United States ) . Virus supernatant was collected at 48 and 72 hr post-transfection , pooled , filtered , and stored at −80°C . HEK293T cells were transduced with virus particles at a multiplicity of infection of five in the presence of 8 μg/ml hexadimethrine bromide . Transduced cells were selected for 2 days with puromycin ( 5 μg/ml ) . To analyze protein expression by Western immunoblotting , 1 million transduced cells were suspended in 80 μl of lysis buffer ( 4% sodium dodecyl sulfate , 7% glycerol , 1 . 25% beta-mercaptoethanol , 0 . 2 mg/ml Bromophenol Blue , 30 mM Tris-HCl , pH 6 . 8 ) . Cells suspensions were lysed using Covaris S220 adaptive focused sonicator , according to the manufacturer's instructions ( Covaris , Woburn , CA , United States ) . Lysates were cleared by centrifugation at 16 , 000×g for 10 min at 4°C . Clarified lysates ( 30 μl ) were resolved using sodium dodecyl sulfate-polyacrylamide gel electrophoresis and electroeluted using the Immobilon FL PVDF membranes ( Millipore ) . Membranes were blocked using the Odyssey Blocking buffer ( Li-Cor , Lincoln , Nebraska ) and blotted using the mouse and rabbit antibodies against GFP ( 1:500 , clone 4B10 ) and β-actin ( 1:5000 , clone 13E5 ) , respectively , both obtained from Cell Signaling Technology ( Beverly , MA , United States ) . Blotted membranes were visualized using goat secondary antibodies conjugated to IRDye 800CW or IRDye 680RD and the Odyssey CLx fluorescence scanner , according to manufacturer's instructions ( Li-Cor ) . The transposon annotation of the human genome ( assembly hg19 ) was downloaded from the UCSC website ( http://hgdownload . soe . ucsc . edu/goldenPath/hg19/database/rmsk . txt . gz ) and converted to the GFF3 annotation format . The sequences of the elements in the piggyBac-like MER75 , MER75A , MER75B , MER85 , UCON29 , and LOOPER families were extracted with 50 flanking base pairs using fastaFromBed from the BedTools genome analysis suite ( http://bedtools . readthedocs . org ) . The set of sequences for each family was aligned using MUSCLE using standard parameters ( Edgar , 2004 ) . ITR sequences for each family were defined as terminal sequences conserved amongst all family members measured using multiple sequence alignments . We used a cutoff of 70% similarity to determine the fist position of the ITR in the alignment . The multiple sequence alignments were then manually curated to identify the set of ‘intact’ elements defined by containing both the TTAA target site duplication as well as both 3′ and 5′ ITRs aligning without gaps with the consensus ITR sequence . Multiple sequence alignment of the ITR sequences was also performed with MUSCLE , and the sequence identity matrix calculated using SIAS ( http://imed . med . ucm . es/Tools/sias . html ) , with the following measure of identity:Identity=100∗ ( Number of Identical ResiduesLength of shortest sequence ) . Chromosome ideograms were made using the NCBI's Genome Decoration Tool ( http://www . ncbi . nlm . nih . gov/genome/tools/gdp/ ) . Multiple sequence alignment of protein sequences was done using Clustal Omega ( Thompson et al . , 1994; Sievers and Higgins , 2014 ) . Pairwise BLAST alignments-based Grishin's sequence distance analysis was done using BLAST ( http://blast . ncbi . nlm . nih . gov/ ) and MEGA6 using standard parameters ( Tamura et al . , 2013 ) . Statistical significance values were determined using two-tailed non-parametric Mann–Whitney tests for continuous variables , and two-tailed Fisher exact test for discrete variables . | Transposons are mobile genetic elements that can be cut out of and inserted into DNA . They are present in most living things and make up almost half of the human genome . Transposons help to rearrange and increase the variety of DNA sequences , which can drive evolution and regulate the expression of genes . Enzymes called transposases help to move transposons , but very few genes that encode these enzymes have been studied in humans . PiggyBac transposase—which was first discovered in the cabbage looper moth—helps to move transposons of the piggyBac family . Humans and many other animals have genes that encode similar enzymes . In particular , the gene that encodes the human PGBD5 transposase is expressed in the developing embryo and particular areas of the brain and is highly similar to genes found in other vertebrate animals . These intriguing features prompted Henssen et al . to investigate PGBD5 . The experiments reveal that PGBD5 is able to move piggyBac-like transposons in human cells and insert them into sites that contain similar DNA sequences that are preferred by other PiggyBac transposases . Henssen et al . compared human PGBD5 to the piggyBac transposases from other organisms , including insects , bats , and frogs . They found that PGBD5 is deeply conserved among vertebrate organisms , and is distinct from other piggyBac transposases . These findings suggest that PGBD5 is indeed a fully working piggyBac transposase . Further work is needed to understand what portions of the human genome may be rearranged by PGBD5 , and how this may contribute to human brain development or disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2015 | Genomic DNA transposition induced by human PGBD5 |
HLA class I presentation of pathogen-derived peptide ligands is essential for CD8+ T-cell recognition of Toxoplasma gondii infected cells . Currently , little data exist pertaining to peptides that are presented after T . gondii infection . Herein we purify HLA-A*02:01 complexes from T . gondii infected cells and characterize the peptide ligands using LCMS . We identify 195 T . gondii encoded ligands originating from both secreted and cytoplasmic proteins . Surprisingly , T . gondii ligands are significantly longer than uninfected host ligands , and these longer pathogen-derived peptides maintain a canonical N-terminal binding core yet exhibit a C-terminal extension of 1–30 amino acids . Structural analysis demonstrates that binding of extended peptides opens the HLA class I F’ pocket , allowing the C-terminal extension to protrude through one end of the binding groove . In summary , we demonstrate that unrealized structural flexibility makes MHC class I receptive to parasite-derived ligands that exhibit unique C-terminal peptide extensions .
CD8 T-cells mediate immunity to Toxoplasma gondii infection ( Khan et al . , 1988; Suzuki and Remington , 1988 ) through recognition of peptide antigens presented by the MHC class I ( MHC I ) molecules of infected cells ( Brown and McLeod , 1990; Deckert-Schlüter et al . , 1994 ) . The majority of peptide ligands identified to date are derived from parasite surface proteins , proteins localized to dense granules , or the rhoptry proteins which are specialized secretory granules whose contents are released either into the host cell cytoplasm or the parasitophorous vacuole ( Blanchard et al . , 2008; Cardona et al . , 2015; Cong et al . , 2011 ) . These secreted proteins are thought to be optimal candidates for MHC I presentation because they have the best access to conventional antigen processing and presentation machinery in the host cell . However , this is a large pathogen , and the full array of parasite proteins that might be sampled and presented remains unknown . Recent advances in immunology and proteomics highlight that non-canonical ligands are presented to T cells by MHC I molecules . While a majority of peptides are 8–11 amino acids in length , MHC I molecules present a considerable number of peptides >11 amino acids ( Hassan et al . , 2015; Schittenhelm et al . , 2014 ) that elicit T-cell responses ( Hassan et al . , 2015; Burrows et al . , 2006 ) . Structural characterizations suggest that these long ligands interact with the MHC I molecule much like canonical peptides: The MHC I alpha chain forms a 10 x 25 angstrom groove in which peptide ligands are anchored by their second ( P2 ) and C-terminal ( PΩ ) residues . In this mode of binding , the middle portion of any oversized peptides can bulge out of the MHC I groove and interact with the receptors of T lymphocytes ( Tynan et al . , 2005 ) . Crystallographic studies have confirmed this bulging model , although there exists a structural example of a 10mer interacting with MHC I molecule HLA-A2 via P2 and P9 with an amino acid extension at P10 ( Collins et al . , 1994 ) . Thus , both peptide extension and peptide bulging have been observed for MHC I ligands , and , as longer ligands become increasingly evident , the interaction of these ligands with MHC I will need to be clarified . The goal of this study was to have the MHC I of infected cells inform the number , breadth , and nature of T . gondii peptide ligands . HLA-A*02:01 was purified from cells infected with T . gondii and peptide ligands eluted from the HLA class I ( human MHC I ) complex were analyzed by two-dimensional LCMS . The resulting data demonstrate that nearly 200 ligands originating from close to 100 different T . gondii proteins are sampled for MHC I presentation . As envisioned , a number of ligands originating from dense granule proteins was observed ( Blanchard et al . , 2008; Cong et al . , 2011 ) , yet MHC I ligands were also derived from a large number parasite cytoplasmic proteins . Surprisingly , T . gondii ligands were significantly longer than existing structural models can accommodate , and a series of peptide analogs demonstrated that these longer peptides are not anchored to MHC I via their C-termini . Crystallographic studies reveal an unreported structural re-arrangement of residues in the MHC I binding groove that accommodate C-terminal peptide extensions , and this structural flexibility is discussed in the context of infection by intracellular pathogens .
The first objective of this study was to identify pathogen-encoded ligands made available by MHC I . To accomplish this objective , HLA-A*02:01 was purified from T . gondii infected THP-1 monocytes as described . ( McMurtrey et al . , 2008; Wahl et al . , 2009 ) . To ensure THP-1 cells were infected , the number of infected cells and free parasites were periodically assessed . Over the course of a 1-week infection , the number of infected cells increased from 12 . 1% day 1 post infection to 71 . 5% on day 7 post infection ( Figure 1 ) . A steady increase in the number of free parasites in the culture media from day 1 to 7 was indicative of a productive infection . The production of 12 mg HLA-A*02:01 from infected cells was sufficient for a comprehensive analysis of T . gondii peptide ligands . 10 . 7554/eLife . 12556 . 003Figure 1 . Kinetics of the T . gondii infection in the bioreactor production . ( A ) Raw flow cytometry data and gates of the samples taken from the bioreactor on each indicated day post infection . ( B ) Histogram of the percent of infected cells ( black bars ) as well as the normalized free parasite counts ( blue line ) . Raw parasite counts were normalized to the total counts of each respective experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 12556 . 003 Peptide ligands eluted from HLA complexes were separated by offline HPLC fractionation and subjected to nanoLCMS . A total of 284 peptide sequences precisely matched the reported sequence of T . gondii . Strikingly , 89 of these peptides were either an exact match or contained an isobaric Ile/Leu ambiguity match to sequences of the H . sapiens host species ( Supplementary file 1 ) , leaving 195 ligands definitively derived from T . gondii ( Supplementary file 2 ) . Of these confirmed T . gondii derived ligands , most ( 159 ) show little ( <50% ) similarity to H . sapiens while 36 peptides have >50% similarity . In summary , 31% of the T . gondii derived ligands are identical to H . sapiens sequences , 13% show host similarity , and 56% of the T . gondii ligands have no similarity to host proteins ( Figure 2 ) . MHC I makes hundreds of parasite-encoded ligands available to the immune system . 10 . 7554/eLife . 12556 . 004Figure 2 . Sequence identity of identified T . gondii ligands to H . sapiens . T . gondii derived sequences were BLAST searched against the NCBInr H . sapiens proteome . Sequence identity was recorded and ligands with <50% sequence identity were considered not significant and were binned together . DOI: http://dx . doi . org/10 . 7554/eLife . 12556 . 004 Previous studies have focused upon dense granule proteins ( GRA ) that are secreted by T . gondii into the cytosol of the host cell , and a handful of GRA ligands have been reported ( Blanchard et al . , 2008; Cardona et al . , 2015; Cong et al . , 2011; 2012 ) . In our dataset , the 195 T . gondii ligands identified originate from 95 different proteins of which 87 are non-GRA proteins , demonstrating that a considerable number of proteins are accessible to MHC I presentation . For 55 . 8% of the T . gondii source proteins a single peptide was presented ( Figure 3A ) , while elongation factor 1 alpha provided 12 peptide ligands . Peptide ligand enrichment from particular proteins was not due to protein length as median protein lengths were not statistically different ( Kruskal-Wallis test , p = 0 . 247 ) regardless of the number of ligands embedded within a protein ( Figure 3—figure supplement 1 ) . Amongst the proteins sampled more than once , a hierarchy emerged whereby several hypothetical proteins were most frequently sampled followed by the dense granule proteins , ribosomal proteins , EF1α , tRNA synthetases , and HSP70 , respectively ( Figure 3B , Supplementary file 2 ) . All together , multiply sampled T . gondii source proteins provided 44 . 2% of the pathogen-derived ligands . Dense granule proteins have been reported as a source of peptides ( Blanchard et al . , 2008; Cardona et al . , 2015; Cong et al . , 2010; 2011; 2012; El Bissati et al . , 2014 ) , and here 30 GRA ligands ( 15% ) were observed with GRA12 providing the most ( Hassan et al . , 2015 ) peptide ligands ( Figure 3C ) . The secreted GRA proteins represent a minority protein source in the rich ligand landscape of this pathogen . 10 . 7554/eLife . 12556 . 005Figure 3 . Ligand sampling of source proteins . ( A ) The number of distinct ligands from a given source protein was counted binned by number of ligands . Gene symbols of the most sampled proteins are shown above the respective bin . ( B ) Distribution of ligands by source protein group or individual source protein . ( C ) Distribution of ligands by source dense granule protein . DOI: http://dx . doi . org/10 . 7554/eLife . 12556 . 00510 . 7554/eLife . 12556 . 006Figure 3—source data 1 . PEAKS export file containing HLA-A*02:01 peptide H . sapiens derived ligands from uninfected THP-1 cells . This is the underlying data for Figures 4 , 5 , and 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 12556 . 00610 . 7554/eLife . 12556 . 007Figure 3—figure supplement 1 . Number of ligands do not correspond to source protein length . Proteins were binned by the number of ligands identified . The median values of the source protein length in each bin are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 12556 . 007 In the search for trends and biases in T . gondii peptide ligands , we tested for significant enrichments in cellular compartments of sampled source proteins . There was a significant enrichment in proteins localized to the apical part of the T . gondii cell ( GO:0045177 , p = 1 . 50 x 10-8 ) and to the parasitophorous vacuole ( GO:0020003 , p = 2 . 79 x 10-6 ) . Unexpectedly , there was a significant enrichment in proteins originating from the parasite cytoplasm ( GO:0005737 , p = 9 . 31 x 10-4 ) with 21 of 95 proteins annotated as cytoplasmic . This enrichment in peptides derived from T . gondii cytoplasm proteins shows that T . gondii does not sequester cytoplasmic proteins from host MHC I antigen processing and presentation . A recent study showed that the C-terminal location of an epitope within the source protein was important for T . gondii ligand presentation and immunodominance ( Feliu et al . , 2013 ) . Given this observation , we assessed the location of ligands within their source proteins to determine if a C-terminal bias was maintained . Normalized ligand position was calculated as described by Kim et al . ( Kim et al . , 2013 ) and were binned into 5 positions with the most N-terminal bin being 0–0 . 2 and the most C-terminal bin being 0 . 8–1 . 0 ( Figure 4 ) . When the T . gondii ligands were compared to the baseline uninfected ligand distribution there was a significant reduction in N-terminal peptides ( bins 0–0 . 2 , p = 2 . 36 x 10-4 and 0 . 2–0 . 4 , p = 2 . 23 x 10-4 , comparison of proportions ) along with significant enrichment in ligands from the C-terminal end of the protein ( bins 0 . 6–0 . 8 , p = 1 . 46 x 10-2 and 0 . 8–1 . 0 , p = 5 . 26 x 10-7 ) . There was no significant change ( p = 0 . 238 ) to the central ligands ( bin 0 . 4–0 . 6 ) . Next , host-derived ligands from the infected and uninfected cells were compared to see if the C-terminal bias of pathogen encoded peptides extended to the infected host . There was significant reduction in the 0 . 2–0 . 4 bin ( p = 0 . 007 ) with a significant increase ( p = 0 . 0095 ) in the very C-terminal bin ( p = 0 . 0095 ) . In summary , T . gondii ligands are significantly enriched from the C-terminal end of their source proteins and host-derived peptide ligands shift towards the C-termini of their respective source proteins following infection . 10 . 7554/eLife . 12556 . 008Figure 4 . Location of ligands within respective source proteins . Normalized ligand location within the respective source protein from the unambiguous T . gondii ligands ( black ) , H . sapiens ligands from infected THP-1 cells ( white ) and , H . sapiens ligands from uninfected THP-1 cells ( grey ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12556 . 008 Our in depth proteomics approach identified thousands of HLA-A*02:01 peptide ligands of a canonical length of 8–11 as well as ligands of non-canonical length . Noteworthy is that a number of the T . gondii ligands were much longer than expected for the HLA-A*02:01 molecule ( Figure 5 ) . When compared with the host-derived ligands , the T . gondii ligands were significantly longer ( t-test , p<0 . 001 ) with an average length of 14 . 6 amino acids compared with 11 . 4 amino acids in the host ligands from the infected cells and with 9 . 8 for uninfected host ligands . This increase in host ligand length following infection was statistically significant ( t-test , p<0 . 001 ) . Thus , infection increases the length of T . gondii and host-derived ligands . 10 . 7554/eLife . 12556 . 009Figure 5 . Length distribution of identified ligands . Length distributions of unambiguous T . gondii ligands ( red ) , H . sapiens ligands from infected THP-1 cells ( purple ) and , H . sapiens ligands from uninfected THP-1 cells ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12556 . 00910 . 7554/eLife . 12556 . 010Figure 5—source data 1 . PEAKS export file containing HLA-A*02:01 peptide T . gondii derived ligands from T . gondii infected THP-1 cells . This is the underlying data for Figures 2 , 3 , 4 , 5 , and 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 12556 . 01010 . 7554/eLife . 12556 . 011Figure 5—source data 2 . PEAKS export file containing HLA-A*02:01 peptide H . sapiens derived ligands from T . gondii infected THP-1 cells . This is the underlying data for Figures 4 , 5 , and 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 12556 . 011 To investigate how long peptides observed following infection might bind to MHC I , in silico algorithms were used to predict HLA-A*02:01 ligand binding affinity . NetMHCpan 2 . 8 predicted that 117 of 195 ligands ( 60% ) bind at a percentile rank score <=10% , leaving 78 predicted non-binders ( Figure 6 ) . As these long peptides were purified from the HLA-A*02:01 of infected cells , we hypothesized that the predicted non-binders interact with HLA-A*02:01 in a non-canonical fashion that escapes algorithms trained on canonical binding data . A subsequent search revealed nested core sequences of 8–11 amino acids within the longer peptides that were not predicted to bind , and these cores were predicted to bind to HLA-A*02:01 . Strikingly , a considerable number ( 52/78 ) of the predicted non-binders contained nested core sequences of 8–11 amino acids that were predicted to bind with high affinity ( percentile rank score <=2% ) . This was significantly more ( p = 1 . 2 x 10-10 , comparison of proportions ) than if the sequences were randomly scrambled ( 13/78 ) . Hence , a binding core appropriate in length and sequence for interaction with HLA-A*02:01 was embedded within long T . gondii peptides . 10 . 7554/eLife . 12556 . 012Figure 6 . Binding prediction analysis of eluted ligands . Percentage of total ligands in indicated dataset that are predicted to be canonical binders ( blue ) , contain a C-terminal binding core ( red ) , contain an N-terminal binding core ( green ) , contain a central binding core ( purple ) , or not predicted to bind ( orange ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12556 . 012 We next assessed the positioning of binding cores within longer ligands . Within the T . gondii ligands , most ( 36/52 ) binding cores were at the very N-terminus with the peptide extending from the core’s C-terminus ( Supplementary file 3 ) . This occurred significantly more than randomly scrambled versions of the peptides ( 1/13 ) ( p = 3 . 1 x 10-5 , comparison of proportions ) . There was no significant increase in proportions of ligands with an N-terminal extension ( 1/52 ) or an extension on both sides ( 15/52 ) compared to scrambled ligand sequences ( 0/13 and 12/13 respectively ) . Noteworthy was that 10/15 of the peptides with predicted extensions on both sides of a central core had an alternate binding core at the N-terminal side with a predicted affinity slightly weaker than the central binding core ( Supplementary file 3 ) . So , most peptides could bind either at the N-terminal end or in the center , and many of the ligands with central cores possessed an alternate N-terminal core . Among the N-terminal binders , the average C-terminal extension was 8 . 7 amino acids and the extension varied considerably from 1 to 30 amino acids ( Supplementary file 3 ) . In summary , most of the long T . gondii predicted non-binders had an N-terminal binding core with a considerable C-terminal extension . The observation of C-terminally extended peptides among the T . gondii derived ligands prompted a similar analysis of the host ligand repertoire from infected cells . Among ligands from uninfected cells , 2 . 3% had a C-terminal extension - the baseline of extended peptides in uninfected THP-1 cells . After T . gondii infection , the percentage of host ligands with a C-terminal extension significantly increased to 7 . 2% ( p<0 . 0001 , comparison of proportions ) ( Figure 6 ) , a percentage that is less than half the T . gondii derived ligands with extensions ( 18 . 5% ) . Thus , T . gondii infection results in C-terminally extended host ligands being nearly 3 times more frequent and T . gondii C-terminally extended ligands being eight times more common . The observation of nested binding cores at the N-termini of extended ligands suggests a novel interaction of T . gondii peptides and MHC I . To confirm that the N-terminal portion of the peptides were binding to HLA-A*02:01 , we synthesized full-length peptides and their corresponding binding cores and determined their affinities for HLA-A*02:01 in a competitive binding assay . Twelve ligands having binding cores with the highest predicted binding affinities ( Supplementary file 3 , bold ) were selected for testing in this manner . Of these 12 peptides , eight binding cores bound with high affinity ( <500 nM ) , three with moderate affinity ( <1000 nM ) , and one had no affinity – the N-terminal cores overwhelmingly bound to HLA-A*02:01 . Two of the twelve full-length ligands ( YLSPIASPLLDGKSLR-RPL7A15-30 ) and FVLELEPEWTVK-UFP16-27 ) also bound with high affinity ( Figure 7A ) even though predictions ranked them as non-binders as their C-termini were incompatible with the HLA-A*02:01 binding motif . The remaining 10 long peptides did not bind in this in vitro assay , perhaps because they are incapable of binding in the absence of molecular chaperones or the distinct conditions of the infected cell . These data confirm that the N-termini of extended peptides have a strong affinity for MHC I . 10 . 7554/eLife . 12556 . 013Figure 7 . Binding affinity of extended ligands and their respective binding cores . ( A ) Measured IC50 of extended peptides ( black fill ) and the respective predicted binding core ( white fill ) . Blue line denotes 500 nM; binding affinities below this are considered binders . ( B , C ) Mutation analysis of FVLELEPEWTVK and YLSPIASPLLDGKSLR with non-permissive F’ pocket residues . Blue letters denote the mutated residue . All data shown are the results of two independent experiments run in triplicate or duplicate . P-values shown are the result of an unpaired two-tailed t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 12556 . 013 Next , the two extended peptides that bound in the in vitro binding assay were tested to see if the binding core of the full-length ligand was essential for binding . In order to assess the contribution of the putative core for peptide binding , HLA-A*02:01 binding assays were completed using the full length peptide , the binding core only , and a series of full length peptide variants containing non-permissive amino acid substitutions at the putative C-terminal anchor of the binding core . For FVLELEPEWTVK , the affinity of all 'mutant' peptides was significantly worse ( t-test , p<0 . 05 ) than either the binding core or the native full-length peptide ( Figure 7B ) . With YLSPIASPLLDGKSLR , all but one ( L10E ) of the mutants had a lower binding affinity in comparison to the wild type sequence ( t-test , p<0 . 05 ) ( Figure 7B ) . As YLSPIASPLLDGKSLR harbors a Leu at P9 and at P10 , it could be that either Leu can serve as an F’ pocket anchor for HLA-A*02:01 and when the P10 was changed to an acidic residue like L10E , the P9 Leu was used as the F-pocket anchor . In summary , the binding core’s F-pocket residue is critical for long-peptide binding to HLA-A*02:01 such that C-terminal amino acid extensions of these longer ligands somehow protrude out of the groove in the vicinity of the F-pocket . Extended ligands YLSPIASPLLDGKSLR-RPL7A15-30 and FVLELEPEWTVK-UFP16-27 bound at high affinity to HLA-A*02:01 , and we next assessed their relative stability when in complex with MHC I . To determine the relative stability of these extended peptide/MHC I complexes , we completed a thermal denaturation assay on the extended peptides and their respective binding cores in complex with HLA-A*02:01 ( Figure 8 ) . The melting temperature ( Tm ) for the two extended ligands exceeded 62°C ( YLSPIASPLLDGKSLR , Tm = 66° C , FVLELEPEWTVK Tm = 63°C ) , temperatures that are consistent with ligands of conventional length ( Hassan et al . , 2015 ) . The Tm of these extended ligands are also within the range of 15 mers that are reported to 'bulge' in the central portion of HLA-A*02:01 binding groove ( Hassan et al . , 2015 ) . It was somewhat surprising that these extended ligands had Tm representative of canonical ligands , and more remarkable was that the thermostability of the binding cores was at least 10°C higher ( YLSPIASPLLDGKSLR , ΔTm = 10°C , FVLELEPEWTVK ΔTm = 12°C ) than their extended counterparts ( Figure 8 ) . Thus , extended ligands have a denaturation temperature consistent with those of conventional ligands , possibly due to the highly thermostable nature of their binding cores . 10 . 7554/eLife . 12556 . 014Figure 8 . Thermal denaturation of extended ligands . ( A ) Raw fluorescence of the melt curve for indicated peptide/HLA-A*02:01 complex . ( B ) First derivative of the melt curve from thermal denaturation experiment for HLA-A*02:01 indicated peptide ligand . The melting temperature for each peptide was calculated from the minima of these curves and is shown in the figure legend . DOI: http://dx . doi . org/10 . 7554/eLife . 12556 . 014 To determine how C-terminal extensions protrude from the F’ pocket of MHC I , we bound the extended ligand FVLELEPEWTVK ( UFP16-27 ) and its binding core FVLELEPEWTV ( UFP16-26 ) to HLA-A*02:01 , crystallized these two complexes , and solved their structures at a resolution of 1 . 5 Å and 1 . 87Å , respectively ( Supplementary file 4 ) . Both peptides bound in a zig-zag fashion in order to accommodate the core 11 amino acids in the antigen-binding groove ( Figure 9A ) . The 11 amino acid core of extended ligand FVLELEPEWTVK interacts with HLA-A*02:01 in an almost identical manner to the shorter peptide , however , both the main chain and side chain of C-terminal Lys12 protrudes out the end of the binding pocket ( Figure 9B ) . Electron density for both peptides was very well defined over the entire peptide length ( Figure 9C , D ) . The N-terminal and C-terminal amino acid residues of the peptide provide the majority of H-bond and van der Waals contacts ( Supplementary file 4 ) . When attention was shifted to the F- pocket of HLA-A*02:01 , with the shorter peptide this pocket was closed by the side chains of Thr80 and Tyr84 with Lys146 of the MHC I providing a lid or cover above the F pocket thereby burying the peptide’s C-terminal 11th residue underneath ( Figure 9A , E ) . However , for HLA-A*02:01 in complex with FVLELEPEWTVK , the side chain of Tyr84 swung up and out by almost 90 degrees ( Figure 9A , F ) , opening the binding groove so that a peptide might protrude from the groove at its C-termini . At the same time , Thr80 adopted a different rotamer , further opening the MHC I pocket toward the end of the α1-helix . Lastly , there was a subtle but noticeable increase in the main chain distance between the α 1 and α2-helices at the F’ pocket ( Figure 9G ) . Together , these structural changes opened the F’ pocket and allowed Lys12 of the peptide to protrude from the pocket . 10 . 7554/eLife . 12556 . 015Figure 9 . Structural details of extended ligand binding to HLA-A*02:01 . Binding of core peptide FVLELEPEWTV ( A , C , E ) and extended ligand FVLELEPEWTVK ( B , D , F ) to HLA-A*02:01 . Peptides are shown as sticks , while HLA-A*02:01 is shown as a molecular surface with electrostatic potential contoured from -30kT/E to +30kT/E ( positive charge in blue , negative in red ) . Peptide FVLELEPEWTV in green , and FVLELEPEWTVK in yellow . 2FoFc electron density is shown as a blue mesh around the peptide ( 2 Å radius ) FVLELEPEWTV ( C ) and FVLELEPEWTVK ( D ) and contoured at 1σ Details of peptide binding to the F’ pocket of MHC ( E , F ) . MHCI residues that form H-bond interactions ( blue dashed lines ) with the peptide are labeled . MHC residues that are critical for the F’ pocket formation are shown with electron density with same settings as in C and D . ( F ) Note how Thr80 ( T80 ) and Tyr84 ( Y84 ) change position upon binding of extended ligand FVLELEPEWTVK . Those structural changes are not seen in PDB ID 2CLR ( orange ) when superimposed with UFP ( 16–26 ) and UFP ( 16–27 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12556 . 015 We compared our structures to a previously reported structure of HLA-A*02:01 in complex with a peptide that also extends its C-terminus from the F’ pocket ( Collins et al . , 1994 ) . While the authors anticipated the structural changes necessary to allow a peptide to protrude from the F’ pocket , the reported structure shows a minor structural change where Tyr84 , the key player in opening the F’ pocket , rotates slightly but does not swing out , and Thr80 did not change its rotamer . A slight movement of the Lys146 side chain , which appeared to be sufficient to allow the small C-terminal glycine to stick up , was reported ( Figure 9G ) . However , in this previously reported model , amino acids other than glycine could not extend from the F’ pocket due to steric clashes with the MHC I . Here , we have identified a previously unreported mechanism that allows pathogen-encoded C-terminally extended peptides to protrude out the end of the MHC I binding grove at the F’ pocket . The binding cores of these C-terminally extended peptides interacts with MHC I using the same peptide register of canonical MHC I binders .
Proteins secreted by T . gondii have been demonstrated to be an important source of immunogenic MHC I peptide ligands ( Grover et al . , 2014 ) . Indeed , the secreted dense granule proteins are the best-characterized source of peptides for presentation by MHC I , and our study confirms these proteins provide a number of MHC I ligands ( Blanchard et al . , 2008; Cardona et al . , 2015; Cong et al . , 2011 ) . However , many of the ligands we observe originate from non-secreted proteins , including parasite cytoplasmic proteins . Enrichment in these proteins is unexpected given that most T . gondii cytoplasmic proteins are not secreted and their cellular location should sequester them from the MHC I peptide processing machinery . Further , we identified many hypothetical proteins sampled by MHC I . Proteomic data had previously been reported for all but one of these hypothetical proteins , and our data substantiates that these proteins are expressed by the tachyzoites within the infected host ( Gajria et al . , 2008; Treeck et al . , 2011; Xia et al . , 2008 ) . Additionally , a little less than half ( 10/22 ) of the hypothetical proteins contain a predicted secretion signal sequence . These data represent a substantial contribution to understanding patterns of T . gondii gene/protein expression and demonstrate that secreted as well as non-secreted tachyzoite-expressed proteins are accessible to host antigen presentation processes . A previous study of MHC I and T . gondii indicated that ligands derived from the C-termini of T . gondii source proteins are immunodominant ( Feliu et al . , 2013 ) , and it was suggested that a proteolytic insufficiency within infected murine cells meant that only peptides derived from the C-termini of T . gondii source proteins would provide ligands of optimal length ( Feliu et al . , 2013 ) . Consistent with this , we found a dramatic C-terminal bias in the T . gondii derived ligands . However , it was completely unexpected that host-derived ligands in the infected cell would also display a C-terminal bias . Gene ontology annotation analysis indicated no bias in the cellular location of the source proteins for these C-terminal ligands ( data not shown ) , so it does not appear to be a protein’s location within the cell that results in a C-terminal processing bias . How T . gondii is mediating this intracellular alteration in antigen processing and presentation is therefore an unsettled question , and it will be interesting to see if protein ligand sampling is also biased in the closely related parasite Plasmodium falciparum . One of the most striking observations in this study is that infection leads to the MHC I presentation of T . gondii ligands that are longer than baseline host ligands . Most ligands of considerable length are thought to be anchored to the B’ and F’ pockets of the MHC I groove via the peptide’s P2 and PΩ side chains , respectively . In such models the long peptide ligands exhibit a central bulge in order for the MHC I groove to accommodate the length of the extra residues ( Hassan et al . , 2015; Tynan et al . , 2005 ) . Consistent with this model , several long T . gondii ligands identified follow this P2/PΩ-anchor means of binding to MHC I . Rather unexpectedly , a group of long ligands did not follow this model but instead were predicted to bind via a canonical N-terminal binding core preceding the aforementioned C-terminal extension . A review of published ligand elution data confirms that C-terminally extended peptides are presented by membrane-bound HLA-A*02:01 ( Hassan et al . , 2015; Chen et al . , 1994; Scull et al . , 2012 ) , yet the considerable number of extended ligands observed in this in vitro model of T . gondii antigen processing and presentation merits future confirmation in vivo . That notwithstanding , structural support for the binding of extended peptide ligands in HLA-A*02:01 can also be found ( Collins et al . , 1994 ) . In this example the nonamer binding core of a calreticulin peptide is extended by a single C-terminal amino acid that fits within the MHC I groove , yet this model was not consistent with T . gondii ligands that had as many as 30 C-terminally appended amino acids . In order to resolve this enigma , an HLA-A*02:01 crystal was solved with FVLELEPEWTVK and with a version of this ligand minus the C-terminal K . This led to the unprecedented observation that the C-terminal lysine at P12 displaced the Tyr84 residue at the end of the HLA-A*02:01 binding groove . As such , Tyr84 emerged as a “swinging gate” whereby a longer T . gondii peptide extended straight out the end of the HLA molecule at its C-terminus when Tyr84 assumed an alternate up and out position – the open gate . With the shorter peptide the Tyr84 was positioned down and in , assuming the traditional closed-groove orientation . A Tyr84 in the open position is consistent with several extended T . gondii ligands and a number of host ligands . This structural configuration is distinct from c-terminal extensions previously reported with C-terminal extension of covalently linked peptides ( Mitaksov et al . , 2007 ) and predicted configurations ( Hörig et al . , 1999 ) where the peptide travels up and over Tyr84 . Note that Tyr84 modestly rotated to facilitate the reported binding of calretulin to HLA-A*02:01 ( Collins et al . , 1994 ) , and the gate-open displacement of Tyr84 observed here confirms that the MHC I binding pocket is less rigid than previously realized . Importantly , this method of binding appears to be a major mechanism by which T . gondii and peptides are bound and presented in contrast to host-derived peptides where this mode of binding significantly less frequent . The increased frequency of extended peptides following infection suggests that T . gondii ligands emerge from a distinct pathway of antigen processing and presentation . Class I MHC peptide ligands are typically derived from the proteasomal degradation of cytosolic proteins , active transport by TAP into the lumen of the ER , further proteolytic trimming in the ER , and chaperone mediated loading into class I MHC prior to egress to the plasma membrane . In the case of T . gondii , parasites are contained within a fusion resistant parasitophorous vacuole ( PV ) , making it necessary that alternative mechanisms contribute to T . gondii protein degradation , transport , and trimming prior to presentation . Current evidence shows that T . gondii proteins secreted into the PV can be retrotransolcated via the endoplasmic reticulum associated degradation ( ERAD ) complex from lumen of the PV to the host cytosol where they are routed to MHC I ( Blanchard et al . , 2008; Feliu et al . , 2013; Grover et al . , 2014; Gubbels et al . , 2005 ) . This process of presenting canonical peptide ligands from exogenous non-cytosolic proteins by class I MHC is referred to as cross-presentation ( Blanchard and Shastri , 2010; Grotzke and Cresswell , 2015 ) . The extended T . gondii ligands observed here do not seem to fit this model of cross-presentation as their C-terminal extensions suggest a lack of interaction with proteolytic agents of the host cytosol and as many T . gondii ligands are derived from proteins not secreted into the PV . Rather , infected cells seem to exhibit a distinct means of cross-presentation , almost as though extended ligands move directly from the PV or the pathogen itself to the host’s ER , forgoing exposure to carboxypeptidases found in the cytosol that are otherwise absent from the lumen of the ER . Indeed evidence of a semi-permeable channel between the PV and the ER might explain the presentation of extended peptides ( Goldszmid et al . , 2009 ) . However it is that ligands of unusual length reach class I MHC , future studies of T cell immunogenicity to T . gondii , and possibly to other large intracellular pathogens , must factor the distinct environment of the infected cell into the identification of immune epitopes . In summary , this study reports how antigens encoded by the large intracellular pathogen T . gondii are processed and presented by the host cell’s MHC I . Studies of peptides that are naturally processed and presented by the MHC I of mammalian cells have historically used healthy , or uninfected , cells to provide the baseline understanding for how antigens are made available for immune recognition . A foundation has emerged whereby peptide ligands of 9 amino acids are enveloped into an MHC I binding groove such that the central portions of the peptide ligand are available for review by adaptive immune receptors , and a legion of structural and functional data support this paradigm . Here we see that thousands of peptide ligands harvested from the MHC I of infected cells also fit this canonical antigen processing and presentation model , but in parallel we observe that alternate and unanticipated mechanisms result from infection and play a role in making T . gondii available for immune recognition . That long ligands can extend from their C-termini in a linear fashion out the end of what was previously recognized as a closed MHC I groove was unexpected . Our findings raise a plethora of important questions that must now be addressed , including whether other large intravacuolar pathogens such as Plasmodium and Mycobacterium species mediate similar changes to MHC I ligand presentation , how infection remodels host cell biology to facilitate the delivery of long extended ligands to MHC I , and the impact that unconventional ligand presentation has on adaptive immune responses to infected cells .
THP-1 cells acquired from ATCC ( ATCC# TIB-202 ) were cultured in RPMI supplemented with 10% FBS . THP-1 cells were routinely authenticated by HLA typing using sequence based typing at the HLA-A , B , C , and DRB1 loci from an American Society for Histocompatibility and Immunogenetics ( ASHI ) accredited laboratory ( ASHI#03-5-OK-07-1 ) ( Lanteri et al . , 2011 ) . The reported HLA type of the THP-1 cells are HLA-A*02 , -B*15 , -C*03 , -DRB1*01 , -DRB1*15 ( Battle et al . , 2013; Tsuchiya et al . , 1980 ) . The observed HLA type of the THP-1 cells used in all experiments is HLA-A*02:01 , -B*15:11 , C*03:03 , DRB1*01:01 , DRB1*15:01 . THP-1 cells were transfected with a soluble form of HLA-A*02:01 as previously described ( McMurtrey et al . , 2008 ) . HLA in the supernatant was measured using a sandwich ELISA using W6/32 as a capture mAb and anti-β2m antibody as a detector . HLA producing cells were subcloned and used for T . gondii infection . Toxoplasma gondii strain RH expressing GFP was propagated on human foreskin fibroblasts cells acquired from ATCC ( ATCC# SCRC-1041 ) cultured in DMEM supplemented with 10% FBS , glutamine and penicillin/streptomycin . Parasites were released from host cells by passage through a 27-gauge needle ( Wiley et al . , 2010 ) All host cell lines and parasites were routinely tested for Mycoplasma contamination with either the MycoAlert Mycoplasma Detection Kit ( Lonza , Basel , Switzerland ) or Venor GeM Mycoplasma Detection Kit ( Sigma-Aldrich , St . Louis MO ) and found to be negative . . HLA from uninfected and infected cells were purified as previously described ( McMurtrey et al . , 2008; Yaciuk et al . , 2014 ) . Briefly , THP-1 cells producing sHLA-A*02:01 were seeded into a hollow fiber bioreactor . For T . gondii infection , cells were expanded to confluence and then infected on day 27 with 3 . 72 x 109 parasites . Bioreactor supernatant containing HLA was collected and pooled over the course of the seven day infection . HLA was purified from both infected and uninfected cells using antibody affinity chromatography with an anti-VLDL antibody . HLA was eluted in 0 . 2 M acetic acid and further acidified to 10% acetic acid . Peptide ligands were dissociated from the alpha chain by heating to 75°C for 15 min . Alpha chain and β-2m were separated from the eluted peptides by 3kDa cutoff ultrafiltration . THP-1 cells were infected with GFP-expressing parasites in the bioreactor and cells were periodically sampled from the extra capillary space of the bioreactor . 1x106 cells were stained with 1 ug of the pan-HLA class I specific antibody W6/32 labeled with Alexafluor 647 and incubated at room temperature for 30 min to differentiate whole cells from cell debris and parasites . Cells were washed with 1% BSA in PBS three times and then fixed with 1% PFA for 15 min at room temperature . Cells and free parasites were measured using a BD FACS Calibur flow cytometer . HLA peptide ligands are were identified with a two-dimensional LCMS system as described ( Yaciuk et al . , 2014 ) . Briefly , peptide pools were fractionated using high pH off-line reverse phase HPLC . Each fraction was dried , resuspended in 10% acetic acid , and placed into an Eksigent NanoLC 400 U-HPLC auto sampler system ( Sciex ) . Approximately twenty percent of each fraction was injected onto a nano-LC column and eluted with a linear acetonitrile water gradient at low pH ( Yaciuk et al . , 2014 ) . Eluate was ionized with a nanospray III ion source and analyzed with a 5600 Triple-TOF mass spectrometer ( Sciex ) . Survey and fragment spectra for all fractions were analyzed using PEAKS ( Bioinformatics Solutions Inc ) and were searched against NCBInr database using Homo sapiens ( ID: 9606 ) or Toxoplasma gondii ( ID: 5811 ) taxonomy filters . For Homo sapiens searches a 1% FDR was applied and for Toxoplasma gondii a 2% FDR was used . All T . gondii unmodified peptide sequences were confirmed with fragmentation of a synthetic peptide . All source proteins derived from T . gondii were manually converted from NCBInr format to ToxoDB ( www . toxodb . org ) format including official protein names and gene symbols . T . gondii source protein gene IDs were used as input for the Gene Ontology Enrichment tool ( Gajria et al . , 2008 ) . Enrichments were calculated using T . gondii strain GT1 , and both annotated as well as predicted terms were considered . Reported p-values are the Bonferronii adjusted p-value . Predicted HLA-A*02:01 binding affinities were generated for all eluted peptides using NetMHCpan-2 . 8 ( Hoof et al . , 2009 ) . The percentage rank score was used for all analysis . The percentage rank score indicates how strong a peptide’s predicted binding affinity is compared to a large pool of naturally occurring peptides . A rank score of 10% indicates that a peptide is amongst the 10% strongest binding random natural peptides for HLA-A*02:01 . Peptides with predicted rank scores <=10% were classified as binders , all other peptides were considered non-binders . All non-binders were screened for potential nested HLA-A*02:01 binders by predicting the binding affinity of all overlapping 8–11mers within the eluted peptide sequence . A more conservative rank score <=2% was used to identify nested binders . If multiple nested binders were identified within the same eluted peptide , the nested binder with the strongest predicted binding affinity was selected . Permuted peptides were generated by scrambling the amino acid sequence of the eluted peptide and predictions were preformed in the same manner . Assays to quantitatively measure peptide binding to HLA-A*0201 ( MHC I ) molecules are performed essentially as detailed elsewhere ( Sidney et al . , 2001; 2008; Sidney , 2013 ) . In brief , 0 . 1–1 nM of radiolabeled peptide is co-incubated at room temperature with 1 µM to 1 nM of purified HLA-A*02:01 in the presence of a cocktail of protease inhibitors and 1 µM β2-microglobulin . Following a two day incubation , HLA-A*02:01 bound radioactivity is determined by capturing the HLA/peptide complexes on W6/32 ( anti-class I ) antibody coated Lumitrac 600 plates ( Greiner Bio-one , Frickenhausen , Germany ) , and measuring bound cpm using the TopCount ( Packard Instrument Co . , Meriden , CT ) microscintillation counter . Under the conditions utilized , where [label]<[HLA] and IC50 ≥ [HLA] , the measured IC50 values are reasonable approximations of the true Kd values ( Cheng and Prusoff , 1973; Gulukota et al . , 1997 ) . Each competitor peptide is tested at six concentrations covering a 100000-fold dose range in three or more independent experiments . As a positive control , the unlabeled version of the radiolabeled probe is tested in each experiment . HLA-A*02:01 class I heavy chain ectodomain ( residues 1–274 ) and human β-2 microglobulin ( hβ2m , 1–99 ) were expressed as inclusions bodies and refolded as reported previously ( Garboczi et al . , 1994 ) with modifications as reported . Briefly , 15 mg of HLA-A heavy chain mixed with 3 mg of peptide ( GenScript ) was then added to the refolding mix and further stirred at 4°C for 72 hr . Final heavy chain:light chain:peptide ratios were 2 . 5:1:12 for peptides FVLELEPEWTVK and FVLELEPEWTV . Following refolding , refolding mixture was spun at 50000g to remove any precipitated protein , supernatant concentrated to about 3 ml and loaded onto a Superdex S200 HR16/60 gel filtration column . Fractions containing refolded HLA-A*02:01-peptide complexes were pooled , concentrated to about 10–12 mg/mL and used for crystallization experiments . HLA-A*02:01-peptide complexes with peptides FVLELEPEWTV , FVLELEPEWTVK , YLSPIASPL and YLSPIASPLLDGKSLR were analyzed for thermal denaturation by differential scanning fluorimetry . HLA-A*02:01-peptide complexes at 100 μM in reaction buffer ( 20 mM Tris-HCl pH 7 . 5 , 150 mM NaCl ) were used as protein stock solution . Each reaction comprised of 1–2 μl protein stock solution , 2 μl of SYPRO Orange dye ( 100X , Invitrogen ) made up to 20 μl in reaction buffer . The experiment was performed in triplicates for individual peptide complexes using a LightCycler 480 ( Roche ) in a 96-well plate format . A temperature gradient from 20°C - 85°C at steps of 0 . 06°C/sec and 10 acquisitions/°C was run . The melt curve of the total fluorescence was plotted against the temperature . The first derivative of the melt curve was obtained from raw fluorescence data ( temperature differential of absolute fluorescence versus temperature ) and plotted as well . The minima in the first derivative of each melt curve , corresponding to the inflection point of the original melt curve , provided the meting temperature ( Tm ) of each protein ( Figure 8B ) ( Walden et al . , 2014 ) . Thin plate-like crystals were obtained for HLA-A*02:01 complex with UFP ( 16–26 ) in 30% PEG 5000 MME , 0 . 1 M Tris-HCl pH 8 . 0 , 0 . 2 M lithium sulfate and HLA-A*02:01 complex with UFP ( 16–27 ) in 30% PEG 4000 , 0 . 1 M Tris-HCl pH 8 . 0 , 0 . 2 M lithium sulfate . Crystals were obtained by sitting drop vapor diffusion by mixing 0 . 15 μl protein and 0 . 15 μl of precipitant at 20°C after 2–4 days . The crystals were flash frozen in cryoprotectant ( Reservoir solution: 100% glycerol - 3:1 ) using liquid nitrogen . Diffraction data for HLA-A2/ UFP ( 16–26 ) and HLA-A2/ UFP ( 16–27 ) were collected remotely at beamline 7 . 1 at the Stanford Synchrotron Radiation Light source ( SSRL ) and processed to 1 . 8 Å and 1 . 5 Å resolution , respectively using HKL2000 ( Otwinowski and Minor , 1997 ) . Phases were obtained using the protein coordinates for HLA-A2 ( PDB ID 3MRE ) using molecular replacement with Phaser MR ( Storoni et al . , 2004 ) in ccp4i ( CCP4 , 1994; Potterton et al . , 2003 ) and provided unambiguous electron density for both the peptides . Model building was carried out using COOT ( Emsley and Cowtan , 2004; Emsley et al . , 2010 ) . Structures were refined using Refmac ( Deckert-Schlüter et al . , 1994 ) to a final Rwork/Rfree of 0 . 164/0 . 222 for HLA-A2/UFP ( 16–26 ) ( PDB ID 5D9S ) and Rwork/Rfree of 0 . 195/0 . 219 for HLA-A2/UFP ( 16–27 ) ( PDB ID 5DDH ) . | Toxoplasma gondii is a parasite that can infect most warm-blooded animals and cause a disease called toxoplasmosis . In humans , toxoplasmosis generally does not cause any noticeable symptoms , but it can cause serious problems in pregnant women and individuals with weakened immune systems . T . gondii is one of many parasites that hide within human cells in an attempt to avoid detection by the immune system . However , proteins called Human Leukocyte Antigens , or HLAs , can reveal hidden parasites by carrying small sections of them from the inside the infected cell to the cell’s surface . The immune system can then recognize the fragments as foreign and attack the parasite . HLAs typically pick up parasite fragments of a certain length , which enables the immune system to recognize that what is being displayed is a piece of parasite . By purifying HLAs from cells that have been infected by T . gondii , McMurtrey et al . have now learned more about which fragments of the parasite are displayed to the immune system . This analysis revealed that the parasite somehow manipulates the HLAs to carry parasite fragments that are considerably longer than can be explained with our current knowledge of how HLAs work . By using a technique called X-ray crystallography , McMurtrey et al . also show that the structure of the HLA assumes a previously unseen configuration when interacting with fragments of T . gondii . In the future , it will be important to understand how infected cells give rise to unusual structural configurations of HLAs and to unravel how these structures affect the immune system’s ability to fight infections . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"immunology",
"and",
"inflammation"
] | 2016 | Toxoplasma gondii peptide ligands open the gate of the HLA class I binding groove |
The bithorax complex ( BX-C ) in Drosophila melanogaster is a cluster of homeotic genes that determine body segment identity . Expression of these genes is governed by cis-regulatory domains , one for each parasegment . Stable repression of these domains depends on Polycomb Group ( PcG ) functions , which include trimethylation of lysine 27 of histone H3 ( H3K27me3 ) . To search for parasegment-specific signatures that reflect PcG function , chromatin from single parasegments was isolated and profiled . The H3K27me3 profiles across the BX-C in successive parasegments showed a ‘stairstep’ pattern that revealed sharp boundaries of the BX-C regulatory domains . Acetylated H3K27 was broadly enriched across active domains , in a pattern complementary to H3K27me3 . The CCCTC-binding protein ( CTCF ) bound the borders between H3K27 modification domains; it was retained even in parasegments where adjacent domains lack H3K27me3 . These findings provide a molecular definition of the homeotic domains , and implicate precisely positioned H3K27 modifications as a central determinant of segment identity .
Cells in higher organisms choose among developmental pathways in response to transitory cues . The choice must be remembered; this is often described as changing the epigenetic state . One prominent mechanism to fix such choices relies on the genes of the Polycomb Group ( PcG ) . The members of this family have been well defined by genetic and biochemical studies ( Simon and Kingston , 2013 ) . The PcG proteins are present in all cells , but they impose long-term repression on different target genes in different lineages . PcG proteins form distinct complexes with distinct functions . These include the PRC1 family of complexes , which compact chromatin and ubiquitylate histone H2A , and the PRC2 complexes , which methylate histone H3 on lysine 27 . The molecular details of how PcG functions are coordinated to lead to repression are poorly understood . The prototypic targets of PcG action are the Drosophila homeotic genes that control segment identity . Indeed , a mutation in Polycomb , the founding member of the PcG , was discovered for its segmental transformations ( Lewis , 1947 ) . In Drosophila melanogaster , the homeotics lie in two clusters , the Antennapedia complex ( Antp-C ) and the bithorax complex ( BX-C ) . The BX-C is the better-studied cluster , due largely to the effort of EB Lewis . Lewis showed that mutations that transform specific segments align on the genetic map in the order of the body segments that they affect ( Lewis , 1978 ) . The evolutionary conservation of this gene order , including that in the mammalian HOX complexes , suggests that chromosomal position dictates the expression pattern . Lewis used nested deletions to show that successively more distal genetic functions of the BX-C are activated in successively more posterior segments along the body axis ( Lewis , 1981 ) . The genetic functions are now thought to encompass nine regulatory domains , each for a different parasegment . The regulatory domains control three homeobox-containing genes of the complex , Ultrabithorax ( Ubx ) , abdominal-A ( abd-A ) , and Abdominal-B ( Abd-B ) ( Maeda and Karch , 2009 ) . The regulatory domains have been roughly defined by mutant lesions ( Lewis , 1978 ) , and by enhancer traps in the BX-C , which show different anterior limits of expression ( Bender and Hudson , 2000 ) . The molecular features that define these regions , as well as the precise locations of their boundaries , have not been defined . Both PRC1 and PRC2 are required to maintain repression of homeotic genes in appropriate parasegments , so these complexes may generate segment-specific chromatin features in the regulatory domains of the BX-C . We initially evaluated H3K27me3 patterns because they reflect the central function of PRC2 . The entire BX-C is heavily marked with H3K27me3 in whole embryos ( Schuettengruber et al . , 2009; Nègre et al . , 2011 ) . This histone modification is necessary for maintaining repression of BX-C genes ( Pengelly et al . , 2013 ) , yet these genes are not repressed throughout the body . It seemed likely that parasegment-specific patterns of histone modification might be obscured when all the parasegments are pooled for analysis . Indeed , Papp and Müller ( 2006 ) observed loss of the K27me3 mark across the Ubx transcription unit in haltere and third leg imaginal discs , tissues where Ubx is transcribed . These discs include cells of two parasegments ( PS5&6 ) , and so it was not clear how the K27me3-free regions might correlate with transcription units or with the genetic regulatory domains . The issue could be resolved if H3K27me3 patterns could be studied in single parasegments .
The chromatin features of the BX-C have not been studied in individual parasegments because of the technical challenges of cell isolation and molecular analysis on small samples . To address this , we marked single parasegments with a combination of Gal4 and Gal80 drivers , and developed a nuclear sorting-chromatin immunoprecipitation-sequencing protocol . Gal4 , a transcriptional activator , was expressed in a series of parasegments with a defined anterior boundary , and Gal80 was expressed in a pattern shifted one parasegment more posterior . Gal80 binds to and inactivates Gal4 , leaving Gal4 activity in a single parasegment ( Figure 1A ) . Gal4 and Gal80 expression domains were established either by enhancer trapping or by selecting an enhancer to achieve the desired expression pattern ( Figure1—figure supplement 1 ) . Various combinations of Gal4 and Gal80 drivers were combined genetically to limit Gal4 activity to each of parasegments 4 , 5 , 6 , and 7 , which approximately correspond to the second thoracic through the second abdominal segments ( Figure 1B , Figure1—figure supplements 1 and 2; ‘Materials and methods’ for details ) . 10 . 7554/eLife . 02833 . 003Figure 1 . Marking single parasegments . ( A ) Drivers for the Gal4 activator and the Gal80 repressor , each with a different anterior limit , are combined genetically . Gal4 activity is thus limited to a single parasegment , and is used to activate transcription of a fluorescent nuclear envelope protein . ( B ) Expression patterns used for isolation of parasegment nuclei are shown , visualized with antibody to the FLAG epitope on the INTACT fusion protein . Each panel shows embryos at about 6 , 8 , 10 , 12 , 14 , and 16 hr after fertilization , with the stained cells marking the indicated parasegments . Embryos between 5 hr and 13 hr old ( or between 4 hr and 10 hr for PS6 ) were harvested for analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 02833 . 00310 . 7554/eLife . 02833 . 004Figure 1—figure supplement 1 . Tools for parasegment-specific ChIP-seq . ( A ) A P element was constructed that initially expresses Gal4 from the P promoter , and transgenes were moved into the homeotic gene complexes by P element swapping . Recombination between the FRT sites flanking Gal4 convert the transgene into one expressing Gal80 . ( B ) A second P element expressed Gal4 or Gal80 , driven by enhancers from the BX-C . The enhancers were positioned upstream of a promoter and a PRE from the engrailed locus . The PRE helps to maintain the restricted expression by the enhancer in older embryos . A cluster of binding sites for the suppressor of Hairy-wing protein flanks the enhancer , to block potential position effects at random chromosomal locations . ( C ) Nuclei were sorted using Hoechst 33 , 342 ( to select only single nuclei ) and mCherry fluorescence . Nuclei from wild type ( Oregon R , OR ) embryos were used to set the fluorescence threshold . ( D ) The flow chart from embryos to sequencing libraries is diagramed; the procedure could be paused at either freezing step . DOI: http://dx . doi . org/10 . 7554/eLife . 02833 . 00410 . 7554/eLife . 02833 . 005Figure 1—figure supplement 2 . Close-up of parasegment-specific expression patterns . Embryos expressing the INTACT marker in parasegments 4–7 were obtained as in Figure 1B . All are at ∼10 hr of development , prior to dorsal closure; they were dissected along the dorsal midline , and flattened to display the epidermis in one focal plane . The last panel shows an embryo stained for UBX ( in blue ) and ABD-A ( in brown ) ( Karch et al . , 1990 ) , to illustrate the expression patterns of these two proteins in parasegments 4–7 . DOI: http://dx . doi . org/10 . 7554/eLife . 02833 . 005 For isolation of marked nuclei , we used Gal4 to drive an mCherry-RanGAP fusion protein , from the INTACT system ( Deal and Henikoff , 2011; Steiner et al . , 2012 ) . Staged embryos with single parasegment stripes of mCherry were formaldehyde-fixed and disrupted , and their nuclei were sorted on a FACS instrument ( Figure1—figure supplement 1C ) . DNA enriched by chromatin immunoprecipitation ( ChIP ) was prepared for paired-end sequencing , using a picogram-scale library protocol ( Figure 1—figure supplement 1D; Bowman et al . , 2013 ) . The largest differences in H3K27me3 enrichment between parasegments occurred at the BX-C ( Figure 2A , Figure 2—figure supplement 1 ) . While the H3K27me3 profiles in our four adjacent parasegments were very similar elsewhere in the genome ( Figure 2—figure supplement 1 ) , the BX-C exhibited a striking ‘stairstep’ pattern . Large expanses of H3K27me3 enrichment were lost as we moved from anterior to posterior along the body axis ( Figure 2C–F ) , revealing the locations of the regulatory domains . This is reminiscent of the temporal activation pattern of the Hox-D cluster in mouse , which correlates with loss of H3K27me3 ( Soshnikova and Duboule , 2009 ) . Our analysis of single parasegments locates distinct domain borders where K27me3 status changes abruptly . The contrast between the whole embryo profile ( Figure 2B ) and the single segment profiles ( Figure 2C–F ) shows how isolated cell types can identify chromatin changes linked to cell identity . 10 . 7554/eLife . 02833 . 006Figure 2 . H3K27me3 , CTCF , and CP190 profiles . ( A ) ChIP-seq profiles across 27 Mb of chromosome 3R are virtually identical , except at the BX-C , in the middle of the chromosome arm . ( B–F ) H3K27me3 profiles across 380 kb encompassing the BX-C . ( G–H ) CTCF binding sites . ( I and J ) CP190 binding sites . Panels B , G , and I were prepared from unsorted nuclei; parasegment-specific nuclei were used for the other panels , as indicated . Transcription units of coding genes are shown below the profiles . The Drosophila reference sequence includes a 6 . 1 kb Diver retroposon insertion at the indicated position; it was not present in the strains used for this analysis . At the bottom are shown the three regulatory domains defined by this analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 02833 . 00610 . 7554/eLife . 02833 . 007Figure 2—figure supplement 1 . Genome-wide comparisons of H3K27me3 patterns . The most striking differences in H3K27me3 enrichment across different parasegments occur at the BX-C . Each plot compares the density of H3K27me3 between two parasegments over all H3K27me3 peaks ( see supplementary methods for additional details ) . A linear line of fit was drawn . H3K27me3 density across the entire genome was largely unchanging between parasegments , with the exception of a few points corresponding to the BX-C that are markedly far off from the line of fit . Within each plot , the comparison of five BX-C segments ( labeled a–e ) are indicated with pink or red circles , the color of which corresponds to the Z-score ( number of observed standard deviations from the line of fit ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02833 . 00710 . 7554/eLife . 02833 . 008Figure 2—figure supplement 2 . H3K27me3 , CTCF , and CP190 profiles in PS7 for the Antennapedia complex , illustrated as in Figure 2 . The H3K27me3 profiles in parasegments 4 through 7 are very similar . ANTP is expressed at high levels in PS4 and PS5 in wild type embryos , and in the absence of the BX-C , parasegments 4–12 all show similar expression patterns ( Carroll et al . , 1986 ) . By this criterion , Antp appears to be free of Pc-G repression in PS4–12 , which is consistent with the lack of H3K27 methylation over its transcription unit in PS7 , as shown . CTCF and CP190 profiles look identical between mixed and PS7 nuclei , as they are in the BX-C . CP190 marks several potential domain boundaries in the Antennapedia complex that lack apparent CTCF binding . DOI: http://dx . doi . org/10 . 7554/eLife . 02833 . 008 In each isolated parasegment , the extent of K27me3 coverage was largely consistent with genetic studies indicating the location of BX-C active and repressed regulatory domains . The PS4 H3K27me3 pattern is mesa-like across the complex ( Figure 2C ) , with only a few narrow gaps at nucleosome-free regions ( Mito et al . , 2005 ) . The BX-C is almost completely repressed in PS4 , with only a few neuronal cells on the midline showing UBX expression in embryos . The PS5 regulatory domain , associated with the bithorax ( bx ) mutant lesions , is defined by the low level of H3K27me3 modification across the leftmost ( proximal ) 92 kb of the complex ( Figure 2D ) . There is a sharp transition to full H3K27me3 coverage at a position upstream of the Ubx transcription start site . The low residual level of H3K27me3 marks in the 92 kb PS5 domain is likely due to partial contamination from more anterior parasegments in our preparation of PS5 nuclei ( Figure 1—figure supplement 2 ) . In PS6 , the leftmost 137 kb of the BX-C is clear of the K27me3 mark ( Figure 2E ) ; the additional K27me3-clear region reveals the PS6 domain . The PS6 regulatory domain is the one best defined by genetic analysis , with a series of rearrangement breakpoints giving bithoraxoid ( bxd ) phenotypes ( Bender and Lucas , 2013 ) . The transition point to full K27 methylation coincides with the genetically defined Fub border between the PS6 and the PS7 regulatory domains ( Bender and Lucas , 2013 ) . In PS7 , an additional 36 kb was depleted of H3K27me3 ( Figure 2F ) . This region includes the abd-A transcription unit , and is associated with the infraabdominal-2 ( iab-2 ) class of genetic mutations . We observed a reproducible mound of H3K27me3 in PS7 nuclei close to the Ubx promoter ( arrow in Figure 2F ) . This may reflect repression of Ubx by abd-A in this parasegment ( Karch et al . , 1990 ) , mediated by the PcG . The regulatory domains defined by the sharp transitions in K27me3 enrichment do not correspond in every detail to expectations from genetic studies . Prior mapping of domains has relied upon random insertion of transgenic reporter elements into the BX-C to interrogate local enhancer activity . These enhancer traps show sharp anterior edges to their expression patterns , with the anterior-most labeled parasegment corresponding to the domain in which the insertion resides ( Bender and Hudson , 2000 ) . In the current study , the border between the PS5 and PS6 domains , defined by the discontinuity in K27me3 enrichment in PS5 nuclei , lies 14 kb upstream of the Ubx promoter . However , transgenic reporter elements inserted between the Ubx promoter and the K27me3-defined border ( presumably part of the PS5 regulatory domain ) show patterns with a PS6 anterior limit . Perhaps the promoters of these reporters interact with the Ubx promoter in distinctive ways . There was also prior evidence of an active BX-C domain in PS4 . An enhancer trap 3 kb downstream of the Ubx start site shows a PS4 anterior border ( Casares et al . , 1997 ) , and a ncRNA with a PS4 anterior expression limit is transcribed from the region 5–15 kb downstream of the Ubx start ( Pease et al . , 2013 ) . Despite the activity of these regulatory regions in PS4 cells , we did not detect a decrease in K27me3 at these locations . The PS4 active state may be early and transient , limited to a subset of PS4 cells , or resistant to H3K27 methylation . We have profiled additional histone modifications and several chromosomal proteins in single parasegments to understand better how domains are established and bounded . The sharp discontinuities in K27 methylation coincide with binding sites of the CCCTC-binding factor ( CTCF ) , which is thought to impose boundaries on regulatory domains in many systems ( Ohlsson et al . , 2010 ) . Figure 2G shows the CTCF ChIP pattern in unsorted nuclei , similar to that previously reported ( Nègre et al . , 2010 ) . CTCF in PS7 nuclei gives a pattern virtually identical to the whole embryo pattern ( Figure 2H ) . Apparently , the activation state of a domain is not coupled to the presence of CTCF at its borders . We note that our experimental conditions do not detect robust CTCF enrichment in some BX-C locations that have been reported by others ( Nègre et al . , 2010 ) , notably at the border between regulatory regions for PS7 and PS8 , and proximal to the 3′ end of the Ubx transcription unit . CTCF is present in a protein complex with Centrosomal Protein 190 ( CP190 ) ( Mohan et al . , 2007 ) , and the binding sites of these proteins often overlap ( Nègre et al . , 2010 ) . The CP190 sites largely coincide with CTCF sites in the BX-C , and , as with CTCF , there is little difference between the CP190 pattern in unsorted nuclei ( Figure 2I; Nègre et al . , 2010 ) and in PS7 nuclei ( Figure 2J ) . There is a noteworthy CP190 site immediately proximal to the 3′ end of the Ubx transcription unit , separating Ubx from the adjacent gene ( modular serine protease ) . In the Antennapedia complex , the other homeotic gene complex in flies , CP190 peaks often appear without coincident CTCF peaks ( Figure 2—figure supplement 2 ) . Acetylation of H3K27 ( H3K27ac ) is correlated with active enhancer regions ( Heintzman et al . , 2009; Kharchenko et al . , 2011 ) , and is anti-correlated with PcG function ( Tie et al . , 2009 ) , prompting us to examine its pattern in the BX-C . H3K27ac showed the opposite pattern to H3K27me3 . There is relatively little H3K27ac across the BX-C in unsorted nuclei , but in PS7 , the active domains show broad enrichment for H3K27ac ( Figure 3B , C ) . The H3K27ac levels differ in the three active domains , for reasons we do not yet understand . Within each domain , however , the coverage is boundary-to-boundary , although many enhancers in the BX-C are not active in embryos . Overall , H3K27ac is a mirror to H3K27me3 , in extent , if not in density . 10 . 7554/eLife . 02833 . 009Figure 3 . Additional features of the BX-C in PS7 . ( A ) The H3K27me3 distribution , repeated from Figure 2F , marks the border between active and repressed domains . ( B and C ) H3K27ac profiles are shown for mixed and PS7 nuclei; the acetylation mark in PS7 is largely restricted to the active domains . ( D and E ) Pol-II profiles show peaks at the promoters of Ubx , abd-A , and Abd-B both in whole embryo and in PS7 nuclei . In PS7 , polymerase is distributed across the transcription units of Ubx and abd-A , but not Abd-B . The arrow in D marks the PS5/PS6 border; the arrow in E marks the major Abd-B promoter . ( F and G ) H3K4me3 profiles show prominent peaks over the Ubx and abd-A promoters , both of which drive strong transcription in PS7 . ( H and I ) POLYCOMB profiles show sharp peaks over all known Polycomb Response Elements . In PS7 , there is a reduction of the PC peak at the ‘bx PRE’ in the active PS5 domain . There is also a PC peak at the abd-A promoter , which is somewhat reduced in PS7 , where Abd-A is transcribed . ( J and K ) POLYHOMEOTIC protein shows peaks at known PREs and the abd-A promoter in both the active and repressed domains . As with PC , the PH peak at the bx PRE is reduced in PS7 . ( L and M ) SU ( Z ) 12 profiles also mark PREs , but in PS7 , the PRE peaks are reduced in all three active domains . The ranges for the vertical axes are indicated above the left edge of each trace . DOI: http://dx . doi . org/10 . 7554/eLife . 02833 . 009 ChIP of serine-5-phosphorylated RNA polymerase II ( Pol-II ) ( transcriptionally engaged polymerase ) in whole embryos shows the transcription units for Ubx , abd-A , and Abd-B ( Figure 3D ) . Embryonic non-coding transcripts ( Bae et al . , 2002; Pease et al . , 2013 ) are not apparent from the Pol-II density , either because they are limited to very early stages of embryogenesis , or because the levels of transcription are too low . There is an unexpected Pol-II peak ∼14 . 5 kb upstream of the Ubx promoter , coincident with the CTCF binding site at the border between the PS5 and PS6 domains ( arrow in Figure 3D ) . There is no reported transcript initiating at this position ( Pease et al . , 2013 ) , although a short transcript could have been missed . Pol-II in PS7 ( Figure 3E ) marks the Ubx and abd-A transcription units , as expected from the UBX and ABD-A protein patterns ( Figure 1—figure supplement 2 ) . There is also a Pol-II peak at the major Abd-B promoter ( arrow in Figure 3E ) , although Abd-B RNA and protein are absent from PS7 ( Celniker et al . , 1990; Boulet et al . , 1991 ) and this region is H3K27 trimethylated . Analogous localization of Pol-II at the Abd-B promoter in wing and haltere discs was reported by Chopra et al . ( 2009 ) , who argued that these polymerase molecules were ‘paused’ . The H3K4me3 mark aligns with the Ubx and abd-A promoters in PS7 nuclei ( but not Abd-B; Figure 3G ) and is reduced in mixed cells ( Figure 3F ) ; this histone mark has been seen at active promoters in other systems ( Ruthenburg et al . , 2007 ) . The POLYCOMB subunit of PRC1 is known to bind to the H3K27me3 modification through its chromodomain ( Fischle et al . , 2003 ) , and so PRC1 localization might be expected to correspond to the stair-step H3K27me3 pattern . Published ChIP–CHIP and ChIP-seq profiles for PC across the BX-C vary from broad coverage matching the H3K27me3 profile ( Kwong et al . , 2008 ) to sharp peaks at the known PREs ( Gonzalez et al . , 2014 ) . The difference may be due to different procedures for chromatin fragmentation prior to immunoprecipitation ( Straub et al . , 2013 ) . Our PC profiles show sharp peaks at PREs in nuclei from whole embryos ( Figure 3H ) , matching the profile for POLYHOMEOTIC , another PRC1 subunit ( Figure 3J ) . In PS7 nuclei , we find both the POLYCOMB and POLYHOMEOTIC subunits of PRC1 are bound to the Polycomb Response Elements of the active PS6 and PS7 domains at levels comparable to those in the whole embryo profile ( Figure 3H–K ) . In the active PS5 domain , the levels of PC and PHO at the PS5 PRE are reduced , relative to the whole embryo profile . Papp and Müller ( 2006 ) have reported reduced retention of PRC1 and 2 components at the PRE in the PS5 domain in haltere and third leg imaginal discs , where the PS5 domain is in the active state and K27me3 is absent . Kwong et al . ( 2008 ) have made analogous observations regarding PRC retention in imaginal discs of the third thoracic segment . PRC1 presence is apparently not a determinant of repressed domains , although its subunits are clearly needed for PcG repression ( Simon and Kingston , 2013 ) . It will be important to map H2A ubiquitylation ( a function of the dRING subunit of PRC1 ) across the domains , especially given recent findings that H2A ubiquitylation promotes PRC2 recruitment and H3K27 trimethylation ( Blackledge et al . , 2014; Cooper et al . , 2014; Kalb et al . , 2014 ) . The profiles for SU ( Z ) 12 , a PRC2 component ( Figure 3L , M ) , look similar to those of the PRC1 proteins , POLYCOMB and POLYHOMEOTIC . There are major peaks at the known PREs in both whole embryo and PS7 nuclei , although in PS7 , the PREs in active domains have reduced levels of SU ( Z ) 12 . This result is again consistent with the prior work of Papp and Müller ( 2006 ) on the PS5 PRE in imaginal discs . SU ( Z ) 12 is also present at the transcription start sites of Ubx , abd-A , and Abd-B; binding at these promoters is more apparent than it is for PRC1 proteins .
The Polycomb Group repression system is often described as a cellular memory mechanism , which can impose lifelong silencing of a gene in response to a transitory signal . That view seems valid , but the concept of a PcG regulatory domain is much richer . In the PS6 domain of the BX-C , for example , there are many enhancers to drive Ubx expression in specific cells at specific developmental times , all of which are blocked in parasegments one through five , but active in parasegments 6 through 12 ( Bender and Lucas , 2013 ) . Individual enhancers need not include a segmental address that is specified , for example , by gap and pair-rule DNA-binding factors; their function is segmentally restricted by the domain architecture . Indeed , these enhancers will drive expression in a different parasegment when inserted into a different domain ( as in the Cbx transposition , Peifer et al . , 1987 ) . Each domain has a distinctive collection of enhancers; the UBX pattern in PS5 is quite different from that in PS6 . Thus , there are two developmental programs for Ubx , one in each of these parasegments , without the need for a duplication of the Ubx gene . Other loci with broad regions of H3K27 methylation may likewise be parsed into multiple domains , once we examine histone marks in specific cell types . The all-or-nothing H3K27me3 coverage of the BX-C parasegmental domains validates and refines the domain model . In particular , K27me3 is uniformly removed across the PS5 and PS7 domains in PS5 and PS7 , even though the activated genes in those parasegments ( Ubx and abd-A , respectively ) are only transcribed in a subset of cells . It is interesting that both PRC1 and PRC2 components have binding patterns that do not fully reflect function ( repression and K27 methylation , respectively ) , indicating the possibility that function of these complexes is regulated separately from binding ( Papp and Müller , 2006 ) . The challenges now are to understand how PcG regulated domains are established , differently in different parasegments , and to describe the molecular mechanisms , including changes in chromosome structure , that block gene activity in H3K27 trimethylated domains .
A P element was built containing both Gal4 and Gal80 ( Figure 1—figure supplement 1A ) . The P promoter initially drives Gal4 transcription . The P-element used the Gal4 gene with an HSP70 terminator derived from pGawB ( Brand and Perrimon , 1993 ) , fused in frame to the P transposase gene after the 130th amino acid . The Gal4 gene was followed by a synthetic FRT site plus a polylinker . The Gal80 coding sequence with an HSP70 terminator , derived from pBPG80Uw ( a gift of B Pfeiffer and G Rubin ) was inserted into this polylinker . Another synthetic FRT site was inserted near the start of the Gal4 coding region at a unique SphI site . Thus , the Gal4 coding sequence is flanked by FRT sites , so that flipase-induced recombination can turn the P element into a Gal80 producer , with the same expression pattern . The P element was transformed into a random chromosomal position , and then used to swap into the position of an existing P enhancer trap in the BX-C ( Bender and Hudson , 2000 ) or at the Antennapedia promoter . P element swapping ( Sepp and Auld , 1999 ) is a low frequency event , but the lines were constructed with UAS/GFP as a marker for Gal4 activity , and thousands of first instar larvae could be quickly screened for one with the expected segmentally-restricted GFP pattern . A derivative of this vector was made by replacing the HSP70 terminator downstream of the Gal80 coding sequence with the poly ( A ) addition site from the Drosophila Alcohol Dehydrogenase ( Adh ) gene , recovered as a 317 bp PCR fragment from genomic DNA . We have also taken known embryonic enhancers from the BX-C and inserted them into a second vector ( Figure 1—figure supplement 1B ) , which was derived from the initial vector with the Adh terminator . At a BamH1 site immediately upstream of the 5′ FRT site , a fragment was inserted containing: ( 1 ) an 841 bp fragment from the 5′ end of the gypsy mobile element ( bases 189-1029 ) , ( 2 ) an AscI site , and ( 3 ) a fragment from the engrailed gene ( −595 bp to +243 bp relative to the major transcription start site ) . The gypsy fragment includes twelve tandem binding sites for SUPPRESSOR OF HAIRY WING ( SU ( HW ) ) . The engrailed fragment includes the proximal PRE ( ‘PSE2’ [DeVido et al . , 2008] ) , and is designed to fuse the first nine amino acids of ENGRAILED to the amino terminal end of Gal4 . Enhancer fragments from the BX-C were cloned by PCR with AscI extensions on the primers , and inserted into the AscI site between the gypsy and engrailed fragments . The engrailed PRE was included to maintain the segmentally-restricted pattern through most of embryonic development , and the SU ( HW ) binding sequences were intended to insulate the promoter from endogenous enhancers near the insertion site . Transformants were recognized by Gal4-driven GFP patterns in young larvae , and , again , any Gal4 producer could be converted to a Gal80 producer . Sequences of these P elements are available on request . P element swapping was used to replace a P element at the Antennapedia P1 promoter ( P[XP] Antpd02480 , target site duplication: 3R: 2 , 825 , 198-2 , 825 , 205 [Thibault et al . , 2004] ) , and a P element in the PS8 ( iab-3 ) domain of the BX-C ( HCJ192 , target site duplication 3R: 12 , 673 , 279-12 , 673 , 286 [Bender and Hudson , 2000] ) . Other swaps were made with P elements in the PS5 ( bx ) and PS6 ( bxd ) domains , but these did not give sufficiently strong or uniform expression patterns . The Antp swap was subjected to P transposase to induce ‘local hopping’ and thereby increase the Gal4 expression level . Fragments used as enhancers inserted into the vector of Figure 1—figure supplement 1B included:abx enhancer ( for PS5 anterior limit ) 3R: 12 , 508 , 951-12 , 513 , 096pbx enhancer ( for PS6 anterior limit ) 3R: 12 , 598 , 546-12 , 600 , 177iab-2 enhancer ( for PS7 anterior limit ) 3R: 12 , 636 , 236-12 , 639 , 140iab-3 enhancer ( for PS8 anterior limit ) 3R: 12 , 664 , 301-12 , 666 , 886 Other fragments tested covered the ‘bx’ and ‘bxd’ enhancers , but these did not give sufficiently strong or uniform expression patterns . One insertion of a P element with the iab-3 enhancer fortuitously inserted 1 . 4 kb upstream of the Ubx transcription start site ( target site duplication , 3R: 12 , 561 , 577-12 , 561 , 584 ) . This gave a strong pattern with a sharp PS6 anterior limit , and was used as the Gal4 source for PS6 nuclei . Insertions with the abx and pbx enhancers were subjected to P transposase to adjust the levels of Gal4 and Gal80 . D . melanogaster genome coordinates are from release 5 . 57 . Flies with the INTACT reporter in a P element on the third chromosome ( w1118; p[w+; UASRG]6 ) were generously provided by Paul Talbert and Steve Henikoff . This P element was crossed to a P transposase source ( P{ry[+t7 . 2] = Delta2-3}99B ) , and offspring were screened initially for darker eye color , most likely due to multiple copies of the INTACT element . Such stocks with INTACT insertions on the second or third chromosome were then crossed to a Gal4 source , and those with brightest mCherry expression were used for subsequent efforts to mark single parasegments . The higher copy INTACT chromosomes are designated ‘INTACTn on II’ and ‘INTACTn on III’ . To mark PS4 , males homozygous for [w; abx en . >Gal80 , pbx en . >Gal80; Antp swap>Gal4* , INTACT on III] were crossed to females homozygous for [w; INTACTn on II; INTACTn on III] . To mark PS5 , males homozygous for [w; abx en . >Gal4* , pbx en . >Gal80* , UAS-GFP*] were crossed to females homozygous for [w; INTACTn on II; INTACTn on III] . To mark PS6 , males homozygous for [w; INTACTn on II; iab-2 en . >Gal80 , iab-3 swap>Gal80 , INTACT on III] were crossed to females [w; bxd insert>Gal4 , iab-2 en . >Gal80 , iab-3 en . >Gal80 , INTACT on III / TM6] . Only half of the resulting embryos had a Gal4-driven stripe . To mark PS7 , males homozygous for [w; iab-2 en . >Gal4 , iab-3 swap>Gal80 , INTACT on III] were crossed to females homozygous for [w; INTACTn on II; INTACTn on III] . Insertions treated with P transposase are marked above with an asterisk ( * ) . Agar plates with yeast paste were put on the PS4 , PS5 or PS7 laying cages for 8 hr at 25°C ( or 16 hr at 18°C ) , and then the eggs were aged for another 5 hr at 25°C ( or 10 hr at 18°C ) . For PS6 , a 6 hr collection was aged for a further 4 hr . The eggs were dechorionated ( 2 min in 50% bleach ) and fixed in fixation buffer ( 5% formaldehyde , 100 mM NaCl , 50 mM Hepes , 1 mM EDTA , 0 . 5 mM EGTA , pH 8 . 0 ) under heptane , on a rotator for 15 min at 25°C . The fixed embryos were washed with stop buffer ( 125 mM glycine , 130 mM NaCl , 7 mM Na2HPO4 , 3 mM KH2PO4 , 0 . 1% Triton X100 , pH 8 . 0 ) for 2 min at 25°C , and then briefly with PBS ( 130 mM NaCl , 7 mM Na2HPO4 , 3 mM KH2PO4 ) plus 0 . 1% Triton X100 . The washed embryos were collected on a nytex filter , and examined under an epifluorescence stereoscope . Embryos that appeared too old , with fluorescent nuclei anterior to the desired parasegment , were manually removed . The remaining eggs were collected and weighed in a 1 . 5 ml microfuge tube , flash frozen in liquid nitrogen , and stored at −75°C . For the isolation of nuclei , ∼100 mg of embryos were suspended in 5 ml of BBT buffer ( 55 mM NaCl , 40 mM KCl , 15 mM MgSO4 , 5 mM CaCl2 , 10 mM Tricine , 20 mM dextrose , 50 mM sucrose , 0 . 1% bovine serum albumin ( Miles PENTEX ) , 0 . 01% Triton X-100 , pH7 . 0 ) plus 100 µl protease inhibitor cocktail ( Roche Complete , EDTA-free ) , and disrupted with 10 strokes in a Dounce homogenizer with a loose pestle , then 10 strokes with a tight pestle . The homogenate was spun at 275 × g for 1 min to remove large debris . The resulting supernatant was spun at 1000 × g for 10 min to pellet the nuclei . The pellet was resuspended in 2 ml BBT buffer plus 40 µl protease inhibitor , and homogenized again , 20 strokes with the tight pestle . The homogenate was passed through a cell strainer cap with a 40 µm nylon mesh and stored at 0°C . Hoechst 33 , 342 dye was added ( 2 µl of 10 mg/ml solution ) before sorting . Nuclei were sorted on a BD Biosciences ( San Jose , CA ) FACSAria IIu instrument . The fluorescence from Hoechst 33342 ( 405 nM excitation , 425–475 nm emission ) was used to select for a 2N DNA content , thus avoiding clumps of nuclei . These single nuclei were secondarily selected for mCherry fluorescence ( 594 nM excitation , 619–640 nM emission ) stronger than 99 . 9% of nuclei from control ( Oregon R ) embryos . Recovered nuclei constituted between 1 and 6% of total single nuclei ( Figure 1—figure supplement 1C ) , depending of the parasegment under selection . Recovered nuclei were kept at 0°C , and used for chromatin preparation within 5 hr . The yield of sorted nuclei in sort buffer ( PBS ) was determined with a hemacytometer , and 50–200 K nuclei were used for each ChIP . Chromatin fragmentation was accomplished either with micrococcal nuclease ( MNase ) or bath sonication . For MNase fragmentation , nuclei in sort buffer were supplemented with PBS containing 0 . 1% Triton-X100 ( PBS-Tx ) to a volume of 400 µl , and CaCl2 to 1 mM . Nuclei were pre-warmed to 37°C and digested with 12U MNase ( Worthington Biochemical ) for 3 min . Digestion was stopped by moving the tubes to ice and adding 10 µl of 250 mM EDTA , 250 mM EGTA , and nuclei were briefly sonicated in a Diagenode Bioruptor ( Denville , NJ ) ( 3 min , high intensity , 30 s on , 30 s off ) . For bath sonication , the nuclear suspension in sort buffer was adjusted to ChIP buffer conditions ( 10 mM Tris pH8 , 100 mM NaCl , 0 . 1% sodium deoxycholate , 0 . 5% sarkosyl , 1% Triton-X100 ) in a volume of 500 µl . Nuclei were sonicated for 30 min at high intensity ( 30 s on , 30 s off ) in a Bioruptor . Fragmented chromatin was snap frozen and stored at −80°C prior to ChIP . MNase fragmented chromatin was adjusted to ChIP buffer conditions in a final volume of 500 µl , and incubated on ice for 5 min . Prior to ChIP , protease inhibitors ( COMPLETE , Roche ) were added to either MNase fragmented chromatin or sonicated chromatin . Chromatin was subjected to a high speed spin at 4°C for 10 min . Supernatant was moved to a new low retention tube , 1% input was removed and stored at 4°C , and and the remaining chromatin was incubated overnight with the appropriate antibody at 4°C . After a high speed spin at 4°C for 10 min , supernatant was moved to a low retention tube with 10 µl prewashed protein A dynabeads ( Life Technologies ) , and incubated at 4°C for 1–2 hr with gentle rotation . After six rinses with ice cold ChIP buffer , immunoprecipitates were eluted from the beads with two successive additions of 125 µl freshly made elution buffer ( 0 . 2% SDS , 0 . 1 M NaHCO3 , 5 mM DTT ) incubated at 65°C for 10 min . Eluates were combined , and adjusted with Tris and EDTA to a final concentration of 10 mM Tris pH8 and 2 mM EDTA . 250 µl elution buffer was added to the reserved input and likewise adjusted . DNA cleanup began by incubating with RNase ( DNase-free , Roche ) at 37°C for 30 min , proteinase K ( PCR-grade , Roche ) at 55°C for 1 hr , and crosslinks were reversed by incubating at 65°C for 1 hr . After two phenol-chloroform extractions and one chloroform extraction , purified DNA was ethanol precipitated in the presence of sodium acetate and Glycoblue ( Ambion ) . Sequencing library construction was performed according to ( Bowman et al . , 2013 ) . Libraries were sequenced on an Illumina ( San Diego , CA ) HiSeq2000 , HiSeq2500 , or MiSeq according to manufacturer's instructions . anti-K27me3 ( Active Motif 39155 ) , anti-K27ac ( Active Motif 39 , 136 ) , anti-K4me3 ( 07473; Millipore ) , anti-CTCF ( Smith et al . , 2009 ) , anti-Ph ( residues 772–984 [Oktaba et al . , 2008] ) , anti-CP190 ( Pai et al . , 2004 ) , anti-Pol-II ( ab5131; Abcam ) , anti-Pc ( Schuettengruber et al . , 2009 ) , anti-SU ( Z ) 12 ( Müller et al . , 2002 ) . ChIP-seq reads were aligned to the dm3 genome using BWA and allowing up to one mismatch . Potential PCR duplicates , tags with multiple alignments , and paired-end reads with insert size greater than 1 kb were removed from further analysis . We calculated normalized positional coverage as previously described ( Pinter et al . , 2012; Yildirim et al . , 2012 ) . Briefly , experimental coverage was normalized by corresponding input coverage for each position: nnorm = [ ( n+1 ) / ( ni+1 ) ] * [Ni/N] , where n , ni , N , and Ni are positional coverages in experiment and input , and total genome coverages in experiment and input , respectively . To determine regions of enrichment , we summed the coverage within sliding windows of size 1 kb and step 50 bp across the entire genome . The same was calculated for tags assigned a random location within the chromosome . The resulting distribution of random window coverages was then used to determine the significance of observed window coverages . Individual p-value cutoffs were selected based on manual inspection of each experiment . For samples with replicates , only regions significant in both replicates were used . For genome-wide analysis of coverage difference between two parasegments for H3K27me3 , we first merged the significant regions of enrichment from the different parasegments to create an aggregate set of regions of interest . Regions greater than 50 kb were split into equally sized sub-regions . For each region we calculated the density , or total normalized coverage divided by region length . Replicate density values were averaged . To find differences , region densities for two parasegments were scatter plotted against one another , and a line of best fit calculated . Points located furthest away from this fit line were identified as the candidate differing regions . This approach allows us to account for the variation in sequencing coverage and noise for each sample and parasegment . SPP ( Kharchenko et al . , 2008 ) was used to generate tag density profiles for visualization . For each chip ChIP experiment , we followed standard methods to estimate the binding peak separation distance , remove low quality or anomalous tags , and generate an input-subtracted , Gaussian-smoothed tag density profile . | Like other insects , the body of the fruit fly is divided into three main parts—the head , the thorax and the abdomen—and each part , in turn , is made up of several smaller segments . The bithorax complex is a cluster of three genes that together control the identity of the segments that make up the back half of the fruit fly's body . This gene cluster has been studied for several decades and these studies have helped to further our understanding of how genetic information is accessed and used to make an animal’s body plan . Early on in a fruit fly embryo , stretches of DNA within the bithorax complex regulate where the complex's genes are switched on , and where they are switched off . Proteins called Polycomb group proteins then keep the silenced genes off , in part by adding small chemical marks to other proteins called histones . Most DNA in a cell is wrapped around histones , and the addition of such chemical marks causes the DNA to become more tightly packed . This prevents the bithorax complex genes from being accessed and switched on . It had previously been suggested that each segment might have a unique pattern of chemical marks on the bithorax complex histones , but evidence to support this idea was lacking . Bowman et al . have now undertaken the technically challenging task of purifying the DNA and its histones from individual segments of fruit fly embryos . This revealed that segments closer to the embryo's head contain larger stretches of bithorax complex DNA covered with histones marked by the Polycomb group proteins . Bowman et al . also found that the coverage of chemical marks on the histones changed dramatically when one segment was compared to its neighboring segments . These sharp boundaries clearly outline which regulatory regions of the DNA are switched on and which are switch off; however the same pattern is not seen for the Polycomb group proteins themselves . Instead , within the bithorax complex , the pattern of these proteins is almost identical in different segments . The challenge now is to understand how the chemical marks and the Polycomb group proteins work together to restrict access to DNA in such precise patterns . Also—since similar gene clusters control the development of the body plans of mammals—this , in turn , might help us to understand how the Polycomb group proteins perform similar functions in human development and disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"developmental",
"biology"
] | 2014 | H3K27 modifications define segmental regulatory domains in the Drosophila bithorax complex |
Adaptations of the lower back to bipedalism are frequently discussed but infrequently demonstrated in early fossil hominins . Newly discovered lumbar vertebrae contribute to a near-complete lower back of Malapa Hominin 2 ( MH2 ) , offering additional insights into posture and locomotion in Australopithecus sediba . We show that MH2 possessed a lower back consistent with lumbar lordosis and other adaptations to bipedalism , including an increase in the width of intervertebral articular facets from the upper to lower lumbar column ( ‘pyramidal configuration’ ) . These results contrast with some recent work on lordosis in fossil hominins , where MH2 was argued to demonstrate no appreciable lordosis ( ‘hypolordosis’ ) similar to Neandertals . Our three-dimensional geometric morphometric ( 3D GM ) analyses show that MH2’s nearly complete middle lumbar vertebra is human-like in overall shape but its vertebral body is somewhat intermediate in shape between modern humans and great apes . Additionally , it bears long , cranially and ventrally oriented costal ( transverse ) processes , implying powerful trunk musculature . We interpret this combination of features to indicate that A . sediba used its lower back in both bipedal and arboreal positional behaviors , as previously suggested based on multiple lines of evidence from other parts of the skeleton and reconstructed paleobiology of A . sediba .
Bipedal locomotion is thought to be one of the earliest and most extensive adaptations in the hominin lineage , potentially evolving initially 6–7 million years ( Ma ) ago . Human-like bipedalism evolved gradually , however , and early hominins appear to have been facultative bipeds on the ground and competent climbers in the trees ( Senut et al . , 2001; White et al . , 2015; Prang , 2019; Prang et al . , 2021 ) . How long climbing adaptations persisted in hominins and when adaptations to obligate terrestrial bipedalism evolved are major outstanding questions in paleoanthropology . Australopithecus sediba – an early Pleistocene ( ~2 Ma ) hominin from the site of Malapa , Gauteng province , South Africa – has featured prominently in these discussions , as well as those concerning the origins of the genus Homo ( Berger et al . , 2010; Berger , 2012; Irish et al . , 2013; Dembo et al . , 2015; Kimbel and Rak , 2017; De Ruiter et al . , 2018; Williams et al . , 2018a; Du and Alemseged , 2019 ) . Previous studies support the hypothesis that A . sediba possessed adaptations to arboreal locomotion and lacked traits reflecting a form of obligate terrestriality observed in later hominins ( Schmid et al . , 2013; Prang , 2015a; Prang , 2015b; Prang , 2016; Holliday et al . , 2018 ) . Malapa Hominin 2 ( MH2 ) metacarpals are characterized by trabecular density most similar to orangutans , which suggests power grasping capabilities ( Dunmore et al . , 2020 ) , and the MH2 ulna was estimated to reflect a high proportion of forelimb suspension in the locomotor repertoire of A . sediba ( Rein et al . , 2017 ) . Evidence from the lower limb also suggests that A . sediba lacked a robust calcaneal tuber ( Prang , 2015a ) and a longitudinal arch ( Prang , 2015b ) , both thought to be adaptations to obligate , human-like bipedalism , and demonstrates evidence for a mobile subtalar joint proposed to be adaptively significant for vertical climbing and other arboreal locomotor behaviors ( Prang , 2016; DeSilva et al . , 2013; Zipfel et al . , 2011; DeSilva et al . , 2018 ) . The upper thorax ( Schmid et al . , 2013 ) , scapula ( Churchill et al . , 2013; Churchill et al . , 2018 ) , and cervical vertebrae ( Meyer et al . , 2017 ) of A . sediba suggest shoulder and arm elevation indicative of arboreal positional behaviors requiring overhead arm positions , and the limb joint size proportions are ape-like ( Prabhat et al . , 2021 ) . Furthermore , analysis of dental calculus from Malapa Hominin 1 ( MH1 ) indicates that this individual’s diet was high in C3 plants like fruit and leaves , similar to savannah chimpanzees and Ardipithecus ramidus ( Henry et al . , 2012 ) . Despite the presence of climbing adaptations , A . sediba also demonstrates clear evidence for bipedal locomotion . The knee and ankle possess human-like adaptations to bipedalism , demonstrating a valgus angle of the femur and a human-like ankle joint ( Zipfel et al . , 2011; DeSilva et al . , 2013; DeSilva et al . , 2018 ) . Evidence for strong dorsal ( lordotic ) wedging of the two lower lumbar vertebrae suggests the presence of a lordotic ( ventrally convex ) lower back ( Williams et al . , 2013; Williams et al . , 2018b ) . However , the initial recovery of just the last two lumbar vertebrae of MH2 limited interpretations of spinal curvature , and a study of the MH2 pelvis reconstruction ( Kibii et al . , 2011 ) suggests that A . sediba was characterized by a small lordosis angle estimated from calculated pelvic incidence ( Been et al . , 2014 ) . A separate pelvis reconstruction of MH2 produces a pelvic incidence angle more in line with other hominins ( Tardieu et al . , 2017 ) . The presence of a long , mobile lower back and a Homo-like lower thorax morphology indicating the presence of a waist further suggest bipedal adaptations in A . sediba ( Schmid et al . , 2013; Williams et al . , 2013 ) . However , missing and incomplete lumbar vertebrae prevented comparative analysis of overall lower back morphology and allowed only limited interpretations of A . sediba back posture and implications for positional behavior . Here , we report the discovery of portions of four lumbar vertebrae from two ex situ breccia blocks that were excavated from an early 20th century mining road and dump at Malapa . The former mining road is represented by a trackway located in the northern section of the site approximately 2 m north of the main pit that yielded the original A . sediba finds ( Dirks et al . , 2010; Figure 1 ) . The trackway traverses the site in an east-west direction and was constructed using breccia and soil removed from the main pit by the historic limestone miners . Specimens U . W . 88–232 , −233 , –234 , and –281 were recovered in 2015 from the upper section of layer 2 ( at a depth of 10 cm ) and formed part of the foundation layer of the mining road . The trackway can be distinguished from the surrounding deposits by a section of compacted soil ( comprising quartz , cherts , and flowstone ) and breccia that extends between layers 1 and 2 . Breccia recovered from the trackway , including the block containing U . W . 88–232 , −233 , –234 , and –281 similarly presented with quantities of embedded quartz fragments and grains . The breccia block containing specimen U . W . 88–280 , along with U . W . 88–43 , –44 , and –114 ( Williams et al . , 2013; Williams et al . , 2018b ) , were recovered from the miner’s dump comprised of excess material ( soil and breccia ) used for the construction of the miner’s road . The composition of the road matrix and associated breccia , as well as the breccia initially recovered from the mine dump , corresponds to the facies D and E identified in the main pit ( Dirks et al . , 2010 ) . Facies D includes a fossil-rich breccia deposit that contained the fossil material associated with MH2 ( Dirks et al . , 2010; Val et al . , 2018 ) . Therefore , the geological evidence suggests that the material used for the construction of the miner’s road was sourced on-site , and most probably originated from the northern section of the main pit . The newly discovered vertebrae ( second and third lumbar ) are preserved in articulation with each other ( Figure 2 , Figure 2—figure supplement 1 ) and refit at multiple contacts with the previously known penultimate ( fourth ) lumbar vertebra ( Figure 3 ) . Together , the new and previously known ( Williams et al . , 2013; Williams et al . , 2018b ) vertebral elements form a continuous series from the antepenultimate thoracic vertebra through the fifth sacral element , with only the first lumbar vertebra missing major components of morphology ( Figure 3—figure supplement 1 ) . The presence of a nearly complete lower back of MH2 allows us to more comprehensively evaluate the functional morphology and evolution of purported adaptations to bipedalism in A . sediba and test the hypotheses that the following fundamental features are similar to modern humans ( Homo sapiens ) and distinct from extant great apes: ( 1 ) lumbar lordosis , ( 2 ) progressive widening of the articular facets and laminae ( pyramidal configuration ) of the lower back , and ( 3 ) overall middle lumbar vertebra shape . Specifically , for these hypotheses , we predict that measurements of combined lumbar wedging ( representing degree of lordosis ascertained from available lumbar vertebrae ) will fall within the human range ( H1 ) , that the configuration of the articular facets and laminae will progressively widen caudally ( rather than remaining constant or becoming increasingly narrow ) as seen in modern humans ( H2 ) , and that the most complete lumbar vertebra of MH2 ( U . W . 88–233 ) will fall within the human range of variation in shape analyses ( H3 ) .
We determine the seriation of the vertebrae described here based on their direct articulation with one another and refits with previously known vertebrae . Most of the sacrum ( U . W . 88–137 ) is preserved in articulation with the neural arch of the last lumbar vertebra ( U . W . 88–138 ) , which articulates in turn with the inferior portion of the neural arch of the penultimate lumbar vertebra ( U . W . 88–154 ) . Corresponding vertebral bodies ( U . W . 88–126 and U . W . 88–127 , respectively ) are preserved together and can be refitted with the neural arches ( Williams et al . , 2013 ) . The new lumbar vertebrae are preserved in partial articulation , including an upper neural arch that refits in two places with U . W . 88–154 . Therefore , portions of five vertebrae are preserved , followed by a sacrum and preceded by at least three lower thoracic vertebrae ( Williams et al . , 2018b ) . U . W . 88–280: This is a partial , superior portion of a vertebral body concealed in the matrix of a previously known block containing lower thoracic vertebrae ( U . W . 88–114 , U . W . 88–43 , and U . W . 88–44 , antepenultimate , penultimate , and ultimate thoracic vertebrae , respectively , of MH2 ) ( Williams et al . , 2018b ) . U . W . 88–280 was revealed in the segmentation of micro-CT ( hereafter , µCT ) data . U . W . 88–280 represents the right side of an upper vertebral body with preservation approaching the sagittal midline . The preserved portions measure 16 . 5 mm dorsoventrally and 14 . 0 mm mediolaterally at their maximum lengths . The lateral portion of the vertebral body is only preserved ~5 . 0 mm inferiorly from the superior surface , but there is no indication of a costal facet on the preserved portion . We identify this as part ( along with U . W . 88–281 ) of the first lumbar vertebra of MH2 based on its position below the vertebral body of what is almost certainly the last thoracic vertebra ( U . W . 88–44 ) ( Williams et al . , 2018b; Figure 2—figure supplement 1 ) . U . W . 88–281: This is the partial neural arch of a post-transitional , upper lumbar vertebra concealed in matrix above the subjacent lumbar vertebra ( U . W . 88–232 ) . It was revealed through the segmentation of µCT data . It consists of the base and caudal portion of the spinous process and parts of the inferior articular processes . The remainder of the vertebra is sheared off and unaccounted for in the block containing the new lumbar vertebrae . U . W . 88–281 is fixed in partial articulation with the subjacent second lumbar vertebra ( L2 ) , U . W . 88–232 . Therefore , we identify U . W . 88–281 as part of the first lumbar vertebra based on its morphology and position within the block . The left inferior articular facet ( IAF ) is more complete than the right , with approximately 6 . 0 mm of its superior-inferior ( SI ) height preserved , and is complete mediolaterally , measuring ~8 . 0 mm in width . The minimum distance between the IAF is 12 . 5 mm , and the maximum preserved distance between them is 21 . 75 mm . The preserved portion of the spinous process is 12 . 75 mm in dorsoventral length . U . W . 88–232: This vertebra is the L2 and remains in articulation with the third lumbar vertebra ( L3 ) , U . W . 88–233 , held together with matrix . Some portions of U . W . 88–232 are covered by adhering matrix or other fossil elements ( U . W . 88–281 and U . W . 88–282 , the latter being the sternal end of a clavicle ) , so µCT data were used to visualize the whole vertebra ( Figure 4 ) . U . W . 88–232 is mostly complete , missing the cranial portions of its superior articular processes and distal portions of its costal ( transverse ) processes . It is distorted due to crushing dorsally from the right side and related breakage and slight displacements of the left superior articular process at the pars interarticularis and the right costal process at its base . Although broken at its base and displaced slightly ventrally , the right costal process is more complete than the left side , which is broken and missing ~10 . 0 mm from its base . Because of crushing , the neural arch is displaced toward the left side , and the vertebral foramen is significantly distorted . A partial mammillary process is present on the left superior articular process , sheared off along with the remainder of the right superior articular process ~8 . 0 mm from its base . The left side is similar but much of the mammillary process is sheared off in the same plane as the right side , leaving only its base on the lateral aspect of the right superior articular process . The vertebral body is complete and undistorted , and the spinous process and inferior articular processes are likewise complete but affected by distortion . Standard measurements of undistorted morphologies are reported in Table 1 . U . W . 88–233: This is the L3 and the most complete vertebra in the lumbar series , although some aspects of the neural arch are distorted , broken , and displaced . It is held in matrix and partial articulation with U . W . 88–234 , the subjacent partial fourth lumbar vertebra ( L4 ) . Due to its position between articulated elements U . W . 88–232 and –234 and some adhering matrix , U . W . 88–233 was visualized using µCT data . U . W . 88–233 is essentially complete; however , like U . W . 88–232 , the neural arch is crushed from the dorsal direction , with breaks and displacement across the right pars interarticularis and the right costal process at its base , with additional buckling around the latter near the base of the of the right superior articular process , resulting in a crushing of the vertebral foramen . The vertebral body , pedicles , spinous process , and superior and inferior articular processes are complete , as are the lamina and costal processes aside from the aforementioned breakage . The left costal process is unaffected by taphonomic distortion . Standard measurements of undistorted morphologies are reported in Table 1 . U . W . 88–234: This is a partial neural arch of the previously known penultimate lumbar vertebra ( L4 ) ( U . W . 88-127/153 ) . U . W . 88–234 refits in two places with the previously known L4: its partial pedicle with the vertebral body ( U . W . 88–127 ) and its spinous process with the inferior base of the spinous process and inferior articular processes ( U . W . 88–153 ) ( Figures 2–3 ) . Only the spinous process and right pedicle , costal process , superior articular processes , and partial lamina are present and in articulation with U . W . 88–233 . Matrix adheres to the spinous process and costal process , so for this element µCT data were used to visualize and virtually refit it with U . W . 88-127/153 , forming a partial L4 missing the left superior articular process , costal process , most of the pedicle , the right lateral aspect of the inferior articular process , a portion of the lamina , the inferior aspect of the costal process , and a wedge-shaped area of the lateral body-pedicle border . Preserved standard measurements are reported in Table 1 . Wedging of articulated vertebrae contribute to the multiple sagittal curvatures of the human spine , with dorsal wedging of lower lumbar vertebrae contributing to a ventrally convex curvature of the lumbar spine ( lumbar lordosis ) . This sinusoidal configuration passively balances the upper body over the pelvis and allows for the unique system of weight bearing and force transmission found in members of the human lineage ( Davis , 1961; Robinson , 1972; Pal and Routal , 1987; Latimer and Ward , 1993; Shapiro , 1993a; Lovejoy , 2005; Whitcome et al . , 2007; Masharawi et al . , 2010; Been et al . , 2014; Tardieu et al . , 2017 ) . Wedging angles for individual lumbar vertebrae ( L2-L5 ) and combined L2-L5 wedging were calculated for A . sediba and the comparative sample and are presented in ( Figure 5 , Figure 5—figure supplement 1 , Figure 5—figure supplement 2 ) and Table 2 and Table 3 . MH2 possesses the greatest ( i . e . , most negative ) combined wedging value of any adult early hominin ( –6 . 8° ) . Although all fossil hominins fall within the 95% prediction intervals of modern humans , only MH2 falls outside the 95% prediction intervals of great apes in combined L2-L5 wedging ( Figure 5 ) . Patterns of change across lumbar levels demonstrate that MH2’s vertebrae transition from ventral ( kyphotic ) to dorsal ( lordotic ) wedging between the L3 and L4 levels; however , all adult fossil hominins fall within the 95% prediction intervals of modern humans ( Figure 5 , Figure 5—figure supplement 1 ) . As shown previously ( Williams et al . , 2013 ) , the last lumbar vertebra of MH2 is strongly dorsally wedged like that of the Kebara 2 Neandertal and the juvenile specimen KNM-WT 15000 , whereas other fossil hominins do not demonstrate this pattern . Although vertebral wedging is characterized by high levels of variation within groups , especially in combined L2-L5 wedging ( Figure 5 , Table 2 ) , the pattern of lumbar wedging angles observed in MH2 ( i . e . , transition from penultimate to ultimate lumbar level ) and its combined L2-L5 wedging fall within the modern human 95% PIs and outside those of great apes ( Figure 5 , Figure 5—figure supplement 1 , Figure 5—figure supplement 2 ) . The hypothesis that A . sediba is human-like in lumbar wedging , therefore , cannot be rejected . The recovery of new lumbar vertebrae of MH2 allows for the quantification and comparison of inter-articular facet width increase in A . sediba . Humans are characterized by a pyramidal configuration of the articular facets such that they increase in transverse width progressively down the lumbar column ( i . e . , from cranial to caudal ) ( Latimer and Ward , 1993; Ward and Latimer , 2005 ) . Using an index of the last lumbar-sacrum inter-articular maximum distance relative to that of lumbar vertebrae three levels higher ( L2-L3 in hominins , L1-L2 in chimpanzees and gorillas ) , we show that Australopithecus africanus ( Sts 14 and StW 431; average = 1 . 42 ) and A . sediba ( 1 . 43 ) fall at the low end of the range of modern human variation in this trait ( Figure 6 ) . We note that A . L . 288–1 ( Australopithecus afarensis ) falls at the low end of human variation near other australopiths if the preserved lumbar vertebra ( A . L . 288-1aa/ak/al ) is treated as an L3 ( Latimer and Ward , 1993; Lovejoy , 2005; Johanson et al . , 1982; Meyer et al . , 2015 ) , but outside the range of human variation and within that of orangutans if it is treated as an L2 ( Cook et al . , 1983 ) . Homo erectus and Neandertals fall well within the range of modern human variation . The presence of a pyramidal configuration of the lumbar articular facets is therefore present in MH2 , supporting our hypothesis that A . sediba was adapted to a human-like configuration of the neural arch . The new middle lumbar vertebra , U . W . 88–233 , is complete , and although the neural arch is compressed ventrally into the vertebral foramen space , it can be reasonably reconstructed from µCT data ( see Materials and methods ) . We used three-dimensional geometric morphometrics ( 3D GM ) to evaluate the shape affinities of U . W . 88–233 among humans , great apes , and fossil hominins . The results of our principal components analysis ( PCA ) on Procrustes-aligned shape coordinates reveal that A . sediba falls within or near the human distribution on the first three principal components ( PC1–3 ) ( Figure 7 ) . PC1 explains 31% of the variance in the dataset , and along it hominins are characterized by more sagittally oriented and concave superior articular facets ( SAF ) , more dorsally oriented costal processes , a dorsoventrally shorter and cranially oriented spinous process , craniocaudally shorter , dorsoventrally longer vertebral body , and more caudally positioned SAF and IAF relative to the vertebral body compared to great apes . PC2 explains 13% of the variance and contrasts long spinous processes and relatively neutrally wedged ( ~0° ± 1° ) vertebral bodies of hominins and African apes with the shorter spinous processes and strongly ventrally wedged vertebral bodies of orangutans . PC3 explains 8% of the variance and largely contrasts dorsoventrally longer vertebral bodies with caudally oriented spinous processes in gorillas with dorsoventrally shorter vertebral bodies and less caudally oriented spinous processes in chimpanzees and orangutans; hominins fall intermediate between these groups . PC4 explains 5% of the variance , and contrasts A . sediba and A . africanus with both humans and great apes . Sts 14 and especially U . W . 88–233 are characterized by longer , taller , more cranially oriented costal processes that do not taper distally and more sagittally oriented ( as opposed to more coronally oriented ) articular facets ( Figure 7 , Figure 7—figure supplement 1 ) . We removed great apes and reran the PCA to ensure that their presence is not affecting the relationship of fossil hominins to modern humans . This hominin-only PCA essentially reproduced the results of PC4 on PC2 ( Figure 7—figure supplement 2 ) , confirming that Sts 14 and U . W . 88–233 are distinct from modern humans in costal process morphology . Therefore , although the A . sediba middle lumbar vertebra is somewhat human-like in overall shape , its vertebral body is intermediate in shape between great apes and modern humans and its costal processes are robust , cranially oriented , and cranially positioned on the pedicles . To provide a more in-depth comparison of the morphometric affinities of U . W . 88–233 , we plot Procrustes distances between U . W . 88–233 and Sts 14 , Shanidar 3 , the modern human sample , and the chimpanzee sample ( Figure 7—figure supplement 1A ) . We also show pairwise comparisons of Procrustes distances for middle lumbar vertebra shape within both human and chimpanzee samples , and between human and chimpanzee samples ( Figure 7—figure supplement 1C ) . These analyses demonstrate that U . W . 88–233 is most similar to Sts 14 among fossil and extant specimens included . To include other fossil hominins with broken processes , we ran a second 3D GM analysis excluding the majority of costal process and spinous process landmarks . This analysis , which includes landmarks on the vertebral body , SAF and IAF , and the bases of the costal and spinous processes , produces a similar pattern compared to the analysis on the full landmark set ( Figure 7 ) . Humans and great apes separate along PC1 , which is largely explained by vertebral body heights ( including vertebral wedging ) and SI position of the articular facets relative to the vertebral body . U . W . 88–233 , like other early hominins included in this study , falls intermediately between modern humans and great apes along PC1 . We used a Procrustes distance-based analysis of variance ( ANOVA ) to evaluate the effect of centroid size on lumbar shape ( Goodall , 1991 ) . The results show significant effects of centroid size ( F = 9 . 83; p < 0 . 001 ) , genus ( with hominins pooled; F = 27 . 7; p < 0 . 001 ) , and an interaction between genus and centroid size ( F = 1 . 48; p = 0 . 01 ) , implying unique shape allometries within genera ( Table 4 ) . We plotted standardized shape scores derived from a multivariate regression of shape on centroid size against centroid size to visualize shape changes ( Drake and Klingenberg , 2008; Figure 7—figure supplement 3 ) . In general , larger centroid sizes are associated with 3D shape changes including dorsoventrally longer and more caudally projecting spinous processes , more cranially oriented and less sagittally oriented costal processes , and less caudally projecting IAF . Importantly , however , the cranially oriented costal processes of U . W . 88–233 ( and Sts 14 ) appear not to be explained by centroid size given its relatively small size and overlap with Pan in standardized shape scores ( Figure 7—figure supplement 3 ) .
The recovery of two new lumbar vertebrae and portions of other lumbar vertebrae of the adult female A . sediba ( MH2 ) , together with previously known vertebrae , form a nearly complete lumbar column ( Figure 3 , Figure 3—figure supplement 1 ) and allows us to test hypotheses based on more limited material . As we outline below , A . sediba demonstrates evidence for lumbar lordosis in the combined pattern of bony wedging of lumbar vertebral bodies , as well as progressive widening of neural arch structures moving caudally ( ‘pyramidal configuration’ ) of lumbar vertebrae and the sacrum , which does not allow us to reject the hypotheses that A . sediba has human-like adaptations to bipedalism . However , the hypothesis that A . sediba’s middle lumbar vertebra ( L3 ) is human-like is not fully supported: although U . W . 88–233 is somewhat human-like in overall shape , its costal processes are long and cranially oriented , unlike modern humans , and its vertebral body is intermediate in shape between those of modern humans ( and Neandertals ) and great apes . Williams et al . , 2013 , predicted strong lumbar lordosis ( ‘hyperlordosis’ ) in MH2 based on the combined wedging values of the penultimate and ultimate lumbar vertebrae . In contrast , Been et al . , 2014 , estimated lumbar lordosis angle using pelvic incidence from a pelvis reconstruction ( Kibii et al . , 2011 ) and found MH2 to produce the least lordotic lumbar column of the sampled members of the genus Australopithecus in their sample , falling well below modern human values and within the distribution of Neandertals . Neandertals are thought to be ‘hypolordotic’ , or characterized by a relatively straight , non-lordotic lumbar column ( Been et al . , 2014; Been et al . , 2017; but see Haeusler et al . , 2019 ) . However , Tardieu et al . , 2017 , report a human-like degree of pelvic incidence ( and therefore lumbar lordosis ) in a new reconstruction of the MH2 pelvis . Therefore , current interpretations of lumbar curvature of A . sediba range from hyperlordotic to hypolordotic . Here , we report that the pattern of vertebral wedging of MH2 and most other fossil hominins are similar to both modern humans and extant great apes except at the last lumbar level , where MH2 is markedly more dorsally wedged ( Figure 5 , Figure 5—figure supplement 1 , Figure 5—figure supplement 2 ) . Like the Neandertal Kebara 2 , the strong dorsal ( lordotic ) wedging of MH2’s last lumbar vertebra is likely countering a strong ventral ( kyphotic ) wedging in the upper lumbar column ( Figure 5 ) . However , MH2 demonstrates much less ventral wedging than Neandertals and produces a human-like combined wedging angles value , falling outside the 95% prediction intervals of great apes . Therefore , it seems likely that MH2 and possibly the juvenile H . erectus individual KNM-WT 15000 demonstrate strong dorsal wedging at the last lumbar level for different reasons than Kebara 2 . It was suggested previously that the morphology of the MH2 last lumbar is part of a kinematic chain linked to hyperpronation of the foot ( DeSilva et al . , 2013; Williams et al . , 2013 ) . With the absence of soft tissue contributions to the kinematic chain ( i . e . , intervertebral discs ) , formal biomechanical testing is beyond the scope of the current paper; however , our results suggest that MH2 was probably neither hypolordotic nor hyperlordotic and produces a combined wedging angles value more similar to modern humans than great apes . Modern humans are characterized by a pyramidal configuration of the lumbar inter-articular facet joints such that the upper lumbar articular facet joints ( and associated laminae ) are transversely more narrowly spaced than those of the lower lumbar vertebrae and especially compared to the lumbosacral inter-articular facet joints ( Latimer and Ward , 1993; Sanders , 1998; Ward and Latimer , 2005 ) . Together with vertebral body and intervertebral disc wedging , this progressive widening facilitates the adoption of lordotic posture during ontogeny and allows for the imbrication of the IAF of a superjacent vertebra onto the laminae of the subjacent vertebra during hyperextension of the lower back ( Latimer and Ward , 1993; Williams et al . , 2013 ) . Like modern humans , known fossil hominin lumbar vertebrae bear ‘imbrication pockets’ , mechanically induced fossae positioned just caudal to the SAF on the lamina ( Latimer and Ward , 1993; Williams et al . , 2013 ) , providing direct evidence for lumbar hyperextension and lordosis . Inadequate spacing of lower lumbar inter-articular facets in modern humans can result in spondylolysis , fracture of the pars interarticularis , and potential separation of the affected vertebra’s spinous process and inferior articular processes ( Ward and Latimer , 2005 ) . Lack of the progressive widening of inter-articular facets of lower lumbar vertebrae in our closest living relatives , the African apes , begs the question of when the pyramidal configuration evolved and to what extent various fossil hominins demonstrated this trait . Although a human-like pattern of interfacet distance was once claimed for the European late Miocene ape Oreopithecus ( Köhler and Moya-Sola , 1997 ) , Russo and Shapiro , 2013 , demonstrated that measures for changes in both interfacet distance and laminar width in this extinct ape fall within ranges of extant apes ( and other suspensory mammals ) and outside those of humans . Among hominins , Latimer and Ward , 1993 , documented the presence of a pyramidal configuration in H . erectus . It has also been demonstrated qualitatively in A . afarensis and A . africanus ( Robinson , 1972; Lovejoy , 2005 ) , and its presence in A . sediba could be inferred previously based on the articulated penultimate and ultimate lumbar vertebrae and sacrum of MH2 ( Williams et al . , 2013 ) . Here , we show that MH2 and other Australopithecus specimens fall at the low end of modern human variation and differ from great apes in having significantly wider inter-articular facets at the lumbosacral junction than higher in the lumbar column ( Figure 6 ) . The overall morphologies of lumbar vertebrae are informative with regard to locomotion and posture in primates ( Slijper , 1946; Shapiro , 1993b; Sanders and Bodenbender , 1994; Granatosky et al . , 2014; Williams and Russo , 2015 ) . Hominoids are characterized by derived vertebral morphologies related to orthogrady and antipronograde positional behaviors , and early hominins have been found to largely resemble modern humans in lumbar vertebra shape , with some retained primitive morphologies ( Robinson , 1972; Schmid , 1991; Shapiro , 1993a ) . Our 3D GM results show that the middle lumbar vertebra of A . sediba ( U . W . 88–233; L3 ) falls with modern humans ( L3 ) to the exclusion of great apes ( L2 ) in overall shape ( Figure 7A–B ) . However , it bears long , cranially and ventrally oriented costal processes unlike those of modern humans ( Figure 7C , Figure 7—figure supplement 1 , Figure 7—figure supplement 2 ) , and the vertebral body is somewhat intermediate in shape between modern humans and great apes ( Figure 7D ) . To evaluate the potential effect of centroid size in driving differences in middle lumbar vertebra shape , we ran a Procrustes distance-based ANOVA on generalized Procrustes analysis ( GPA ) shape scores to test whether shape differences between this fossil specimen and any extant taxon , such as H . sapiens , could be explained by differences in size ( ‘allometry’ ) . Since body mass scales as the cube of linear dimensions and the physiological cross-sectional area of skeletal muscle – a major determinant of isometric force production – scales as the square of linear dimensions , larger-bodied individuals should be relatively weaker with all else held equal . Thus to compensate , we would expect to see changes in bony morphology based on differences in body size . We find that the spinous and costal processes are longer in specimens with larger centroid sizes ( Figure 7—figure supplement 3 ) . These changes would increase the moment arms of the erector spinae and quadratus lumborum muscles , respectively , resulting in greater moments that contribute to lumbar extension , ventral flexion , and lateral flexion to cope with increases in body mass . Our results suggest that , while we detect a statistically significant effect of centroid size on middle lumbar vertebral shape within each group , the differences in costal process size and orientation observed between A . sediba and modern humans appear not to be explained by size alone . Long costal processes give the psoas major and quadratus lumborum muscles an effective leverage in acting on the vertebral column , increasing their moment arms and torque generation capabilities to assist the erector spinae in lateral flexion of the spine , back extension , and stabilizing the trunk during upright posture and bipedalism and ape-like vertical climbing ( Robinson , 1972; Waters and Morris , 1972; Schmid , 1991; Sanders , 1998; Figure 7—figure supplement 2 ) . Psoas major acts with iliacus as a powerful flexor of the thigh and trunk , while quadratus lumborum is a trunk extensor and a lateral flexor of the vertebral column and pelvis unilaterally ( Robinson , 1972; Drake et al . , 2019 ) . The lumbar vertebral morphology of A . sediba , therefore , is that of a biped equipped with especially powerful trunk musculature for stabilizing the hip and back during walking and/or vertical climbing . Further work on back morphology and function in A . sediba and other early hominins is required to explore the efficacy of these possible functional explanations for the observed morphology of MH2’s lumbar vertebrae . Previous work has shown that the adult , presumed female individual from Malapa ( MH2 ) demonstrates clear adaptations to bipedal locomotion ( Zipfel et al . , 2011; DeSilva et al . , 2013; DeSilva et al . , 2018; Williams et al . , 2013; Williams et al . , 2018b ) , as do other Australopithecus specimens , despite their retention of features linked to suspensory behavior and other arboreal proclivities ( Zipfel et al . , 2011; Henry et al . , 2012; Churchill et al . , 2013; Churchill et al . , 2018; DeSilva et al . , 2013; DeSilva et al . , 2018; Prang , 2015a; Prang , 2015b; Prang , 2016; Meyer et al . , 2017; Rein et al . , 2017; Holliday et al . , 2018; Prabhat et al . , 2021 ) . The new fossils here reinforce these conclusions , signaling a lower back in MH2 as that of an upright biped equipped with powerful trunk musculature potentially used in both terrestrial and arboreal locomotion . The recovery and study of new fossil material , including juvenile material such that the ontogeny of bipedal features can be examined ( Ward et al . , 2017; Nalley et al . , 2019 ) , along with experimental biomechanical work and additional comparative analyses , will allow for testing hypotheses of form and function in the hominin fossil record .
Original fossil material was studied in all cases with the exception of two Neandertal specimens ( Kebara 2 and Shanidar 3 ) , for which high-quality casts were used . The A . sediba fossils belonging to MH2 ( U . W . 88-280/281 , L1; U . W . 88–232 , L2; U . W . 88–233 , L3; U . W . 88-127/153/234 , L4; U . W . 88-126/138 , L5 ) were studied at the University of the Witwatersrand ( Johannesburg ) , as was LES1 Homo naledi ( U . W . 102a-154B , L1 ) and fossils purportedly belonging to A . africanus: StW 431 ( StW 431r , L1; StW 431s , L2; StW 431t , L3; StW 431u , L4; StW 431v , L5 ) , StW 8 ( StW8a , L1; StW8b , L2; StW8c , L3; StW8d , L4 ) , StW 572 ( L2 ) , StW 656 ( L3 ) , and StW 600 ( L5 ) . The A . africanus specimen Sts 14 ( Sts 14e , L1; Sts 14d , L2; Sts 14c , L3; Sts 14b , L4; Sts 14a , L5 ) and possible Paranthropus robustus or early Homo specimen SK 3981b ( L5 ) were studied at the Ditsong National Museum of Natural History , A . afarensis specimens ( A . L . 288-1aa/ak/al , L3; A . L . 333–73 , L3 ) at the National Museum of Ethiopia , the H . erectus juvenile individual KNM-WT 15000 ( AV/AA , L1; Z/BW , L2; AB , L3; BM , L4; AC , L5 ) at the National Museums of Kenya , and La Chapelle-aux-Saints 1 at Musée de l’Homme ( Paris ) . Our comparative sample consisted in total of 43 chimpanzees ( Pan troglodytes ) , 31 western gorillas ( Gorilla gorilla ) , 14 orangutans ( Pongo sp . ) , and 54 modern humans ( H . sapiens ) . To ensure that adequate space between elements was taken into account , we only included great apes with four lumbar vertebrae . Eastern gorillas ( Gorilla beringei ) , which mostly possess just three lumbar vertebrae ( Williams et al . , 2019 ) , are not included here , nor are other great ape individuals with only three lumbar vertebrae . The human sample includes data from an archaeological sample representing individuals from Africa , Asia-Pacific , and South America studied at the American Museum of Natural History ( New York City ) , Musée de l’Homme , the Natural History Museum ( London ) , and the University of the Witwatersrand . Measurements ( listed in Appendix 1 ) were collected with Mitutoyo digital calipers ( Mitutoyo Inc , Japan ) and recorded at 0 . 01 mm; however , we report measurements at 0 . 1 mm . Following Digiovanni et al . , 1989 , we calculated wedging angles for lumbar vertebrae 2–5 using the arctangent of difference between the dorsal and ventral height of the vertebral body and its dorsoventral length ( see Appendix 1 ) . We also summed those values into a combined lumbar wedging value . For both great apes and male and female humans , 95% prediction intervals of the mean ( 1 . 96 * standard deviation ) were calculated for each vertebral level and for the combined wedging values . Inter-articular facet spacing was measured across the lateral borders of the IAF of lumbar vertebrae three levels apart: on the last lumbar vertebra and on L1 in great apes with four lumbar vertebrae and on L2 in hominins . This is done to estimate the difference in inter-articular facet width at upper and lower lumbar levels and thus quantify neural arch configuration . Due to preservation , this measurement was estimated from the SAF of the L3 vertebra and/or the sacrum in a selection of fossils ( A . L . 288–1 , Sts 14 ) . In instances of partial preservation , the relevant adjacent elements were articulated to estimate the measurement ( MH2 , StW 431; KNM-WT 15000 ) . An index was created by dividing the last lumbar-sacrum interarticular facet mediolateral width by that of the upper lumbar vertebrae . For 3D GM analyses , we used subsets of middle lumbar vertebrae that were scanned at the aforementioned institutions using an Artec Space Spider 3D scanner ( Source Graphics , Anaheim , CA ) . The middle lumbar vertebra of hominins with five lumbar vertebrae is the third lumbar vertebra , and that of chimpanzees and gorillas with three lumbar vertebrae is L2 . Many chimpanzees and bonobos , western gorillas , and orangutans have four lumbar vertebrae ( Williams et al . , 2019 ) , and we use L2 in these individuals as well for consistency . Thirty-six modern humans , 28 chimpanzees , 26 western gorillas , and 8 orangutans were included . For this analysis , we also utilized a sample of 23 bonobos ( Pan paniscus ) and 7 eastern gorillas ( G . beringei ) . U . W . 88–233 is a complete third lumbar vertebra , but it is partially encased in breccia , which obscures some morphologies . The lumbar new vertebrae ( U . W . 88-232-234 ) were µCT scanned in partial articulation ( Figure 2 , Figure 2—figure supplement 1 ) at the University of the Witwatersrand using a Nikon Metrology XTH 225/320 LC system . Scan settings were 70 kV , 120 μA , 1 s exposure time , and 3000 projections . Voxel size was 0 . 049 mm and scans included 2000 voxels . The high-resolution µCT scans were processed to yield virtual 3D models . Each vertebra was segmented using Amira 6 . 2 ( Thermo Fisher Scientific , Waltham , MA ) . After importing µCT scan slices ( TIFF files ) and creating a volume stack file ( . am ) , an Edit New Label Field module was attached to the stack file . Voxels were selected and assigned to each model separately using the magic wand and brush tools after verification in all three orthogonal views . A Generate Surface module was used to produce a labels file ( . labels . am ) once an individual element was completely selected . A 3D surface model was created from the labels file using an unconstrained smoothing setting of 5 . Models of each element were then saved as polygon ( . ply ) files . Using GeoMagic Studio software ( 3D Systems , Rock Hill , SC ) , broken portions of U . W . 88–233 were refitted and the specimen was reconstructed accordingly . The affected portions of the neural arch were pulled dorsally to refit the fractured portion of the left lamina; additionally , the broken and deflected costal process was refitted . The result is a reconstructed 3D model ( Figure 4 ) . Due to crushing of the right SAF , we collected landmarks on the left side of U . W . 88–233 and our comparative sample of middle lumbar vertebrae ( Table 1 ) . Our 3D landmark set consisted of 48 landmarks distributed across the vertebra to reflect the gross morphology ( Appendix 1 ) . Landmarks were collected using the Landmarks tool in Amira on the surface model of U . W . 88–233 and on 3D models of middle lumbar vertebrae produced using Artec Studio 14 software ( Source Graphics , Anaheim , CA ) . We used ‘geomorph’ package version 4 . 0 ( Adams et al . , 2021 ) in R version 4 . 0 . 2 ( Core Team , 2020 ) to carry out 3D GM analyses . The geomorph package was then used to subject the raw landmark data to GPA to correct for position , rotation , and size adjustment . The GPA shape scores were then subjected to PCA using the covariance matrix . We evaluated the effects of centroid size on shape using Procrustes distance-based ANOVA on GPA shape scores as implemented in the geomorph package ( Goodall , 1991; Adams et al . , 2021 ) . Specifically , we evaluated the effect of centroid size as a predictor of middle lumbar shape coordinates within each genus by including a genus interaction term ( shape~centroid size * genus ) . Finally , we analyzed two datasets: one on the full set of 48 landmarks in which only complete ( reconstructed ) fossils ( U . W . 88–233 , Sts 14c , Shanidar 3 ) were included , and one on a 37 landmark subset with 11 landmarks on the costal and spinous processes removed so that additional , less well-preserved fossils could be included ( A . L . 288-1aa/ak/al , StW 431 , Kebara 2 ) . | One of the defining features of humans is our ability to walk comfortably on two legs . To achieve this , our skeletons have evolved certain physical characteristics . For example , the lower part of the human spine has a forward curve that supports an upright posture; whereas the lower backs of chimpanzees and other apes – which walk around on four limbs and spend much of their time in trees – lack this curvature . Studying the fossilized back bones of ancient human remains can help us to understand how we evolved these features , and whether our ancestors moved in a similar way . Australopithecus sediba was a close-relative of modern humans that lived about two million years ago . In 2008 , fossils from an adult female were discovered at a cave site in South Africa called Malapa . However , the fossils of the lower back region were incomplete , so it was unclear whether the female – referred to as Malapa Hominin 2 ( MH2 ) – had a forward-curving spine and other adaptations needed to walk on two legs . Here , Williams et al . report the discovery of new A . sediba fossils from Malapa . The new fossils are mainly bones from the lower back , and they fit together with the previously discovered MH2 fossils , providing a nearly complete lower spine . Analysis of the fossils suggested that MH2 would have had an upright posture and comfortably walked on two legs , and the curvature of their lower back was similar to modern females . However , other aspects of the bones’ shape suggest that as well as walking , A . sediba probably spent a significant amount of time climbing in trees . The findings of Williams et al . provide new insights in to our evolutionary history , and ultimately , our place in the natural world around us . Our lower back is prone to injury and pain associated with posture , pregnancy and exercise ( or lack thereof ) . Therefore , understanding how the lower back evolved may help us to learn how to prevent injuries and maintain a healthy back . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2021 | New fossils of Australopithecus sediba reveal a nearly complete lower back |
Resting-state networks offer a unique window into the brain’s functional architecture , but their characterization remains limited to instantaneous connectivity thus far . Here , we describe a novel resting-state network based on the delayed connectivity between the brain and the slow electrical rhythm ( 0 . 05 Hz ) generated in the stomach . The gastric network cuts across classical resting-state networks with partial overlap with autonomic regulation areas . This network is composed of regions with convergent functional properties involved in mapping bodily space through touch , action or vision , as well as mapping external space in bodily coordinates . The network is characterized by a precise temporal sequence of activations within a gastric cycle , beginning with somato-motor cortices and ending with the extrastriate body area and dorsal precuneus . Our results demonstrate that canonical resting-state networks based on instantaneous connectivity represent only one of the possible partitions of the brain into coherent networks based on temporal dynamics .
The parsing of the brain into resting-state networks ( RSNs ) has been widely exploited to study the brain’s functional architecture in health and disease ( Fox and Raichle , 2007 ) . With long-time scales , RSNs closely match the anatomical backbone of the brain ( van den Heuvel et al . , 2009; Honey et al . , 2009; Shen et al . , 2015 ) . With short-time scales ( ~10–100 s ) , spontaneous brain activity is characterized by the emergence and dissolution of network patterns encompassing and extending classical RSN topologies ( Ponce-Alvarez et al . , 2015; Shine et al . , 2016 ) with rich temporal trajectories ( Mitra et al . , 2015 ) . Temporal trajectories indicate the existence of delays between regions , whereas the methods most often used to parse brain activity into functional networks ( seed-based correlation and independent component analysis ) make the implicit assumption that RSNs are characterized by instantaneous or zero delay connectivity . Therefore , we analyzed delayed connectivity in resting-state BOLD signals using techniques widely used in electrophysiological studies of large-scale brain dynamics ( Lachaux et al . , 1999 ) that quantify the stability of temporal delays between time series . More specifically , we studied the delayed coupling between resting-state brain activity and a visceral organ , the stomach . The stomach continuously produces a slow electrical rhythm ( 0 . 05 Hz , one cycle every 20 s ) that can be non-invasively measured ( electrogastrogram , EGG [Koch and Stern , 2004] ) . The gastric basal rhythm is continuously ( Bozler , 1945 ) and intrinsically ( Suzuki et al . , 1986 ) generated in the stomach wall by a network of specialized cells , the interstitial cells of Cajal ( Sanders et al . , 2014 ) , which form synapse-like connections not only with gastric smooth muscle but also with afferent sensory neurons ( Powley and Phillips , 2011 ) . The stomach is an interesting candidate for large-scale brain coordination for several reasons . First , visceral inputs can reach a number of cortical targets ( Critchley and Harrison , 2013; Park and Tallon-Baudry , 2014 ) . Second , gastric frequency ( ~0 . 05 Hz ) falls within the range of BOLD fluctuations that are used to define RSNs and that are free from known cardiac and respiratory artifacts ( Glerean et al . , 2012 ) . Finally , the amplitude of alpha rhythm , the dominant rhythm in the human brain at rest , depends on the phase of gastric rhythm ( Richter et al . , 2017 ) . We simultaneously recorded brain activity with fMRI and stomach activity with EGG ( Figure 1a ) in 30 human participants at rest with open eyes . We then determined the regions in which spontaneous fluctuations in brain activity were phase synchronized with gastric basal rhythm; we refer to these regions as the gastric network .
We first determined gastric frequency ( Figure 1b ) in each participant as the frequency of the largest spectral peak within the normogastric range ( 0 . 033–0 . 066 Hz ) . The mean EGG peak frequency across the 30 participants was 0 . 047 Hz ( ±SD 0 . 003 , range 0 . 041–0 . 053 ) . EGG peak frequency measured inside and outside the scanner did not differ ( EGG outside the scanner measured in 29 of the 30 participants , mean 0 . 046 Hz ± SD 0 . 006; two-sided paired t-test , t ( 28 ) =0 . 35 , p=0 . 725 Bayes Factor <0 . 001 , indicating decisive evidence for the null hypothesis ) . In each participant and at each voxel , we quantified the degree of phase synchrony between the EGG signal and BOLD time series filtered around gastric frequency ( Figure 1c ) . We computed the phase-locking value ( PLV ) ( Lachaux et al . , 1999 ) , a measure widely used in electrophysiology that varies between zero when two time series show no consistent phase relationship ( Figure 1c , bottom panel ) and one when two time series have a consistent phase relationship over time ( Figure 1c , upper panel ) . PLV has three important properties: PLV is high for any lag between the time series as long as this lag is constant over time , PLV is independent of signal amplitude , and PLV gives no indication on the directionality of interactions between the two time series . In each participant and at each voxel , we estimated the PLV that could be expected by chance from EGG signals that were shifted with respect to the BOLD time series . The empirical PLVs were then compared with chance-level PLVs using a cluster-based statistical procedure that intrinsically corrects for multiple comparisons ( Maris and Oostenveld , 2007 ) . Significant phase coupling between the EGG and resting-state BOLD time series occurred in twelve nodes ( voxel threshold p<0 . 01 , two-sided paired t-test between observed and chance PLV; cluster threshold corrected for multiple comparisons , Monte-Carlo p<0 . 05 ) . Exact p-values are reported for each cluster in Table 1 . The gastric network ( Table 1 , Figure 2a ) comprises the right primary somatosensory cortex ( SIr ) , bilateral secondary somatosensory cortices ( SII ) , medial wall motor regions ( MWM ) , comprising the caudate cingulate motor zone ( CCZ ) , posterior rostral cingulate motor zone ( RCZp ) , and right supplementary motor area ( SMA ) , a region of the right occipito-temporal cortex overlapping the extrastriate body area ( EBA ) , as well as nodes in the posterior cingulate sulcus ( pCS ) , dorsal precuneus ( dPrec ) , occipital cortex ( ventral and dorsal portions , vOcc and dOcc ) , retrosplenial cortex ( RSC ) , and superior parieto-occipital sulcus ( sPOS ) . Estimating chance-level PLV by computing gastric-BOLD coupling between the BOLD signal of one participant with the EGG of the other 29 participants resulted in a qualitatively similar network , with coupling occurring either in the same or neighboring voxels ( Supplemental Figure 2 ) . The average shared variance between the EGG and BOLD signals across participants , as estimated from squared coherence , ranged from 12% in the left anterior dorsal precuneus to 16 . 9% in the posterior cingulate sulcus ( Table 1 ) . An analysis of covariance across nodes did not reveal significant links between gastric-BOLD coupling strength ( defined as the difference between empirical and chance PLV ) and gender ( F ( 1 , 28 ) =1 . 02 , p=0 . 46 ) , body mass index ( BMI ) ( F ( 1 , 28 ) =1 . 3 , p=0 . 3 ) or trait anxiety scores ( F ( 1 , 28 ) =1 . 02 , p=0 . 47 . Statistics ( including Bayes Factor ) per node are reported in Table 2 . Note that there is less variation in BMI in our sample than in the general population since all participants had a BMI smaller than 25 . To assess the robustness of the gastric network , we ran several controls . First , we verified that EGG-BOLD coupling was specific to gastric frequency . We filtered both EGG and BOLD time series at frequencies that were slightly offset from the peak gastric frequency of each participant and recomputed cluster statistics . Summary statistics ( sum of the absolute t-values resulting from the paired t-test between empirical and chance-level PLV at each voxel , either summed across the whole brain or within the gastric network ) decreased when shifting below or above the gastric peak frequency ( Figure 2b ) . This result indicates that the gastric network corresponds to BOLD fluctuations specifically occurring at gastric frequency . Second , we estimated the likelihood of false positives with our statistical procedure . We randomly sampled surrogate datasets in which a random time shift was applied to the EGG of each participant a thousand times . Next , we tested whether any of those 1000 combinations would generate summary statistics as large as the original data when compared with the chance-level estimate we used to determine significantly coupled regions at the group level ( Figure 2c ) . This result was never observed , indicating that the probability of our results being a false positive is less than 0 . 001 . Third , we verified that gastric-BOLD coupling strength was unrelated to BOLD power at gastric frequency . We computed the correlation between BOLD power at gastric frequency and coupling strength for each participant and voxel , and found the two measures to be unrelated ( Fisher z-transformed Pearson correlation coefficients tested against zero , t ( 29 ) =1 . 19 , p=0 . 24; Bayes factor <0 . 001 , indicating decisive evidence for the absence of a link between coupling strength across the brain and BOLD power at gastric frequency ) . Finally , we investigated whether submillimeter head movements might have influenced the results . We defined voxel motion susceptibility as the regression coefficient of head movement ( Power et al . , 2012 ) from the BOLD time series . Coupling strength and voxel motion were unrelated ( Fisher z-transformed Pearson correlation coefficients tested against zero , t ( 29 ) =-0 . 34 , p=0 . 73; Bayes factor <0 . 001 , indicating decisive evidence for the absence of a link between coupling strength and head movement ) . Stomach contractions might also lead to small head movements that could be phase locked to gastric rhythm . Although gastric rhythm is continuously produced even during fasting , it is larger during stomach contractions . Thus , we tested whether the effects we found were due to differences in EGG power ( or frequency ) across participants . We found no link between coupling strength in the 12 nodes and EGG power ( ANCOVA , F ( 1 , 28 ) =0 . 9 , p=0 . 51; all Bayes factor <0 . 33 , indicating substantial evidence for the null hypothesis ) nor between coupling strength and EGG peak frequency ( ANCOVA , F ( 1 , 28 ) =1 . 6 , p=0 . 17; Bayes Factor <0 . 33 , indicating substantial evidence for the null hypothesis in 9 of 12 nodes; Bayes Factor <1 . 3 in the three remaining nodes , indicating anecdotal or no evidence ) . The gastric network is thus specific to individual gastric peak frequency , is highly unlikely to be a chance finding , is not dependent on BOLD power , and is not linked to spurious effects of head movement on the BOLD signal . We then examined the areas comprising the gastric network in more detail . By definition , the gastric network is composed of regions with activity that co-fluctuates with gastric basal rhythm . Five nodes of the gastric network also share a common functional feature , somatotopic organization , as detailed in Figure 2d . The gastric network includes the following regions with a well-known body representation based on touch: the right primary somatosensory cortex in the hand and mouth region and bilateral secondary somatosensory cortices . We quantified the overlap between these gastric network nodes and known cytoarchitectonic subdivisions of the somatosensory cortices ( Geyer et al . , 2000; Grefkes et al . , 2001 ) . The gastric network mostly overlapped with area 1 ( 60 . 2% of the SIr node ) and to a lesser extent , with area 2 ( 13 . 1% ) and area 3b ( 9 . 9% ) . The SII nodes of the gastric network overlapped with the secondary somatosensory cortices and more precisely with the somatotopically organized subdivisions of the parietal operculum OP1 and OP4 ( 22 ) . The right SII node mostly overlapped with area OP1 ( 35 . 2% of the node ) , while the left SII node overlapped with both OP1 ( 21 . 7% ) and OP4 ( 14 . 9% ) . Additionally , both left and right SII nodes extended more ventrally to the temporal cortex . The gastric network also includes three medial wall motor regions ( CCZ , RCZp , and SMA ) that reveal their somatotopic organization when participants are required to move specific body parts ( Amiez and Petrides , 2014 ) . Note that gastric-BOLD coupling also included a more posterior area in the cingulate sulcus ( pCS ) . Finally , the gastric network overlapped with the EBA , a region of the lateral occipital cortex activated when participants view images of body parts ( Downing et al . , 2001 ) with a clear somatotopic organization ( Orlov et al . , 2010 ) . The overlap between the gastric network and EBA occurred in the lower face region , which includes the mouth . Thus , the gastric network overlaps with body maps classically associated with different modalities , including touch in somatosensory cortices , action in MWM and vision in the EBA . Finally , we found gastric-BOLD coupling in the posterior bank of the parieto-occipital sulcus ( dOcc and vOcc ) and retrosplenial cortex . In a previous study using magneto-encephalography ( Richter et al . , 2017 ) , the amplitude of the alpha rhythm in these regions was modulated by gastric phase ( Figure 2e ) . The insula is one region that receives visceral inputs ( Critchley and Harrison , 2013; Park and Tallon-Baudry , 2014 ) , but it did not appear to be significantly phase synchronized to the EGG using our whole-brain , statistically conservative procedure . Thus , we performed post-hoc region-of-interest analysis of the three insular subdivisions ( anterior dorsal , anterior ventral , posterior [Deen et al . , 2011] ) in both hemispheres . Only the right posterior insula showed evidence for gastric-BOLD coupling across participants ( empirical vs . chance-level PLV , paired t-test , two sided , t ( 29 ) =2 . 78 , p=0 . 043 , Bonferroni corrected; all other regions , p>0 . 21 ) . The use of statistical thresholds results in binary outputs . To get a finer grained picture , we computed effect sizes in the 6 insular subdivisions and in the 12 gastric network nodes ( Cohen’s d for the difference between empirical and chance PLV on the mean time series in each region of interest ) . Mean Cohen’s d across gastric network nodes was 1 . 19 ± 0 . 21 STD , ranging from 0 . 80 in the dorsal occipital cortex to 1 . 62 in the right secondary somatosensory cortex . The right posterior insula had an effect size of 0 . 84 , within the lower range of the gastric network . All other insula subdivisions displayed smaller effect sizes ( right: dorsal anterior 0 . 61 , ventral anterior 0 . 54; left: posterior 0 . 60 , dorsal anterior 0 . 41 , ventral anterior = 0 . 52 ) . Thus , the right posterior insula does show evidence for coupling with the stomach , with an effect size comparable to that of the weakest nodes of the gastric network , provided signal-to-noise ratio is first increased by averaging within a region of interest . Is the gastric network specific to the stomach , or is it also linked to other organs such as the heart ? We determined brain regions ( FWE corrected p<0 . 05; Figure 3A ) fluctuating with high- and low-frequency heart rate variability , that represent parasympathetic and a mixture of sympathetic and parasympathetic outputs , respectively . Of the gastric network , 30% was also related to heart rate variability , mostly in medial motor regions and in the posterior cingulate cluster ( low-frequency heart rate variability ) , and , to a lesser extent , in the dorsal occipital cluster ( high-frequency heart rate variability ) . Because we did not record any measure that isolates sympathetic output , we additionally analyzed the overlap between the gastric network and known sympathetic areas ( Beissner et al . , 2013 ) . This overlap was very limited ( 34 voxels , 4 . 7% of the gastric network ) and confined to SIr and the anterior parts of medial motor regions ( Figure 3—figure supplement 1 ) . We also determined brain regions that correlate with pupil diameter ( n = 20 due to data loss or artefacts; Figure 3B ) . The strongest correlations were found in occipital regions , somato-motor cortices and medial wall motor regions . 17% of the gastric network ( SI , SIIr , MWM and EBA ) overlaps with regions correlating with pupil diameter . Shared variance between pupil diameter and EGG , estimated from squared coherence , was 9 . 7 ± 2 . 5% . Coupling strength averaged across SI , SIIr , MWM and EBA did not correlate with shared pupil-EGG variance ( mean r = 0 . 05 , p=0 . 82 , BF = 0 . 17 which indicates substantial evidence for the null hypothesis ) . In the different nodes of the gastric network , gastric-brain coupling occurred with different phase delays with respect to the gastric cycle . We analyzed between-participant phase-delay consistency and found temporal delays of ~3 . 3 s between the earliest nodes ( somatosensory cortices ) and latest nodes ( dorsal precuneus and EBA ) of the gastric network ( Figure 4a , b ) . The delay in the right posterior insula was in the range of the earliest nodes of the gastric network ( Figure 4a ) . The Watson-Williams test for circular data confirmed that different nodes of the gastric network were coupled to the gastric rhythm with different phase delays ( F ( 11 , 29 ) =5 . 22 , p<10−6 ) , indicating a precise temporal sequence of activations within each gastric cycle . Thus , each node of the gastric network appears to be characterized by a specific temporal delay with respect to gastric phase . These temporal delays were accompanied by delayed functional connectivity ( FC ) between the nodes of the gastric network . We first illustrated this point with an example in a single participant ( Figure 4c ) , with two 200 s time series of the gastric network ( MWM and EBA ) . The two time series systematically co-varied with a temporal delay . The existence of temporal delays between the nodes of the gastric network is one of the reasons why the gastric network could not be observed in prior studies . Indeed , fMRI RSN studies are typically based on measures of instantaneous FC , such as shared variance estimated from the squared Pearson correlation coefficient , which does not detect the temporally delayed interactions revealed here . These measures differ from delayed FC measures based on the consistency of phase delays over time , such as shared variance estimated from squared coherence . In the example illustrated in Figure 4c , instantaneous FC between the two time series is 56% , whereas delayed FC is 86% . If we advance the timing of the medial wall time series by 2 s , instantaneous FC increases to 86% . This finding shows that the difference between the two FC estimates is due to temporal delays only . We then estimated both instantaneous and delayed FC between all nodes of the gastric network in all participants . Delayed FC between gastric nodes ( mean 40 . 8% ± SD 8% , ranging from 26 . 5% between the right primary somatosensory cortex and RSC , up to 63 . 9% between the ventral and dorsal occipital cortices ) was systematically larger ( paired t-test , t ( 29 ) =9 . 02 , p<10−10 ) than instantaneous FC ( mean 30 . 2% ± SD 11% , ranging from 9 . 3% between the dorsal precuneus and right SII , up to 61 . 2% between right and left SII ) . Next , we verified ( Figure 4d , e ) that two regions belonging to both the gastric network and the same RSN ( i . e . two regions of the gastric network with little temporal delay , such as MWM and SIr ) would display large values of both delayed and instantaneous FC , whereas two regions belonging to the gastric network but not to the same classical RSN ( i . e . two regions of the gastric network with a large temporal delay , such as MWM and EBA ) would show large delayed FC and small instantaneous FC . Thus , in contrast to classical RSNs , the gastric network appears to be characterized by between-node delayed connectivity . Thus far , we have identified a sequence of activation that occurs at each gastric cycle , which characterizes gastric-BOLD coupling . We then investigated whether slow temporal fluctuations in the strength of gastric-BOLD coupling were accompanied by fluctuations in BOLD amplitude . As illustrated in Figure 5a , we found that episodes of elevated gastric-BOLD synchronization corresponded to episodes of increased BOLD amplitude . Indeed , time-varying PLV and BOLD time series , computed in sliding time windows of 60 s ( approximately three gastric cycles ) , were significantly correlated ( Fisher z-transformed Pearson correlation coefficients t-tested against zero , Bonferroni corrected p<0 . 006 in all gastric nodes , mean r across nodes 0 . 18 STD =± 0 . 02 , ranging from 0 . 15 in MWM to 0 . 22 in SIIl ) . Next , we tested whether slow temporal fluctuations in gastric-BOLD synchronization occurred simultaneously or independently in the different nodes of the gastric network ( Figure 5b ) . We computed the correlation between time-varying PLVs for all possible node pairs in each participant and found that at the group level , this correlation was significantly positive ( Fisher z-transformed average Pearson correlation coefficients against zero , t ( 29 ) =9 . 22 , p<10−10 , mean r = 0 . 129 ± STD 0 . 075 , range across participants 0 . 02–0 . 35 ) . To determine whether the overall pattern of synchronous fluctuations in gastric-BOLD coupling strength was driven by specific node pairs , we investigated correlations between node pairs . All node pairs but RSC-SIr , sPOS-RSC , sPOS-SIIr and sPOS-ladPrec showed a significant positive correlation at the group level ( Fisher z-transformed Pearson correlation coefficients in all pairs tested against zero , Bonferroni corrected ) . The three nodes showing the largest covariation in time-varying PLV with the other nodes were dPrec , pCS and vOcc , and the three nodes showing the least covariation in time-varying PLV with the other nodes were RSC , SIr and sPOS . Thus , slow temporal fluctuations in gastric-BOLD coupling are associated with changes in BOLD amplitude and occur simultaneously in all nodes .
SIr , SII and medial wall motor regions likely receive direct gastric inputs . The stimulation of the splanchnic ( spinal ) nerve that innervates the stomach evokes responses in contralateral SI and bilateral SII in several mammals ( Amassian , 1951 ) , and the spinothalamic tract was recently shown to target MWM in monkeys ( Dum et al . , 2009 ) . Vagal stimulation can also evoke responses in somato-motor cortices ( Ito et al . , 2003 ) . In addition , single neurons with convergent visceral and hand inputs have been observed in SI ( Brüggemann et al . , 1997 ) , in line with the overlap we observed between the gastric network and hand representation . SIr , SII and MWM are not only targeted by documented ascending visceral pathways , they are also early nodes of the gastric network , with a phase advance compared with that of other nodes . Thus , these areas could be the entry point of gastric afferences . We found the right posterior insula , a region that receives direct cardiac inputs in monkeys ( Zhang et al . , 1998 ) and is considered as a visceral cortex , to be coupled with the stomach , with coupling similar to the weakest node of the gastric network . In addition , the right posterior insula appeared with a phase advance , in line with its role in visceroception . To be revealed , gastric-BOLD coupling in the right posterior insula required a region-of-interest approach , that is an increase of signal-to-noise ratio by averaging across voxels . The modest involvement of the insula in the present data might be due to the absence of an interoceptive task . Indeed , BOLD signal in the insula increases when participants explicitly monitor a visceral variable ( see , e . g . [Critchley et al . , 2004] ) . Regions receiving direct visceral inputs are also early nodes of the gastric network . This suggests that the BOLD fluctuations locked to the gastric rhythm have a neural origin . An additional argument for a neural origin is that we found gastric-BOLD coupling in parieto-occipital regions , where neural activity in the alpha range is modulated by gastric phase ( Richter et al . , 2017 ) . However , below , we examine the possibility that other non-neural mechanisms might contribute to gastric-BOLD coupling . Artefactual BOLD fluctuations caused by head movements driven by stomach contractions seem unlikely . Indeed , gastric-BOLD coupling was neither related to head movement nor to EGG power that increases during stomach contractions . Another possibility is a vascular artifact . During digestion , gastric blood flow does indeed vary ( Matheson et al . , 2000 ) , but cerebral blood flow is unaltered ( Gallavan et al . , 1980 ) . Artificial distension of the stomach can cause increases in peripheral blood pressure ( Min et al . , 2011 ) , but this peripheral increase is mostly due to the insertion of a bag catheter , not to its inflation ( Cantù et al . , 2008 ) . Finally , spontaneous fluctuations in blood pressure in humans occur at approximately 0 . 1 Hz ( so-called Mayer waves ) , which is much faster than gastric rhythm . Thus , a vascular effect seems unlikely , and the hypothesis that activity in the gastric network is driven by neural activity in areas directly receiving ascending inputs appears more plausible . Twenty years and thousands of articles after the discovery of the default network , the debate on its functional role at rest or during tasks is still open . Thus , any discussion of the functional role the gastric network can only be tentative and speculative at this stage . Several non-mutually exclusive interpretations can nevertheless be considered . The functional role of the coupling between stomach and body maps might be related to homeostatic regulations , which would account for the partial overlap between the gastric network and regions involved in heart rate variability . More specifically , the gastric network might be involved in the regulation of digestion , which is accompanied by changes in cardiac output and heart rate ( Kelbaek et al . , 1989 ) . In addition , the unusual experimental setting with abdominal electrodes and a moderate fasting state might have drawn participants' attention toward their internal state , notably of hunger . Since participants had been fasting for only 2 hr , their state of hunger was probably rather moderate and unlikely to have dominated their spontaneous thoughts for 20 min . We find areas containing body maps in the gastric network . This could simply indicate that the stomach , as any other body part , such as for example the hand , is represented in any body map . The body maps of the gastric network are classically associated with different sensory modalities and resting-state networks . In addition to the primary and secondary somatosensory cortices , the gastric network includes MWM ( CCZ , RCZp and SMA ) that are involved in motor preparation and display a clear somatotopic organization ( Amiez and Petrides , 2014; Picard and Strick , 2001 ) . The gastric network also comprises the EBA , a functional region in the occipito-temporal cortex that selectively responds to visual images of the human body ( Downing et al . , 2001; Weiner and Grill-Spector , 2011 ) and is causally involved in body visual recognition ( Urgesi et al . , 2007 ) , with a fine topographical organization ( Orlov et al . , 2010 ) . The stomach , an organ that cannot be easily touched , moved or seen , thus appears to be mapped in body maps related to touching , moving or seeing the body . However , the areas where the stomach is represented are more multi-sensory than usually held . The EBA is not purely visual since it is also activated when participants move or imagine body parts without visual feedback ( Astafiev et al . , 2004 ) , as well as during haptic recognition of body parts ( Kitada et al . , 2009; Costantini et al . , 2011 ) . Primary somatosensory cortex combines internal and external bodily information since it receives both tactile and visceral afferents ( Brüggemann et al . , 1997; Follett and Dirks , 1994 ) . Medial wall motor regions do not only contain motor maps but also receive visceral inputs ( Levinthal and Strick , 2012 ) . Thus , activity in 5 out of the 12 nodes of the gastric network could be simply explained by a representation of the stomach in the brain body maps . However , the gastric network is not limited to body maps , it also comprises regions that play a role in mapping the external space in bodily coordinates , namely , the right superior parieto-occipital sulcus , dorsal precuneus and RSC . The superior parieto-occipital sulcus region is a visuo-motor area that encodes visual stimuli in bodily coordinates during action ( Bernier and Grafton , 2010 ) . The dorsal precuneus and RSC both implement the integration of information into an egocentric reference frame ( i . e . centered on the body ) , a key basic mechanism involved in many different situations ( Burgess et al . , 2001; Vann et al . , 2009 ) , including foraging . All 12 nodes of the gastric network but three ( the posterior cingulate sulcus and ventral and dorsal occipital clusters ) either contain body maps or map external information in bodily coordinates . One could thus speculate that the gastric network coordinates these different body-centered maps . Indeed , the gastric rhythm is continuously produced and originates in the center of the body . In this view , the function of gastric-BOLD coupling in those nine areas would be to co-register body-centered maps of the body and of the external space . In which type of tasks would the gastric network play a role ? Foraging and feeding behaviors are likely candidates , since they involve both the coordination of different egocentric maps and the homeostatic regulation of digestion . Besides , in SI and EBA gastric-BOLD coupling is maximal in the hand and mouth region , suggesting a potential link with the stereotypical actions of feeding behavior , where food goes from hand to mouth , and from mouth to stomach . Still , the coordination of different systems of bodily coordinates is important for many actions besides feeding , such as navigating in the environment or grasping any object . Whether the gastric network plays a role in food-related , but also nonfood-related behaviors , remains to be determined . The gastric network is characterized by temporal fluctuations with delays between the gastric rhythm and brain regions . Delays in resting-state functional connectivity have been highlighted only recently ( Yellin et al . , 2015; Mitra and Raichle , 2016 ) , but have long been documented in stimulus-induced BOLD responses ( Saad et al . , 2001; Kruggel et al . , 1999 ) . Within functionally coherent systems such as the visual ( Saad et al . , 2001 ) or auditory ( Kruggel et al . , 1999 ) systems , delays of 2 s are common . In this light , our finding of up to 3 s delays between areas much further apart thus not appear so surprising . Still , the interpretation of long delays is not straightforward . They are unlikely to directly reflect synaptic delays of fast sequential neural transmission between areas since feed-forward transfer , with only minimal local computations , can be as fast as 10 to 15 ms per processing stage ( Thorpe et al . , 1996 ) . However , if local recurrent processing is involved , longer delays might occur . Delays might additionally reflect regional differences in the timing of the vascular response ( Saad et al . , 2001; Kruggel et al . , 1999 ) , in slow changes of neural activity over time , as in accumulation processes ( Yellin et al . , 2015 ) , or in the involvement of neuromodulatory influences . The different factors may further be combined , that is neuromodulation might affect cerebrovascular reactivity ( Krimer et al . , 1998 ) . What is the directionality of the brain-stomach interactions ? The methods used here do not allow to answer this question since PLV is not a directed measure . If the gastric network plays a role in homeostasis , interactions are likely to be bidirectional , because homeostasis implies both the monitoring of ascending inputs , to evaluate the peripheral state , and the production of descending control commands . Medial wall motor regions , which both receive inputs from the spino-thalamic tract , and generate sympathetic outputs , might fit with this schema . On the other hand , the gastric-locked modulation of the alpha rhythm in the ventral and dorsal occipital clusters was previously shown to be mostly due to ascending influences from the stomach to the brain ( Richter et al . , 2017 ) . RSNs have been defined as segregated systems that show synchronous fluctuations during rest ( Fox and Raichle , 2007 ) . The gastric network , albeit distinct from classical RSNs falls under this definition . In terms of dynamics , the gastric network is defined by its phase synchronization with the stomach and its delayed connectivity between nodes . The gastric network can thus be considered a novel RSN that could not be previously observed due to methodological reasons . As opposed to classical RSNs , the gastric network is characterized by delayed connectivity , with temporal delays that can extend to several seconds but are stable over time and captured by coherence and phase synchrony . Delays are an intrinsic characteristic of brain dynamics unfolding in anatomically connected networks ( Deco et al . , 2011 ) and pervasive even at the timescale of BOLD signal fluctuations ( Mitra et al . , 2015 ) . Canonical RSNs based on instantaneous connectivity represent only one of the possible partitions of the brain into coherent networks based on temporal dynamics . Therefore , we propose the addition of delayed connectivity to the operational definition of RSNs .
We first removed the MRI gradient artefact from the ECG data using the FMRIB plug-in ( Iannetti et al . , 2005; Niazy et al . , 2005 , version 1 . 21 ) for EEGLAB ( Delorme and Makeig , 2004 , version 14 . 1 . 1 ) , provided by the University of Oxford Centre for Functional MRI of the Brain ( FMRIB ) . Data from the three ECG channels was then 1–100 Hz bandpass filtered using a FIR filter , designed with Matlab function firws . We then retrieved the inter-beat-interval ( IBI ) time series by identifying R peaks using a custom semi-automatic algorithm , which combined automatic template matching with manual selection of R peaks for extreme IBIs . This procedure was performed in the ECG channel of each participant that required the least manual identification of R peaks . The resulting IBI time series were then interpolated at 1 Hz using a spline function ( order 3 ) , and band-pass filtered at high ( 0 . 04–0 . 15 Hz ) and low frequencies ( 0 . 15–0 . 4 Hz ) of heart rate variability using a FIR filter ( designed with Matlab function FIR2 , center frequency for LFHRV 0 . 1 Hz ± 0 . 06 Hz , HFHRV centered at 0 . 275 ± 0 . 125 Hz ) and then downsampled at MRI frequency ( 0 . 5 Hz ) . The amplitude envelope of HF- and LF-HRV were then computed using the Hilbert transform and used as regressors of interest ( without convolution with the HRF as in [Critchley et al . , 2003] ) in two separate first level GLMs , which also included six movements parameters as regressors . The MRI pre-processing parameters were the same as for the gastric-BOLD coupling analysis ( slice-timing and motion correction , co-registration to MNI space and spatial smoothing of FWHM = 3 mm ) . The BOLD time series were high-pass filtered ( cutoff: 128 s ) for the GLM analysis . GLM analysis was performed using SPM8 ( Friston et al . , 1994 ) . Contrast images from the first level were entered into two separate second level random-effects analysis to test for consistent effects across the 30 participants separately for HF and LF HRV . The contrast images were spatially smoothed ( FWHM = 8 mm ) , and submitted to a one-sample T-test . Statistical inference was performed at the voxel level , family-wise-error-corrected ( pFWE < 0 . 05 ) for multiple comparisons over the whole brain . Pupil size during blinks and saccades ( as automatically detected by the EyeLink software ) was estimated by interpolating between pupil size 100 ms before and 100 ms after each event . Artefacted windows separated by less than 200 ms were combined and treated as a single epoch . Data from seven participants were excluded due to a high ( >20% ) amount of artefacted data . Data from three participants were excluded because MRI and pupil data could not be synchronized , due to missing triggers . Pupil data from the remaining 20 participants were downsampled at MRI frequency ( 0 . 5 Hz ) , bandpassed filtered ( 0 . 0078–0 . 1 Hz ) using a butterworth infinite impulse response filter and used as a regressor ( convolved with the canonical HRF as in [Yellin et al . , 2015; Schneider et al . , 2016] ) in a first level GLM , which also included six movement nuissance regressors . The BOLD time series were high-pass filtered ( cutoff: 128 s ) for the GLM analysis ( SPM8 ) . Contrast images from the first level were entered into a second level random-effects analysis to test for consistent effects of pupil size across the 20 participants . The contrast images were spatially smoothed ( FWHM = 8 mm ) , and submitted to a one-sample T-test . Statistical inference was performed at the voxel level , ( p<0 . 001 , uncorrected for multiple comparisons ) . Bayesian statistics on correlation coefficients were computed and interpreted according to ( Wetzels and Wagenmakers , 2012 ) and ( Kass and Raftery , 1995 ) , and Bayesian statistics on two sample ( unpaired ) comparisons according to ( Rouder and Morey , 2011 ) . Regarding the specific test of an absence of effect of voxel motion susceptibility on coupling strength ( H0 ) , submillimeter voxel motion was estimated as in ( Power et al . , 2012 ) , and H1 was modeled as the minimum effect size required to detect a significant difference from zero , given one-sample t-test of 29 degrees of freedom on a normal distribution with a mean of 0 and a standard deviation of 1 . The same method was used to test for the absence of a difference between the EGG peak frequency measured inside and outside the scanner . Functional group-level images were overlaid on a 3D rendering of the MNI template using MRIcroGL ( https://www . nitrc . org/frs/ ? group_id=889 , June 2015 ) . The results from the literature were converted when necessary from Talairach coordinates to MNI coordinates using the nonlinear transform provided by ( http://imaging . mrc-cbu . cam . ac . uk/imaging/MniTalairach ) and visualized using Caret software ( Van Essen et al . , 2001 ) ( http://www . nitrc . org/projects/caret/ , v5 . 65 ) . Overlap of the gastric network with cytoarchitectonic subdivisions of primary and secondary somatosensory cortices was determined with the anatomy toolbox for SPM ( Eickhoff et al . , 2005; Eickhoff et al . , 2007; Eickhoff et al . , 2006 ) . Unthresholded t maps of empirical vs chance PLV comparisons ( intermediate step for Figure 2a ) , mask of significant clusters ( Figure 2a ) , unthresholded and significant mask of HF and LF HRV and pupil diameter ( Figure 3 ) and average phase-locking angle of each significant cluster ( Figure 4b ) are available at Neurovault ( Gorgolewski et al . , 2015 ) at the following address: http://neurovault . org/collections/GMHHGEXA/ The custom code used for this article ( Rebollo , 2018 ) can be accessed online at the following address: https://github . com/irebollo/stomach_brain_Scripts . A copy is archived at https://github . com/elifesciences-publications/stomach_brain_Scripts . | The brain is always active . Even when it is not receiving sensory input , it generates its own spontaneous activity . This activity shapes how we interpret future sensory signals and creates our inner mental world . Moreover , this spontaneous activity is not random . When a healthy volunteer lies inside a brain scanner without performing any task , his or her brain shows predictable patterns of activity . Specific groups of brain regions – often with related roles – become active at the same time as one another . Each set of regions is referred to as a resting state network . Of course , the brain does not operate in isolation from the rest of the body . Our internal organs continuously send signals to the brain via the spinal cord and cranial nerves . Specialized cells in the stomach wall in particular produce a slow rhythmic pattern of electrical activity . Known as the gastric rhythm , this activity helps ensure that the stomach muscles contract at the correct speed for digestion . But the stomach also produces this rhythm even when empty , suggesting that it has other roles too . To find out what these might be , Rebollo et al . placed electrodes on the abdomen of healthy volunteers lying inside brain scanners . By examining the volunteers’ spontaneous brain activity , Rebollo et al . identified a new resting state network that is active in synchrony with the gastric rhythm . The regions within this so-called gastric network are not active at the same time as each other , but instead become active in a specific sequence that is repeated at each gastric cycle . Many of the regions within the gastric network belong to other resting state networks too . Some of the regions help regulate automatic bodily functions such as heart rate , while others process information about the body’s position in space . The existence of the gastric network suggests a link between the automatic regulation of processes such as digestion , and spontaneous brain activity . Future studies could examine whether this link impacts perception and cognition , and whether this link plays a role in disorders where the connection between the digestive system and the brain appears to be altered . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Stomach-brain synchrony reveals a novel, delayed-connectivity resting-state network in humans |
Memory retrieval is considered to have roles in memory enhancement . Recently , memory reconsolidation was suggested to reinforce or integrate new information into reactivated memory . Here , we show that reactivated inhibitory avoidance ( IA ) memory is enhanced through reconsolidation under conditions in which memory extinction is not induced . This memory enhancement is mediated by neurons in the amygdala , hippocampus , and medial prefrontal cortex ( mPFC ) through the simultaneous activation of calcineurin-induced proteasome-dependent protein degradation and cAMP responsive element binding protein-mediated gene expression . Interestingly , the amygdala is required for memory reconsolidation and enhancement , whereas the hippocampus and mPFC are required for only memory enhancement . Furthermore , memory enhancement triggered by retrieval utilizes distinct mechanisms to strengthen IA memory by additional learning that depends only on the amygdala . Our findings indicate that reconsolidation functions to strengthen the original memory and show the dynamic nature of reactivated memory through protein degradation and gene expression in multiple brain regions .
Memory retrieval is not a passive phenomenon . Previous studies have presented evidence that memory retrieval is a dynamic process during which memories can be made stronger , weaker , or their content altered ( Misanin et al . , 1968; Schneider and Sherman , 1968; Lewis , 1979; Mactutus et al . , 1979; Gordon , 1981; Nader et al . , 2000; Nader and Hardt , 2009; Dudai , 2012 ) . Recent studies have shown that reactivated memory becomes labile after retrieval and is re-stabilized through a gene expression-dependent process known as memory reconsolidation . Memory reconsolidation after retrieval may be used to maintain or update long-term memories , reinforcing or integrating new information into them ( Nader et al . , 2000; Dudai , 2002; Tronel et al . , 2005 ) . However , the function of memory reconsolidation still remains unclear; especially , whether memory reconsolidation strengthens the original memory ( Tronson et al . , 2006; Inda et al . , 2011; Pedroso et al . , 2013 ) . Importantly , the reinforcement of traumatic memory after retrieval ( i . e . , re-experience such as flashbacks or nightmares ) may be associated with the development of emotional disorders such as post-traumatic stress disorder ( PTSD ) . In classical Pavlovian fear conditioning paradigms , the reactivation of conditioned fear memory by re-exposure to the conditioned stimulus ( CS ) in the absence of the unconditioned stimulus ( US ) also initiates extinction as a form of new learning that weakens fear memory expression ( i . e . , a new CS-no US inhibitory memory that competes with the original CS-US memory trace ) ( Pavlov , 1927; Rescorla , 2001; Myers and Davis , 2002 ) . Therefore , in the majority of reconsolidation paradigms , reactivation also includes extinction learning , which could confound how reconsolidation functions; the dominance of the original or new memory traces is thought to determine the fate of memory through their competition ( Eisenberg et al . , 2003; Pedreira and Maldonado , 2003; Suzuki et al . , 2004 ) . Therefore , it is necessary to investigate the function of reconsolidation in an experimental condition in which memory extinction is not induced following memory retrieval . To this end , we developed a procedure using an inhibitory avoidance ( IA ) task that can engage reconsolidation in the absence of extinction . In the IA task , mice receive an electrical footshock after they enter a dark ( shock ) compartment from a light ( safe ) compartment and form a memory to avoid the dark compartment . Re-exposure to the light compartment could elicit fear memory without giving the mice the opportunity to acquire extinction of the dark compartment memory . This is because the mice can only learn that the dark compartment does not lead to shock only after they have experienced this event by re-entering the dark compartment . This procedural control allows this paradigm to discriminate the reconsolidation and extinction phases at the time point when the mice enter the dark compartment from the light compartment . In the present study , we established a mouse model in which fear memory is enhanced after retrieval using the IA task and investigated the mechanisms of fear memory enhancement through reconsolidation .
The mice were first placed in the light compartment . A brief electrical footshock was delivered ( Training ) 5 s after they entered the dark compartment . The mice were re-exposed to the light compartment 24 hr later and we assessed their crossover latency to enter the dark compartment ( Reactivation session ) . Immediately after they entered the dark compartment , the mice were returned to their home cage and crossover latency was assessed 48 hr later ( post-reactivation long-term memory test , PR-LTM ) . The control group ( treated with vehicle , VEH ) displayed significantly increased crossover latency in Reactivation ( 398 . 9 ± 48 . 31 s ) compared to Training ( 19 . 7 ± 1 . 68 s ) , indicating that the mice formed and retrieved IA memory ( Figure 1A ) . Interestingly , the VEH group displayed further and significantly increased crossover latency at PR-LTM ( 1553 . 5 ± 193 . 81 s ) compared to Reactivation , suggesting that memory retrieval enhanced IA memory in our experimental condition . This ability of retrieval to enhance memories is consistent with previous work ( Gordon , 1981 ) . 10 . 7554/eLife . 02736 . 003Figure 1 . Memory retrieval can enhance inhibitory avoidance memory in a manner that is blocked by inhibiting protein synthesis . ( A ) Re-exposure to the light compartment until mice entered the dark compartment at 1 d after training . The VEH group showed enhancement of inhibitory avoidance ( IA ) memory ( n = 10 ) . The ANI group showed disruption of reactivated IA memory ( n = 9 ) . ( B ) At 3 d after training , a similar pattern was observed ( VEH , n = 8; ANI , n = 12 ) . ( C ) At 7 d after Reactivation ( VEH , n = 8; ANI , n = 8 ) . ( D ) Re-exposure to the light compartment for 3 min , but not 0 min ( NR ) , led to IA memory enhancement ( 0 min: VEH , n = 8 , ANI , n = 8; 3 min: VEH , n = 8 , ANI , n = 9 ) . ( E ) Positive correlation of crossover latency between the Reactivation and PR-LTM sessions ( n = 96 , r2 = 0 . 646 ) . ( F ) Re-exposure to the dark compartment for 3 min following re-exposure to the light compartment . The VEH group showed long-term extinction of IA memory , while ANI blocked this ( VEH , n = 10; ANI , n = 10 ) . ( G ) CREB-mediated transcription is required for memory reconsolidation in the protocol used in Figure 1A ( WT/VEH , n = 9; WT/TAM , n = 9; CREBIR/VEH , n = 9; CREBIR/TAM , n = 9 ) . ANI: anisomycin; CREB: cAMP responsive element binding protein; CREBIR: inducible CREB repressor ( CREBIR ) transgenic mice; IA: inhibitory avoidance; NR: non-reactivated; PR-LTM: post-reactivation long-term memory test; TAM: tamoxifen; VEH: vehicle; WT: wild-type mice . Error bars , SEM . *p<0 . 05 , **p<0 . 005; paired t test . The results of the statistical analyses are presented in Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 00310 . 7554/eLife . 02736 . 004Figure 1—source data 1 . Summary of statistical analyses with F values . The asterisks indicate a significant difference . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 00410 . 7554/eLife . 02736 . 005Figure 1—figure supplement 1 . Correlational analyses of crossover latency between the Training and Reactivation sessions . The crossover latency of individual mice of all VEH groups used in our current study ( n = 96 ) was compared between Training and Reactivation . No significant positive correlation of crossover latency was observed between Training and Reactivation ( r2 = 0 . 106; p>0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 005 Previous studies have shown that the inhibition of protein synthesis immediately after re-exposure to the light compartment disrupts reconsolidation of IA memory ( Milekic and Alberini , 2002; Milekic et al . , 2007 ) . To this end , the mice received a systemic injection of anisomycin ( ANI ) immediately after Reactivation ( Figure 1A ) . The ANI group showed decreased crossover latency at PR-LTM compared to Reactivation , although the VEH and ANI groups showed comparable crossover latency at Reactivation , suggesting that the inhibition of protein synthesis disrupted IA memory . These results suggest that memory retrieval induces reconsolidation , raising the intriguing hypothesis that memory enhancement may be mediated by reconsolidation . To test whether this enhancement was specific only to these experimental parameters used in the previous experiment , we performed Reactivation or PR-LTM at 3 or 7 d after Training or Reactivation , respectively ( Figure 1B , C ) . Consistent with the results shown in Figure 1A , the VEH and ANI groups displayed enhancement and disruption , respectively , of the reactivated memory at PR-LTM , although both groups displayed comparable crossover latency at Reactivation compared to the VEH and ANI groups , respectively , in Figure 1A ( Figure 1B or Figure 1C vs Figure 1A ) . These results indicate that the enhancement associated with the reconsolidation of fear memory was not transient or specific to the age of the memory . We also examined the effects of a shorter duration of re-exposure to the light compartment on IA memory ( Figure 1D ) . Trained mice were either not exposed to the light compartment ( non-reactivated , NR ) or were re-exposed to the light compartment for 3 min . The VEH and ANI groups in the NR condition displayed comparable crossover latency at PR-LTM ( VEH , 457 . 25 ± 55 . 25 s; ANI , 450 . 38 ± 78 . 26 s ) . In contrast , re-exposure of the VEH and ANI groups to the light compartment for 3 min resulted in the enhancement and disruption , respectively , of the reactivated memory compared to the controls ( 0 min ( no ) re-exposure groups in Reactivation and control group re-exposed 24 hr after training , respectively ) . Taken together , our observations suggest that memory enhancement is associated with memory reconsolidation . To clarify further the effects of Reactivation on the enhancement of IA memory , the crossover latency of individual mice from all VEH groups used in the present study ( n = 96 ) was compared between Reactivation and PR-LTM . We observed a significant positive correlation of crossover latency between Reactivation and PR-LTM ( Figure 1E ) . These observations indicate that longer reactivation of IA memory results in increased enhancement of this memory and strongly support our conclusion that IA memory is enhanced after retrieval . It is important to note that no correlation of crossover latency was observed between Training and Reactivation ( Figure 1—figure supplement 1 ) , indicating that enhancement of IA memory is associated with memory retrieval of this memory . Finally , we examined whether our experimental paradigm dissociates the reconsolidation and extinction phases . To do this , the mice stayed in the dark compartment for 3 min after they entered from the light compartment at Reactivation ( Figure 1F ) . The VEH and ANI groups showed decreased and comparable , respectively , crossover latency compared to Reactivation . These observations indicate that memory reactivation in the dark compartment induces long-term memory extinction that requires new gene expression . Thus , consistent with our hypothesis , the IA task enables us to discriminate between the reconsolidation and extinction phases at the time point when the mice enter the dark compartment from the light compartment . Previous studies indicated that cAMP responsive element binding protein ( CREB ) -mediated transcription is required for the reconsolidation of reactivated fear memory ( Kida et al . , 2002; Mamiya et al . , 2009 ) . Using inducible CREB repressor ( CREBIR ) transgenic mice ( Kida et al . , 2002 ) , we tested what effect inhibition of CREB-mediated transcription would have in the protocol described in Figure 1A ( Figure 1G ) . CREBIR mice and wild-type ( WT ) littermates received a systemic injection of tamoxifen ( TAM ) or VEH to inhibit CREB activity at 6 hr before Reactivation ( Kida et al . , 2002 ) . The CREBIR mice injected with TAM and the other control groups displayed disrupted and enhanced , respectively , IA memory at PR-LTM . These observations suggest that CREB-mediated transcription is required for the reconsolidation/enhancement of IA memory . To characterize the brain systems that contribute to these memory representations , we tried to identify the brain regions in which CREB-mediated gene expression is activated . We measured the expression levels of the immediate-early genes c-fos and Arc , which are CREB-dependent genes , using immunohistochemistry ( Sheng et al . , 1990; Kaczmarek and Robertson , 2002; Kawashima et al . , 2009 ) . The following four groups were used in this experiment: two groups which were trained with a footshock ( Trained [T] groups ) and the remaining two groups which did not receive a footshock ( Untrained [U] groups ) . During Reactivation , the animals were either returned to the light compartment ( T/R and U/R ) or not ( T/NR and U/NR ) . Significant increases in the number of c-fos-positive cells were observed in the lateral ( LA ) and basolateral amygdala ( BA ) , CA1 and CA3 regions of the hippocampus , and prelimbic ( PL ) and infralimbic ( IL ) regions of the medial prefrontal cortex ( mPFC ) , but not in the central amygdala ( CeA ) regions or hippocampal dentate gyrus ( DG ) area , in the T/R group compared to the other control groups ( Figure 2A–D ) . In contrast , there was no increase in c-fos expression in the anterior cingulate cortex ( ACC ) , visual cortex ( VC ) , temporal cortex ( TC ) , perirhinal cortex ( PRh ) , or entorhinal cortex ( EC ) regions of the T/R group ( Figure 2—figure supplement 1 ) . Similarly , significant increases in Arc expression were observed in the amygdala ( LA and BA ) , hippocampus ( CA1 and CA3 ) , and mPFC ( PL and IL ) in the T/R group compared to the other control groups ( Figure 2—figure supplement 2 ) . 10 . 7554/eLife . 02736 . 006Figure 2 . Roles of gene expression in the amygdala , hippocampus , and mPFC in the enhancement of reactivated inhibitory avoidance memory . ( A–D ) c-fos induction when IA memory is enhanced after reactivation . ( A ) Representative immunohistochemical staining of BA , CA3 , and PL c-fos-positive cells from the indicated group . Scale bar , 50 μm . Two groups were trained with a footshock: one group received memory reactivation ( T/R ) , while the other group did not ( T/NR ) . The remaining two groups did not receive a footshock . During the Reactivation session , the animals were returned to the light compartment ( U/R ) or not ( U/NR ) . ( B–D ) c-fos expression in the LA , BA , and CeA regions of the amygdala ( B ) , CA1 , CA3 , and DG regions of the hippocampus ( C ) , and PL and IL of the mPFC ( D ) ( n = 13–21 for each group ) . ( E–G ) Effects of anisomycin micro-infusions immediately after Reactivation in the amygdala ( E , VEH , n = 8 , ANI , n = 9 ) , hippocampus ( F , VEH , n = 10 , ANI , n = 10 ) , and mPFC ( G , VEH , n = 10 , ANI , n = 11 ) . Micro-infusion of ANI into the amygdala blocked IA memory as seen by the reduction in performance between Reactivation and PR-LTM . In contrast , micro-infusion of ANI into the hippocampus or mPFC blocked the enhancement , but not the underlying performance . ( H and I ) Phosphorylation of GluA1 at Ser831 and Ser845 was induced in the amygdala , hippocampus , and mPFC following memory retrieval . ( H ) Representative western blot analysis of the amygdala , hippocampus , and mPFC showing phosphorylated GluA1 and total GluA1 levels . ( I ) Levels of Ser831 and Ser845-phosphorylated GluA1 in the amygdala , hippocampus , and mPFC ( n = 9 per group ) . The levels of Ser831- and Ser845-phosphorylated GluA1 for each group are expressed as the ratio of the U/NR group to the other groups . ANI: anisomycin; BA: basolateral amygdala; CeA: central amygdala; DG: dentate gyrus; IA: inhibitory avoidance; IL: infralimbic region; LA: lateral amygdala; mPFC: medial prefrontal cortex; PL: prelimbic region; PR-LTM: post-reactivation long-term memory test; VEH: vehicle . Error bars , SEM . *p<0 . 05 , compared with the other control groups ( B–D and I ) . **p<0 . 005; paired t test ( E–G ) . The results of the statistical analyses are presented in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 00610 . 7554/eLife . 02736 . 007Figure 2—source data 1 . Summary of statistical analyses with F values ( including the data for the figure supplements ) . The asterisks indicate a significant difference . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 00710 . 7554/eLife . 02736 . 008Figure 2—figure supplement 1 . No c-fos induction in the ACC , VC , TC , PRh , and EC regions of the T/R group when inhibitory avoidance memory is enhanced after Reactivation . ( A ) Representative immunohistochemical staining of anterior cingulate cortex ( ACC ) c-fos-positive cells from the indicated group . Scale bar , 50 μm . ( B ) c-fos expression in the ACC , visual cortex ( VC ) , temporal cortex ( TC ) , perirhinal cortex ( PRh ) , and entorhinal cortex ( EC ) ( n = 13–21 for each group ) . Error bars , SEM . The expression of c-fos in each group is expressed as the ratio of the U/NR group to the other groups . The results of the statistical analyses are presented in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 00810 . 7554/eLife . 02736 . 009Figure 2—figure supplement 2 . Arc induction when inhibitory avoidance memory is enhanced after Reactivation . ( A ) Representative immunohistochemical staining of BA , CA3 , PL , and ACC Arc-positive cells from the indicated group . Scale bar , 50 μm . ( B–E ) Arc expression in the LA , BA , and CeA regions of the amygdala ( B ) , CA1 , CA3 , and DG regions of the hippocampus ( C ) , prelimbic ( PL ) and infralimbic ( IL ) regions of the mPFC ( D ) , and ACC , VC , TC , PRh , and EC ( E ) . Arc induction was observed in the amygdala ( LA and BA ) , hippocampus ( CA1 and CA3 ) , and mPFC ( PL and IL ) , but not in the ACC , VC , TC , PRh , or EC , in the T/R group compared to the other control groups ( n = 8–9 for each group ) . The expression of Arc in each group is expressed as the ratio of the U/NR group to the other groups . ACC: anterior cingulate cortex; BA: basolateral amygdala; CeA: central amygdala; DG: dentate gyrus; EC: entorhinal cortex; IL: infralimbic region; LA: lateral amygdala; mPFC: medial prefrontal cortex; PL: prelimbic region; PRh: perirhinal cortex; TC: temporal cortex; VC: visual cortex . Error bars , SEM . *p<0 . 05 , compared with the other control groups . The results of the statistical analyses are presented in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 00910 . 7554/eLife . 02736 . 010Figure 2—figure supplement 3 . Effects of inhibiting protein synthesis in the hippocampus and mPFC on the enhancement of inhibitory avoidance memory . ( A ) Protein synthesis inhibition in both the hippocampus and the mPFC blocked the enhancement of IA memory , but not the disruption of IA memory . Mice infused with ANI into the hippocampus , mPFC , or hippocampus and mPFC displayed comparable crossover latency between Reactivation and PR-LTM ( hippocampus-VEH/mPFC-VEH , n = 8; hippocampus-ANI/mPFC-VEH , n = 8; hippocampus-VEH/mPFC-ANI , n = 8; hippocampus-ANI/mPFC-ANI , n = 9 ) . ( B and C ) The effect of ANI micro-infusion immediately after Reactivation into the hippocampus ( B , VEH , n = 7 , ANI , n = 8 ) or mPFC ( C , VEH , n = 8 , ANI , n = 7 ) on PR-LTM-1 and PR-LTM-2 ( 48 hr after PR-LTM-1 ) . The ANI groups displayed comparable crossover latency at PR-LTM-1 compared with Reactivation . However , the ANI groups showed significantly increased crossover latency at PR-LTM-2 compared to PR-LTM-1 . ( D–F ) Cannula tip placement from mice infused with each drug for ( A ) , ( B ) , and ( C ) , respectively . Schematic drawing of coronal sections from all micro-infused animals ( hippocampus , 1 . 94 mm posterior to the bregma; mPFC , 1 . 94 mm anterior to the bregma ) . Only mice with needle tips within the boundaries of the hippocampus and mPFC were included in the data analysis . ANI: anisomycin; IA: inhibitory avoidance; mPFC: medial prefrontal cortex; PR-LTM: post-reactivation long-term memory test; VEH: vehicle . Error bars , SEM . *p<0 . 05 , **p<0 . 005; paired t test . The results of the statistical analyses are presented in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 01010 . 7554/eLife . 02736 . 011Figure 2—figure supplement 4 . Time course analysis of the phosphorylation levels of GluA1 at Ser831 and Ser845 following re-exposure to the light compartment . ( A–D ) Time course analysis of the phosphorylation levels of GluA1 at Ser831 and Ser845 in the amygdala , dorsal hippocampus , and medial prefrontal cortex ( mPFC ) of the T/R and U/R groups at 0 , 15 , 30 , 60 , or 180 min after re-exposure to the light compartment . The synaptic membrane fractions were analyzed . ( A ) Representative western blot analysis of the amygdala , hippocampus , and mPFC showing the levels of phosphorylated GluA1 at Ser831 and Ser845 and GluA1 protein at 30 min after re-exposure to the light compartment . ( B–D ) Increases in Ser831 and Ser845 phosphorylation in the amygdala ( B ) , hippocampus ( C ) , and mPFC ( D ) of the T/R groups peaked at 30 min and returned to basal levels by 180 min after re-exposure to the light compartment ( n = 3–7 ) . The levels of phosphorylated GluA1 at Ser831 and Ser845 for the T/R group are expressed as the ratio of the U/NR group at each time point . Error bars , SEM . *p<0 . 05 , compared with the U/R group at each time point . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 011 To understand further the functional roles of induced gene expression in the reconsolidation/enhancement of IA memory , we performed correlational analyses on the number of c-fos-positive cells and/or crossover latency among the brain regions of individual mice in the T/R condition . Interestingly , significant positive correlations of the number of c-fos positive cells were observed between the amygdala ( LA and BA ) , hippocampus ( CA1 and CA3 ) , and mPFC ( PL and IL ) ( Table 1 ) , suggesting that c-fos expression in these regions was activated synchronously in response to memory reactivation . Significant positive correlations were also observed between crossover latency and the number of c-fos-positive cells ( Table 2 ) . Taken together with the finding shown in Figure 1E , our observations suggest that the increase in crossover latency at PR-LTM compared to Reactivation is associated with the synchronized induction of c-fos in the amygdala , hippocampus , and mPFC . 10 . 7554/eLife . 02736 . 012Table 1 . Significant positive correlations of the number of c-fos positive cells among amygdala ( LA and BA ) , hippocampus ( CA1 and CA3 ) and mPFC ( PL and IL ) ( n = 18 ) DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 012PLILACCCA1CA3DGLABACeAVCTCPRhECPL0 . 831*0 . 520*0 . 639*0 . 576*0 . 624*0 . 640*0 . 650*0 . 639*−0 . 1230 . 205−0 . 1340 . 015IL0 . 831*0 . 578*0 . 649*0 . 685*0 . 703*0 . 674*0 . 749*0 . 654*0 . 1560 . 478−0 . 0440 . 105ACC0 . 520*0 . 578*0 . 561*0 . 422*0 . 323*0 . 615*0 . 476*0 . 765*−0 . 0870 . 5570 . 2630 . 227CA10 . 639*0 . 649*0 . 561*0 . 530*0 . 401*0 . 676*0 . 629*0 . 577*0 . 1540 . 536−0 . 1140 . 145CA30 . 576*0 . 685*0 . 422*0 . 530*0 . 589*0 . 426*0 . 457*0 . 557*−0 . 1680 . 7200 . 1940 . 391DG0 . 624*0 . 703*0 . 323*0 . 401*0 . 589*0 . 395*0 . 326*0 . 517*−0 . 2100 . 3360 . 075−0 . 165LA0 . 640*0 . 674*0 . 615*0 . 676*0 . 426*0 . 395*0 . 919*0 . 820*−0 . 1560 . 2940 . 0000 . 090BA0 . 650*0 . 749*0 . 476*0 . 629*0 . 457*0 . 326*0 . 919*0 . 768*0 . 0740 . 074−0 . 1180 . 164CeA0 . 639*0 . 654*0 . 765*0 . 577*0 . 557*0 . 517*0 . 820*0 . 768*−0 . 1870 . 4200 . 1000 . 207VC−0 . 1230 . 156−0 . 0870 . 154−0 . 168−0 . 210−0 . 1560 . 074−0 . 1870 . 073−0 . 2200 . 063TC0 . 2050 . 4780 . 5570 . 5360 . 7200 . 3360 . 2940 . 2960 . 4200 . 0730 . 5830 . 748PRh−0 . 134−0 . 0440 . 263−0 . 1140 . 1940 . 0750 . 000−0 . 1180 . 100−0 . 2200 . 5830 . 579EC0 . 0150 . 1050 . 2270 . 1450 . 391−0 . 1650 . 0900 . 1640 . 2070 . 0630 . 7480 . 579*indicate a significant positive correlation ( p<0 . 05 ) . 10 . 7554/eLife . 02736 . 013Table 2 . Significant positive correlations between crossover latency and the number of c-fos-positive cells after the Reactivation session ( n = 18 ) DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 013RegionPLILACCLABACeACA1CA3DGVCTCPRhECCorrelation coefficient0 . 574*0 . 627*0 . 2020 . 612*0 . 630*0 . 3460 . 646*0 . 503*0 . 3600 . 5710 . 3620 . 0740 . 202*Significant positive correlation . ACC: anterior cingulate cortex; BA: basolateral amygdala; CeA: central amygdala; DG: dentate gyrus; EC: entorhinal cortex; IL: infralimbic region; LA: lateral amygdala; PL: prelimbic region; PRh: perirhinal cortex; TC: temporal cortex; VC: visual cortex . To examine the roles of new gene expression in the amygdala , hippocampus , and mPFC in the reconsolidation/enhancement of IA memory , we examined the effects of protein synthesis inhibition in these brain regions . We performed a similar experiment as in Figure 1A , except that the mice received a micro-infusion of ANI or VEH into the amygdala , hippocampus , or mPFC immediately after Reactivation . Similarly with the results for the systemic injection of ANI ( Figure 1A ) , inhibition of protein synthesis in the amygdala resulted in the disruption of the reactivated IA memory ( Figure 2E ) . Interestingly , protein synthesis inhibition in the hippocampus or mPFC failed to disrupt the reactivated IA memory , but blocked its enhancement ( Figure 2F , G ) ; the ANI groups displayed comparable crossover latency between Reactivation and PR-LTM . It is important to note that protein synthesis inhibition in both the hippocampus and mPFC blocked the enhancement of IA memory , but did not disrupt IA memory; furthermore , mice infused with ANI to the hippocampus or mPFC displayed a comparable enhancement of IA memory at PR-LTM-2 at 48 hr after PR-LTM-1 compared to the VEH group at PR-LTM-1 ( Figure 2—figure supplement 3 ) . Thus , these observations suggest that the inhibition of protein synthesis in the hippocampus or mPFC at Reactivation simply blocks the enhancement of IA memory without modulating it . These results are consistent with a previous finding that protein synthesis in the amygdala , but not the hippocampus , is required for the reconsolidation of IA memory ( Taubenfeld et al . , 2001; Milekic et al . , 2007 ) , but more interestingly , suggest distinct roles for the amygdala and hippocampus/mPFC in the enhancement and reconsolidation of IA memory; new gene expression in the amygdala is required for the reconsolidation and enhancement of IA memory , while new gene expression in the hippocampus and mPFC is required only for its enhancement . The α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor ( AMPAR ) GluA1 subunit undergoes distinct phosphorylation and dephosphorylation following the induction of long-term potentiation ( LTP ) and long-term depression ( LTD ) ( Lee et al . , 2000 ) ; serine 831 ( Ser831 ) and serine 845 ( Ser845 ) are phosphorylated by LTP induction or dephosphorylated after LTD induction , respectively . This phosphorylation is thought to alter the function of AMPAR and contribute to the expression of LTP and LTD . To examine the possibility that synaptic plasticity was changed following Reactivation , the phosphorylation of Ser831 and Ser845 in the synaptic membrane fraction of the amygdala , dorsal hippocampus , and mPFC regions was measured at 30 min after Reactivation . Significant increases in Ser831 and Ser845 phosphorylation were observed in the amygdala , hippocampus , and mPFC regions of the T/R groups compared to the control groups ( Figure 2H , I ) . Furthermore , time course analyses indicated that increases in Ser831 and Ser845 phosphorylation in the T/R groups peaked at 30 min and returned to basal levels by 180 min after Reactivation ( Figure 2—figure supplement 4 ) . Our observations that Ser831 and Ser845 phosphorylation was induced in the amygdala , hippocampus , and mPFC following Reactivation suggest that synaptic plasticity was changed in these regions when IA memory was reconsolidated/enhanced . A previous study showed that proteasome-dependent protein degradation plays critical roles in the destabilization of reactivated contextual fear memory ( Lee et al . , 2008 ) . Therefore , we investigated the roles of this protein degradation process in the brain systems regulating reconsolidation/enhancement . To do this , we compared c-fos induction and Lys48-linked poly-ubiquitin chain ( Ub-Lys48 ) , an activation marker of proteasome-dependent protein degradation , in the amygdala , hippocampus , and mPFC using immunohistochemistry . Consistent with the results shown in Figure 2 , we observed increases in the number of c-fos-positive cells in the LA and BA of the amygdala , CA1 and CA3 of the hippocampus , and PL and IL of the mPFC only in the T/R group ( Figure 3A–F ) . Interestingly , significant increases in the number of Ub-Lys48-positive cells were observed in the same regions of the T/R group where c-fos was induced ( Figure 3A–F ) . Most importantly , more than 70–80% of the c-fos-positive neurons of the T/R group were also Ub-Lys48-positive ( Figure 3D–F ) . These observations indicate that gene expression and proteasome-dependent degradation were induced in the same neurons following IA memory retrieval , suggesting that IA memory is reconsolidated/enhanced through gene expression and protein degradation in the same neurons . 10 . 7554/eLife . 02736 . 014Figure 3 . Roles of proteasome-dependent protein degradation in the amygdala , hippocampus , and mPFC in the enhancement of reactivated inhibitory avoidance memory . ( A–F ) Ub-Lys48 levels were increased following IA memory retrieval . ( A–C ) Representative immunohistochemical staining of BA ( A ) , CA3 ( B ) , and PL ( C ) Ub-Lys48- , c-fos- , and Ub-Lys48/c-fos-positive cells from the indicated mice . Scale bar , 100 μm . ( D–F ) Ub-Lys48 , c-fos , and Ub-Lys48/c-fos expression in the LA and BA regions of the amygdala ( D ) , CA1 and CA3 regions of the hippocampus ( E ) , and PL and IL of the mPFC ( F ) ( n = 5–6 for each group ) . ( G–I ) Effects of inhibition of proteasome-dependent protein degradation by micro-infusion of β-lac with or without ANI immediately after Reactivation into the amygdala ( G ) , hippocampus ( H ) , or mPFC ( I ) on the enhancement of IA memory ( amygdala: VEH , n = 8 , ANI , n = 8 , β-lac , n = 8 , ANI + β-lac , n = 9; hippocampus: VEH , n = 7 , β-lac , n = 8; mPFC , VEH , n = 7 , β-lac , n = 9 ) . Error bars , SEM . *p<0 . 05 , compared with the other control groups in Ub-Lys48- , c-fos- , or Ub-Lys48/c-fos-positive cells , respectively ( D–F ) . ANI: anisomycin; β-lac , clasto-lactacystin β-lactone; BA: basolateral amygdala; IA: inhibitory avoidance; IL: infralimbic region; mPFC: medial prefrontal cortex; PL: prelimbic region; VEH: vehicle . *p<0 . 05 , **p<0 . 005; paired t test ( G–I ) . The results of the statistical analyses are presented in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 01410 . 7554/eLife . 02736 . 015Figure 3—source data 1 . Summary of statistical analyses with F values ( including the data for the figure supplements ) . The asterisks indicate a significant difference . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 01510 . 7554/eLife . 02736 . 016Figure 3—figure supplement 1 . Roles of gene expression and proteasome-dependent protein degradation in the consolidation of inhibitory avoidance memory . ( A–D ) No increase in Ub-Lys48-positive cells when IA memory is consolidated after IA training . ( A ) Representative immunohistochemical staining of BA , CA3 , and PL Ub-Lys48-positive cells from the indicated group . Scale bar , 50 μm . The mice were trained in the presence ( Trained ) or absence ( Untrained ) of a footshock . The mice stayed in their home cage during training ( Home cage ) . Ub-Lys48-positive cells were measured at 90 min after training . ( B–D ) Ub-Lys48 expression in the LA , BA , and CeA regions of the amygdala ( B ) , CA1 , CA3 , and DG regions of the hippocampus ( C ) , and PL and IL of the mPFC ( D ) ( n = 6 for each group ) . Ub-Lys48 expression for each group is expressed as the ratio of the home cage group to the other groups . ( E and F ) Effects of micro-infusion of ANI ( 62 . 5 μg ) or β-lac ( 9 . 6 ng ) into the hippocampus immediately after training on IA memory . ( E ) Protein synthesis inhibition in the hippocampus blocked the consolidation of IA memory . The ANI group displayed significantly less crossover latency at Test compared to the VEH group ( VEH , n = 7 , ANI , n = 7 ) . ( F ) Inhibition of proteasome-dependent protein degradation in the hippocampus did not affect the consolidation of IA memory ( VEH , n = 7 , β-lac , n = 7 ) . ( G and H ) Cannula tip placement in the hippocampus . Schematic drawing of coronal sections from all micro-infused animals ( 1 . 94 mm posterior to the bregma ) . Only mice with needle tips within the boundaries of the hippocampus were included in the data analysis . ANI: anisomycin; β-lac , clasto-lactacystin β-lactone; BA: basolateral amygdala; CeA: central amygdala; DG: dentate gyrus; IA: inhibitory avoidance; LA: lateral amygdala; PL: prelimbic region; VEH: vehicle . Error bars , SEM . *p<0 . 05 . The results of the statistical analyses are presented in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 016 To understand the roles of proteasome-dependent protein degradation in the reconsolidation/enhancement of IA memory , we examined the effects of inhibiting proteasome-dependent protein degradation by a micro-infusion of clasto-lactacystin β-lactone ( β-lac ) into the amygdala , hippocampus , or mPFC ( Figure 3G–I ) . Interestingly , micro-infusions of β-lac into the amygdala , mPFC , or hippocampus blocked the enhancement of IA memory , indicating that activating proteasome-dependent protein degradation in these brain regions is required for the enhancement of IA memory . Furthermore , a micro-infusion of β-lac with ANI into the amygdala prevented the disruption of IA memory by protein synthesis inhibition ( Figure 3G ) , suggesting that protein degradation in the amygdala is required for the destabilization of reactivated IA memory . Taken together , our observations suggest that IA memory is reconsolidated/enhanced in the amygdala through protein degradation and synthesis , whereas protein synthesis and degradation in the hippocampus and mPFC are required for the enhancement of IA memory . A previous study using contextual fear conditioning showed that memory reconsolidation mediates the strengthening of memories following additional training ( re-learning ) via the activation of proteasome-dependent protein degradation and gene expression ( Lee , 2008 ) . To compare the mechanisms of memory enhancement through retrieval and additional training , we examined the mechanisms of IA memory enhancement by additional training in our experimental condition ( Figure 4 ) . We performed a similar experiment as in Figure 1A , except that the mice received a footshock at Reactivation ( Training-2 ) at 5 s after they entered the dark compartment and then received micro-infusions of drugs immediately after Training-2 ( Figure 4A ) . Crossover latency increased dramatically in the VEH group at PR-LTM ( 3940 ± 571 . 5 s ) at 48 hr after Training-2 compared to following Reactivation ( without a footshock in Figure 1A ) , suggesting that additional training enhanced IA memory more than retrieval without a footshock . Similarly with a previous report , inhibition of protein synthesis in the amygdala disrupted IA memory , whereas inhibition of proteasome-dependent protein degradation with or without protein synthesis inhibition blocked the enhancement and disruption induced by ANI , respectively , of IA memory ( Figure 4B ) . These results were consistent with previous findings that protein degradation and synthesis in the amygdala are required for strengthening fear memory mediated by additional training . More interestingly , in contrast to the results shown in Figure 3 , the inhibition of protein synthesis in the hippocampus or mPFC did not affect the enhancement of IA memory by additional training ( Figure 4C , D ) . Taken together , our observations that gene expression in the hippocampus and mPFC is required only for IA memory enhancement following retrieval indicate that these brain regions contribute to the enhancement of fear memory following retrieval , but not additional training , and more importantly , memory enhancement by retrieval utilizes a mechanism distinct from that used for additional learning . 10 . 7554/eLife . 02736 . 017Figure 4 . Effects of inhibiting protein synthesis and degradation in the amygdala , hippocampus , and mPFC on additional training . ( A ) Experimental design . ( B ) Effects of inhibiting protein synthesis and/or degradation immediately after Training-2 in the amygdala ( VEH , n = 8 , ANI , n = 8 , β-lac , n = 8 , ANI/β-lac , n = 9 ) . ( C and D ) Effects of protein synthesis inhibition immediately after Training-2 in the hippocampus ( C , VEH , n = 10 , ANI , n = 10 ) or mPFC ( D , VEH , n = 10 , ANI , n = 10 ) . The VEH group showed a dramatic enhancement of IA memory . Micro-infusion of ANI into the amygdala blocked IA memory as seen by the reduction in performance between Training-2 and PR-LTM . In contrast , micro-infusion of ANI into the hippocampus or mPFC did not affect additional training . ANI: anisomycin; β-lac , clasto-lactacystin β-lactone; IA: inhibitory avoidance; mPFC: medial prefrontal cortex; PR-LTM: post-reactivation long-term memory test; VEH: vehicle . Error bars , SEM . *p<0 . 05 , ***p<0 . 001; paired t test . The results of the statistical analyses are presented in Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 01710 . 7554/eLife . 02736 . 018Figure 4—source data 1 . Summary of statistical analyses with F values . The asterisks indicate a significant difference . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 018 Furthermore , it is important to note that: ( 1 ) protein synthesis inhibition in the hippocampus immediately after Training blocked the consolidation of IA memory ( Taubenfeld et al . , 2001; Zhang et al . , 2011; Figure 3—figure supplement 1 ) ; ( 2 ) this consolidation was not affected by the inhibition of proteasome-dependent protein degradation in the hippocampus ( Figure 3—figure supplement 1 ) ; and ( 3 ) an increase in c-fos ( Zhang et al . , 2011 ) , but not Ub-Lys48 , in the hippocampus was observed when IA memory was consolidated ( Figure 3—figure supplement 1 ) . Taken together , these observations indicate that the reconsolidation/enhancement observed in this study utilize a mechanism distinct from that used for additional learning and consolidation . Previous studies have shown that molecules , such as LVGCCs and CB1 , and protein degradation , which are required for memory extinction , also play critical roles in the destabilization of reactivated memory ( Suzuki et al . , 2008; Lee et al . , 2008 ) . Interestingly , a recent study showed that calcineurin ( CaN ) activation in the hippocampus is required for the extinction of contextual fear memory ( de la Fuente et al . , 2011 ) . Therefore , we examined the roles of CaN in the reconsolidation/enhancement of IA memory . We performed a similar experiment as in Figure 3 , except that the mice received a micro-infusion of the CaN inhibitor FK506 at 5 min before Reactivation . Similar results were observed with those using the proteasome inhibitor β-lac ( Figure 5A–C ) . The micro-infusion of FK506 into the amygdala blocked the disruption of reactivated IA memory by protein synthesis inhibition and enhancement of IA memory ( Figure 5A ) , whereas the infusion of FK506 into the hippocampus or mPFC blocked the enhancement of IA memory ( Figure 5B , C ) . These observations suggest that , similar to protein degradation , CaN activation in the amygdala is required for the destabilization and enhancement of IA memory , while CaN activation in the hippocampus and mPFC is required for the enhancement of IA memory . 10 . 7554/eLife . 02736 . 019Figure 5 . Roles of calcineurin in the amygdala , hippocampus , and mPFC on the enhancement of inhibitory avoidance memory and memory retrieval-induced protein degradation . ( A–C ) Effect of micro-infusion of FK506 with or without ANI before Reactivation into the amygdala ( A ) , hippocampus ( B ) , or mPFC ( C ) on the enhancement of IA memory ( n = 10 for each group ) . ( D–I ) The calcineurin inhibitor FK506 blocked the increase of Ub-Lys48 following IA memory retrieval . ( D–F ) Representative immunohistochemical staining of BA ( D ) , CA3 ( E ) , and PL ( F ) Ub-Lys48- , c-fos- , and Ub-Lys48/c-fos-positive cells from the indicated mice . Scale bar , 100 μm . ( G–I ) Ub-Lys48 , c-fos , and Ub-Lys48/c-fos expression in the LA and BA regions of the amygdala ( G ) , CA1 and CA3 regions of the hippocampus ( H ) , and PL and IL of the mPFC ( I ) ( n = 8–11 for each group ) . ANI: anisomycin; BA: basolateral amygdala; IA: inhibitory avoidance; IL: infralimbic region; LA: lateral amygdala; mPFC: medial prefrontal cortex; PL: prelimbic region . Error bars , SEM . *p<0 . 05 , **p<0 . 005; paired t test ( A–C ) . *p<0 . 05 , compared with the other control groups ( G–I ) . #p<0 . 05 , compared with the U/R VEH group . The results of the statistical analyses are presented in Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 01910 . 7554/eLife . 02736 . 020Figure 5—source data 1 . Summary of statistical analyses with F values ( including the data for the figure supplements ) . The asterisks indicate a significant difference . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 02010 . 7554/eLife . 02736 . 021Figure 5—figure supplement 1 . Roles of calcineurin on the enhancement of inhibitory avoidance memory after retrieval . ( A and B ) Systemic injection of a high dose of FK506 ( 5 mg/kg ) or cyclosporin A ( 5 mg/kg ) blocked the disruption of IA memory by protein synthesis inhibition and enhancement of IA memory ( n = 10–11 for each group ) . *p<0 . 05 , **p<0 . 005; paired t test . ( C–H ) The calcineurin inhibitor cyclosporin A blocked the increase of Ub-Lys48 , but not c-fos , following IA memory retrieval . ( C–E ) Representative immunohistochemical staining of BA ( C ) , CA3 ( D ) , and PL ( E ) Ub-Lys48- , c-fos- , and Ub-Lys48/c-fos-positive cells from the indicated mice . Scale bar , 100 μm . ( F–H ) Ub-Lys48 , c-fos , and Ub-Lys48/c-fos expression in the LA and BA regions of the amygdala ( F ) , CA1 and CA3 regions of the hippocampus ( G ) , and PL and IL of the mPFC ( H ) ( n = 4 for each group ) . The T/R groups treated with or without FK506 showed a comparable increase in the number of c-fos-positive cells compared to the U/R groups ( C–H ) . In contrast , the T/R groups treated with Cyc failed to increase the number of Ub-Lys48-positive cells; only the T/R groups treated with VEH showed significantly more Ub-Lys48-positive cells compared to the other groups ( C–H ) ( n = 4 for each group ) . BA: basolateral amygdala; Cyc: cyclosporin A; FK: FK506; IA: inhibitory avoidance; IL: infralimbic region; LA: lateral amygdala; PL: prelimbic region; VEH: vehicle . Error bars , SEM . *p<0 . 05 , compared with the other control groups ( F–H ) . #p<0 . 05 , compared with the U/R VEH group . The results of the statistical analyses are presented in Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 021 To understand the roles of CaN in the activation of gene expression and proteasome-dependent protein degradation , we examined the effects of CaN inhibition on the number of c-fos- and Ub-Lys48-positive cells following Reactivation , similar to the experiment shown in Figure 3 , except that the mice received a systemic injection of FK506 ( 5 mg/kg ) or VEH at 5 min before Reactivation ( Figure 5D–I ) . It is important to note that , similar to the results of the micro-infusions , the systemic injection of FK506 blocked the disruption of the reactivated IA memory by protein synthesis inhibition and the enhancement of IA memory ( Figure 5—figure supplement 1 ) . The T/R groups treated with or without FK506 showed a comparable increase in the number of c-fos-positive cells compared to the U/R groups ( Figure 5D–I ) , suggesting that CaN activation is not required for c-fos induction . In contrast , the T/R groups treated with FK506 did not show an increase in the number of Ub-Lys48-positive cells; only the T/R groups treated with VEH showed significantly more Ub-Lys48-positive cells than the other groups ( Figure 5D–I ) . Thus , the inhibition of CaN blocked the increase of Ub-Lys48-positive cells , but not c-fos-positive cells , in the amygdala , hippocampus , and mPFC . It is important to note that similar results were observed using another CaN inhibitor , cyclosporine A ( Figure 5—figure supplement 1 ) . These observations suggest that CaN functions as an upstream regulator of proteasome-dependent protein degradation , but not c-fos induction , when reactivated memory is reconsolidated/enhanced .
Retrieval has been thought to have a role in enhancing memory that requires activation of gene expression and signal transduction pathways such as mTOR ( Gordon , 1981; Nader et al . , 2000; Nader and Hardt , 2009; Inda et al . , 2011; Pedroso et al . , 2013 ) . In this study , we developed IA paradigms in which the retrieval of fear memory triggers reconsolidation without inducing memory extinction; the reconsolidation and extinction phases are discriminated at the time point when the mice enter a dark compartment from a light compartment . Using this paradigm , we found that memory retrieval enhances IA memory through reconsolidation . In contrast to the mechanisms for the reinforcement of contextual fear memory by additional learning , the enhancement of IA memory by retrieval required CREB-mediated gene expression and calcineurin-induced proteasome-dependent protein degradation not only in the amygdala but also in the hippocampus and mPFC . Consistently , IA memory retrieval was suggested to induce synaptic plasticity in these brain regions through the phosphorylation of AMPAR . Interestingly , we further found that the amygdala is required for the reconsolidation and enhancement of IA memory , whereas the hippocampus and mPFC are required for the enhancement , but not reconsolidation , of IA memory . These findings suggest that an IA memory is enhanced/reconsolidated through the reactivation of memory circuits consisting of multiple brain regions including the amygdala , hippocampus , and mPFC and that the amygdala plays central and distinct roles from the hippocampus and mPFC in the enhancement/reconsolidation of IA memory . It is important to investigate further the differences in the roles of the hippocampus and mPFC in the enhancement of an IA memory . A previous study showed that gene expression is required for the re-stabilization ( reconsolidation ) of reactivated memory , whereas proteasome-dependent protein degradation is required for its destabilization ( Nader et al . , 2000; Taubenfeld et al . , 2001; Debiec et al . , 2002; Kida et al . , 2002 , Lee et al . , 2008 ) . We found that the amygdaloid neurons reactivated by the retrieval of an IA memory activated proteasome-dependent protein degradation and gene expression , suggesting that these reactivated neurons regulate the destabilization and re-stabilization of IA memory . Importantly , our finding that inhibition of CaN blocked proteasome-dependent protein degradation , but not gene expression , suggests that destabilization and re-stabilization are regulated independently at the early stages of the signal transduction pathways activated after memory retrieval . Furthermore , our finding that the activation of gene expression and proteasome-dependent protein degradation in the hippocampus and mPFC is required for the enhancement , but not reconsolidation , of an IA memory also suggests that the activation of proteasome-dependent protein degradation is not only required for the destabilization of a reactivated memory but also plays additional roles in the modification or alteration of memory without the induction of destabilization . Previous and current studies have shown that CB1 , proteasome-dependent protein degradation , and CaN are required for not only the destabilization of reactivated fear memory but also the extinction of fear memory ( Suzuki et al . , 2008; Lee et al . , 2008; de la Fuente et al . , 2011 ) . These findings suggest that destabilization and extinction , both of which are triggered by memory retrieval , share similar signal transduction cascades . It is important to investigate further and compare the molecular mechanisms that underlie the destabilization and extinction of reactivated fear memory . Understanding such mechanisms will enable the identification of the mechanism by which the fate of a memory is determined , that is , reconsolidation or extinction .
All experiments were conducted according to the Guide for the Care and Use of Laboratory Animals ( Japan Neuroscience Society ) and the Guide for the Tokyo University of Agriculture . All of the animal experiments performed in this study were approved by the Animal Care and Use Committee of Tokyo University of Agriculture ( authorization number: 250013 ) . Male C57BL/6N mice were obtained from Charles River ( Yokohama , Japan ) . Transgenic mice expressing an inducible CREB repressor ( CREBIR mice ) were backcrossed to C57BL/6N mice ( National Institutes of Health ) ( Kida et al . , 2002; Suzuki et al . , 2008; Mamiya et al . , 2009 ) . The mice were housed in cages of five or six animals each , maintained on a 12 hr light/dark cycle , and allowed ad libitum access to food and water . The mice were at least 8 weeks of age when tested . Testing was performed during the light phase of the cycle . All experiments were conducted blind to the treatment condition of the mice . The step-through IA apparatus ( OHARA Pharmaceutical , Tokyo , Japan ) consisted of a box with separate light and dark compartments ( both 15 . 5 × 12 . 5 × 11 . 5 cm ) . The light compartment was illuminated by a fluorescent light ( 2500 lux ) ( Fukushima et al . , 2008; Zhang et al . , 2011 ) . Before the commencement of IA , the mice were handled individually for 2 min each day for 1 week . During the training sessions , each mouse was allowed to habituate to the light compartment for 30 s , and the guillotine door was raised to allow access to the dark compartment . Latency to enter the dark compartment was considered as a measure of acquisition . As soon as the mice had entered the dark compartment , the guillotine door was closed . After 5 s , a footshock ( 0 . 2 mA ) was delivered for a total period of 2 s ( training ) . At 24 hr after the training session , the mice were placed back in the light compartment ( Reactivation ) for a varying length of time ( 0 min , 3 min , or until the mice entered the dark compartment without a footshock ) . Memory was assessed at 48 hr later ( PR-LTM ) as the crossover latency for the mice to enter the dark compartment when replaced in the light compartment , as in Reactivation . For the first experiment , we examined the effect of protein synthesis inhibition after Reactivation ( re-exposure to the light compartment ) . The protein synthesis inhibitor anisomycin ( ANI; Wako , Osaka , Japan ) was dissolved in saline ( pH adjusted to 7 . 0–7 . 4 with NaOH ) . The mice were trained as described above , and at 24 hr or 3 d later they received VEH or ANI ( 150 mg/kg , i . p . ) immediately after re-exposure to the light compartment for 0 min , 3 min , or until they entered the dark compartment without a footshock ( Reactivation ) . At 48 hr or 7 d after the Reactivation session , the mice were once again placed in the light compartment , and crossover latency was assessed . At this dose , ANI inhibits >90% of protein synthesis in the brain during the first 2 hr ( Flood et al . , 1973 ) . To examine the effects of re-exposure to the dark compartment on memory extinction , the mice were re-exposed to the light compartment until they entered the dark compartment . Immediately after the mice had entered the dark compartment , the guillotine door was closed and the mice stayed in the dark compartment for 3 min without a footshock ( Reactivation ) . The mice received VEH or ANI ( 150 mg/kg , i . p . ) immediately after Reactivation . At 48 hr later , crossover latency was assessed again ( PR-LTM ) . To examine the effects of disrupting CREB function on memory reconsolidation , we used transgenic mice that express an inducible CREB repressor ( CREBIR ) in the forebrain , where a dominant-negative CREB protein is fused with the ligand binding domain of a mutant estrogen receptor ( ER ) . Previous studies have shown that the systemic injection of TAM , the artificial ligand for ER , into these transgenic mice inhibits CREB activity in the forebrain ( Kida et al . , 2002 ) . At 24 hr after the training session , CREBIR and WT mice were placed back in the light compartment and then crossover latency was assessed ( Reactivation ) . The mice were administered an intraperitoneal injection of 16 mg/kg 4-hydroxytamoxifen ( TAM; Sigma , MO , USA ) , which was dissolved in 10 ml peanut oil ( Sigma ) or VEH ( a similar volume of peanut oil ) , at 6 hr before retrieval ( Kida et al . , 2002; Suzuki et al . , 2008; Mamiya et al . , 2009 ) . At 48 hr after Reactivation , the mice were once again placed in the light compartment , and crossover latency was assessed ( PR-LTM ) . For the second experiment ( c-fos , Arc , and Ub-Lys48 immunohistochemistry ) , we examined the brain regions that are activated after re-exposure to the light compartment . The mice were divided into four groups: ( 1 ) T/R and T/NR groups; two groups of mice were trained as described above , and at 24 hr later , were or were not re-exposed to the light compartment . The animals were then anesthetized with Nembutal ( 750 mg/kg , i . p . ) at 90 min after Reactivation; ( 2 ) U/R and U/NR groups; two groups received a training session in the absence of footshock , and at 24 hr later , were or were not re-exposed to the light compartment . The animals were then anesthetized , as described above , at 90 min after Reactivation . For the third experiment ( micro-infusion of drugs ) , we examined the effects of the inhibition of protein synthesis , proteasome-dependent protein degradation , and CaN in the amygdala , hippocampus , or mPFC on memory reconsolidation/enhancement . ANI was dissolved in artificial cerebrospinal fluid ( ACSF ) and adjusted to pH 7 . 4 with NaOH . The proteasome inhibitor β-lac ( Sigma ) was dissolved in 2% dimethyl sulfoxide ( DMSO ) in 1 M HCl diluted in ACSF and adjusted to pH 7 . 0–7 . 4 with NaOH ( Lee et al . , 2008 ) . The CaN inhibitor FK506 monohydrate ( FK506; Sigma ) was dissolved in ACSF containing three drops of Tween 80 in 2 . 5 ml of 7 . 5% DMSO and adjusted to pH 7 . 4 with NaOH . The mice were trained as described above , and at 24 hr later , they were placed back in the light compartment ( Reactivation ) . The mice were micro-infused with ANI ( 62 . 5 µg ) , β-lac ( 9 . 6 ng ) , FK506 ( 10 µg ) , or VEH immediately into the various brain regions after ( Figures 2–4 ) or 5 min before ( Figure 5 ) Reactivation . At 48 hr after Reactivation , the mice were once again placed in the light compartment and crossover latency was assessed ( PR-LTM ) . Micro-infusions into the hippocampus and mPFC ( 0 . 5 µl ) were performed at a rate of 0 . 25 μl/min . Micro-infusions into the amygdala ( 0 . 2 µl ) were performed at a rate of 0 . 1 μl/min . The injection cannula was left in place for 2 min after micro-infusion and the mice were then returned to their home cages . For the fourth experiment ( additional training ) , we compared the mechanisms underlying memory enhancement through retrieval and additional training . The mice received a footshock in the reactivation ( Training-2 ) session at 5 s after they entered the dark compartment and then received micro-infusions of drugs immediately after Training-2 . At 48 hr later , crossover latency was assessed ( PR-LTM ) . The dose of locally infused ANI used inhibits 90% of protein synthesis for at least 4 hr ( Rosenblum et al . , 1993 ) . For the fifth experiment ( treatment with the CaN inhibitor ) , we examined the effects of CaN inhibition on the expression of c-fos and Ub-Lys48 following Reactivation . The CaN inhibitor FK506 and cyclosporin A ( Cyc; Wako ) were dissolved in saline containing one drop of Tween 80 in 3 ml of 2 . 5% DMSO and 10% Cremophor EL ( Sigma ) . The mice received a systemic injection of FK506 ( 5 mg/kg ) , Cyc ( 5 mg/kg ) , or VEH at 5 min before Reactivation ( re-exposure to the light compartment ) . For immunohistochemistry , the animals were anesthetized at 90 min after Reactivation , as described above . Another group of mice was assessed for crossover latency at 48 hr after Reactivation ( PR-LTM ) . Immunohistochemistry was performed as described previously ( Mamiya et al . , 2009; Suzuki et al . , 2011; Zhang et al . , 2011 ) . After anesthetization , all mice were perfused with 4% paraformaldehyde . The brains were then removed , fixed overnight , transferred to 30% sucrose , and stored at 4°C . Coronal sections ( 30 μm ) were cut in a cryostat . For c-fos or Arc staining , the sections were washed and preincubated in 3% H2O2 in methanol for 1 hr , followed by incubation in a blocking solution ( phosphate-buffered saline [PBS] plus 1% goat serum albumin , 1 mg/ml bovine serum albumin , and 0 . 05% Triton X-100 ) for 3 hr . Consecutive sections were incubated with a polyclonal rabbit primary antibody for anti-c-fos ( Ab-5; 1:5000; Millipore , MA , USA ) or anti-Arc ( 1:1000; Santa Cruz Biotechnology , TX , USA ) in the blocking solution overnight . Subsequently , the sections were washed with PBS and incubated for 3 hr at room temperature with biotinylated goat anti-rabbit IgG ( SAB-PO Kit; Nichirei Biosciences , Tokyo , Japan ) , followed by 1 hr at room temperature in the streptavidin-biotin-peroxidase complex ( SAB-PO Kit ) . For Ub-Lys48 and/or c-fos staining , free floating sections were treated with 1% H2O2 and then incubated overnight with the rabbit monoclonal anti-Ub-Lys48 antibody ( 1:100; Millipore ) and/or rabbit polyclonal anti-c-fos antibody ( 1:1000; Millipore ) in the blocking solution as described above . The sections were washed with PBS and then incubated with horseradish peroxidase-conjugated donkey anti-rabbit IgG ( 1:500; Jackson ImmunoResearch Laboratories , PA , USA ) for c-fos or biotinylated donkey anti-rabbit IgG ( 1:500; Jackson ImmunoResearch Laboratories ) for Ub-Lys48 for 1 hr at room temperature . Ub-Lys48 signals were amplified and visualized using a VECTASTAIN Elite ABC Kit ( Vector Laboratories , CA , USA ) and Alexa Fluor-conjugated streptavidin ( Invitrogen , OR , USA ) . c-fos signals were amplified with TSA-FCM ( Invitrogen ) . The sections were mounted on slides and coverslipped using mounting medium ( Millipore ) . Structures were defined anatomically according to the atlas of Franklin and Paxinos ( 1997 ) . All immunoreactive neurons were counted by an experimenter blind to the treatment condition . Quantification of c-fos- or Arc-positive cells in sections ( 100 × 100 μm ) of the mPFC ( bregma between 2 . 10 and 1 . 98 mm ) , amygdala ( bregma between −1 . 22 and −1 . 34 mm ) , dorsal hippocampus ( bregma between −1 . 46 and −1 . 82 mm ) , VC ( bregma between −3 . 88 and −4 . 00 ) , ACC ( bregma between 0 . 8 and 1 . 0 mm ) , TC ( bregma between −3 . 88 and −4 . 00 mm ) , PRh ( bregma between −3 . 88 and −4 . 00 mm ) , and EC ( bregma between −3 . 88 and −4 . 00 mm ) was performed using a computerized image analysis system , as described previously ( WinROOF version 5 . 6 software; Mitani Corporation , Fukui , Japan ) ( Frankland et al . , 2006; Suzuki et al . , 2008 , 2011; Mamiya et al . , 2009; Zhang et al . , 2011 ) . Immunoreactive cells were counted bilaterally with a fixed sample window across at least three sections . Fluorescence images were acquired using a confocal microscope FV300 ( Olympus , Tokyo , Japan ) or TCS SP8 ( Leica , Wetzlar , Germany ) . For Ub-Lys48 and/or c-fos staining , confocal 2 μm z-stack images were obtained using LAS AF software ( Leica ) . Equal cutoff thresholds were applied to all slices . We quantified the number of Ub-Lys48- or c-fos-positive cells using a 20× ( for the mPFC and amygdala ) or 40× ( for the hippocampus ) objective . The cells in the field of view within the mPFC and amygdala ( 581 × 581 μm ) and the hippocampus ( 290 × 290 μm ) across at least two sections were counted using WinROOF version 5 . 6 software , as described above . Surgery was performed as described previously ( Frankland et al . , 2006; Suzuki et al . , 2008 , 2011; Mamiya et al . , 2009; Kim et al . , 2011; Zhang et al . , 2011; Nomoto et al . , 2012 ) . Under Nembutal anesthesia and using standard stereotaxic procedures , a stainless steel guide cannula ( 22 gauge ) was implanted into the mPFC , dorsal hippocampus , or amygdala . Stereotaxic coordinates for mPFC , dorsal hippocampus , or amygdala placement based on the brain atlas of Franklin and Paxinos ( 1997 ) were as follows: mPFC ( 2 . 7 mm , ±0 mm , −1 . 6 mm ) , dorsal hippocampus ( −1 . 8 mm , ±1 . 8 mm , −1 . 9 mm ) , or amygdala ( −1 . 3 mm , ±3 . 3 mm , −4 . 4 mm ) . The mice were allowed to recover for at least 1 week after surgery . After this , they were handled for 1 week before the commencement of IA . Only mice with a cannulation tip within the boundaries of the amygdala , hippocampus , or mPFC were included in the data analysis . Cannulation tip placement is shown in Figure 6 . 10 . 7554/eLife . 02736 . 022Figure 6 . Cannula tip placement in the amygdala , hippocampus , and mPFC . ( A–L ) Cannula tip placement from mice infused with each drug shown in Figure 2E ( A ) , Figure 2F ( B ) , Figure 2G ( C ) , Figure 3G ( D ) , Figure 3H ( E ) , Figure 3I ( F ) , Figure 4B ( G ) , Figure 4C ( H ) , Figure 4D ( I ) , Figure 5A ( J ) , Figure 5B ( K ) , and Figure 5C ( L ) . Schematic drawing of coronal sections from all micro-infused animals ( amygdala , 1 . 34 mm posterior to the bregma; hippocampus , 1 . 94 mm posterior to the bregma; mPFC , 1 . 94 mm anterior to the bregma ) . Only mice with needle tips within the boundaries of the amygdala , hippocampus , or mPFC were included in the data analysis . ANI: anisomycin; β-lac: clasto-lactacystin-β-lactone; FK: FK506; VEH: vehicle . DOI: http://dx . doi . org/10 . 7554/eLife . 02736 . 022 Mouse brains were sliced using a Rodent Brain Matrix ( RBM-2000; ASI Instruments , MI , USA ) . The amygdala ( bregma between −1 . 06 and −2 . 06 mm ) , dorsal hippocampus ( bregma between −1 . 06 and −2 . 06 mm ) , and mPFC ( bregma between 2 . 4 and 1 . 4 mm ) regions were punched with a Sample corer ( 0 . 8 mm inner diameter; Muromachi , Tokyo , Japan ) and stored at −80°C . Synaptic membrane fractions were isolated on a discontinuous sucrose gradient , as described previously ( Van den Oever et al . , 2008; Counotte et al . , 2011 ) . Brain tissues were homogenized in ice-cold homogenization buffer ( 4 mM HEPES , pH 7 . 4; 320 mM sucrose ) containing a protease inhibitor mixture ( Complete; Roche Diagnostics , IN , USA ) . The homogenized samples were centrifuged twice at 500×g at 4°C for 2 min to remove nuclei and other debris . The supernatant was centrifuged at 20 , 000×g at 4°C for 30 min . The pellets were suspended in a sodium dodecyl sulfate–polyacrylamide gel loading buffer and analyzed by western blotting . Western blotting , using a rabbit polyclonal anti-GluR1 antibody ( 1:1000; Millipore ) , anti-glutamate receptor 1 phospho-Ser831 antibody ( 1:1000; Millipore ) , or anti-glutamate receptor 1 phospho-Ser845 antibody ( 1:1000; Millipore ) , was performed as described previously ( Hosoda et al . , 2004 ) . Positive antibody binding was visualized using an ImmunoStar LD system ( Wako ) , and protein-transferred PVDF membranes ( Bio-Rad , CA , USA ) were analyzed using the Lumi-imager TM chemiluminescence detection system ( Roche Diagnostics , IN , USA ) . The phosphorylation levels of GluA1 were calculated by normalizing the levels of phosphorylated GluA1 at Ser831 or Ser845 to the total amount of GluA1 ( relative phospho-GluA1 [Ser831 or Ser845]/GluA1 levels ) . One-way analysis of variance ( ANOVA ) followed by post hoc Newman–Keuls and two- or three-way ANOVA followed by post hoc Bonferroni's comparisons were used to analyze the effects of group , genotypes , training , reactivation , and drugs . A repeated ANOVA followed by post hoc Bonferroni's comparisons were used to analyze the effects of drugs and times on crossover latency . A paired t test was used to analyze the differences in the crossover latency within each group between two sessions ( Training vs Reactivation , Reactivation vs PR-LTM , or PR-LTM-1 vs PR-LTM-2 ) . A two-tailed , paired Student's t test was used to analyze the GluA1 phosphorylation levels at each time point . All values in the text and figure legends are means ± standard error of the mean ( SEM ) . | Video cameras allow us to record events as they happen . When we look back at a video clip , what we see is an exact replica of what was originally recorded . We tend to assume that our memories work in a similar manner . However , recent research suggests that our memories may be more malleable than we realize . Once a memory has been reactivated , it goes through a process known as reconsolidation that can make it stronger or weaker , or that can change its content . Now , Fukushima et al . have carried out a series of experiments which shed light on the process of memory reconsolidation . Mice were trained to remember a negative event , and later tested on their memory of this event . Some of the mice were also given a ‘reactivation’ session , during which they were reminded of the original memory . These mice were more fearful of the event during the memory test than those who had not been reminded of it . This suggests that the process of reconsolidating the memory after it had been retrieved had the effect of making the memory stronger . Fukushima et al . then demonstrated that this enhancement depended on the synthesis of proteins in particular regions of the brain . When the mice were given an injection to block protein synthesis immediately after reactivation of the memory , their memory of the negative event was weakened . Crucially , this effect only happened when the injection was given immediately after reactivation of the memory; if the memory had not been reactivated , the injection did not change its strength . Fukushima et al . went on to show that three regions of the brain—the amygdala , the hippocampus , and the medial prefrontal cortex—are involved in memory enhancement . However , only one of them , the amygdala , is involved in the other aspects of reconsolidation . This research could support clinical work by elucidating the potential role of reconsolidation in conditions such as post-traumatic stress disorder . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2014 | Enhancement of fear memory by retrieval through reconsolidation |
Changes to the structure and function of neural networks are thought to underlie the evolutionary adaptation of animal behaviours . Among the many developmental phenomena that generate change programmed cell death ( PCD ) appears to play a key role . We show that cell death occurs continuously throughout insect neurogenesis and happens soon after neurons are born . Mimicking an evolutionary role for increasing cell numbers , we artificially block PCD in the medial neuroblast lineage in Drosophila melanogaster , which results in the production of ‘undead’ neurons with complex arborisations and distinct neurotransmitter identities . Activation of these ‘undead’ neurons and recordings of neural activity in behaving animals demonstrate that they are functional . Focusing on two dipterans which have lost flight during evolution we reveal that reductions in populations of flight interneurons are likely caused by increased cell death during development . Our findings suggest that the evolutionary modulation of death-based patterning could generate novel network configurations .
Nervous systems are exquisitely adapted to the biomechanical and ecological environments in which they operate . How they evolve to be this way is largely unknown . Such changes can occur through modifications in receptor tuning , transmitter/receptor repertoires , neuronal excitability , neuromodulation , structural connectivity , or in the number of neurons within specific regions of the central nervous system ( CNS ) . The differences seen in networks , over an evolutionary timescale , ultimately result from heritable changes in developmental processes ( Horder , 1989 ) . Advancing our knowledge of the mechanisms of neural development using comparative approaches will help us understand how specific elements can be modified , how new ‘circuits’ and behaviours evolve , and will ultimately lead to a better understanding of how nervous systems function ( Ramdya and Benton , 2010 ) . Studies comparing the nervous systems of mammalian species that occupy diverse ecological niches reveal clear differences in the number of cells within homologous brain regions ( Herculano-Houzel et al . , 2014 ) . Such differences have occurred either through expansion or reduction of specific cell populations , through changes in proliferation or apoptotic programmed cell death ( PCD ) during development ( Charvet et al . , 2011 ) . Most studies of nervous system evolution have focused on stem cell identity and the role of differential proliferation dynamics ( Biffar and Stollewerk , 2014; Rakic , 2009; Truman and Ball , 1998 ) . While one recent study has elegantly shown a role for PCD in the evolution of peripheral olfactory sensory neurons in drosophilids and mosquitoes ( Prieto-Godino et al . , 2020 ) , how changes in cell death can modify central circuits still remains an open question . In insects , the number and arrangement of neural progenitor cells that generate central neurons ( termed neuroblasts , NBs ) are highly conserved despite the remarkable diversity of insect body plans and behaviours ( Bate , 1976; Biffar and Stollewerk , 2014; Booker and Truman , 1987; Doe , 1992; Doe and Goodman , 1985; Hartenstein and Campos-Ortega , 1984; Nordlander and Edwards , 1969; Shepherd and Bate , 1990; Tamarelle et al . , 1985; Truman , 1996; Truman and Ball , 1998; Truman and Bate , 1988; Wheeler , 1891 ) . In the ventral nerve cord ( VNC – functionally equivalent to the vertebrate spinal cord ) all but one NBs are arranged in a bilaterally symmetric array across the midline , while an unpaired , single medial neuroblast ( MNB ) stands out in the posterior end of each segment ( Figure 1A , B ) . In the Drosophila embryo a first wave of neurogenesis generates the larval nervous system after which the majority of NBs become quiescent . Following reactivation from quiescence NBs produce neurons throughout larval life until the early pupal stages ( Booker and Truman , 1987; Truman and Bate , 1988 ) . These postembryonic neurons – which make up most of the adult CNS – extend simple neuritic processes into the neuropil and stall until the pupal-adult transition when they grow complex arborisations , synapsing with their target cells ( Truman , 1990 ) . In the VNC , NBs bud off a ganglion mother cell ( GMC ) which undergoes a terminal division to generate two neurons with distinctly different cell fates ( an A cell and a B cell ) . As the A and B cells result from a single division , one cannot be produced without the other . After several rounds of GMC divisions , a lineage produced by a single NB is composed of two half-lineages: ‘hemilineage A’ made up of all the A cells and ‘hemilineage B’ made up of B cells ( Figure 1C ) . Hemilineages act as functional units in adult flies ( Harris et al . , 2015; Lacin et al . , 2019; Lin et al . , 2010; Shepherd et al . , 2016; Shepherd et al . , 2019; Truman et al . , 2010; Truman et al . , 2004 ) . For example , in the MNB lineage , hemilineage A cells mature into GABAergic local interneurons while hemilineage B cells become efferent octopaminergic neurons . Our previous work showed that a common fate of postembryonic neurons is PCD affecting approximately 40% of VNC hemilineages ( Figure 1D , E; Truman et al . , 2010 ) , this is also seen in the brain ( Bertet et al . , 2014; Kumar et al . , 2009; Lin et al . , 2010 ) . The pattern of PCD is stereotypical and targets the same hemilineages across individuals . Taken together , the breadth of PCD suggests it plays a major role in shaping the final makeup of the adult nervous system , while its stereotypy points towards a heritable genetic basis . We therefore propose that changes in neural circuits may result from heritable alterations in the extent and pattern of PCD in hemilineages . To mimic such an evolutionary role for PCD , we use the powerful genetic tools available in Drosophila to block death in one doomed hemilineage . We chose to target the MNB lineage for the following reasons; Its easy-to-locate position made the MNB identifiable in all developing insects described from as early as 1891 by Wheeler , 1891 , and spanning all insect orders from wingless silverfish to locusts , beetles , moths and flies ( Bate , 1976; Biffar and Stollewerk , 2014; Booker and Truman , 1987; Doe , 1992; Doe and Goodman , 1985; Hartenstein and Campos-Ortega , 1984; Shepherd and Bate , 1990; Tamarelle et al . , 1985; Truman and Ball , 1998; Truman and Bate , 1988 ) . The MNB gives rise to two distinct populations of neurons , one GABAergic and one octopaminergic , which are also homologous across insects ( Campbell et al . , 1995; Jia and Siegler , 2002; Lacin et al . , 2019; Pflüger and Stevenson , 2005; Rowell , 1976; Siegler and Pankhaniya , 1997; Siegler et al . , 2001; Siegler et al . , 1991; Stevenson and Spörhase-Eichmann , 1995; Thompson and Siegler , 1991; Witten and Truman , 1998 ) . There appears to be a relationship between cell number and function in these populations . Flying insects have greater numbers of octopaminergic neurons within segments that control wings ( Stevenson and Spörhase-Eichmann , 1995 ) , while grasshoppers have more GABAergic neurons in the fused metathoracic/abdominal ganglia , where they receive auditory input from the abdomen ( Witten and Truman , 1998; Thompson and Siegler , 1991 ) . Alongside differences in numbers of the same cell type between segments and species , numbers of GABAergic and octopaminergic neurons found in one segment are never equal . This is especially intriguing as during development each GABAergic neuron is a sister cell to an octopaminergic neuron , arising from one cell division and are produced in equal number ( see Figure 1C ) . The greater number of GABAergic cells in each segment results from PCD targeting octopaminergic neurons in both grasshoppers ( Jia and Siegler , 2002 ) and fruit flies ( Truman et al . , 2010 ) ( see Figure 1D ) . Pieced together , these data suggest that , at least in part , the evolution of some behaviours can be explained by variation in the number of octopaminergic neurons caused by PCD during MNB development . Octopamine release in the thoracic ganglion has been reported to induce and maintain rhythmic behaviours such as stepping movements and flight muscle contractions in locusts ( Sombati and Hoyle , 1984 ) and walking , wing flicking and hindleg grooming in decapitated fruit flies ( Yellman et al . , 1997 ) . All octopaminergic neurons produce tyramine as well , the precursor of octopamine , and tyramine has also been shown to induce fictive walking and flight in a thoracic preparation in locusts ( Rillich et al . , 2013 ) . Throughout our work , we do not discriminate between the role of tyramine and octopamine release from hemilineage 0B and collectively refer to these neurons as octopaminergic . Consistent with its role in both ( 1 ) walking and ( 2 ) flight , we show that ( 1 ) blocking PCD in the octopaminergic hemilineage produced by the MNB in Drosophila melanogaster results in mature differentiated ‘undead’ neurons that survive into adulthood , elaborate complex arborisations and induce walking when activated; and ( 2 ) PCD may be responsible for reducing hemilineage 0B in the mesothorax of the flightless swift louse Crataerina pallida . Alongside , we propose that PCD may have caused reductions in flight hemilineages within thoracic networks in another true fly , Braula coeca ( the bee louse ) , during the evolution of flightlessness . Additionally , using new tools in D . melanogaster , we demonstrate that PCD takes place in these neurons early , very soon after they are born . We find evidence of this early PCD in primitively wingless firebrats and hippoboscid louseflies suggesting that it is deployed widely . This ‘early’ death is categorically different to the neuronal death described in the majority of studies in insects , that focus on hormonally gated PCDs occurring at moults ( Pinto-Teixeira et al . , 2016 ) . Our work highlights the importance of viewing hemilineages as functional units of neurodevelopment in all insects and shows that their alteration through an early mode of PCD can lead to adaptive changes in central circuits during evolution .
First , we wondered what specific type of PCD is responsible for sculpting VNC lineages in D . melanogaster , reasoning that only by gaining insight into the exact developmental process involved can we understand its role in nervous system evolution . The majority of studies on neuronal PCD in insects have focused on its role at metamorphic transitions , where death eliminates fully differentiated neurons either at puparium formation ( Truman et al . , 1994 ) or in adults post-eclosion ( Draizen et al . , 1999; Kimura and Truman , 1990 ) . Both of these remodelling events are gated by ecdysteroids . However , our previous observations in Drosophila ( Truman et al . , 2010 ) , together with studies in the fly brain ( Nordlander and Edwards , 1968; Kumar et al . , 2009; Lin et al . , 2010; Lovick et al . , 2017 ) , made us consider that hemilineage-specific PCD takes place early , in newly born neurons . So far , the dynamics of cell death has been difficult to evaluate on a cell-by-cell basis within a complex nervous system . To interrogate postembryonic neuronal death , we have built a novel genetically encoded effector caspase probe called SR4VH ( Figure 2A , B ) . SR4VH consists of a membrane-bound red fluorescent protein ( Src::RFP ) and a yellow fluorescent protein with a strong nuclear localisation signal from histone H2B ( Venus::H2B ) separated by four tandem repeats of the amino acid sequence DEVD . When effector caspases cleave the DEVD site , Venus accumulates in the nucleus while RFP remains bound to the cell membrane ( Figure 2B ) . This reporter is similar in design to Apoliner ( Bardet et al . , 2008 ) , but has different subcellular localisation signals as well as four tandem caspase cleavage sites instead of one . We also found that tethering the probe to the membrane with the myristoylation signal from Src means that there is no excess signal accumulation in the Golgi apparatus ( Mukherjee et al . , in preparation ) . The nuclear localisation signal from H2B allows for highly efficient sequestration of cleaved Venus in the nucleus even in late stages of apoptosis , when the nuclear membrane is likely compromised . Using the GAL4/UAS system and the NB driver Worniu-GAL4 , we found we could visualise postembryonic neurogenesis and label up to 20 of the most recently born progeny from a single NB ( this is due to GAL4 and reporter perdurance ) . The number of progeny we can detect at any one time using Worniu-GAL4 varies from 10 to 20 , most likely as a result of differential proliferation rates across lineages . We confirmed that SR4VH is reliable as a reporter for cell death in larvae by analysing its expression pattern in all lineages of postembryonic neurons in the thoracic VNC and comparing it to our previous work on MARCM homozygous mutant clones of the initiator caspase Dronc ( Truman et al . , 2010; Figure 2C , D , E , F and Figure 2—figure supplement 1 ) . We found dying cells associated with lineages in the brain and VNC throughout the whole of postembryonic neurogenesis ( Figure 2C , D , E , F ) , which lasts for 3 . 5 days , from mid-2nd instar ( L2 ) to 12 hr after pupariation . As previously suggested ( Truman et al . , 2010 ) , the time course of PCD indicates that cells die early – very soon after they are born – often before they have even extended a neuritic process . This death appears to be unlike the ‘trophic’ PCD found in vertebrates , where a neuron extends a process , interacts with its target cell and dies in the absence of appropriate survival signals . In support of an early onset of PCD , we were able to see sequential stages of cell death , dependent on the distance from the NB ( Figure 2G , H , I and Figure 2—figure supplement 1B ) . Older cells located further away from the NB appear to be at a more advanced stage of PCD indicated by the complete translocation of Venus from the membrane to the nucleus ( Cell three in Figure 2G , H ) and by the accumulation of RFP-positive dead cell membranes close to the lineage bundle ( arrowheads in Figure 2E , F ) . The number of dying cells within a doomed lineage varied from 1 to 8 , with most lineages containing 1–2 dying cells from a total of 10–20 cells labelled with Worniu-GAL4 ( dying cells/lineage: 1 . 3 ± 1 . 5 given as average ± standard deviation; n = 444 doomed lineages from 5 VNCs ) . From a total of 444 doomed lineages , 243 harboured more than one dying cell , of which 148 displayed a progression of cell death ( Figure 2G , H ) . Truman and Bate , 1988 approximated the cell cycle of an NB to 55 min and that of a GMC to 6 . 5 hr , with 7 GMCs present in a proliferating lineage at all times . Therefore , after subtracting the NB and GMCs from clusters of 10–20 Worniu-GAL4-labelled cells , 2–12 will be neurons which resulted from 1 to 6 divisions , each separated in time by 55 min . This means that PCD was initiated early , at some time between 0 and 5 . 5 hr after neurons were born . To look at death specifically during the development of the MNB lineage ( lineage 0 ) we imaged SR4VH in wandering L3 larvae and used molecular markers to identify members of hemilineage 0A and 0B . The transcription factors Engrailed/Invected ( En/Inv ) are known to be expressed in immature and fully differentiated interneurons of hemilineage A ( Allen et al . , 2020; Lacin et al . , 2019; Lacin et al . , 2014; Truman et al . , 2004 ) . As previously reported , the mature differentiated octopaminergic neurons found in hemilineage B express the transcription factor Vestigial ( Vg ) ( Landgraf et al . , 2003 ) . Here , we find that a small number of immature postembryonic neurons ( about 3–5 ) in close proximity to the MNB also express Vg ( Figure 2J ) . Within these immature neurons the expression of Vg and En are mutually exclusive . Using Worniu-GAL4 to drive SR4VH we found that only the engrailed-negative cells are undergoing apoptosis ( Figure 2K ) , i . e . the same small number of cells that express Vg . Their proximity to the MNB suggests that Vg-positive B cells ( i . e . immature octopaminergic neurons ) undergo an early death , very soon after they are born . After observing the extent of early PCD during development , we wondered if , by reducing PCD , we could generate novel functional expansions of a hemilineage . To explore this , we made use of the powerful genetic tools available in Drosophila to block PCD in the MNB lineage to determine if ‘undead’ cells survive into adulthood , elaborate their neurites and acquire a distinctive neurotransmitter identity . From our previous work ( Truman et al . , 2010 ) , we know that during postembryonic neurogenesis MNB hemilineage A survives , expresses Engrailed ( Truman et al . , 2004 ) and differentiates into GABAergic interneurons ( Lacin et al . , 2019 ) . Because during embryonic development the MNB hemilineage B produces a small number of octopaminergic neurons , we hypothesised that preventing PCD would generate additional octopaminergic neurons in the later postembryonic phase of neurogenesis . In postembryonic nomenclature , all the neurons generated by the MNB are collectively called lineage 0 and therefore we will refer to octopaminergic neurons generated by the MNB as hemilineage 0B . Using the octopaminergic neuron driver , TDC2-GAL4 , we observed a 4- to 9-fold increase in the number of octopaminergic neurons in the thoracic VNC of H99/XR38 adult flies deficient for proapoptotic genes ( hid+/- , grim+/- , rpr-/- and skl+/- ) ( Peterson et al . , 2002; White et al . , 1994 ) compared with wild-type control animals ( Figure 3A , B , C , D ) , ( T1: 20 . 9 ± 2 . 3 , Mann-Whitney U = 0 , p = 0 . 0002; T2: 26 . 3 ± 4 . 4 , Mann-Whitney U = 0 , p=0 . 0004; T3: 27 . 5 ± 3 . 5 , Mann-Whitney U = 0 , p = 0 . 0004; n = 11 each ) . These ‘undead’ neurons also express the vesicular glutamate transporter VGlut ( Figure 3E , F ) , just like wild-type octopaminergic neurons ( Greer et al . , 2005 ) . Ideally , to label and manipulate dying neurons from hemilineage 0B , we require a specific driver line expressed only in the newly born doomed neurons . To test if we could use TDC2-GAL4 to label and manipulate dying neurons from hemilineage 0B during their development , we performed a timeline of expression in wild-type and H99/XR38 flies ( Figure 3—figure supplement 1 ) . Unfortunately , even though undead neurons are generated from L2 onwards in H99/XR38 flies , the TDC2-GAL4 is only active in the undead cells days later . Gradually , in pupae , TDC2-GAL4 expression reveals the remaining undead B cells ( Figure 3—figure supplement 1C , D ) . We concluded that the TDC2 driver line cannot be used to visualise and manipulate newly born postembryonic ‘doomed cells’ . Instead , TDC2-GAL4 allowed us to accurately reveal ‘undead’ hemilineage 0B neurons but only in the adult ( see cartoon Figure 3G ) . To ensure sparse labelling and the precise manipulation of only doomed cells from hemilineage 0B , we generated postembryonic TDC2-GAL4-expressing MARCM clones homozygous for the loss-of-function allele DroncΔA8 ( in which PCD is inhibited ) ( Kondo et al . , 2006; Truman et al . , 2010 ) . This strategy guarantees that , even though cell death can be rescued in other lineages , it is only within the TDC2-positive postembryonic neurons that UAS-based tools are expressed . Analysis of the projection patterns of undead neurons revealed that they display both common and distinct features compared to their wild-type embryonically born counterparts . Similar to wild-type octopaminergic cells ( Monastirioti et al . , 1995 ) , the undead neurons have cell bodies located ventrally at the midline , at the posterior border of the thoracic segment ( Figure 3H , I , J ) , project a primary neurite in the dorsal-most region of the neuropil , the tectulum ( Court et al . , 2020 ) and join thoracic nerves ( Pauls et al . , 2018; Figure 3H , I , J , Figure 3—figure supplement 2 and 3 ) . Unlike wild-type cells which bifurcate and branch extensively in the tectulum , the primary neurite of undead neurons fails to bifurcate , branches in both dorsal and ventral regions of the neuropil and sends projections to neighbouring segments ( Figure 3J and Figure 3—figure supplement 3 ) . As we describe in Figure 3—figure supplements 1 and 2 , a few wild-type octopaminergic neurons are produced in all thoracic segments during postembryonic neurogenesis in lineage 0 . We propose that the very few bilateral projecting neurons we encounter in our clones are wild-type cells ( Figure 3—figure supplement 2E , F , G ) . To avoid any uncertainty when performing our behavioural experiments ( below ) , we excluded flies which contained a bifurcating neuron in undead MARCM clones . Thus , using MARCM clonal approaches we show that undead neurons in hemilineage 0B become octopaminergic , elaborate complex neurites and join thoracic nerves . We next asked if these differentiated undead neurons are functional . To address this , we tested if activating undead neurons with the warm temperature-gated ion channel TrpA1 in headless adult Drosophila could elicit behaviours ( Figure 4 and Video 1 ) . For this purpose , we deployed the same MARCM-based technique detailed above which ensured that only postembryonic octopaminergic neurons expressed CD8::GFP and TrpA1 . The stochastic nature of MARCM allowed for generating both controls and flies with undead neurons in one mating cross using the same genotype , rearing and heat-shock conditions , i . e . there would be animals that would have experienced the heat- shock but have no octopaminergic neurons labelled . This further meant that behavioural experiments were performed blindly and each fly was matched to its control or undead neuron group only following dissection and imaging of the VNC . As mentioned above , we excluded flies with MARCM clones containing bilaterally symmetric neurons from our analysis , as these may be wild-type ( see Figure 3—figure supplement 1 and Figure 3—figure supplement 2E , F , G ) . Additionally , we examined the effects of heat exposure in negative control flies with the genotype UAS-TrpA1 and the effects of heat-activation of wild-type octopaminergic neurons in positive controls expressing UAS-TrpA1 driven by TDC2-GAL4 . In TDC2-GAL4 positive controls ( expressing TrpA1 in wild-type embryonic-born octopaminergic neurons ) , thermogenetic stimulation induced long bouts of locomotion ( Figure 4C , D and Video 1 ) in 13/18 flies . Importantly , we found UAS-TrpA1 negative controls ( i . e . an absence of a GAL4 ) and MARCM control flies ( containing no GAL4-positive clones; see Figure 3—figure supplement 2A ) did not walk in response to temperature elevation ( Figure 4B , C , D and Video 1 ) ( 1/19 negative control , 2/22 MARCM control ) . We found that the activation of undead neurons expressing TrpA1 caused decapitated males to walk in 9/17 samples ( Figure 4B , C , D , E and Video 1; also see Figure 3—figure supplement 2B , C , D for examples of MARCM clones of undead neurons ) ( negative control versus positive control , χ2 = 17 . 6 , p = 0 . 0002; negative control versus MARCM control , χ2 = 0 . 2 p = 6; negative control versus MARCM undead neurons , χ2 = 10 . 2 , p = 0 . 0135; MARCM control versus positive control , χ2 = 16 . 8 , p = 0 . 0002; MARCM control versus MARCM undead neurons , χ2 = 9 . 1 , p = 0 . 0242; MARCM undead neurons versus positive control , χ2 = 1 . 4 , p = 1 . 4283; All comparisons were performed using Pearson’s chi-squared and p values were adjusted using a Bonferroni correction ) . A further analysis of the occurrence of walking after splitting the MARCM undead neuron group into the three anatomical subgroups T1 , T2 and T3 according to the location of undead neurons in the pro- , meso- or metathoracic segment , yielded no significant differences: 4/6 T1 , 3/4 T2 and 2/7 T3 ( χ2 = 2 . 9 , p = 0 . 262855 , Pearson’s chi-squared ) ( Figure 4E ) . As previously reported , decapitated flies walked slowly by moving their limbs in a seemingly erratic manner , without having a tripod gait ( Harris et al . , 2015; Yellman et al . , 1997 ) . Flies were considered to be walking if they covered a distance of at least one body length during recordings and if they moved their legs in the order T3-T2-T1 at least once on each side , as evidence of intersegmental coordination ( Strauss and Heisenberg , 1990 ) . The direction of walking was either forward , sideways or backwards and most flies turned or walked in circles ( Figure 4C ) , probably caused by variation in step size ( Yellman et al . , 1997; Harris et al . , 2015 ) . Our data are consistent with the observation that octopamine applied to the exposed anterior notum of decapitated flies causes walking ( Yellman et al . , 1997 ) and suggests that undead neurons are functional and capable of releasing neurotransmitters in the CNS . The extent of walking was greater in positive control flies than in the undead MARCM condition ( compare panels in Figure 4C and Video 1 ) . This is likely because , alongside activating thoracic octopaminergic neurons in the VNC , in the positive controls we also stimulate the severed axons of octopaminergic cells in the brain which send descending projections to the VNC . These are not present in our TDC2-GAL4 MARCM flies . To determine if undead neurons are integrated into thoracic motor circuits , we recorded the activity of mixed undead and wild-type octopaminergic neuron populations expressing GCaMP6s ( an activity reporter ) and tdTomato ( an anatomical fiduciary ) in intact H99/XR38 flies during tethered behaviour on a spherical treadmill ( Chen et al . , 2018; Figure 5A , B ) . The complexity of our calcium imaging experiment ( see Figure 5A ) , together with a rate of success for obtaining MARCM clones with undead neurons of 15% , prompted us to approach our question whether undead neurons are functional during natural walking by using H99/XR38 flies . Keeping in mind that , instead of bifurcating , undead neurons collectively take a turn ( see Figure 3I , J and Figure 3—figure supplement 2B , C , D ) , we interpret activity from the thickest bundle in the bifurcation as belonging to both undead and wild-type neurons . In these animals we observed conspicuous increases in neural activity during air-puff-induced walking in both wild-type controls ( Figure 5—figure supplement 1 ) and in H99/XR38 flies ( Figure 5C , D , E and Videos 2 , 3 and 4 ) . Because undead neurons outnumber their wild-type counterparts by a ratio of 6 . 5 to 1 ( see Figure 3A , B , C , D ) , these results imply that both neuronal types are active in H99/XR38 flies . Supporting this , we observed an increase in GCaMP6s fluorescence across all subregions along the width of the thickest primary neurite bundle in the bifurcation , which contains all undead neurons together with three wild-type cells ( Figure 5—figure supplement 2 and Video 4 ) . Taken together , these data reveal that ‘undead’ neurons in the adult fly are functional and can integrate into motor networks . Our observation that undead neurons functionally integrate into the CNS of adult flies strongly supports the possibility that PCD could be leveraged to modify neural circuits over the course of evolution . Alongside walking , octopaminergic neurons in the MNB lineage have a well-known function in flight ( Roeder , 2005 ) , and differences in neuron numbers correlate well with varying degrees of flight performance ( Figure 6 ) . Proficient fliers such as locusts and most flies have more octopaminergic neurons in winged thoracic segments ( highlighted in yellow in Figure 6 ) , while clumsy fliers such as cockroaches and crickets have similar numbers of neurons across thoracic segments . Figure 6 reviews our current knowledge of the number of GABAergic ( hemilineage A ) and octopaminergic ( hemilineage B ) neurons in the MNB of insects compiled from multiple studies spanning decades of research ( see references in Figure 6 ) . The lack of data for one population or the other is indicated with a question mark . Alongside our own data from swift louseflies ( see Figure 7 ) , here we also include our unpublished observations in the horse lousefly ( for 0B ) and the bee lousefly ( for 0A ) . Having limited samples and antibody , we successfully labelled one preparation each and therefore resort to depicting these as a cartoon in Figure 6 . Because octopaminergic population size reflects flight ability across insects , and dipterans generally display larger populations in the mesothorax , we wondered if flies which have lost flight during evolution show reduced numbers of octopaminergic neurons in this segment . To this end , we described the MNB lineage in the flightless dipteran Crataerina pallida , the swift lousefly ( Figure 7A ) , a viviparous haematophagous ectoparasite of the swift Apus apus ( Bequaert , 1952; Hagan , 1951; Hutson , 1984; Walker and Rotherham , 2010a; Walker and Rotherham , 2010b ) . We labelled lousefly octopaminergic neurons from hemilineage 0B using antibodies for tyramine β-hydroxylase ( Monastirioti et al . , 1996 ) and by comparing the ratio of octopaminergic cells in the mesothorax ( winged segment ) and the prothorax ( lacks wings ) , we found that , unlike flying dipterans , the swift lousefly has lost segment-specific variability of cell numbers ( fruit fly [2 . 2 ± 0 . 2 , n = 7] versus swift lousefly [1 . 1 ± 0 . 2 , n = 3] , p=0 . 012 , Mann-Whitney U = 0 , Mann- Whitney t-test ) ( Figure 7B , C , D , E , F ) . In the sister hemilineage 0A we found a considerably larger number of GABAergic neurons ( Figure 7F ) , suggesting that PCD may be responsible for the selective elimination of the octopaminergic hemilineage . We next wondered if we can find evidence of early PCD in hemilineages in the swift lousefly . Swift louseflies are viviparous , with only one progeny being produced and carried by the female at any one time ( see Figure 7A , lower panel ) . Larvae hatch and remain inside the uterus for their entire larval life , feeding on lipid-rich secretions from milk glands until pupariation , when they are deposited in the swift nests and pupate . Adults only emerge the following spring , when swifts return from North Africa ( Hagan , 1951; Bequaert , 1952 ) . To capture postembryonic development , we dissected both larvae from inside female abdomens and pupae staged from Day 0 onwards , indicating days passed since laying ( Figure 7G ) . Similar to the tsetse fly ( Truman , 1990 ) , neurodevelopment is significantly delayed compared to ‘typical’ dipteran flies - the nervous system only acquires dipteran larval-like features many days after pupariation ( Figure 7H ) . Using EdU to label proliferating cells and immunostaining for active Dcp-1 ( Figure 7H , I , J ) , we found dying cells located close to NBs throughout the 24 days of pupal neurogenesis ( Pop et al . , in preparation ) . In the MNB lineage , which is easily identified by its medial position and projection pattern in the neuropil , we found cell death in thoracic segments at all time points examined , from Day 4 after pupariation to Day 23 ( Figure 7I , J ) . Because cell death is present during neurogenesis in winged insects: selectively eliminating hemilineages in the fruit fly ( Truman et al . , 2010 ) , killing off immature octopaminergic neurons produced by the MNB in the grasshopper ( Jia and Siegler , 2002 ) , appearing to sculpt neural networks in swift louseflies ( see Figure 7H , I , J ) ; we wondered if PCD also occurs during CNS development in a ‘primitive’ wingless insect . Using TUNEL labelling in the firebrat Thermobia domestica ( Figure 8A , B ) , we found dying cells close to many NBs in all thoracic neuromeres at 50–55% of embryonic development ( Figure 8C , D , E ) . Similar to what we see in Drosophila ( see Figure 2 and Figure 2—figure supplement 1 ) , our observation that dying cells are found close to NBs in firebrats and louseflies suggests that this early PCD may be a universal and ancestral feature that sculpts the nervous system of all insects . To further explore if changes in PCD may have been deployed during evolution to accommodate adaptive modifications to behaviour , we next searched for evidence of increased PCD in other flight hemilineages of flightless dipterans . To explore the possibility that changes in the pattern and/or extent of PCD are adaptive , we looked for evidence of evolutionary modifications in the VNC of yet another species , the bee lousefly Braula coeca ( Figure 9A , B , C , D ) . Braula , a close relative of drosophilids , is wingless , lacks halteres and has an extremely reduced thorax ( Figure 9A ) . Bee louseflies spend their entire adult life as kleptoparasites on the honeybee Apis mellifera ( Imms , 1942; McAlister , 2018 ) . We specifically asked whether lineages known to function in flight circuitry might be modified in flightless insects . In Drosophila , a thorough anatomical study recently described the pattern of innervation for each lineage into known functional domains of the adult neuropil and categorised them accordingly as being involved in leg , wing and both leg and wing control ( Shepherd et al . , 2019 ) . In addition , the functional role of most hemilineages was previously assessed by thermogenetic activation in headless flies and those involved in flight-associated behaviours , such as wing wave , wing buzz or take-off , were identified ( Harris et al . , 2015 ) . Together , these studies provide an excellent starting point for anatomical comparisons of homologous hemilineages which may have served an ancestral role in flight . Using antibodies for Neuroglian , we compared homologous hemilineages involved in the flight circuits of the two flightless dipterans , the swift lousefly Crataerina pallida and the bee lousefly Braula coeca . Comparing the Neuroglian-labelled axon bundle width of homologous hemilineages between flightless and flying species , we found that hemilineages 3B , 7B , 11B and 12A , which produce wing waving , wing buzzing or take-off in the fruit fly ( Harris et al . , 2015 ) , hemilineages 6A and 19B , which innervate the wing neuropil , and 5B , which innervates both leg and wing neuropils ( Shepherd et al . , 2019 ) , are reduced in bee louseflies , but not in swift louseflies ( Figure 9B , C , D ) . Among the latter , hemilineages 5B and 6A are known to contribute to leg movements and changes in posture , while the role for 19B is yet to be determined ( Harris et al . , 2015 ) . Importantly , 3B , 6A , 11B , 12A and 19B belong to lineages in which both hemilineages survive in fruit flies ( see schematic in Figure 1E ) . A difference in axon bundle diameter between sister hemilineages could indicate a difference in cell number , possibly established by PCD during development . We chose to quantify the ratio of axon bundle diameters in lineage three because we expected a reduction in hemilineage 3B , which innervates the wing neuropil , but no change in hemilineage 3A , which projects into the leg neuropil ( Shepherd et al . , 2019 ) . The fibre tracts of 3A and 3B originate as a common bundle and split only in the intermediate neuropil . After they split , the sister fibre tracts sit in the same plane before they defasciculate , making them easy to trace and compare ( Figure 9E , F , G ) . We found that the ratio between sister hemilineages A and B is significantly higher in bee louseflies compared to fruit flies , indicating that hemilineage 3B , which controls flight-related behaviours , is severely reduced in these flightless flies ( T1: fruit fly [0 . 6 ± 0 . 1 , n = 7] versus bee louse [1 . 8 ± 0 . 4 , n = 8] , p < 0 . 0001 , t = −8 . 084 , independent samples t-test; T2: fruit fly [0 . 7 ± 0 . 1 , n = 7] versus bee louse [1 . 7 ± 0 . 4 , n = 8] , p=0 . 0001 , F = 42 . 22 , Welch’s t-test; T3: fruit fly [1 . 1 ± 0 . 4 , n = 7] versus bee louse [1 . 7 ± 0 . 2 , n = 8] , p = 0 . 0044 , F = 14 . 84 , Welch’s t-test ) . Even though we cannot make a precise inference of cell numbers in each hemilineage , as the fibre tract of hemilineage 3B appears frayed while 3A is more compact in both species , our results clearly show that 3B is greatly reduced , associated with the loss of flight machinery .
To help us understand more about the patterning of PCD , we designed and used a new effector caspase probe SR4VH , which allowed us to interrogate the extent and dynamics of hemilineage-specific cell death . It shows us that an early onset PCD is responsible for the elimination of postembryonic neurons in the fly VNC , that this happens throughout the entire 3 . 5 days of postembryonic neurogenesis , and is hemilineage-specific . Although PCD has been reported as a fate within lineages in the embryo ( Karcavich and Doe , 2005; Rogulja-Ortmann et al . , 2007 ) , the impact of this ‘early’ and hemilineage-specific PCD on the construction of the adult network has yet to be fully appreciated . This type of PCD is responsible for removing almost half of all postembryonic neurons that are born in the fly ( Truman et al . , 2010 ) . Until now , the most frequently reported type of neuronal death described in insects has been the hormonally regulated PCD that removes mature neurons during the narrow developmental windows at the beginning of metamorphosis and within the first day after adult eclosion ( Draizen et al . , 1999; Lee et al . , 2013 ) . We now know from our data that these make up only a small fraction , compared to the total number of neuronal deaths in the fly . Our SR4VH probe allows us to see that newly born neurons initiate cell death within the first 5 . 5 hr after birth . We can capture different stages of cell death: with young cells at very initial stages of PCD located closer to the NB , while older cells at more advanced stages of PCD with RFP labelled cell membranes found close to the lineage bundle . Importantly , death happens before neurites have extended , strongly suggesting that this PCD is not an analogue of neurotrophic death , found in vertebrates - where neuron-target interactions play a major role in the decisions of cell survival ( Dekkers and Barde , 2013 ) . In Thermobia domestica and Crataerina pallida , where the use of sophisticated genetic reporters such as SR4VH was not yet possible , EdU incorporation to mark dividing cells and immunolabelling for the active effector caspase Dcp-1 as a proxy for cell death revealed dying cells close to sites of division . Dying cells were found in the proximity of NBs ( identifiable in all insects by their large size and position in the outermost layer of the CNS cortex ) and far removed from mature neurons which congregate in the innermost layer of the cortex , adjacent to the neuropil . Therefore , we speculate that these cells are immature neurons and propose that they too undergo a death with rapid onset following division , similar to the early hemilineage-specific cell death we see with SR4VH in fruit flies . The critical question that these data bring into focus is how early onset PCD is orchestrated , especially that an early intrinsically determined mode of cell death seems to be widespread across animals , from C . elegans to mice ( Fricker et al . , 2018; Southwell et al . , 2012 ) . Early PCD likely involves a combination of intrinsic patterning and cell-cell interactions between sibling neurons which ultimately deploy the activity of proapoptotic genes rpr , hid , grim and/or skl . Previously , patterning genes such as Ubx have been shown to contribute to the survival of hemilineages in the thoracic VNC in a parasegment-specific manner ( Marin et al . , 2012 ) , while the transcription factor Unc-4 has been recently demonstrated to provide neural identity to hemilineages involved in flight ( Lacin et al . , 2020 ) . Such spatial patterning is also required to establish NB identity which , in turn , can determine which hemilineage is maintained and which dies . In the developing optic lobe Bertet et al . , 2014 have shown that the temporal sequence of transcription factors expressed in NBs ( inherited by the GMC and newly born neurons ) produces a switch in the selective survival of one hemilineage over the other . In this manner , the changes to NB identity which have been documented in other insects ( Biffar and Stollewerk , 2014 ) , despite a conserved NB array and progeny , could in turn influence the pattern of PCD . Alongside , cell-cell interactions between newly born neurons could influence fate choices . The requirement of interactions between newly born siblings in determining asymmetric fates has been shown in the grasshopper VNC ( Doe et al . , 1985; Kuwada and Goodman , 1985 ) , although this exact same mechanism has not been demonstrated in Drosophila . The spatio-temporal pattern of death revealed with SR4VH shows us that understanding the molecular control of hemilineage-based death is the key question going forward and is likely to provide insight into how networks evolve . Here we blocked PCD within the MNB lineage ( called lineage 0 in the postembryonic literature ) and found that ‘doomed’ neurons become octopaminergic , generate arborisations and target the tectulum neuropil ( Court et al . , 2020 ) . Previous work has shown that octopamine can induce and maintain walking in locusts ( Sombati and Hoyle , 1984 ) and decapitated fruit flies ( Yellman et al . , 1997 ) . Consistently , we find that our ‘undead’ hemilineage 0B cells can induce walking when activated thermogenetically in headless flies and show calcium activity during naturalistic bouts of locomotion . Thus , by blocking death , we have ‘resurrected’ functional neurons that are able to integrate into thoracic motor networks . Although this ‘dialling up’ of cell numbers , in our system , is artificial , it reveals how doomed neurons possess cryptic cellular phenotypes that can emerge when death is blocked , advocating for the evolvability of such a hemilineage-based system . While these undead hemilineage 0B octopaminergic neurons share many conserved features with wild-type cells , the variations we see in their morphology could act as a substrate for evolutionary change . A recent study has linked structural changes in the VNC with changes in behaviour between strains of Drosophila melanogaster ( Mellert et al . , 2016 ) . Mellert et al . , show that hemilineage 12A in the mesothorax has variable bundle morphologies and that these correlate well with the time of flight initiation . Flight is used as an escape response and can be instrumental for predator evasion , one of the major evolutionary forces which have selected for flight in insects in the first place ( Dudley , 2002 ) . With this in mind , it seems plausible that changes in either neuron number and/or innovations in ‘undead’ neuron structure could affect adult behaviour and be ultimately adaptive . Recently it has been shown that undead sensory neurons that are functional , integrate and appear to be tuned to specific odours ( Prieto-Godino et al . , 2020 ) . Importantly , Prieto-Godino et al . also show that there is cell number variation in this neuronal population across drosophilids and that blocking PCD in melanogaster results in the survival of mosquito-like CO2-sensing neurons in the maxillary palps . This suggests that both the central and peripheral nervous system may use similar modes of early PCD to sculpt circuits during the evolution of true flies . Our data show that PCD is extensive and widespread during neurogenesis in the CNS of insects , from the primitive firebrats Thermobia domestica , to true flies Drosophila melanogaster and the swift lousefly Crataerina pallida . We wondered whether alterations in cell death may have contributed to adaptations in the VNC of flightless dipterans and found that a greater extent of PCD in the MNB lineage may be responsible for abolishing the segment-specific difference in octopaminergic cell numbers in the swift lousefly Crataerina pallida – which has lost flight and gained adaptations to a parasitic lifestyle ( Bequaert , 1952; Hagan , 1951; Hutson , 1984; Lehane , 2008; Petersen et al . , 2018; Walker and Rotherham , 2010b ) . As octopaminergic neurons are involved in flight-related behaviours ( Brembs et al . , 2007; Duch and Pflüger , 1999; Roeder , 2005 ) , we suggest that PCD has been co-opted in this lineage during the evolution of flightlessness in the swift louse . We believe that this PCD takes place early , in newly born neurons , as we have seen dying cells close to the MNB in all of our 20 pupae , from Day 4 to Day 23 . Therefore , the decrease in octopaminergic cell number in the mesothoracic segment of swift louseflies is likely the result of increased hemilineage-specific PCD during evolution . Alternatively , an increase in the number of neurons in the non-flying prothorax and metathorax could lead to a uniform population size across thoracic segments in swift louseflies . This could be achieved either by additional MNB divisions , or by reduced PCD . If changes in MNB proliferation play a role remains to be determined . The midline neurons within the VNC of insects have long been a source of interest because they are homologous across species , yet show a diversity in cell numbers correlated with body form and function ( Lacin et al . , 2019; Stevenson and Spörhase-Eichmann , 1995; Witten and Truman , 1998 ) . For example , flying insects have greater numbers of midline octopaminergic neurons within segments that control wings ( Campbell et al . , 1995; Eckert et al . , 1992; Jia and Siegler , 2002; Konings et al . , 1988; Monastirioti et al . , 1995; Schlurmann and Hausen , 2003; Siegler and Pankhaniya , 1997; Siegler et al . , 2001; Siegler et al . , 1991; Spörhase-Eichmann et al . , 1992; Stevenson et al . , 1992; Thompson and Siegler , 1993 ) , while grasshoppers have more GABAergic neurons in the metathoracic/abdominal ganglia , which receives auditory input ( Witten and Truman , 1998; Thompson and Siegler , 1991 ) . These two midline neuronal populations are derived from the same NB , the MNB , which buds off multiple GMCs , each dividing once to generate a GABAergic ( A cell ) and an octopaminergic ( B cell ) neuron . Both A and B neurons are generated in equal numbers but , in all cases , the numbers of GABAergic ( ‘hemilineage A’ ) and octopaminergic ( ‘hemilineage B’ ) neurons within one segment are never the same . The greater numbers of GABAergic cells within each segment has been shown in grasshoppers ( Jia and Siegler , 2002 ) and fruit flies ( Truman et al . , 2010 ) to be the result of removal by PCD of large numbers of cells from hemilineage B . As suggested by our previous work , the ‘hemilineage’ emerges as a discrete developmental unit that shows common features of gene expression and function ( Harris et al . , 2015; Lacin et al . , 2019; Lacin and Truman , 2016; Shepherd et al . , 2019; Truman et al . , 2010 ) . Therefore , we refer here to the PCD found in the MNB lineage of grasshoppers as ‘hemilineage-specific PCD’ . Following our observations of PCD during development in other insects , together with a vast body of knowledge on homology in insect nervous systems ( Kutsch and Breidbach , 1994; Thomas et al . , 1984 ) , we suggest that variations in neural circuits between species is very likely set up by modifying hemilineages , with PCD playing a major role . In the wingless bee lousefly Braula coeca , we found clear reductions in the thickness of fibre tracts in several hemilineage bundles which in Drosophila are associated with flight-related behaviours ( Harris et al . , 2015 ) . It remains for us to determine if early PCD takes place in these specific bee lice lineages during development and causes the reduction in bundle diameter that we see in flight hemilineages . As with most parasitic insects , bee and swift louseflies are impossible to maintain in the laboratory in the absence of their hosts and procuring them is a challenge ( e . g . bee louseflies are now only found on two islands in the UK , while collecting swift louseflies is restricted to the summer months due to Apus apus migrations ) . Nonetheless , our observations that PCD is widespread across insects complements our findings in bee louseflies , strongly suggesting that an extensive PCD in flight hemilineages may have accompanied the loss of flight during evolution . Interestingly , the reduction in fibre diameter we see in bee louseflies was not evident in swift louseflies . This difference is likely due to the more significant changes to body plan in bee louseflies , i . e . a complete loss of the flight apparatus during evolution . Swift louseflies however still maintain vestigial wings and halteres ( Walker and Rotherham , 2010b ) , whereas bee louseflies have a severely reduced thorax , completely lacking wings , halteres and flight muscles ( McAlister , 2018 ) . Here we have shown that undead neurons elaborate complex arborisations , express distinct transmitter identities and function . We find that ‘early’ PCD is widespread during the development of the CNS of insects from the primitive firebrats , to most derived true flies . Early cell death appears to be a specific subtype of PCD present across animals . Understanding how early PCD is specified across species should help us elucidate how nervous systems are built and evolve . Our exploration of homologous lineages in flightless dipterans shows that changes in body plan may accompany changes in the extent and pattern of PCD . As the evolutionary changes seen in neural networks ultimately result from heritable differences in developmental processes , our future endeavours will be directed towards elucidating how genetic programs are deployed to establish the pattern of PCD . The cellular leitmotif of hemilineage-based cell death , we present here , provides us with something tangible that we can search for . Thus , we suggest that viewing the evolution of insect nervous systems through the lens of the ‘hemilineage’ will be critical for understanding how development brings about adaptive changes in neural network motifs .
We used the following Drosophila melanogaster stocks: Worniu-GAL4; Dr/TM3 , Ubx-LacZ , Sb ( BDSC_56553 ) , TDC2-GAL4 ( BDSC_9313 ) , OK371-GAL4 ( BDSC_26160 ) , UAS-SR4VH ( described here ) , UAS-CD8::GFP ( BDSC_5137 ) , UAS-tdTomato-p2A-GCaMP6s ( Chen et al . , 2018 ) ( kind gift from M . Dickinson ) , H99/TM3 , Sb ( BDSC_1576 ) , XR38/TM3 ( Peterson et al . , 2002 ) , Sb , If/CyO; dronc∆A8 , FRT2A/TM6β , Tb , Hu ( Kondo et al . , 2006 ) and hs-flp;; TubP-GAL80 , FRT2A/TM3 , Sb ( Truman et al . , 2010 ) . Firebrat adults of Thermobia domestica were obtained from Buzzard Reptile and Aquatics ( buzzardreptile . co . uk ) and reared on a diet of fish flakes and wholemeal bran at 40°C in darkness inside a humid plastic container . A staging series was calculated by time to hatching . Crataerina pallida swift lousefly adults were collected from swift ( Apus apus ) nesting boxes fitted behind the louvres of belfry windows from churches in Cambridgeshire and Suffolk ( UK ) with the help of local conservationists Simon Evans , Richard Newell and Bill Murrells . Swift louseflies were kept at 20°C on a 12 hr dark:12 hr light cycle until dissected . Pregnant females , recognised by their enlarged and translucent abdomen through which larvae or prepupae could be detected , were kept separately and checked daily for pupa ejection . The day in which a pupa was laid was defined as ‘Day 0’ of external development ( outside the mother’s abdomen ) . Braula coeca bee lousefly adults were obtained from a black bee ( Apis mellifera mellifera ) colony on the Isle of Colonsay , UK ( kind gift from A . Abrahams ) . Bee louseflies were shipped by post in small cages containing worker bees feeding on bee fondant . The black bees and bee louseflies were anaesthetised by placing the cage on a CO2 pad and the bee louseflies were removed for dissection . SR4VH was constructed by standard molecular biology procedures . It comprises the myristylation signal of Drosophila Src64B ( amino acids 1–95 ) , a monomeric red fluorescent protein mRFP1 ( Campbell et al . , 2002 ) , a linker that contains four DEVD sites , a yellow florescent protein Venus ( Nagai et al . , 2002 ) , and a nuclear localisation signal of Drosophila histone H2B ( amino acids 1–51 ) . While the design is similar to the previously reported caspase probe Apoliner ( Bardet et al . , 2008 ) , the Src64B myristylation signal and the H2B NLS offers better membrane and nuclear localisation , respectively , and four DEVD sites are expected to provide higher sensitivity . The probe was cloned in pUAST ( Brand and Perrimon , 1993 ) and introduced into the Drosophila genome by P element-mediated transformation . Drosophila larvae were dissected in PBS without anaesthesia . Firebrat embryos were removed from their chorion and dissected using minuten pins . Drosophila , swift lousefly and bee lousefly adults were anaesthetised on ice , briefly submerged in absolute ethanol and dissected in PBS . Swift lousefly pupae were immobilised on double sided sticky tape , removed from their pupal case using forceps and dissected in PBS without anaesthesia . Samples were fixed in 3 . 6% paraformaldehyde in PBS for 30 min ( larvae and pupae ) or 1 hr ( adults ) , washed 3 times in 0 . 3% PBST ( 0 . 3% Triton-X100 in PBS , Sigma-Aldrich ) , blocked in 5% goat serum ( Sigma-Aldrich ) in PBST for 1 hr and incubated with primary antibodies in block for 1–3 days at 4°C ( Drosophila , bee louseflies , swift lousefly pupae ) , room temperature ( firebrats ) or 37°C to increase antibody penetration ( swift lousefly adults; block supplemented with 0 . 02% NaN3 to prevent microbial growth ) . Samples were then washed four times throughout the day in PBST and incubated with secondary antibodies in block for a further 1–3 days , followed by final washes in PBST and PBS . Brains and VNCs were mounted on poly-L-lysine-coated coverslip , dehydrated in increasing serial concentrations of ethanol ( 15% , 30% , 70% , 80% , 90% and twice in 100% ) for 5 min each , dipped once in xylene , then incubated twice for 5 min in fresh xylene . A droplet of DePeX ( EMS ) was added on top of the mounted sample and the coverslip was placed face-down on a glass slide . We used the following primary antibodies: chicken anti-GFP ( 1:500; ab13970 , Abcam ) , mouse anti-Neuroglian ( 1:50; BP 104 , Developmental Studies Hybridoma Bank ) , rabbit anti-cleaved Drosophila Dcp1 ( 1:100; 9578 , Cell Signaling ) , guinea pig anti-Syncrip ( 1:100; kind gift from I . Davis; to label NBs and early progeny in lineages - JW Truman , personal communication , January 2019 ) , mouse anti-Engrailed/Invected ( 1:2; 4D9 , Developmental Studies Hybridoma Bank ) , rabbit anti-DVGLUT C-terminus ( Mahr and Aberle , 2006 ) ( 1:5000; AB_2490071 , kind gift from H . Aberle ) , rat anti-tyramine β-hydroxylase ( Monastirioti et al . , 1996 ) ( 1:50; AB_2315520 , kind gift from M . Monastirioti ) , rabbit anti-vestigial ( 1:400; kind gift from Sean Carroll and Kirsten Gruss ) and rabbit anti-GABA ( 1:100; 20094 , ImunoStar ) . Secondary antibodies were Alexa Fluor 488-conjugated goat anti-chicken ( 1:500; A11039 , Invitrogen , Thermo Fisher Scientific ) , Alexa Fluor 488-conjugated goat anti-rabbit ( 1:500; A11070 , Invitrogen , Thermo Fisher Scientific ) , Cy3-conjugated donkey anti-rabbit ( 1:500; 711-006-152 , Jackson ImmunoResearch ) , Cy5-conjugated donkey anti-mouse ( 1:500; 715-006-151 , Jackson ImmunoResearch ) , Alexa Fluor 488-conjugated donkey anti-rat cross-adsorbed against mouse ( 1:100; 712-545-153 , Jackson ImmunoResearch ) , Alexa Fluor 488-conjugated donkey anti-guinea pig ( 1:500; A11073 , Invitrogen , Thermo Fisher Scientific ) . In firebrat embryos we detected dying cells using the Click-iT Plus TUNEL assay kit ( C10618 , life technologies ) . To stain cell nuclei and neuropil , firebrat samples were incubated with DAPI ( 1 µg/mL; D9542 , Sigma-Aldrich ) and Phalloidin-488 ( 1:100 , A12379 , life technologies ) in PBST for 30 min at room temperature . Incubations were carried out following secondary antibody treatment . Samples were then washed in PBST and PBS , and mounted . To label proliferating cells and their progeny we used the Click-iT EdU imaging Kit ( C10337 , life technologies ) . Freshly dissected nervous systems from swift lousefly pupae were incubated in EdU 1:1000 in PBS at room temperature for 1–3 hr on a shaker , rinsed with PBS and fixed in cold buffered formaldehyde 3 . 6% in PBS for 30 min . Samples were then stained using the immunohistochemistry protocol described above . The colour reaction for EdU was carried out as instructed by the vendor after the secondary antibodies were washed out . To induce mitotic clones of undead neurons , rescued from PCD , we used the mosaic analysis with a repressible cell marker technique ( Lee and Luo , 1999 ) . 0–4 hr first instar larvae resulting from crossing females of the genotype hs-flp;; TubP-GAL80 , FRT2A/TM3 , Sb with; TDC2-GAL4 , UAS-CD8::GFP , UAS-TrpA1; dronc∆A8 , FRT2A/TM6β , Tb , Hu males were heat-shocked at 37°C in a plastic food vial placed in a water bath for either 1 hr or 45 min , followed by 45 min at room temperature and a second incubation period at 37°C for 30 min . After heat-shock , larvae were immediately returned to 23°C or 25°C . Cell death was blocked in clones homozygous for the loss-of-function allele of the initiator caspase Dronc . Because we used the octopaminergic driver line TDC2-GAL4 to induce the expression of CD8::GFP and TrpA1 , we were able to visualise and thermogenetically activate only postembryonic neurons of hemilineage 0B . A small number of wild-type octopaminergic neurons are born during postembryonic neurogenesis ( one in T1 and T3 , 4–5 in T2 , see Figure 3—figure supplement 1 ) . To ensure the characterisation of undead neurons only , MARCM clones including a bilaterally symmetrical primary neurite were excluded from analysis . Prior to recordings , 2–6 day-old males of Drosophila melanogaster ( hs-flp/+; TDC2-GAL4 , UAS-CD8::GFP , UAS-dTRPA1/+; dronc∆A8 , FRT2A/TubP-GAL80 , FRT2A ) reared at 23°C or 25°C in a 12 hr:12 hr light:dark cycle were anaesthetised on ice and decapitated using a pair of micro spring scissors in under 3 min . We used males as we found they are more responsive to octopamine release by thermogenetic activation than females ( data not shown ) . The headless flies were brushed back into a food vial placed on its side and left to recover for at least 1 hr . To generate the heat ramp required to thermogenetically activate undead neurons , we used a 12V thermoelectric Peltier plate ( model: TEC1-12706 , size: 40 mm x 40 mm x 3 . 6 mm ) connected to a DC power supply ( HY3005D , Rapid Electronics ) set at a constant current of 0 . 46A , with a variable voltage , calibrated using an infrared laser thermometer ( N92FX , Maplin ) . These settings generated a temperature ramp which lasted 70 s from 22°C to 34°C . Videos were recorded at 25 fps using a Sony NEX-5N digital camera ( kindly provided by Ian Wynne ) mounted to a stereo microscope . A piece of graph paper was used for spatial calibration . To match the presence of undead neurons with behaviour , each decapitated fly used for thermogenetic activation was indexed and prepared for dissection and immunostaining . The method for in vivo two-photon imaging of the VNC in behaving adult Drosophila is described in Chen et al . , 2018 . Briefly , flies were anaesthetised through cooling and then mounted onto custom imaging stages . The dorsal thoracic cuticle was removed and indirect flight muscles were left to degrade over the course of 1 hr . Subsequently , the proventriculus and salivary glands were resected to gain optical access to the VNC . Horizontal sections of the T1 leg ganglion were imaged using galvo-galvo scanning . For control animals , the bifurcation point of TDC-positive neurites were imaged to circumvent ROI disappearances caused by movement . For animals harbouring undead TDC2-GAL4-positive neurons , the thickest branch of the axonal bifurcation was chosen because they were most likely to contain undead neurites . Image dimensions ranged between 512 × 512 and 320 × 320 , resulting in 1 . 6 to 3 . 4 fps data acquisition . Imaging areas ranged between 92 × 92 µm and 149 × 149 µm . Laser power was held at ~8 mW . Python scripts ( modified from Chen et al . , 2018 ) were used to extract ROI fluorescence traces and to compute spherical treadmill ball rotations . Walking epochs were determined by placing a threshold on ball rotations , which were first converted into anterior-posterior ( vforward ) and medial-lateral ( vside ) speeds ( one rot s−1 = 31 . 42 mm s−1 ) and into degrees s−1 ( 1 rot s−1 = 360° s−1 ) for yaw ( vrotation ) movements . Thresholds were 0 . 12 mm , 0 . 12 mm and 5° , respectively . Periods below these thresholds were considered ‘resting’ while other periods were considered ‘walking’ . Fluorescence traces for epochs with the same behaviour were aligned by start point to compute average %∆R/R traces for specific actions . To calculate fluorescence traces for small subregions-of-interest across neuritic bundles containing both undead and wild-type neurites , images were registered using an optic flow method described in Chen et al . , 2018 . This registration served to minimise motion artefacts . Analysis was limited to a period with no warping artefacts and no ROI disappearance . Subregions were manually selected as small circular ROIs across the neuritic bundle of the registered image . Fluorescence values were then computed from each sub-ROI . Images were acquired using a Zeiss LSM 510 or a Zeiss LSM 800 confocal microscope at a magnification of 20x or 40x with optical sections taken at 1 µm intervals . The resulting images were examined and processed using Fiji ( https://imagej . net/Fiji ) . Some images were manually cropped using the Freehand Selection tool to remove debris or to cut out neuronal lineages in Worniu-GAL4 , UAS-SR4VH samples . To generate fluorescence intensity along Line plots , we used the Plot Profile tool in Fiji to extract raw fluorescence intensity values for the RFP and Venus channels . The values were imported into MATLAB ( R2018a , MathWorks ) and normalised by dividing all fluorescence intensity values to the maximum value encountered along each Line . In this manner , all fluorescence intensity along Line plots have a common scale from 0 to 1 , with one being the highest value encountered along that Line . Decapitated flies were considered to be walking if they covered a distance greater than one body length and moved their legs in a coordinated sequence from T3 to T2 to T1 at least once on each side ( Harris et al . , 2015 ) . Forward , backward and sideway movements were all interpreted as walking when both aforementioned conditions were respected . To generate fly body traces video recordings were imported in MATLAB ( R2018a , MathWorks ) and the centroid of the decapitated fly ( located on the scutellum ) was extracted from each frame using a custom-written script which can be found at github . com/snznpp/undead-walking ( Pop , 2020; copy archived at https://github . com/elifesciences-publications/undead-walking ) . Each frame was converted into a greyscale image , its contrast enhanced using contrast-limited adaptive histogram equalisation , filtered using a Gaussian smoothing kernel with a standard deviation of 4 , binarised using a custom threshold and the geometric centre of the fly body automatically extracted and stored in an array . To confirm that the centroid detection was accurate , a red dot with the centroid coordinates was superimposed onto each frame of the original recording and the annotated movie was saved for manual inspection . For calculating 3A/3B hemilineage bundle diameter ratios in fruit flies and bee louseflies , we generated transverse rendered maximum intensity projections of inverted greyscale confocal stacks for the pro- and mesothorax ( T1 and T2 ) and frontal projections for the metathorax ( T3 ) . Optical sections were selected to include the common lineage bundle and the individual hemilineage bundles after their split . Diameter measurements were taken at the widest point within 5 µm of the bundle split using the Straight Line tool in Fiji and ratios were calculated by dividing the diameter of hemilineage 3A to that of 3B . For comparing neuron numbers , 3A/3B bundle diameter and T2/T1 number of neurons , data were tested for normal distribution using the Kolmogorov-Smirnov test and visualisation of Normal Q-Q plots . Differences between groups were analysed using either the independent samples t-test for normally distributed data , Welch’s test if data failed to meet the homogeneity of variances assumption or Mann-Whitney t-tests if data failed to meet the normality and homogeneity of variances assumptions of the independent samples t-test . For comparing the number of flies which walked in each experimental group , we performed a Pearson chi-squared test and interpreted the resulting exact significance if the minimum expected count was greater than 5 , or the Fisher’s Exact Test 2-sided significance if the minimum expected count was lower than five in at least one cell of the contingency table . To correct for multiple comparisons we performed a Bonferroni correction ( i . e . p values were multiplied by 6 , the total number of pairwise tests ) . All statistical tests were performed in SPSS Statistics 23 ( IBM ) with an α set at 0 . 05 . In all figures , bars represent means ± standard deviation; *p < 0 . 05 , ***p < 0 . 001 , nsp = not significant . | Just like a sculptor chips away at a block of granite to make a statue , the nervous system reaches its mature state by eliminating neurons during development through a process known as programmed cell death . In vertebrates , this mechanism often involves newly born neurons shrivelling away and dying if they fail to connect with others during development . Most studies in insects have focused on the death of neurons that occurs at metamorphosis , during the transition between larva to adult , when cells which are no longer needed in the new life stage are eliminated . Pop et al . harnessed a newly designed genetic probe to point out that , in fruit flies , programmed cell death of neurons at metamorphosis is not the main mechanism through which cells die . Rather , the majority of cell death takes place as soon as neurons are born throughout all larval stages , when most of the adult nervous system is built . To gain further insight into the role of this ‘early’ cell death , the neurons were stopped from dying , showing that these cells were able to reach maturity and function . Together , these results suggest that early cell death may be a mechanism fine-tuned by evolution to shape the many and varied nervous systems of insects . To explore this , Pop et al . looked for hints of early cell death in relatives of fruit flies that are unable to fly: the swift lousefly and the bee lousefly . This analysis showed that early cell death is likely to occur in these two insects , but it follows different patterns than in the fruit fly , potentially targeting the neurons that would have controlled flight in these flies’ ancestors . Brains are the product of evolution: learning how neurons change their connections and adapt could help us understand how the brain works in health and disease . This knowledge may also be relevant to work on artificial intelligence , a discipline that often bases the building blocks and connections in artificial ‘brains’ on how neurons communicate with one another . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2020 | Extensive and diverse patterns of cell death sculpt neural networks in insects |
Shotgun metagenomic sequencing is a powerful approach to study microbiomes in an unbiased manner and of increasing relevance for identifying novel enzymatic functions . However , the potential of metagenomics to relate from microbiome composition to function has thus far been underutilized . Here , we introduce the Metagenomics Genome-Phenome Association ( MetaGPA ) study framework , which allows linking genetic information in metagenomes with a dedicated functional phenotype . We applied MetaGPA to identify enzymes associated with cytosine modifications in environmental samples . From the 2365 genes that met our significance criteria , we confirm known pathways for cytosine modifications and proposed novel cytosine-modifying mechanisms . Specifically , we characterized and identified a novel nucleic acid-modifying enzyme , 5-hydroxymethylcytosine carbamoyltransferase , that catalyzes the formation of a previously unknown cytosine modification , 5-carbamoyloxymethylcytosine , in DNA and RNA . Our work introduces MetaGPA as a novel and versatile tool for advancing functional metagenomics .
Advances in next-generation sequencing technology have been reshaping metagenomics , making it possible to explore all microbes within a sample , including the overwhelming majority of those unculturable in standard growth conditions ( Quince et al . , 2017 ) . From studying human gut microbiome to marine viral communities ( Sunagawa et al . , 2015 ) , metagenomic analyses are being utilized across a broad spectrum of life science disciplines , contributing to novel clinical diagnoses ( Chiu and Miller , 2019 ) , antibiotics and small molecule discovery ( Hu et al . , 2013; Charlop-Powers et al . , 2014 ) , food safety ( Cao et al . , 2017 ) , biofuel generation ( Hess et al . , 2011 ) , and environment stewardship ( Nesme et al . , 2014; Cavicchioli et al . , 2019 ) . Of the two primary questions metagenomic studies strive to address , profiling the taxonomic biodiversity – ‘what is out there ? ’ is easier to answer compared to inferring the biological functions – ‘what do they do ? ’ . Indeed , there have been many well-established strategies to quantify taxonomic diversity such as analyzing known marker genes , binning contigs and assembling sequences into taxonomic groups or genomes ( Sharpton , 2014; Quince et al . , 2017 ) . However , a large fraction of the functional potential of microbiomes remains to be discovered . Microbiomes encode a large number of evolutionarily diverse genes coding for millions of peptides and proteins for distinctive functions . The functional diversity is particularly enormous for secondary metabolites and epigenetic modifications ( Sharon et al . , 2014 ) . Taking DNA modifications as an example , environmental microbiomes are a plentiful source of DNA-modifying enzymes with diverse mechanisms ( Weigele and Raleigh , 2016; Hiraoka et al . , 2019 ) . In bacteriophages , DNA modifications are used to evade the restriction modification systems of the host bacterial cell . Notably , a handful of bacteriophages have been described to completely modify cytosines in their genomic DNA , for example , 5’ methylcytosine in XP12 phage ( Kuo et al . , 1968 ) and glycosylated 5’ hydroxymethylcytosine in T4 phage ( Revel and Georgopoulos , 1969 ) . Recently , additional base modifications have been discovered including 5- ( 2-aminoethoxy ) methyluridine , 5- ( 2-aminoethyl ) uridine , and 7-deazaguanine ( Lee et al . , 2018; Hutinet et al . , 2019 ) . Current methods to harness such functional potential of microbiomes , for the most part , came from computational prediction of gene function which is based on homology search to existing databases . Nonetheless , because of the poor completeness and accuracy of microbial annotation , homology searches often fail to impute the correct functions ( Huson et al . , 2009; Nayfach and Pollard , 2016 ) . Novel functions can also be found using functional screens; however , such effort entails the construction and high-throughput screening of a large number of clones which can be very time-consuming ( Martinez et al . , 2004 ) . Novel approaches are therefore needed to link genetic information from microbial metagenomes to function . For specific bacteria , determining the genetic basis of phenotypes has been addressed using genome-wide association studies ( GWAS ) ( Falush and Bowden , 2006 ) combined with specific phylogenetic methods to account for the unique population structure of microbes ( Collins et al . , 2018 ) . Examples of microbial GWAS have explored hundreds of isolates to identify genomic elements that are statistically associated with , for example , antibiotic resistance ( Chewapreecha et al . , 2014 ) , host specificity ( Sheppard et al . , 2013 ) , or virulence ( Laabei et al . , 2014 ) . Nonetheless , these studies are limited to known isolates and have not yet been extended to entire complex microbial communities . Here , we developed the Metagenomics Genome-Phenome Association ( MetaGPA ) framework to bridge the gap between genetic information and functional phenotype in complex microbial communities . MetaGPA is conceptually close to GWAS as it associates genotypic data with phenotypic traits . Contrasting with microbial GWAS which uses sequence variations as genetic markers , MetaGPA incorporated association analyses at the level of protein domains to reveal genes that are significantly associated with the phenotype of interest . By applying this workflow on DNA modifications as our phenotypic trait , we discovered a number of candidate enzyme families . From these candidates , we validated a novel DNA/RNA cytosine modification , 5-carbamoyloxymethylcytosine ( 5cmdC ) , and the enzyme responsible for this modification . From this example , we show that MetaGPA is a powerful and versatile method to improve metagenome functional analysis .
Like GWAS in individual species ( Hirschhorn and Daly , 2005 ) , MetaGPA requires definition of two cohorts , a ‘case’ cohort , that is , a group of organisms that share a specific phenotype under study in a given microbiome , and a ‘control’ cohort composed of all organisms within that microbiome ( Figure 1 and Figure 1—figure supplement 1 ) . While both cohorts are sequenced independently , all organisms included in a cohort are sequenced together without the need to isolate organisms . The ‘case’ group is derived from the control group after applying a selection to only retain a given phenotype . The association is computed using a computational workflow composed of the core MetaGPA pipeline that defines genetic units associated with the phenotype and further analysis tools to refine these associations . The core MetaGPA pipeline described in Figure 1 and Figure 1—figure supplement 1 takes sequencing reads from both cohorts and performs de novo assemblies into contigs . Contigs from both cohorts are combined and duplicated contigs are discarded . Reads from the case and control experiments are mapped back to the combined contig set . Each contig is either labeled as enriched or depleted from the case cohort based on the enrichment score calculated using the relative number of reads mapping to the contig . Genes identified in the contig set are annotated using homology search to known protein domains and domains that are found significantly enriched in the enriched contig set are considered to be associated with the studied phenotype . Finally , genes in the enriched contig set annotated with one or more associated domain ( s ) are defined as candidate genes . Both associated domains and their corresponding candidate genes are further refined using evidence such as phylogenetic clustering and co-occurrence with other candidate domain families/genes . Phylogenetic clustering assesses for each associated domain , whether candidate genes are phylogenetically closer to each other relative to genes containing the same domain family in the depleted contigs . This analysis can be done at domain or residue resolution . Co-occurrence with other candidate genes strengthen the association by attempting to identify the entire metabolic pathway responsible for the specific phenotype under study . Taken together , this multilayer analysis effectively identifies protein domain families and their corresponding candidate genes that are truly related to the phenotypes of interest . Among the 110 domains that are significantly associated with DNA modifications , carbamoyltransferase domains are ranking within the top 20 most significant domains in our MetaGPA study ( Figure 4A ) and are exhibiting all three of our refinement metrics . Carbamoyltransferases are part of a large protein family catalyzing the addition of a carbamoyl group to various substrates but so far none of them has been shown to act on any form of cytosine , potentially revealing a new function and a new cytosine modification . We therefore sought to further explore these enzymes for their ability to modify cytosine . The co-occurrence of carbamoyltransferase and thymidylate synthase homologs specifically in modified contigs ( Figure 4C ) resembles the arrangement of the T4 phages for which genes coding for the dCMP hydroxymethylase and β-glucosyltransferase co-occur on the genome ( Miller et al . , 2003 ) . The T4 dCMP hydroxymethylase is homologous to thymidylate synthase and transfers a carbon from methyltetrahydrofolate ( mTHF ) to C5 of the pyrimidine ring producing an exocyclic methylene in the active site of the enzyme ( Graves et al . , 1992 ) . However , unlike thymidylate synthase , the methylene intermediate undergoes nucleophilic attack by water producing a hydroxymethyl group . Following incorporation of 5hmC into DNA during replication , T4 β-glucosyltransferase transfers a glucose to the hydroxyl moiety of 5hmC . Thus , the pairing of a carbamoyltransferase with dCMP hydroxymethylase led us to hypothesize a novel form of DNA modification , in which the carbamoyltransferase catalyzes the transfer of a carbamoyl group to the nucleophilic hydroxyl acceptor group of 5hmdC producing 5cmdC ( Figure 6A ) . Our composite dataset contains 62 genes annotated as carbamoyltransferase for which 17 are found in the modified contigs . We selected a carbamoyltransferase candidate gene from a modified contig originally sequenced in sewage #2 sample containing both the thymidylate synthase and the carbamoyltransferase genes ( Figure 6—figure supplement 1A ) . The ORF was cloned into pET28b vector , and the 63 kDa protein product was expressed in E . coli and purified ( Figure 6—figure supplement 1B , Materials and Methods ) . The predicted reaction was tested by enzymatic assays and results showed that every substrate , namely carbamoyl phosphate , ATP , 5hmdC ( genomic T4gt DNA was used as substrate in these experiments ) , and the enzyme were indispensable for the reaction ( Figure 6—figure supplement 1C , D ) . The choice of carbamoyl phosphate and ATP substrates were guided by the enzymatic characterization of TobZ previously published ( Parthier et al . , 2012 ) . The expected product was detected by liquid chromatography-mass spectrometry ( LC-MS ) and confirmed with corresponding fragmentation patterns ( Figure 6—figure supplement 2A , B ) . Nearly 70% of 5hmdC were converted into 5cmdC in the denatured single-stranded T4gt genomic DNA . Interestingly , the activity of our carbamoyltransferase was several fold lower on double-stranded DNA , suggesting the preference of this enzyme for single-stranded DNA ( Figure 6—figure supplement 1C ) . When using synthesized single-stranded DNA oligo containing a single internal 5hmdC site as substrate , the conversion rate was nearly 100% ( Figure 6B ) . We also tested if the carbamoyltransferase could react with free deoxynucleoside triphosphate . LC-MS results demonstrated about 60% conversion of 5-hydroxymethyl-2’-deoxycytidine triphosphate ( 5hmdCTP ) ( Figure 6C ) . As expected , no activity was seen for 5-methyl-2’-deoxycytidine triphosphate ( 5mdCTP ) or 5-hydroxymethyl-2’-deoxyuridine triphosphate ( 5hmdUTP ) , indicating the carbamoyltransferase is specific to 5hmdCTP . The fact that the carbamoyltransferase is active on 5hmdCTP ( Figure 6C and Figure 6—figure supplement 3A ) opens up the possibility that the reaction could take place before the nucleotide is incorporated into the phage DNA . To examine if the carbamoyltransferase favors certain DNA sequences , we performed the enzymatic assay on a mixed genomic DNA library containing lambda ( dC ) , XP12 ( 5mdC ) , and T4gt ( 5hmdC ) . Both untreated ( control ) and treated libraries were subjected to APOBEC3A deamination ( see Materials and methods ) . Carbamoylation protects cytosine derivatives from deamination by APOBEC and thus , the difference in deamination rate between control and treated libraries is indicative of carbamoylation . We saw a decrease in deamination rate only in the T4gt genome indicating specific carbamoylation on 5hmdC . This result further validates the LC-MS results regarding the specificity of the enzyme ( Figure 6D ) . Furthermore , all combinations of NCN motifs containing 5hmdC displayed comparable deamination levels compared to the control library , suggesting a general binding mechanism with no preferred context . This result was also consistent with enzymatic assays performed on free nucleotides , which further suggests that the in vivo carbamoylation reaction may take place prior to DNA replication . Based on our finding that the carbamoyltransferase prefers single-stranded DNA , we further investigated whether RNA containing 5-hydroxymethylcytosine can be modified by the enzyme . LC-MS and fragmentation pattern confirmed that 5cmC is formed during the reaction , albeit at much lower yields ( Figure 6E and Figure 6—figure supplement 3B-C ) . Carbamoylation of free nucleotide 5-hydroxymethylcytidine triphosphate ( 5hmCTP ) was also detected ( Figure 6F and Figure 6—figure supplement 3A ) . We thus concluded that this novel nucleic acid-modifying enzyme identified from MetaGPA studies acts on DNA , RNA , and nucleotide triphosphates .
All raw and processed sequencing data generated in this study have been submitted to the NCBI Sequence Read Archive ( SRA; https://www . ncbi . nlm . nih . gov/sra ) under accession number PRJNA714147 .
The E . coli K-12 MG1655 , XP12 , and T4gt genomic DNA used in this study were obtained from NEB . For each batch , 2–4 L of sewage or coastal seawater were used for phage collection . Large debris and bacterial cells were pelleted and removed by centrifuging at 5000× g for 30 min at 4°C . Phage particles in the supernatant were precipitated by adding PEG8000 to 10% ( w/v ) and NaCl to 1 M and let stand at 4°C overnight . Aggregates of phage particles were pelleted at 10 , 000× g for 30 min at 4°C , washed with 1 mL solution containing 10 % PEG8000 and 1 M NaCl , and resuspended in 2–4 mL of phage dilution buffer ( 10 mM Tris-HCl at pH 8 . 0 , 10 mM MgCl2 , 75 mM NaCl ) . The crude phage particle suspension was stored at 4°C for subsequent phenol-chloroform DNA extraction . Two to four milliliters of crude phage suspension was divided in 400 μL aliquots . For each aliquot , phage particles were lysed at 56°C for 2 hr in 550 μL of lysis buffer ( 100 mM Tris-HCl at pH 8 . 0 , 27 . 3 mM EDTA , 2% SDS , ~1 . 6 U Proteinase K [NEB #P8107] ) . After lysis , RNase A was added to 10 μg/mL and the reaction was incubated at 37°C for 30 min; 1× volume ( ~550 μL ) of phenol-chloroform ( Tris-HCl buffered at pH 8 . 0 ) was mixed with the lysis solution and vortexed vigorously for ~1 min , and centrifuged at 10 , 000× g for 5 min for phase separation . The top aqueous layer ( ~500 μL ) was collected and mixed with 1× volume ( ~500 μL ) of chloroform , vortex vigorously , and centrifuged for phase separation . The top aqueous layer ( ~450 μL ) was collected; 1× volume of isopropanol was slowly added on top of the aqueous solution . Phage DNA was ‘spooled’ with a glass capillary by swirling and mixing isopropanol with the aqueous solution . The spooled DNA was washed in 70% ethanol , dried at room temperature for ~30 min , and dissolved in ~600–800 μL of TE buffer ( 10 mM Tris pH 7 . 5 , 1 mM EDTA ) . The phage DNA solution was further purified by ethanol precipitation . Briefly , DNA was precipitated by adding 0 . 1× volume of 3 M sodium acetate and 2 . 5× volume of 100% ethanol and incubated at −20°C overnight . Precipitated DNA was pelleted at 16 , 000× g for 20 min , washed twice with 1 mL of 70% ethanol , dried at room temperature , and finally dissolved in 200 μL of TE buffer for storage at −20°C . On average more than 20 μg of DNA was extracted in each batch . For each environmental sample , 1 μg of metagenomic DNA was sheared to 300 bp in 130 μL of TE buffer ( 10 mM Tris pH 7 . 5 , 1 mM EDTA ) using Covaris S2 Focused Ultrasonicator; 1 . 3 μL of 10 mg/mL RNase A ( Qiagen #1007885 ) was added and incubated at 37°C for 30 min to remove RNA . To remove EDTA , the sheared DNA was purified with Zymo Oligo Clean & Concentrator Kit ( #D4061 ) and eluted in 50 μL of 1 mM Tris buffer ( pH 7 . 5 ) . One reaction of NEBNext Ultra II DNA Library Prep Kit for Illumina ( NEB #E7645 ) was used for 1 μg of input DNA , with the following modifications to the standard protocol: 5mdC Y-shaped Illumina adaptors were used to protect the adaptor from subsequent enzymatic treatment . The dCTP was replaced with 5mdCTP in the end repair reaction ( 5mdCTP was used instead of regular dCTP to protect end-repaired fragments from subsequent enzymatic treatment ) . The DNA library was purified with 1× volume of NEBNext Sample Purification Beads ( NEB #E7103 ) and eluted with 40 μL of 1 mM Tris buffer ( pH 7 . 5 ) . For each of the two sewage DNA samples , experiments were performed in duplicate . Each one contained two pairs of replicate libraries subjected to enzymatic selection or control , respectively . The coastal sample generated only one pair: one library for enzymatic selection and one for control . First to test the recovery of modified DNA with enzymatic selection , mixed genomic DNA ( E . coli and T4gt ) were prepared at various dilutions . A total of 250 ng mixed DNA was used per reaction; 1 µL TET2 ( NEB #7120S ) and 1 µL T4-BGT ( NEB #M0357S ) were added to the 50 µL reaction mixture containing 1× TET2 reaction buffer , 40 µM UDP-glucose , and 40 µM iron ( II ) sulfate hexahydrate . After 60 min incubation at 37°C , proteinase K was added at 0 . 4 mg/mL to inactivate the enzymes . Products were purified with Zymo Oligo Clean & Concentrator kit ( #D4061 ) and eluted in 16 µL water . To denature double-stranded DNA , 4 µL formamide ( Sigma #11814320001 ) was added . The 20 µL mixture was then incubated at 95°C for 10 min and immediately transferred to an ice bath . One µL APOBEC3A ( NEB #E7120S ) was added directly to the reaction with 10 µL of 10× APOBEC3A reaction buffer and 1 µL BSA ( 10 mg/mL ) . The reaction volume was brought up to 100 µL with water . APOBEC3A-mediated deamination was conducted at 37°C for 3 hr . Purification was performed using Zymo Oligo Clean & Concentrator kit and elution with 43 µL of water . In the final step , the reaction was incubated with 2 µL of USER ( NEB #5508 ) in 1× CutSmart Buffer at 37°C for 15 min before final purification with Zymo Oligo Clean & Concentrator kit . Purified samples were then used for qPCR in the next step . For each prepared phage library sample , 100 ng spiked-in genomic DNA mixture ( E . coli:XP12:T4gt = 1:1:1 by molarity ) were added to the library before being subjected to enzymatic selection protocol described above . The final library was eluted in 50 μL of 1 mM Tris buffer ( pH 7 . 5 ) . The qPCR reactions were performed with enzymatic selection or control samples using Luna Universal qPCR Master Mix ( NEB #M3003S ) on a Bio-Rad CFX96 real-time PCR detection system . Two µL of purified DNA ( diluted 100-folds ) were added per reaction . Primers used in the experiments were the following: E . coli F: 5’-TTGCTGAGTTTCACGCTTGC , E . coli R: 5’-AAAACCGCTTGTGGATTGCC , T4gt F: 5’-TCGCGAAACGGTTTTCCAAG , T4gt R: 5’-AAAGCGCTTGACCCAACAAC , XP12 F: 5’-TGCGATGTTGGATTCGTTGG , and XP12 R: 5’-ACAACCCGCCATAATGGAAC . Recovery was normalized to control using the delta-delta Ct method . Libraries were indexed ( with NEBNext Multiplex Oligos for Illumina E7335 ) , amplified using NEBNext Ultra II Q5 Master Mix ( six cycles for control library and 12 cycles for selection library ) and pooled for sequencing on an Illumina Nextseq instrument with paired end reads of 75 bp . De novo assembly of contigs for each sample was performed with SPAdes v3 . 13 . 0 ( Nurk et al . , 2017 ) with the --meta option . We selectively reported contigs longer or equal to 1000 bp . To remove redundant contigs between case and control samples , we used CD-HIT v4 . 8 . 1 ( Fu et al . , 2012 ) nucleotide mode cd-hit-est with sequence identity threshold set to 0 . 95 ( sequences with more than 95% similarity are considered redundant ) . We set this threshold due to the fact that many microorganism genomes are related . Other options used were -n 10 -d 0 M 0 T 4 . The remaining non-redundant contigs were annotated with HMM-based Pfam entries ( Pfam-A ) using HMMER v3 . 3 ( http://hmmer . org/ ) . Alignment of reads onto contigs was done with BOWTIE2 v2 . 3 . 5 . 1 ( Langmead and Salzberg , 2012 ) together with SAMTOOLS v1 . 9 ( Li et al . , 2009 ) to generate , sort , and index bam files for later analysis . ORFs on each contig were predicted with GLIMMER 3 . 02 ( Delcher et al . , 2007 ) . We used the long-ORFs program with options --cutoff 1 . 15 and --linear to identify long ORFs that are very likely to contain genes . Other options used in glimmer3 program were --max_olap 100 , --gene_len 110 , and --threshold 30 . The enrichment score for each contig was calculated using the normalized mapped reads ( reads per kb per million , RPKM ) from selection and control as follows: enrichment score = RPKM ( selection ) / RPKM ( control ) . The mapped reads counts were generated with Multicov using BEDTOOLS v2 . 29 . 2 ( Quinlan and Hall , 2010 ) . Contigs with higher enrichment score represent more mapped reads in case library relative to control library , therefore , are more likely to be associated with modification . We considered contigs with an enrichment score greater or equal to three to be modified and the rest unmodified . The calculation was done individually for three independent experiments . The information including the number and type of Pfams on each contig was obtained with hmmsearch in the annotation step . We then separately counted the numbers of modified and unmodified contigs containing each type of Pfam . To avoid redundant counting , Pfams occurred multiple times on the same contig were counted only once . Fisher’s exact test was performed for each Pfam to identify if the count difference between the modified and unmodified contig group is significant . Because large-scale multiple testing was conducted for each Pfam , we did the Bonferroni correction to adjust the p-value . Both tests were performed in python with SciPy or Statsmodels modules . For each Pfam of interest , the protein sequences from contigs containing the Pfam were aligned with MUSCLE v3 . 8 . 1551 ( Edgar , 2004 ) . The resulting aligned fasta files were subjected to construct phylogenetic trees using the maximum likelihood method in the phylogenetic analysis program RAxML v8 . 2 . 12 ( Stamatakis , 2014 ) . We chose the -f a option to do rapid bootstrap analysis and the -m PROTGAMMAAUTO model to automatically determine the best protein substitution model to be used for the dataset . The parsimony trees were built with random seeds 1237 . The online tool iTOL ( https://itol . embl . de/ ) was used to visualize trees . The presence-absence matrix with rows being the Pfams and columns being the contigs was generated with annotation output file from the previous step . We specifically performed co-occurrence analysis in the R package co-ocur v1 . 3 ( Griffith et al . , 2016 ) for the top 20 Pfams associated with modified contigs . Significant positive correlations ( p-value < 0 . 05 ) were exported and the network was visualized in Cytoscape v3 . 8 . 0 ( Shannon et al . , 2003 ) with prefuse force directed layout . Protein sequences were assigned to two groups according to whether they were encoded on modified or unmodified DNA . After multiple sequence alignment , positions that have less than 50% residues present were ignored . Differential conservation score was calculated at each aligned position . For each position in the alignment , intra-group similarity scores were calculated by the average of all possible ‘within-group’ pairwise similarities , while the inter-group similarity score was calculated from all possible ‘across-group’ pairwise similarities using the BLOSUM80 matrix . For a given multiple sequence alignment column , let N1 and N2 be the number of residues for the modified and unmodified groups , respectively , the two intra-group similarity scores ( Imodified and Iunmodified ) were defined as:Imodified=∑i=1N1∑j§amp;gt;iN1Mai , aj×2N1 ( N1-1 ) Iunmodified=∑i=1N2∑j§amp;gt;iN2Mai , aj×2N2 ( N2-1 ) where Mai , aj is the value of amino acid pair ai and aj in the BLOSUM80 matrix . The inter-group similarity score ( J ) was defined as:J=∑i=1N1∑j=1N2Mai , aj×1N1N2 The differential conservation score ( S ) was defined as the average of two intra-group similarity scores subtracted by the inter-group similarity score . S=Imodified+Iunmodified2-J The following carbamoyltransferase sequence : MSDLLLTLGHNASAIAISVGDDGAAKVENAYELERLTGKKSDSAFPIDAIIALKERGMDKIDRVYVSHWSPTGRVDDLKAKYWDRSIFPPHVPVITQESMNLTHHDCHAQAAMAFAGSSFPTKDTGVLVVDGFGNLAEHLSYYRVQAGGQLHLMRRWYGYGTSLGLMYQYATSFLGLKMHEDEYKLLGYGARVATIGCDMDVLNQRIFTEAQAFLKRFRSLNSFEMSPDLAGLPAVQEKWAERFAAILDDVGFKGSSSTYEARCIVGYAVQQLLEIVIRNLFMADLPKPTNLIVTGGVAFNVELNRMLLGLIPGKLCVMPLAGDQGNALGLWAFSNRRAKLDFGDLCWGRREMTLGEPGPDTPLPDGMIVVEHDTPAVYEAIAEQLKTVGFINIVRGNMEFGPRALCNTTTLARADDRAVVEEINRINGRDTVMPFAPVVSAHEWLRYFPDASRLHRSAEFMICAVQYAPGLGEQVPGAALRTVKGLYTGRPQVYSSKYEWDSVTRILDDYGLLINTSFNVHGVPICLDLKHVVHSHQFQRERNPNVRTIVIAN* was extracted from de novo assembled contigs . The expression plasmid was synthesized from GenScript . Two 6× His-tags were co-expressed at both the N-terminus and the C-terminus of the recombinant protein using T7 Express Competent E . coli ( NEB #C2566 ) . Cells were cultured in LB media until an OD600 of 0 . 6 was reached and induced with 0 . 4 mM IPTG ( Growcells #MESP-2002 ) for protein expression . One µM iron ( II ) was also added to facilitate protein folding . The induced cultures were maintained at 16°C in a shaker at 200 rpm for 23 hr . Cells were harvested by spinning down cell pellets at 3500 rpm at 4°C for 30 min . Cell pellets from 4 L culture were resuspended in 160 mL buffer A containing 20 mM Tris pH 7 . 5 , 500 mM NaCl , 0 . 05% Tween-20 , 20 mM imidazole , and sonicated using a Misonix S-4000 sonicator with 20 s on and 20 s off cycles until an OD260 plateau was reached . Cell lysates were spinned down at 13 , 000 rpm for 30 min in a pre-chilled centrifuge at 4°C . The supernatant was separated and combined with 0 . 2 mM PMSF ( Sigma #78830 ) ; 50 mL of supernatant was loaded on AKTA ( GE Healthcare Life Sciences ) with 1 mL Histrap column ( GE Healthcare Life Sciences ) pre-equilibrated with buffer A . The column was washed with 50 mL buffer A and eluted with a gradient of buffer B containing 20 mM Tris pH 7 . 5 , 500 mM NaCl , 0 . 05% Tween-20 , and 500 mM imidazole . Aliquots containing concentrated proteins were pooled and diluted 1:1 with 20 mM Tris pH 7 . 5 , 5% glycerol and 0 . 05% Tween-20 . The diluent was reloaded on AKTA with 5 mL Hitrap Q HP column ( GE Healthcare Life Sciences ) , followed by a wash with 35 mL buffer containing 20 mM Tris pH 7 . 5 , 100 mM NaCl , 5% glycerol , and 0 . 05% Tween-20 and eluted with a gradient of buffer containing 20 mM Tris pH 7 . 5 , 1 M NaCl , 5% glycerol , and 0 . 05% Tween-20 . Finally , collected fractions with concentrated proteins were pooled and mixed with equal volume glycerol for storage at −20°C . For enzyme assay using T4gt genomic DNA as substrate , 10 min incubation at 95°C was performed to denature double-stranded DNA and the sample was immediately transferred to ice bath to prevent re-annealing . Then 0 . 38 nM denatured DNA was used for each 50 µL reaction with 1× NEBBuffer2 . 1 ( NEB #B7202S ) , freshly prepared 10 µM iron ( II ) sulfate hexahydrate ( Sigma #203505 ) , freshly prepared 10 mM carbamoylphosphate and 5 mM ATP . Carbamoyltransferase was added to the reaction at 7 . 2 µM . The reaction mixture was incubated at 30°C for 3 hr before adding 2 µL Proteinase K to inactivate the enzyme . After 30 min incubation at 37°C with proteinase K , DNA was purified with Zymo Oligo Clean & Concentrator kit . For assays with synthesized single-stranded DNA oligos containing 5hmdC , the heat-denaturing step was omitted . Oligos were added at 1 . 6 µM per 50 µL reaction with the same concentration of carbamoyltransferase and other components added as listed before . Purification was performed using Norgenbiotek Oligo Clean-up and Concentrator kit ( #34100 ) . For assays with free nucleotides , 0 . 5 mM of the corresponding nucleotide was used per reaction . For assays with synthesized RNA oligos containing 5hmC , 1 . 57 µM RNA was added per reaction . Genomic DNA and synthetic oligonucleotides were digested to nucleosides by treatment with the Nucleoside Digestion Mix ( NEB #M0649S ) at 37°C for 3 hr . The resulting nucleoside mixtures were directly analyzed by reversed-phase LC/MS or LC-MS/MS without further purification . Nucleoside and nucleotide analyses were performed on an Agilent LC/MS System 1200 Series instrument equipped with a G1315D diode array detector and a 6120 Single Quadrupole Mass Detector operating in positive ( +ESI ) and negative ( -ESI ) electrospray ionization modes . LC was carried out on a Waters Atlantis T3 column ( 4 . 6 mm × 150 mm , 3 μm ) at a flow rate of 0 . 5 mL/min with a gradient mobile phase consisting of 10 mM aqueous ammonium acetate ( pH 4 . 5 ) and methanol . MS data acquisition was recorded in total ion chromatogram mode . LC-MS/MS was performed on an Agilent 1290 UHPLC equipped with a G4212A diode array detector and a 6490A triple quadrupole mass detector operating in the positive electrospray ionization mode ( +ESI ) . UHPLC was performed on a Waters XSelect HSS T3 XP column ( 2 . 1 × 100 mm , 2 . 5 µm particle size ) at a flow rate of 0 . 6 mL/min with a binary with a gradient mobile phase consisting of 10 mM aqueous ammonium formate ( pH 4 . 4 ) and methanol . MS/MS fragmentation spectra were obtained by collision-induced dissociation in the positive product ion mode with the following parameters: gas temperature 230°C , gas flow 13 L/min , nebulizer 40 psi , sheath gas temperature 400°C , sheath gas flow 12 L/min , capillary voltage 3 kV , nozzle voltage 0 kV , and collision energy 5–65 V . Library preparation was performed except the following modifications to the standard protocol: ( 1 ) we did not perform RNase A treatment for this experiment; ( 2 ) we used pyrrolo-dC Y-shaped adaptor instead of regular adaptor so that they are protected from subsequent enzymatic treatment . For each library , 1 µg genomic DNA mixture ( Lambda:XP12:T4gt = 1:1:1 by molarity ) was used . After adapter ligation , DNA libraries were purified with 1× volume of NEBNext Sample Purification Beads ( NEB #E7103 ) and eluted with 20 μL nuclease-free water . The sample was then denatured by heating to 95°C for 10 min and subjected to carbamoyltransferase reaction as described above . Carbamoyltransferase was added to the reaction at 7 . 2 µM for every 1 µg DNA library . Purification was performed using Zymo Oligo Clean & Concentrator kit and eluted with 16 µL water . Purified DNA samples were heated at 90°C with 4 µL formamide to generate single-stranded fragments for the deamination reaction . One µL APOBEC3A was added per reaction to both carbamoyltransferase-treated or control ( untreated ) samples with 10 µL of 10× APOBEC3A reaction buffer and 1 µL BSA ( 10 mg/mL ) . The reaction mixture was incubated at 37°C overnight . Final libraries were purified using Zymo Clean & Concentrator kit , indexed ( with NEBNext Multiplex Oligos for Illumina E7335 ) and amplified with NEBNext Q5U Mater mix ( NEB #M0597 ) . Sequencing was performed on an Illumina Mi-seq instrument with pair-end reads ( 2 × 75 bp ) . Raw reads were trimmed with TrimGalore . Methylation was analyzed with Bismark v0 . 22 . 3 and plotted with RStudio v3 . 6 . 3 . Forward and reverse DNA templates were annealed at 95°C for 4 min and slowly cooled for 20 min . RNA synthesis was performed with HiScribe T7 High Yield RNA Synthesis Kit ( NEB #E2040 ) . One µg of annealed DNA template was used per reaction with 1 . 5 µL T7 RNA Polymerase Mix . 5hmCTP was used with the other three nucleotides ATP , UTP , and GTP at 7 . 5 mM each . The reaction was incubated at 37°C for 4 hr . Two µL nulease-free DNase I were added to each reaction to digest DNA templates , followed by incubation at 37°C for 15 min . Synthesized RNA was purified with Norgenbiotek Oligo and Concentrator kit and stored at −80°C . Single-stranded DNA oligos used in enzymatic assays were purchased from IDT . The sequences are as follows: The DNA templates for synthesizing RNA were purchased from IDT as follows ( T7 promoter sequence was underlined ) : 5hmdCTP ( D1045 ) and 5mdCTP ( D1035 ) were purchased from Zymo Research . 5hmdUTP ( N-2059 ) and 5hmCTP ( N-1087 ) were purchased from Trilink Biotechnologies . Custom-built bioinformatics pipelines are available at https://github . com/linyc74/MetaGPA ( Lin , 2021 , copy archived at swh:1:rev:28c23f47fcf108b0e7b8851b92f37921358c8e8e ) . | Many industrial processes , such as starch processing and oil refinement , use chemicals that cause harm to the environment . These can often be switched to more sustainable biological processes that are powered by proteins called enzymes . Enzymes are micro-factories that speed up biochemical reactions in most living things . Communities of microorganisms ( also known as microbiomes ) are an amazing but often untapped resource for discovering enzymes that can be harnessed for industrial purposes . To gain a better picture of the microbes present within a population , researchers often extract and sequence the genetic material of all microorganisms in an environmental sample , also known as the metagenome . While current methods for analyzing the metagenome are good at identifying new species , they often provide limited information about the microorganism’s functional role within the community . This makes it difficult to find new enzymes that may be useful for industry . Here , Yang , Lin et al . have developed a new technique called Metagenomics Genome-Phenome Association , or MetaGPA for short . The method works in a similar way to genome-wide association studies ( GWAS ) which are used to identify genes involved in human disease . However , instead of disease associated genes in humans , MetaGPA finds microbial genes that are associated with a biological process useful for biotechnology . Like GWAS , the new approach created by Yang , Lin et al . compares two groups: the first contains microorganisms that carry out a specific process , and the second contains all organisms in the microbiome . The metagenome of each group is extracted and a computational pipeline is then applied to identify genes , including those coding for enzymes , that are found more often in the group performing the desired task . To test the technique , Yang , Lin et al . used MetGPA to find new enzymes involved in DNA modification . Microbiome samples were collected from coastal water and sewage , and the computational pipeline was applied to discover genes that are associated with this process . Further analysis revealed that one of the identified genes codes for an enzyme that introduces a previously unknown change to DNA . MetaGPA could be applied to other processes and microbiomes , and , if successful , may help researchers to identify more diverse enzymes than is currently available . This could scale up the discovery of new enzymes that can be used to power industrial reactions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"genetics",
"and",
"genomics"
] | 2021 | A genome-phenome association study in native microbiomes identifies a mechanism for cytosine modification in DNA and RNA |
Calcineurin is responsible for mediating a wide variety of cellular processes in response to dynamic calcium ( Ca2+ ) signals , yet the precise mechanisms involved in the spatiotemporal control of calcineurin signaling are poorly understood . Here , we use genetically encoded fluorescent biosensors to directly probe the role of cytosolic Ca2+ oscillations in modulating calcineurin activity dynamics in insulin-secreting MIN6 β-cells . We show that Ca2+ oscillations induce distinct temporal patterns of calcineurin activity in the cytosol and plasma membrane vs at the ER and mitochondria in these cells . Furthermore , we found that these differential calcineurin activity patterns are determined by variations in the subcellular distribution of calmodulin ( CaM ) , indicating that CaM plays an active role in shaping both the spatial and temporal aspects of calcineurin signaling . Together , our findings provide new insights into the mechanisms by which oscillatory signals are decoded to generate specific functional outputs within different cellular compartments .
Calcium ( Ca2+ ) is a ubiquitous and universal intracellular signal whose remarkable biological versatility is a product of diverse patterns of spatial and temporal regulation ( Thomas et al . , 1996; Berridge et al . , 2000; Dupont et al . , 2011 ) , most notably in the form of repetitive , transient elevations in cytosolic Ca2+ concentrations , or Ca2+ oscillations . In general , oscillatory signals are thought to function as a critical biological regulator by allowing a single message to encode multiple types of information through variations in the frequency , amplitude , and spatial characteristics of the signal ( Cheong and Levchenko , 2010; Ganesan and Zhang , 2012 ) , thereby promoting specificity in the regulation of downstream targets . Ca2+ oscillations in particular are known to regulate numerous processes including gene expression ( Negulescu et al . , 1994; Lewis , 2003 ) , exocytosis ( Tse et al . , 1993; Pasti et al . , 2001 ) , and excitation-contraction coupling ( Viatchenko-Karpinski et al . , 1999; Maltsev and Lakatta , 2007 ) , and Ca2+ oscillations have been shown to significantly enhance the specificity and efficiency of Ca2+-regulated processes ( Dolmetsch et al . , 1998; Li et al . , 1998; Kupzig et al . , 2005 ) . Cells are primarily dependent on a single effector protein , the Ca2+ sensor calmodulin ( CaM ) , to transduce Ca2+ signals . CaM sits at the epicenter of Ca2+ signaling , modulating the activity of a vast array of target proteins throughout the cell ( Persechini and Stemmer , 2002 ) . CaM is also thought to play a prominent role in the decoding of Ca2+ oscillations , largely via the differential activation of target proteins such as the Ca2+- and CaM-dependent phosphatase calcineurin ( Saucerman and Bers , 2008; Song et al . , 2008; Parekh , 2011; Slavov et al . , 2013 ) . One of the major targets of CaM in almost all eukaryotic cells ( Hilioti and Cunningham , 2003 ) , calcineurin is involved in regulating diverse physiological processes , including cell proliferation , differentiation , and death , as well as gene expression , secretion , immune function , learning , and memory ( reviewed in Rusnak and Mertz , 2000; Aramburu et al . , 2004 ) . However , the precise spatiotemporal regulation of calcineurin signaling remains poorly understood . Ca2+ oscillations have previously been shown to enhance calcineurin-mediated transcriptional regulation ( Dolmetsch et al . , 1998; Li et al . , 1998; Tomida et al . , 2003; Wu et al . , 2012 ) , and studies have also shown that the Ca2+ oscillatory frequency is a critical determinant of calcineurin-dependent hypertrophic signaling in cardiomyocytes ( Colella et al . , 2008; Saucerman and Bers , 2008 ) and long-term depression in neurons ( Mehta and Zhang , 2010; Li et al . , 2012b; Fujii et al . , 2013 ) . Nevertheless , the precise relationship between Ca2+ oscillations and calcineurin signaling has yet to be elucidated . Similarly , calcineurin dephosphorylates multiple target proteins located throughout the cell ( Cameron et al . , 1995; Wang et al . , 1999; Bandyopadhyay et al . , 2000; Cereghetti et al . , 2008; Tandan et al . , 2009; Bollo et al . , 2010 ) , and although spatial compartmentalization is suspected to play an important role in regulating calcineurin signaling ( Heineke and Ritter , 2012 ) , this phenomenon has yet to be directly examined . Both Ca2+ oscillations and calcineurin signaling are known to play important roles in pancreatic β-cells . Ca2+ oscillations are responsible for driving the pulsatile secretion of insulin ( Hellman et al . , 1992; Tengholm and Gylfe , 2008 ) , and the chronic inhibition of calcineurin , which is a common form of immunosuppressive therapy , is often accompanied by the onset of diabetes ( Heisel et al . , 2004; Chakkera and Mandarino , 2013 ) . In the present study , we use a variety of genetically encoded fluorescent reporters to directly investigate the spatiotemporal dynamics of calcineurin signaling in response to cytosolic Ca2+ oscillations in MIN6 β-cells . We were able to observe distinct subcellular patterns of calcineurin activity in the cytosol and plasma membrane , where calcineurin appeared to integrate the oscillatory input , vs at the ER and mitochondrial surfaces , where calcineurin activity was observed to oscillate . Furthermore , an exploration of the molecular determinants involved in regulating calcineurin activity revealed that significant differences in the subcellular distribution of free Ca2+/CaM are responsible for generating these discrete activity patterns . Our findings provide the first evidence that oscillatory signals are capable of differentially regulating calcineurin activity and suggest a more active role for CaM in transducing oscillatory Ca2+ signals .
To investigate the spatiotemporal regulation of calcineurin signaling in response to Ca2+ oscillations in β-cells , we engineered an improved version of our previously described FRET-based calcineurin activity reporter ( CaNAR ) ( Newman and Zhang , 2008 ) by optimizing the donor and acceptor fluorescent protein pair ( Figure 1 ) . We then targeted this reporter , called CaNAR2 , to the cytosol ( cytoCaNAR2 ) , plasma membrane ( pmCaNAR2 ) , mitochondrial outer membrane ( mitoCaNAR2 ) , and ER surface ( erCaNAR2 ) via in-frame fusion with a C-terminal nuclear export signal ( NES ) ( Ullman et al . , 1997 ) , an N-terminal motif derived from Lyn kinase ( Lyn ) ( Gao and Zhang , 2008; Depry et al . , 2011 ) , an N-terminal motif derived from DAKAP1a ( DAKAP ) ( DiPilato et al . , 2004 ) , and an N-terminal motif derived from cytochrome P450 ( CYP450 ) ( Szczesna-Skorupa et al . , 1998 ) , respectively ( Figure 2A–E ) . Each targeted CaNAR2 variant was co-expressed in MIN6 β-cells along with a diffusible version of the genetically encoded , red-fluorescent Ca2+ indicator RCaMP ( Akerboom et al . , 2013 ) , allowing us to simultaneously visualize and characterize the coordination of subcellular calcineurin activity with cytosolic Ca2+ levels . Membrane depolarization induced by tetraethylammonium chloride ( TEA ) treatment produced robust oscillations in RCaMP fluorescence intensity , which were consistent with the cytosolic Ca2+ oscillations previously observed in MIN6 cells ( Landa et al . , 2005; Ni et al . , 2010 ) ( Figure 2F–I , red curves ) . 10 . 7554/eLife . 03765 . 003Figure 1 . Development and characterization of CaNAR2 . ( A ) Schematic depicting the CaNAR variants tested . FRET pair optimization was performed by replacing the original ECFP and circularly permuted Venus ( cpV ) L194 of CaNAR1 ( top ) with Cerulean , Cerulean2 , Cerulean3 , cpV E172 , or YPet . The responses from each construct are indicated as follows: + , ∼5–10%; ++ , ∼10–15%; +++ , ∼15–20%; ++++ , >20% . ( B and C ) Representative time-courses comparing the yellow/cyan ( Y/C ) emission ratio changes from CaNAR1 and CaNAR2 in HEK293 cells treated with ( B ) 1 μM thapsigargin ( TG ) or ( C ) 1 μM ionomycin ( iono ) . CaNAR2 exhibits an at least twofold greater response in each condition . ( D and E ) Pseudocolored images showing the responses of CaNAR1 and CaNAR2 to ( D ) 1 μM TG or ( E ) 1 μM iono in HEK293 cells . Warmer colors correspond to higher FRET ratios . Cyan fluorescence images ( left ) show the cellular distribution of CaNAR1 and CaNAR2 fluorescence . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 00310 . 7554/eLife . 03765 . 004Figure 2 . Subcellular calcineurin activity dynamics in response to Ca2+ oscillations in MIN6 cells . ( A ) Schematic illustrating the domain structures of the subcellularly targeted variants of the CaNAR2 biosensor . ( B and C ) Yellow fluorescence images showing the biosensor distribution in transiently transfected MIN6 cells expressing ( B ) cytoCaNAR2 and ( C ) pmCaNAR2 . ( D and E ) Fluorescence images showing the localization of mitoCaNAR2 and erCaNAR2 . MIN6 cells expressing ( D ) mitoCaNAR2 or ( E ) erCaNAR2 were stained with MitoTracker Red or ER-Tracker Red , respectively . Image series corresponds to biosensor fluorescence ( YFP , left ) , dye fluorescence ( middle ) , and merged ( right ) . ( F–I ) Representative time-courses showing the yellow/cyan ( Y/C ) emission ratio changes from ( F ) cytoCaNAR2 ( n = 29 ) , ( G ) pmCaNAR2 ( n = 22 ) , ( H ) mitoCaNAR2 ( n = 11 ) , and ( I ) erCaNAR2 ( n = 15 ) ( black curves ) , along with the red fluorescence intensity changes from RCaMP ( red curves ) , in MIN6 cells stimulated with 20 mM TEA . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 00410 . 7554/eLife . 03765 . 005Figure 2—figure supplement 1 . CaNAR expression levels do not affect subcellular calcineurin response dynamics . ( A ) Representative time-courses showing the TEA-stimulated response in MIN6 cells expressing high ( 1812 ) , medium ( 521 ) , or low ( 42 ) levels of cytoCaNAR2 . ( B ) Representative time-courses showing the TEA-stimulated response in MIN6 cells expressing high ( 1441 ) , medium ( 485 ) , or low ( 158 ) levels of pmCaNAR2 . ( C ) Representative time-courses showing the TEA-stimulated response in MIN6 cells expressing high ( 658 ) , medium ( 384 ) , or low ( 91 ) levels of mitoCaNAR2 . ( D ) Representative time-courses showing the TEA-stimulated response in MIN6 cells expressing high ( 969 ) , medium ( 599 ) , or low ( 120 ) levels of erCaNAR2 . Values in parentheses correspond to background-subtracted YFP intensities . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 005 Using both cytosolic and plasma membrane-targeted CaNAR2 , we were able to observe integrating , step-like patterns of calcineurin activity in response to TEA-induced Ca2+ oscillations , with each step-increase in calcineurin activity synchronized to a cytosolic Ca2+ peak , as measured by RCaMP ( Figure 2F , G , Figure 2—figure supplement 1A , B ) . We have seen previously that the FRET signal from CaNAR , which is based on calcineurin-dependent dephosphorylation of the N-terminal domain of nuclear-factor of activated T-cells ( NFAT ) , is not easily reversed and can remain elevated even after cytosolic Ca2+ has returned to basal levels ( Newman and Zhang , 2008 ) . Indeed , these observations appear to confirm previous reports suggesting that calcineurin and NFAT behave as signal integrators to form a working memory of oscillatory Ca2+ signals , owing to the rapid dephosphorylation and slow rephosphorylation kinetics of NFAT family members ( Tomida et al . , 2003; Colella et al . , 2008 ) . To our surprise , however , mitochondrial and ER-targeted CaNAR2 both exhibited far more reversible responses to cytosolic Ca2+ oscillations ( Figure 2H , I , Figure 2—figure supplement 1D , E ) . This difference was particularly striking for erCaNAR2 , which exhibited almost perfect calcineurin activity oscillations upon TEA stimulation . Furthermore , these response patterns were independent of the CaNAR expression level ( Figure 2—figure supplement 1 ) , indicating that they reflect genuine differences in endogenous calcineurin signaling . Our findings suggest the existence of two discrete subcellular zones ( e . g . , cytosol/plasma membrane and ER/mitochondria ) with distinct calcineurin signaling activities in MIN6 β-cells . We therefore investigated the molecular mechanisms responsible for defining these differential signaling zones . In particular , we focused on the reversibility of the CaNAR response at the ER surface , which suggested that calcineurin activity within this region of the cell is being precisely balanced by the action of endogenous kinases . A number of kinases have been shown to re-phosphorylate NFAT ( reviewed in Crabtree and Olson , 2002; Hogan et al . , 2003 ) and are thus potentially capable of reversing the response from CaNAR . Notably , PKA phosphorylates multiple residues in NFAT ( Beals et al . , 1997; Sheridan et al . , 2002 ) , and PKA has repeatedly been shown to antagonize calcineurin signaling in the regulation of a variety of cellular processes such as exocytosis ( Lester et al . , 2001 ) , excitation-contraction coupling ( Santana et al . , 2002 ) , and mitochondrial division ( Cribbs and Strack , 2007 ) . We therefore tested whether PKA activity is involved in reversing the ER-localized CaNAR response . In addition , we also tested the role of the Ca2+- and diacylglycerol-stimulated protein kinase ( PKC ) , which may also phosphorylate NFAT ( San-Antonio et al . , 2002 ) . Blocking PKA activity by treating TEA-stimulated MIN6 cells with the PKA inhibitor H89 appeared to abolish erCaNAR2 oscillations , leading to a delayed but pronounced increase in the CaNAR response ( Figure 3A , B ) . In addition , H89 treatment completely abolished the Ca2+ oscillations , which is consistent with the behavior of a previously described Ca2+/cAMP/PKA oscillatory circuit in MIN6 β-cells ( Ni et al . , 2010 ) . Curiously , the increased response from erCaNAR2 coincided with the attenuation of Ca2+ oscillations and the return of Ca2+ to basal levels . Despite the apparent lack of Ca2+ signaling , the observed increase in the erCaNAR2 response was nonetheless specifically caused by calcineurin activity , as the increase could be blocked by addition of the calcineurin inhibitor cyclosporin A prior to H89 treatment ( Figure 3E ) . Moreover , addition of the PKC inhibitor Gö6983 to TEA-stimulated MIN6 cells also resulted in a steadily increasing erCaNAR2 response , accompanied by a transition from Ca2+ oscillations to persistently elevated cytosolic Ca2+ levels ( Figure 3C , D ) . 10 . 7554/eLife . 03765 . 006Figure 3 . Effect of PKA and PKC inhibition on TEA-induced ER calcineurin activity oscillations . ( A ) Representative time-course showing the effect of 20 μM H89 treatment on the TEA-stimulated responses from erCaNAR2 ( black curve ) and RCaMP ( red curve ) in MIN6 cells ( n = 11 ) . ( B ) Expanded time-course showing the TEA-stimulated responses from ( A ) . ( C ) Representative time-course showing the effect of 10 μM Gö6983 on the TEA-stimulated responses from erCaNAR2 ( black curves ) and RCaMP ( red curve ) in MIN6 cells ( n = 6 ) . ( D ) Expanded time-course showing the TEA-stimulated responses from ( C ) . ( E ) Representative time-course showing the responses from erCaNAR2 ( black curve ) and RCaMP ( red curve ) in MIN6 cells treated with 20 mM TEA , 6 μM cyclosporin A ( CsA ) , and 20 μM H89 at the indicated times ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 006 The fact that both inhibitor treatments altered the underlying Ca2+ dynamics prevented us from reaching any firm conclusions regarding the roles of PKA and PKC based on these experiments . We therefore sought an alternative method for generating repetitive Ca2+ spikes in MIN6 cells . TEA functions by blocking plasma membrane K+ channels , leading to membrane depolarization , and thereby activating voltage-gated Ca2+ channels ( Hille , 1967; Wang and Greer , 1995 ) . The direct addition of KCl also promotes depolarization-induced Ca2+ influx in electrically excitable cells such as β-cells ( Bianchi and Shanes , 1959; Powell et al . , 1984; Bading et al . , 1993; Graef et al . , 1999; Macías et al . , 2001; Everett and Cooper , 2013 ) , and repeated cycles of KCl addition and wash-out were successfully able to mimic TEA-stimulated oscillations in cytosolic Ca2+ levels and ER-localized calcineurin activity in MIN6 cells ( Figure 4A ) . 10 . 7554/eLife . 03765 . 007Figure 4 . PKA antagonizes ER calcineurin activity in MIN6 cells . ( A ) Manual induction of Ca2+ oscillations in MIN6 cells via repeated addition and washout of 15 mM KCl . Addition of KCl rapidly increases the responses from RCaMP ( red curve ) and erCaNAR2 ( black curve ) , which are then both reversed upon washout . Repeating this process generates oscillatory responses . ( B ) Representative time-course showing the effect of 20 μM H89 treatment on the KCl-induced erCaNAR2 ( black curve ) and RCaMP ( red curve ) response in MIN6 cells ( n = 5 ) . H89 was added prior to the initial KCl treatment , and the H89 concentration in the experiment was maintained by re-addition of 20 μM H89 after washing out KCl . Although RCaMP did not detect Ca2+ responses in this experiment , Ca2+ spikes could clearly be seen using a high-affinity Ca2+ probe ( Figure 4—figure supplement 1 ) . ( C ) Representative time-course showing the effect of 10 μM Gö6983 treatment on the KCl-induced erCaNAR2 ( black curve ) and RCaMP ( red curve ) response in MIN6 cells ( n = 4 ) . Gö6983 was added prior to the initial KCl treatment , and the Gö6983 concentration in the experiment was maintained by re-addition of 10 μM Gö6983 after washing out KCl . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 00710 . 7554/eLife . 03765 . 008Figure 4—figure supplement 1 . YC-Nano50 detects KCl-induced Ca2+ influx in the presence of H89 . Representative curves showing the responses of both YC-Nano50 ( black curves ) and RCaMP ( red curves ) to repeated addition and wash-out of KCl in the ( A ) absence and ( B ) presence of 20 μM H89 treatment . H89 was added prior to the initial KCl treatment , and the H89 concentration in the experiment was maintained by re-addition of 20 μM H89 after washing out KCl . H89 abolishes the RCaMP response , whereas YC-Nano50 is still able to detect KCl-induced Ca2+ spikes . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 008 Using this approach , we found that the addition of H89 resulted in an integrative , step-like response from erCaNAR2 ( Figure 4B ) , much like that seen with cytoCaNAR2 in TEA-stimulated cells ( Figure 2F ) , although RCaMP did not appear to respond under these conditions . However , we were able to confirm the generation of Ca2+ spikes using a higher affinity probe , YC-Nano50 ( Figure 4—figure supplement 1; Horikawa et al . , 2010 ) . In contrast to H89 treatment , the inclusion of Gö6983 to inhibit PKC activity did not appear to alter the response of erCaNAR2 to successive KCl treatments ( Figure 4C ) . Here , the addition of the inhibitor resulted in an immediate increase in the erCaNAR2 FRET signal , though this was most likely due to non-specific increases in the background fluorescence due to the addition of Gö6983 , which is fluorescent . Nevertheless , KCl-induced erCaNAR2 oscillations were still clearly observable above this increased background , as were the RCaMP Ca2+ spikes . Our findings indicated that PKA activity , but not PKC activity , antagonizes calcineurin near the ER surface , thereby giving rise to ER-localized calcineurin activity oscillations in MIN6 cells . Based on these results , we tested whether differences in the level of PKA activity present in the cytosol and ER might contribute to differential calcineurin response patterns . To do so , we utilized variants of the FRET-based PKA activity reporter AKAR4 ( Depry et al . , 2011 ) that were localized to the cytoplasm or ER via fusion to the targeting sequences described above ( Figure 5A ) . We then compared the relative amounts of PKA activity in these two compartments by normalizing the TEA-stimulated PKA responses ( Figure 5C , D ) with respect to the total amount of PKA activity available in the cytosol and ER , which was defined as the maximum subcellular response observed upon combined treatment with the adenylyl cyclase activator forskolin ( Fsk ) and the general phosphodiesterase inhibitor 3-isobutyl-1-methylxanthine ( IBMX ) . Interestingly , we detected slightly less PKA activity at the ER , with TEA-stimulated AKAR4 responses reaching 55 . 4 ± 5 . 9% and 39 . 8 ± 3 . 3% ( p = 0 . 0249 ) of the maximum dynamic range in the cytosol and at the ER , respectively ( Figure 5B ) . We also found that setting PKA activity to maximum levels using IBMX pretreatment in conjunction with KCl stimulation , thus clamping PKA activity in these regions ( Figure 5—figure supplement 1 ) , did not affect the subcellular CaNAR response patterns ( Figure 5E–H ) . 10 . 7554/eLife . 03765 . 009Figure 5 . Characterization of cytosolic and ER-localized PKA activity in MIN6 cells . ( A ) Schematic illustrating the domain structures of cyto- and erAKAR4 . ( B ) Comparison of fractional PKA activity levels in the cytosol and ER of MIN6 cells . To determine the relative fraction of PKA activity being induced by TEA stimulation in each subcellular compartment , the FRET ratio change from each TEA-induced cytoAKAR4 or erAKAR4 response was divided by the maximum FRET ratio change observed upon co-stimulation with 50 μM Fsk and 100 μM IBMX . Data shown are presented as mean ± SEM , with n = 36 and 34 for cytoAKAR4 and erAKAR4 , respectively . ( C and D ) Representative time-courses showing the responses from ( C ) cytoAKAR4 and ( D ) erAKAR4 ( black curves ) , along with RCaMP ( red curves ) , in MIN6 cells treated with 20 mM TEA . ( E-H ) Effect of maximal PKA activity on subcellular CaNAR2 responses . Representative time-courses showing the KCl-induced cytoCaNAR2 response in the ( E ) absence ( n = 26 ) or ( G ) presence ( n = 14 ) of 100 μM IBMX treatment in MIN6 cells , and representative time-courses showing the KCl-induced erCaNAR2 response in the ( F ) absence ( n = 18 ) or ( H ) presence ( n = 8 ) of 100 μM IBMX treatment in MIN6 cells . IBMX was added prior to the initial KCl treatment , and the IBMX concentration in the experiment was maintained by re-addition of 100 μM IBMX after washing out KCl . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 00910 . 7554/eLife . 03765 . 010Figure 5—figure supplement 1 . Blocking oscillatory PKA activity in the cytosol and at the ER surface in MIN6 cells . ( A and B ) Representative time-courses showing the KCl-induced cytoAKAR4 response in the absence ( A ) or presence ( B ) of 100 μM IBMX treatment in MIN6 cells . ( C and D ) Representative time-courses showing the KCl-induced erCaNAR2 response in the absence ( C ) or presence ( C ) of 100 μM IBMX treatment in MIN6 cells . IBMX was added prior to the initial KCl treatment , and the IBMX concentration in the experiment was maintained by re-addition of 100 μM IBMX after washing out KCl . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 010 Given that Ca2+ oscillations actually induce a lower amount of PKA activity at the ER surface compared with the cytosol , the apparent inability of PKA to reverse the cytosolic CaNAR response is puzzling . One possible explanation is that the kinetics of calcineurin activation itself may in fact differ in these two subcellular zones . Specifically , we reasoned that calcineurin may remain in an activated state for longer periods in the cytosol than at the ER surface in response to the individual Ca2+ pulses that occur during Ca2+ oscillations , thus potentially rendering cytosolic calcineurin activity less prone to the antagonistic effects of cytosolic PKA activity . To test this hypothesis , we took advantage of the intrinsic conformational changes that are associated with calcineurin activation . Calcineurin exists as a stable heterodimer between a regulatory subunit ( CNB ) and a catalytic subunit ( CNA ) and becomes activated when Ca2+-bound CaM ( Ca2+/CaM ) binds to the regulatory arm of CNA , thus driving a conformational change that removes calcineurin from its basal , auto-inhibited state ( Rusnak and Mertz , 2000; Wang et al . , 2008; Rumi-Masante et al . , 2012 ) . Guided by this information , we constructed a FRET-based reporter for monitoring calcineurin activation by sandwiching the catalytic subunit of calcineurin between Cerulean ( i . e . , CFP ) and Venus ( i . e . , YFP ) ( Figure 6A ) . The resulting Calcineurin Activation Ratiometric indicator , or CaNARi , exhibited a robust increase in the cyan-to-yellow fluorescence emission ratio , which tracked closely with the increase in Ca2+ detected using RCaMP , in response to cytosolic Ca2+ elevation in HEK293 cells ( Figure 6B ) . This Ca2+-induced FRET change could be blocked by pretreatment with the membrane-permeable CaM antagonist W7 , indicating that the CaNARi response is in fact dependent on the binding of Ca2+/CaM ( Figure 6C ) . 10 . 7554/eLife . 03765 . 011Figure 6 . Development of CaNARi , a FRET-based reporter of calcineurin activation . ( A ) Schematic illustrating the domain structure of CaNARi and the proposed Ca2+/CaM-induced molecular switch . ( B ) Representative time-course showing the cyan/yellow ( C/Y ) emission ratio change from CaNARi ( black curve ) , along with the RCaMP response ( red curve ) , in HEK293 cells treated with 1 μM ionomycin ( iono ) . ( C ) Summary of C/Y emission ratio responses from CaNARi in HEK293 cells stimulated using 1 μM iono with ( +W7 , n = 18 ) or without ( −W7 , n = 21 ) pretreating for 30 min with 50 μM of the CaM antagonist W7 . Data shown are presented as mean ± SEM . *p < 0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 011 We then tested whether the kinetics of calcineurin activation display subcellular variations in response to Ca2+ oscillations by localizing CaNARi to the cytosol and plasma membrane , as well as to the mitochondrial and ER surfaces , in MIN6 cells via fusion to the aforementioned targeting sequences ( Figure 7A ) . As above , TEA stimulation induced cytosolic Ca2+ oscillations in the CaNARi-expressing MIN6 cells , as determined using co-expressed RCaMP . However , in contrast to the subcellular CaNAR responses , the responses from the targeted CaNARi variants were similar in all of the subcellular compartments that we examined ( Figure 7C–F ) . Specifically , the responses from CaNARi in each subcellular region closely matched the cytosolic Ca2+ dynamics , rapidly increasing as Ca2+ levels rose and subsequently decreasing as Ca2+ fell back to basal levels , and revealed no major subcellular differences in how quickly calcineurin was turned on or off in response to oscillating Ca2+ levels . These results indicate that the kinetics of calcineurin activation are similar throughout the cell under these conditions . 10 . 7554/eLife . 03765 . 012Figure 7 . Ca2+ oscillations induce uniform subcellular calcineurin activation patterns . ( A ) Schematic illustrating the domain structures of the subcellularly targeted CaNARi variants . ( B ) Summary of the C/Y emission ratio responses from each subcellularly targeted CaNARi variant . Data shown are presented as mean ± SEM , with n = 16 , 13 , 10 , and 12 for cytoCaNARi , pmCaNARi , mitoCaNARi , and erCaNARi , respectively . ( C–F ) Representative time-courses showing the responses from ( C ) cytoCaNARi , ( D ) pmCaNARi , ( E ) mitoCaNARi , and ( F ) erCaNARi ( black curves ) , along with the response from RCaMP ( red curves ) , in MIN6 cells treated with 20 mM TEA . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 01210 . 7554/eLife . 03765 . 013Figure 7—figure supplement 1 . Subcellular CaNARi responses are not affected by the reporter expression level . Scatter plots showing the maximum C/Y FRET ratio change vs YFP intensity ( i . e . , reporter expression level ) for ( A ) cytoCaNARi , ( B ) pmCaNARi , ( C ) mitoCaNARi , and ( D ) erCaNARi . Weakly positive correlations were detected using linear regression analyses , though none were statistically significant . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 013 Although we were unable to observe any subcellular differences in the kinetics of calcineurin activation using CaNARi , our findings do not completely rule out a possible role for variations in subcellular calcineurin activation in regulating subcellular calcineurin activity . In fact , a closer inspection of the subcellularly targeted CaNARi responses revealed that both mitochondrial and ER-targeted CaNARi exhibited slightly smaller TEA-stimulated responses compared with the cytosolic and plasma membrane probes ( Figure 7B ) . We also observed no correlation between the CaNARi response amplitude and the biosensor expression level ( Figure 7—figure supplement 1 ) , suggesting this difference was not an artifact of reporter expression . Although the trend was not statistically significant , these results lend credence to the hypothesis that different amounts of calcineurin are being activated in different parts of the cell . We then investigated whether variations in upstream signaling components could be generating differences in subcellular calcineurin activation . Using subcellularly targeted versions of the green-fluorescent Ca2+ sensor GCaMP3 ( Tian et al . , 2009 ) , we were unable to detect any obvious differences in TEA-stimulated Ca2+ dynamics in the cytosol or at the ER surface with respect to the RCaMP response , although we did observe a steady rise in the basal Ca2+ level using ER-targeted GCaMP3 ( Figure 8 ) . These results largely agree with our subcellular CaNARi data and suggest that the divergent subcellular calcineurin activity patterns are not caused by local differences in the underlying Ca2+ dynamics ( e . g . , influx or efflux ) . In addition , only a small difference was observed when we compared the magnitude of these local Ca2+ signals , with a slightly lower amount of Ca2+ near the ER surface than in the cytosol ( 71 . 4% of max Ca2+ in the cytosol vs 61 . 2% at the ER , p = 0 . 0049 ) . 10 . 7554/eLife . 03765 . 014Figure 8 . Subcellular Ca2+ dynamics match global Ca2+ dynamics in MIN6 cells . Representative curves showing the response from ( A ) cytoGCaMP3 ( n = 23 ) or ( B ) erGCaMP3 ( n = 38 ) ( green curves ) and RCaMP ( red curves ) in MIN6 cells treated with 20 mM TEA . ( A ) To compare the Ca2+ dynamics in different subcellular compartments in MIN6 cells , GCaMP3 and RCaMP were first co-expressed in the cytosol . TEA stimulation induced completely overlapping responses from both probes , indicating that they have similar properties , though the RCaMP response amplitude often decreased over time . ( B ) GCaMP3 was then targeted to the ER surface while RCaMP was kept in the cytosol . Upon TEA stimulation , the GCaMP3 response again tracked closely with the RCaMP response , with no noticeable differences in the timing of Ca2+ increases or decreases between the ER and cytosol . The basal GCaMP response drifted upwards at the ER surface , but this did not affect the overall dynamics of the Ca2+ oscillations . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 014 Calcineurin activation is also completely dependent on the binding of Ca2+/CaM , and multiple studies have shown that CaM is a limited cellular resource ( reviewed in Persechini and Stemmer , 2002; Saucerman and Bers , 2012 ) . Indeed , the concentration of free CaM often reaches only 75 nM ( Persechini and Stemmer , 2002; Wu and Bers , 2007; Saucerman and Bers , 2012 ) , whereas CaM targets can be in excess of 20 μM ( Saucerman and Bers , 2012 ) . Similarly , based on the kinetics of Ca2+ binding to and dissociation from the N- and C-lobes of CaM , it was calculated that Ca2+/CaM would only diffuse ∼0 . 1 μm before Ca2+ begins to dissociate , suggesting that Ca2+/CaM primarily acts as a highly localized signal ( Saucerman and Bers , 2012 ) . Rapid , long-range signaling may even require actively transporting CaM to other parts of the cell ( Deisseroth et al . , 1998 ) . Thus , taking into account both the scarcity of Ca2+/CaM and its limited range of action , along with the fact that calcineurin activity increases as a direct function of CaM concentration ( Quintana et al . , 2005 ) , it is conceivable that even subtle variations in the availability of Ca2+/CaM could have a significant effect on local calcineurin activity ( Figure 9A ) . 10 . 7554/eLife . 03765 . 015Figure 9 . Subcellular Ca2+/CaM levels determine local calcineurin activity dynamics . ( A ) Model for the regulation of subcellular calcineurin activity in MIN6 cells by local variations in free Ca2+/CaM levels . As depicted in this illustration , CaM is predicted to be relatively abundant in the cytosol , thereby leading to strong activation of calcineurin ( CaN ) in the cytosol ( red curve ) . Conversely , CaM is predicted to be present in much lower quantities near the ER surface , thereby leading to weaker levels of calcineurin activation at any given Ca2+ concentration in this part of the cell ( green curve ) . ( B ) Summary of the C/Y emission ratio responses of BSCaM-2 expressed in the cytosol ( Cyto ) and at the ER surface without ( ER ) or with ( ER+CaM ) the overexpression of CaM in MIN6 cells . To achieve maximal levels of free Ca2+/CaM , cells were treated with 5 mM CaCl2 and 5 μM ionomycin . Data shown are presented as mean ± SEM , with n = 31 , 33 , and 34 for Cyto , ER , and ER+CaM , respectively . *p < 0 . 0001 . ( C and D ) Representative time-courses showing the KCl-induced responses from erCaNAR2 in the ( C ) absence ( n = 18 ) and ( D ) presence ( n = 7 ) of CaM overexpression in MIN6 cells . ( E and F ) Representative time-courses showing the KCl-induced responses from cytoCaNAR2 in the ( E ) absence ( n = 26 ) and ( F ) presence ( n = 23 ) of 20 μM W7 treatment in MIN6 cells . W7 was added prior to the initial KCl treatment , and the W7 concentration in the experiment was maintained by re-addition of 20 μM W7 after washing out KCl . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 01510 . 7554/eLife . 03765 . 016Figure 9—figure supplement 1 . Subcellular BSCaM-2 responses are not affected by the reporter expression level . Scatter plots showing the maximum C/Y FRET ratio change vs YFP intensity ( i . e . , reporter expression level ) for ( A ) BSCaM-2 , ( B ) erBSCaM-2 , and ( C ) erBSCaM-2 + CaM , which correspond to ‘Cyto’ , ‘ER’ , and ‘ER+CaM’ in Figure 9B . Very weak positive or negative correlations were detected using linear regression analyses , though none of the correlations were statistically significant . DOI: http://dx . doi . org/10 . 7554/eLife . 03765 . 016 To investigate subcellular differences in Ca2+/CaM levels , we used the FRET-based biosensor BSCaM-2 , which has been used previously to measure free Ca2+/CaM levels in living cells ( Persechini and Cronk , 1999; Tran et al . , 2003 ) , as well as to investigate subcellular differences in free Ca2+/CaM levels ( Teruel et al . , 2000 ) . This reporter features a modified version of the Ca2+/CaM-binding sequence from avian smooth muscle myosin light chain kinase that binds Ca2+/CaM with a Kd of 2 nM . Calcineurin has been shown to bind Ca2+/CaM with sub-nanomolar affinity ( Hubbard and Klee , 1987; Quintana et al . , 2005 ) , therefore we reasoned that BSCaM-2 should offer a fair approximation of how much Ca2+/CaM is locally accessible by calcineurin . Our model predicts that free Ca2+/CaM is less abundant in the vicinity of the ER compared with the bulk cytosol ( Figure 9A , left ) , and indeed we observed a significantly lower ( p < 0 . 0001 ) FRET response when BSCaM-2 was tethered to the ER surface than when it was able to diffuse freely through the cytosol ( Figure 9B ) , suggesting that less Ca2+/CaM is present near the ER . These responses are similar to those observed by Teruel and colleagues , who reported smaller increases in free nuclear Ca2+/CaM levels compared with the cytosol in response to Ca2+ transients ( Teruel et al . , 2000 ) . Our results also appear to reflect actual differences in subcellular Ca2+/CaM levels as opposed to Ca2+/CaM buffering , given that biosensor expression levels did not appear to affect the FRET responses ( Figure 9—figure supplement 1 ) . More importantly , we found that overexpressing mCherry-tagged CaM was able to rescue most of the difference between the ER and cytosolic FRET responses ( Figure 9B , ‘ER+CaM’ ) . Our model suggests that lower levels of Ca2+/CaM will result in weaker calcineurin activation near the ER surface ( Figure 9A , red and green curves ) , which would in turn translate into lower levels of calcineurin activity that are more susceptible to antagonism by PKA activity . To test this model directly , we reasoned that if calcineurin activity is in fact being affected by local Ca2+/CaM levels , it should then be possible to generate cytosol-like CaNAR responses with erCaNAR2 by overexpressing CaM . Remarkably , combining mCherry-tagged CaM overexpression with the application of repeated , KCl-induced Ca2+ transients in CaNAR-expressing cells reveals that this is indeed the case . In contrast to cells expressing erCaNAR2 alone , which exhibit oscillatory FRET responses in response to repeated KCl stimulation and washout ( Figure 9C ) , the co-expression of CaM-mCherry alongside erCaNAR2 clearly results in integrated calcineurin activity responses similar to those seen in the cytosol ( Figure 9D ) . Conversely , reducing the amount of available Ca2+/CaM should lead to ER-like CaNAR oscillations in the cytosol . Indeed , pretreating cells with a low dose ( 20 μM ) of the CaM antagonist W7 gave rise to oscillatory responses from cytoCaNAR2 , in contrast to integrating responses in cells lacking W7 pretreatment ( Figure 9E , F ) . Taken together , our results strongly suggest that free concentrations of Ca2+/CaM are limiting near the ER surface and thus significantly modulate the local , Ca2+ oscillation-induced calcineurin activity dynamics in this subcellular region .
The spatiotemporal regulation of calcineurin signaling has come under increased scrutiny of late . Recently , calcineurin responses in cortical neurons treated with the amyloid-β peptide were shown to differ subcellularly , with more rapid calcineurin activation occurring in dendritic spines than in the cytosol and nucleus ( Wu et al . , 2012 ) . Calcineurin dynamics are also predicted to differ significantly within the dyadic cleft and cytosol in cardiomyocytes ( Saucerman and Bers , 2008 ) . In keeping with these findings , our investigation revealed subcellular differences in the temporal pattern of calcineurin activity in response to Ca2+ oscillations in pancreatic β-cells . Specifically , cytosolic and plasma membrane calcineurin activity was observed to integrate Ca2+ oscillations , whereas Ca2+ oscillations evoked intermittent , oscillating calcineurin activity at the ER and mitochondria . Given the wide variety of cellular functions regulated by calcineurin signaling and the significant role of subcellular compartments in modulating signaling molecule behavior ( see Mehta and Zhang , 2010 ) , this phenomenon is likely to shape calcineurin activity patterns in other cell types as well . Our investigation into the spatiotemporal dynamics of calcineurin signaling turns on the use of a pair of FRET-based biosensors , each giving distinct responses based on its specific properties and thereby offering a multifaceted view of calcineurin behavior in living cells . The CaNAR family , including CaNAR2 and its precursor CaNAR1 ( Newman and Zhang , 2008 ) , utilizes the well-characterized dephosphorylation of NFAT to report on the substrate-level dynamics of calcineurin activity and is sensitive to multiple cellular factors , such as both phosphatase and kinase activity . We also generated CaNARi , which reports on the activation of calcineurin upon the binding of Ca2+/CaM . The CaNARi response is exclusively determined by the intrinsic affinities between Ca2+ , CaM , and calcineurin , and CaNARi revealed largely uniform subcellular calcineurin activation patterns during Ca2+ oscillations . On the other hand , CaNAR was able to detect clear subcellular variations in calcineurin activity , which stems from the fact that CaNAR monitors the net endogenous calcineurin activity at each location . The enzymatic nature of CaNAR , wherein calcineurin dephosphorylates many probe molecules , also makes it more sensitive for detecting weak calcineurin signals . However , given the somewhat peculiar dephosphorylation and rephosphorylation behavior of NFAT ( Okamura et al . , 2000; Tomida et al . , 2003 ) , CaNAR may not reflect calcineurin activity towards all targets . Additional approaches , including redesigned CaNARs , are therefore needed to provide a more complete picture . The differential calcineurin activity dynamics suggest that different cellular compartments are tuned for regulating distinct calcineurin targets . For example , mice specifically lacking calcineurin in β-cells show impaired insulin production and decreased expression of several critical genes , all of which are regulated by the calcineurin-dependent transcription factor NFAT ( Heit et al . , 2006 ) . The efficient activation and nuclear translocation of NFAT in response to Ca2+ oscillations requires periods of sustained calcineurin activity to produce a cytoplasmic pool of dephosphorylated NFAT that is available for nuclear import ( Tomida et al . , 2003 ) . Correspondingly , the sustained , integrating calcineurin activity patterns we observed in the cytosolic and plasma membrane regions of MIN6 cells indicated that this signaling domain is optimized for the oscillatory control of transcriptional regulation . In fact , the plasma membrane may play a particularly crucial role in promoting calcineurin/NFAT signaling . The scaffolding protein AKAP79/150 has specifically been shown to recruit calcineurin into a signaling complex located at the C-terminus of the L-type voltage-gated Ca2+ channel ( VGCC ) in hippocampal neurons ( Oliveria et al . , 2007 ) , and the presence of anchored calcineurin within this complex was required for proper NFAT-dependent nuclear signaling in these cells ( Oliveria et al . , 2007; Li et al . , 2012a ) . Both Ca2+ influx through the L-type VGCC and AKAP anchoring of calcineurin are also important for insulin secretion ( Bokvist et al . , 1995; Wiser et al . , 1999; Barg et al . , 2001; Lester et al . , 2001; Hinke et al . , 2012 ) , therefore it is possible that an AKAP-calcineurin-VGCC complex is similarly involved in calcineurin/NFAT signaling in β-cells . In contrast to the cytosol and plasma membrane , however , the oscillatory calcineurin activity we observed near the ER and mitochondria in MIN6 cells suggests weak and intermittent NFAT activation ( Tomida et al . , 2003 ) . As such , transcriptional signaling via NFAT is likely not the primary function of calcineurin in these subcellular regions . Our previous findings indicate that oscillatory signals can spatially restrict enzyme activity ( Ni et al . , 2010 ) ; thus , calcineurin activity oscillations may alternatively be directed towards local ER and mitochondrial targets . For instance , calcineurin was recently shown to dephosphorylate the Ca2+-dependent chaperone calnexin and interact with the ER kinase PERK to modulate protein folding and ER stress ( Bollo et al . , 2010 ) , while calcineurin and PERK also jointly regulate insulin secretion ( Wang et al . , 2013 ) . Similarly , calcineurin dephosphorylates and activates the mitochondrial fission protein Drp1 ( Cribbs and Strack , 2007; Cereghetti et al . , 2008 , 2010; Slupe et al . , 2013 ) . Interestingly , Drp1-mediated mitochondrial division involves direct contact between the outer mitochondrial membrane and ER tubules ( Friedman et al . , 2011 ) . It is conceivable that such contact points are hotspots for local calcineurin signaling; the ER and mitochondria are in fact characterized by extensive physical and functional coupling ( reviewed in de Brito and Scorrano , 2010; Rowland and Voeltz , 2012 ) . Further analyses revealed that PKA , which opposes calcineurin activity towards Drp1 ( Cribbs and Strack , 2007 ) , antagonizes calcineurin at the ER and helps give rise to calcineurin activity oscillations . Curiously , inhibiting PKA in TEA-stimulated MIN6 cells led to steadily increasing ER calcineurin activity , despite cytosolic Ca2+ apparently returning to basal levels . This effect was blocked by calcineurin inhibition and may be due to residual activity from ER Ca2+ release channels or to passive Ca2+ leak from the ER ( Camello et al . , 2002; Szlufcik et al . , 2012; Hammadi et al . , 2013 ) . However , PKA activity was not sufficient to produce calcineurin activity oscillations , as even saturating amounts of PKA activity did not lead to CaNAR oscillations in the cytosol . AKAPs may be involved in locally promoting the PKA-mediated antagonism of calcineurin signaling , and DAKAP1 ( AKAP121 ) , which binds calcineurin ( Abrenica et al . , 2009 ) , in fact targets to both the mitochondrial and ER membranes ( Ma and Taylor , 2008 ) . AKAP79 is also known to anchor PKA and calcineurin to the plasma membrane in β-cells ( Lester et al . , 2001 ) , yet calcineurin responses nevertheless appear integrative in this region . We also detected less free Ca2+/CaM near the ER surface compared with the cytosol , and CaM overexpression led to integrating rather than oscillating ER calcineurin activity , whereas inhibiting CaM produced the opposite effect in the cytosol . Both of these results are consistent with a model in which limited access to CaM at the ER leads to weak calcineurin activity . Since excess calcineurin activity can lead to β-cell dysfunction and death ( Bernal-Mizrachi et al . , 2010 ) , often involving ER stress and apoptosis , this combination of PKA activity and limited access to CaM may help constrain local calcineurin signals within physiologically permissible limits . Our results clearly show that the distribution of CaM directly determines subcellular calcineurin activity during β-cell Ca2+ oscillations . CaM acts as both a dedicated regulatory subunit and a promiscuous binding partner depending on the specific target ( Saucerman and Bers , 2012 ) , and whereas promiscuous CaM is free to associate with and dissociate from target proteins as a function of Ca2+ , dedicated CaM remains bound irrespective of changes in the Ca2+ concentration . Subcellular variations in the amounts of dedicated CaM targets may therefore affect the levels of promiscuous CaM that can activate calcineurin in response to Ca2+ signals . Moreover , the Ca2+-binding kinetics of CaM limit its range of action and potentially require CaM to be activated within Ca2+ microdomains ( Saucerman and Bers , 2012 ) . In β-cells , Ca2+ oscillations are often driven by VGCCs in the plasma membrane ( Ashcroft and Rorsman , 1989; Hellman et al . , 1992; Bokvist et al . , 1995; Mears , 2004; Tengholm and Gylfe , 2008 ) , and CaM has been found to be locally enriched near these channels ( Mori , 2004 ) , ready to be activated by local Ca2+ . AKAP79 has also been shown to bind CaM ( Gold et al . , 2011 ) , thereby serving as another potential source for local Ca2+/CaM and also perhaps lowering the ability of anchored PKA to antagonize local calcineurin activity . However , additional studies are needed to unravel the precise mechanisms underlying the spatial differences in CaM levels . In neurons and cardiomyocytes , calcineurin is implicated to behave as an integrator or peak number counter in response to Ca2+ oscillations ( Saucerman and Bers , 2008; Song et al . , 2008; Li et al . , 2012b; Fujii et al . , 2013 ) . In these cells , rapid ( i . e . , > 1 s−1 ) Ca2+ transients , combined with the slow dissociation of calcineurin and Ca2+/CaM , facilitate persistent calcineurin activation ( Song et al . , 2008 ) . This also engenders the preferential activation of calcineurin at lower oscillatory frequencies compared with other targets ( e . g . , CaMKII ) , resulting in frequency decoding . In β-cells , on the other hand , Ca2+ oscillations are slow ( i . e . , < 1 min−1 ) ( Hellman et al . , 1992; Tengholm and Gylfe , 2008 ) , and our results indicate that Ca2+/CaM fully dissociates from calcineurin in between each Ca2+ peak , suggesting that frequency modulation may not be a prominent feature in β-cell calcineurin signaling . Rather , we found that CaM plays a critical role in spatial signaling under these conditions , which has been hinted at previously ( Teruel et al . , 2000 ) . Furthermore , the frequency decoding observed in neurons and cardiomyocytes is based entirely on the intrinsic kinetics of CaM-target interactions where CaM plays a passive role . Conversely , our analyses highlight a novel mechanism whereby cells utilize CaM to actively encode spatial information , which is then decoded by calcineurin to ensure that oscillatory Ca2+ signals are transduced properly within specific local contexts .
Each CaNAR variant was generated by sandwiching the substrate domain of CaNAR1 , which corresponds to amino acids 1–297 from the N-terminus of NFAT1 ( Newman and Zhang , 2008 ) , between different cyan ( CFP ) and yellow fluorescent protein ( YFP ) variants . Cerulean ( Rizzo et al . , 2004 ) , Cerulean2 , Cerulean3 ( Markwardt et al . , 2011 ) , circularly permuted Venus ( E172 ) ( Nagai et al . , 2004 ) , and YPet ( Nguyen and Daugherty , 2005 ) were each subcloned using BamHI and SphI restriction sites ( for CFP ) or SacI and EcoRI restrictions sites ( for YFP ) to replace the original ECFP or circularly permuted Venus ( L194 ) in CaNAR1 . All CaNAR constructs were generated in the pRSETB vector ( Invitrogen , Carlsbad , CA ) and then subcloned into pCDNA3 ( Invitrogen ) for subsequent mammalian expression . Plasma membrane- , outer mitochondrial membrane- , and ER-targeted CaNAR2 constructs were generated by in-frame fusion of full-length CaNAR2 with the N-terminal 11 amino acids from Lyn kinase , the N-terminal 30 amino acids from DAKAP1 , and the N-terminal 27 amino acids from CYP450 , respectively . Similarly , cytosolic CaNAR2 was generated by in-frame fusion of a C-terminal NES ( LPPLERLTL ) immediately prior to the stop codon . All constructs were verified by sequencing . CaNARi was generated in pCDNA3 . 1 ( + ) ( Invitrogen ) . The full-length α isoform of CNA was PCR amplified from pETCNα ( Mondragon et al . , 1997 ) , a gift of Fan Pan ( Johns Hopkins School of Medicine , Baltimore , MD ) , using primers incorporating a 5′ BamHI site and 3′ XhoI site . Cerulean was PCR amplified from plasmid DNA using a forward primer encoding a 5′ HindIII restriction site followed by a Kozak consensus sequence for mammalian expression ( Kozak , 1987 ) and a reverse primer encoding a 3′ BamHI site . Similarly , Venus was PCR amplified from plasmid DNA using primers encoding a 5′XhoI site and 3′ XbaI site . The resulting PCR fragments were digested with the corresponding restriction enzymes , gel purified , and ligated into pCDNA3 . 1 ( + ) . To generate the plasma membrane- , outer mitochondrial membrane- , and ER-targeted CaNARi constructs , Cerulean was PCR amplified from plasmid DNA using nested forward primers encoding a 5′ HindIII site , a Kozak translation initiation sequence , and the N-terminal targeting sequence from either Lyn kinase , DAKAP1 , or CYP450 , along with a reverse primer encoding a 3′ BamHI site . The PCR fragments were then subcloned into CaNARi in pCDNA3 . 1 ( + ) using HindIII and BamHI to replace the original Cerulean sequence . Similarly , cytosol-targeted CaNARi was generated by PCR amplification of Venus using a forward primer encoding a 5′ XhoI site and nested reverse primers encoding a 3′ XbaI site and an NES . The PCR fragment was subcloned into CaNARi in pCDNA3 . 1 ( + ) using XhoI and XbaI to replace the original Venus sequence . All constructs were verified by sequencing . ER-targeted AKAR4 was generated from AKAR4 ( Depry et al . , 2011 ) as above . The Ca2+/CaM biosensor BSCaM-2 ( Persechini and Cronk , 1999; Tran et al . , 2003 ) was a gift of Dr Anthony Persechini ( University of Missouri–Kansas City , Kansas City , MO ) . ER-targeted BSCaM-2 was generated by subcloning full-length BSCaM-2 between the HindIII and XbaI sites of pCDNA3 . 1 ( + ) containing the N-terminal targeting sequence from CYP450 . The CaM-mCherry construct was generated by PCR amplification of full-length CaM ( without a stop codon ) from plasmid DNA using primers encoding a 5′ NheI site and a 3′ BamHI site and PCR amplification of mCherry ( with a stop codon ) from plasmid DNA using primers encoding a 5′ BamHI site and 3′ EcoRI site . The resulting PCR fragments were digested using the corresponding restriction enzymes , gel purified , and ligated into pCDNA3 . 1 ( + ) . The red-fluorescent Ca2+ sensor RCaMP ( Akerboom et al . , 2013 ) and the green-fluorescent Ca2+ sensor GCaMP3 ( Tian et al . , 2009 ) were kind gifts of Dr Loren Looger ( Janelia Farm Research Campus , HHMI , Ashburn , VA ) . Subcellularly targeted versions of GCaMP3 were generated by PCR amplification of GCaMP3 ( with or without a stop codon ) using primers encoding a 5′ BamHI site and a 3′ EcoRI site and subcloning into a plasmid containing either an NES or an ER-targeting sequence as above . The high affinity FRET-based Ca2+ sensor YC-Nano50 ( Horikawa et al . , 2010 ) was provided Dr Takeharu Nagai ( Hokkaido University , Sapporo , Hokkaido , Japan ) . All constructs were verified by sequencing . HEK293 cells were cultured in Dulbecco modified Eagle medium ( DMEM; Gibco , Grand Island , NY ) containing 1 g/l glucose and supplemented with 10% ( vol/vol ) fetal bovine serum ( FBS , Sigma , St . Louis , MO ) and 1% ( vol/vol ) penicillin-streptomycin ( Pen-Strep , Sigma-Aldrich , St . Louis , MO ) . MIN6 β-cells were cultured in DMEM containing 4 . 5 g/l glucose and supplemented with 10% ( vol/vol ) FBS , 1% ( vol/vol ) Pen-Strep , and 50 μM β-mercaptoethanol . All cells were maintained in a humidified incubator at 37°C with a 5% CO2 atmosphere . Prior to transfection , cells were plated onto sterile , 35-mm glass-bottomed dishes and grown to 50–70% confluence . Cells were then transfected using calcium phosphate and grown an additional 24–48 hr ( HEK ) or using Lipofectamine 2000 ( Invitrogen ) and grown an additional 48–96 hr ( MIN6 ) before imaging . MIN6 β-cells expressing either mitochondrial or ER-targeted CaNAR2 were stained for 30 min with MitoTracker RED ( Molecular Probes , Eugene , OR ) or ER-Tracker RED ( Molecular Probes ) , respectively , at a final concentration of 1 μM in Hank’s Balanced Salt Solution ( HBSS ) . These cells , as well as cells expressing cytosolic or plasma membrane-targeted CaNAR2 , were imaged on a Nikon Eclipse Ti inverted fluorescence microscope ( Nikon Instruments , Melville , NY ) equipped with a perfect focus system ( Nikon ) , a 100x/1 . 49 NA oil-immersion objective lens , an electron multiplying charge coupled device camera ( Photometrics , Tucson , AZ ) , and a laser TIRF system ( Nikon ) . YFP and RFP images were acquired using a 514-nm argon laser ( Melles Griot , Rochester , NY ) and a 561-nm Sapphire solid-state laser ( Coherent , Santa Clara , CA ) , respectively . The system was controlled using the NIS-Elements software package ( Nikon ) . Exposure times were between 50 and 200 ms . Images were analyzed using ImageJ software ( http://imagej . nih . gov/ij/ ) . Cells were washed twice with HBSS and subsequently imaged in the dark at 37°C . Tetraethylammonium chloride ( TEA; Sigma ) , thapsigargin ( TG; Sigma ) , ionomycin ( iono; Calbiochem , San Diego , CA ) , calcium chloride ( CaCl2; Sigma ) , W-7 ( Cayman Chemical , Ann Arbor , MI ) , potassium chloride ( KCl; JT Baker , Phillipsburg , NJ ) , H89 ( Sigma ) , Gö6983 ( Sigma ) , forskolin ( Fsk; Calbiochem ) , and 3-isobutyl-1-methylxanthine ( IBMX; Sigma ) were added as indicated . Images were acquired using an Axiovert 200M inverted fluorescence microscope ( Carl Zeiss , Thornwood , NY ) with a 40x/1 . 3 NA oil-immersion objective lens and a cooled charge-coupled device camera ( Roper Scientific , Trenton , NJ ) controlled by Metafluor 7 . 7 software ( Molecular Devices , Sunnyvale , CA ) . Dual cyan/yellow emission ratio imaging was performed using a 420DF20 excitation filter , a 450DRLP dichroic mirror , and two emission filters ( 475DF40 for CFP and 535DF25 for YFP ) . RFP intensity was imaged using a 568DF55 excitation filter , a 600DRLP dichroic mirror , and a 653DF95 emission filter . Filter sets were alternated using a Lambda 10–2 filter changer ( Sutter Instruments , Novato , CA ) . Exposure times were between 10 and 500 ms , and images were taken every 20–30 s . Raw fluorescence images were corrected by subtracting the background fluorescence intensity of a cell-free region from the emission intensities of biosensor-expressing cells . Emission ratios ( yellow/cyan or cyan/yellow ) were then calculated at each time point . All time-courses were normalized by dividing the emission ratio or , in the case of RCaMP , the intensity at each time point by the basal value immediately preceding drug addition . | Cells need to be able to communicate with other cells , and they employ a variety of molecules and ions to send messages to each other . When calcium ions are used for these communications , the concentration of the ions typically rises and falls in a wave-like pattern . The size and shape of these ‘calcium waves’ contains information that is needed by organs as diverse as the heart and the brain . Most cells detect calcium waves using a sensor molecule called calmodulin . This , in turn , activates an enzyme called calcineurin . However , relatively little is known about the ways in which calcium waves shape the activity of calcineurin , even though calcium signaling is very common . Mehta et al . have now clarified this relationship by studying how calcium ions affect the activity of calcineurin molecules inside pancreatic cells . The response of calcineurin to calcium depends on position inside the cell . In the cytosol and at the plasma membrane that encloses the cell , calcium waves trigger a very fast ‘step-like’ increase in calcineurin activity . By contrast , at the surface of certain organelles within the cell , the calcium waves cause the calcineurin activity to rise and fall in a wave-like pattern . Experiments designed to identify the molecular mechanism behind this difference revealed that the answer lies in the distribution of calmodulin , the intermediate between calcium and calcineurin . At the surface of organelles , there is less calmodulin available to activate calcineurin than in the cytosol or at the plasma membrane . As a result , calcineurin activity in the vicinity of organelles is vulnerable to being canceled out by the actions of other enzymes . When more calmodulin is available , this canceling out does not occur , which is how wave-like input can lead to step-like output . By identifying the mechanism by which a single signal—a calcium wave—generates distinct responses in the same target molecule—calcineurin—depending on subcellular location , Mehta et al . have identified a process that is relevant to a wide range of biological systems . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2014 | Calmodulin-controlled spatial decoding of oscillatory Ca2+ signals by calcineurin |
Conflict detection in sensory input is central to adaptive human behavior . Perhaps unsurprisingly , past research has shown that conflict may even be detected in the absence of conflict awareness , suggesting that conflict detection is an automatic process that does not require attention . To test the possibility of conflict processing in the absence of attention , we manipulated task relevance and response overlap of potentially conflicting stimulus features across six behavioral tasks . Multivariate analyses on human electroencephalographic data revealed neural signatures of conflict only when at least one feature of a conflicting stimulus was attended , regardless of whether that feature was part of the conflict , or overlaps with the response . In contrast , neural signatures of basic sensory processes were present even when a stimulus was completely unattended . These data reveal an attentional bottleneck at the level of objects , suggesting that object-based attention is a prerequisite for cognitive control operations involved in conflict detection .
Every day we are bombarded with sensory information from the environment , and we often face the challenge of selecting the relevant information and ignoring irrelevant – potentially conflicting – information to maximize performance . These selection processes require much effort and our full attention , sometimes rendering us deceptively oblivious to irrelevant sensory input ( e . g . , chest-banging apes ) , as illustrated by the famous inattentional blindness phenomenon ( Simons and Chabris , 1999 ) . However , unattended events that are not relevant for the current task might still capture our attention or interfere with ongoing task performance , for example , when they are inherently relevant to us ( e . g . , our own name ) . This is illustrated by another famous psychological phenomenon: the cocktail party effect ( Cherry , 1953; Moray , 1959 ) . Thus , under specific circumstances , task-irrelevant information may capture attentional resources and be subsequently processed with different degrees of depth . It is currently a matter of debate which processes require top-down attention ( Dehaene et al . , 2006; Koch and Tsuchiya , 2007; Koelewijn et al . , 2010; Lamme , 2003; Lamme and Roelfsema , 2000; Rousselet et al . , 2004; VanRullen , 2007 ) . It was long thought that only basic physical stimulus features or very salient stimuli are processed in the absence of attention ( Treisman and Gelade , 1980 ) due to an ‘attentional bottleneck’ at higher levels of analysis ( Broadbent , 1958; Deutsch and Deutsch , 1963; Lachter et al . , 2004; Wolfe and Horowitz , 2004 ) . However , there is now solid evidence that several tasks may in fact still unfold in the ( near ) absence of attention , including perceptual integration ( Fahrenfort et al . , 2017 ) , the processing of emotional valence ( Sand and Wiens , 2011; Stefanics et al . , 2012 ) , semantical processing of written words ( Schnuerch et al . , 2016 ) , and visual scene categorization ( Li et al . , 2002; Peelen et al . , 2009 ) . Although one should be cautious in claiming complete absence of attention ( Lachter et al . , 2004 ) , these and other studies have pushed the boundaries of input processing that is task-irrelevant ( without attention ) and may even question the existence of an attentional bottleneck at all , at least for relatively low-level information . Conceivably , the attentional bottleneck is only present at higher , more complex , levels of cognitive processing , like cognitive control functions . Over the years , various theories have been proposed with regard to this attentional bottleneck among which are the load theory of selective attention and cognitive control ( Lavie et al . , 2004 ) , the multiple resources theory ( Wickens , 2002 ) , and the hierarchical central executive bottleneck theory and formalizations thereof in a cortical network model for serial and parallel processing ( Sigman and Dehaene , 2006; Zylberberg et al . , 2010; Zylberberg et al . , 2011 ) . These theories all hinge on the idea that resources for the processing of information are limited and that the brain therefore has to allocate resources to processes that are currently most relevant via selective attention ( Broadbent , 1958; Treisman , 1969 ) . Resource ( re- ) allocation , and thus flexible behavior , is thought to be governed by an executive network , most prominently involving the prefrontal cortex ( Goldman-Rakic , 1995; Goldman-Rakic , 1996 ) . Information that is deemed task-irrelevant has fewer resources at its disposal and is therefore processed to a lesser extent . When more resources are necessary for processing the task-relevant information , for example , under high perceptual load , processing of task-irrelevant information diminishes ( Lavie et al . , 2003; Lavie et al . , 2004 ) . Yet even under high perceptual load , task-irrelevant features can be processed when they are part of an attended object ( when object-based attention is present ) ( Chen , 2012; Chen and Cave , 2006; Cosman and Vecera , 2012; Kahneman et al . , 1992; O'Craven et al . , 1999; Schoenfeld et al . , 2014; Wegener et al . , 2014 ) . There is currently no consensus which type of information can be processed in parallel by the brain and which attentional mechanisms determine what information passes the attentional bottleneck . One unresolved issue is that most empirical work has investigated the bottleneck with regard to sensory features; however , it is unknown if the bottleneck and the distribution of processing resources also take place for more complex , cognitive processes . Here , we test whether such a high-level attentional bottleneck indeed exists in the human brain . Specifically , we aim to test whether cognitive control operations , necessary to identify and resolve conflicting sensory input , are operational when that input is irrelevant for the task at hand ( and hence unattended ) and what role object-based attention may have in conflict detection . Previous work has shown that the brain has dedicated networks for the detection and resolution of conflict , in which the medial frontal cortex ( MFC ) plays a pivotal role ( Ridderinkhof et al . , 2004 ) . Conflict detection and subsequent behavioral adaptation is central to human cognitive control , and , hence , it may not be surprising that past research has shown that conflict detection can even occur unconsciously ( Atas et al . , 2016; D'Ostilio and Garraux , 2012a; Huber-Huber and Ansorge , 2018; van Gaal et al . , 2008 ) , suggesting that the brain may detect conflict fully automatically and that it may even occur without paying attention ( e . g . , Rahnev et al . , 2012 ) . Moreover , it has been shown that this automaticity can be enhanced by training , resulting in more efficient processing of conflict ( Chen et al . , 2013; MacLeod and Dunbar , 1988; van Gaal et al . , 2008 ) . Conclusive evidence regarding the claim that conflict detection is fully automatic has , to our knowledge , not been provided , and therefore , the necessity of attention for cognitive control operations remains open for debate . Previous studies have shown that cognitive control processes are operational when to-be-ignored features from either a task-relevant or a task-irrelevant stimulus overlap with the behavioral response to be made to the primary task , causing interference in performance ( Mao and Wang , 2008; Padrão et al . , 2015; Zimmer et al . , 2010 ) . In these circumstances , the interfering stimulus feature carries information related to the primary task and is therefore de facto not task-irrelevant . Consequently , it is currently unknown whether cognitive control operations are active for conflicting sensory input that is not related to the task at hand . Given the immense stream of sensory input we encounter in our daily lives , conflict between two ( unattended ) sources of perceptual information is inevitable . Here , we investigated whether conflict between two features of an auditory stimulus ( its content and its spatial location ) would be detected by the brain under varying levels of task relevance of these features . The main aspect of the task was as follows . We presented auditory spoken words ( ‘left’ and ‘right’ in Dutch ) through speakers located on the left and right side of the body . By presenting these stimuli through either the left or the right speaker , content-location conflict arises on specific trials ( e . g . , the word ‘left’ from the right speaker ) but not on others ( e . g . , the word ‘right’ from the right speaker ) ( Buzzell et al . , 2013; Canales-Johnson et al . , 2020 ) . A wealth of previous studies has revealed that conflict arises between task-relevant and task-irrelevant features of the stimulus in these type of tasks ( similar to the Simon task and Stroop task; Egner and Hirsch , 2005; Hommel , 2011 ) . Here , these potentially conflicting auditory stimuli were presented during six different behavioral tasks , divided over two separate experiments , multiple experimental sessions , and different participant groups ( both experiments N = 24 ) . In all tasks , we focus on the processing of content-location conflict of the auditory stimulus . There were several critical differences between the behavioral tasks: ( 1 ) task relevance of a conflicting feature of the stimulus , ( 2 ) task relevance of a non-conflicting feature that was part of a conflicting stimulus , and ( 3 ) whether the response to be given mapped onto a conflicting feature of the stimulus . Note that in all tasks only one feature could be task-relevant and that all the other feature ( s ) had to be ignored . The systematic manipulation of task relevance and the response-mapping allowed us to explore the full landscape of possibilities of how varying levels of attention affect sensory and conflict processing . Electroencephalography ( EEG ) was recorded and multivariate analyses on the EEG data were used to extract any neural signatures of conflict detection ( i . e . , theta-band neural oscillations; Cavanagh and Frank , 2014; Cohen and Cavanagh , 2011 ) and sensory processing for any of the features of the auditory stimulus . Furthermore , in both experiments we measured behavioral and neural effects of task-irrelevant conflict before and after training on conflict-inducing tasks , aiming to investigate the role of automaticity in the detection of ( task-irrelevant ) conflict .
In the first experiment , 24 human participants performed two behavioral tasks ( Figure 1A ) . In the auditory conflict task ( from hereon: content discrimination task I ) , the feature ‘sound content’ was task-relevant . Participants were instructed to respond according to the content of the auditory stimulus ( ‘left’ vs . ‘right’ ) , ignoring its spatial location that could conflict with the content response ( presented from the left or right side of the participant ) . For the other behavioral task , participants performed a demanding visual random dot-motion ( RDM ) task in which they had to discriminate the direction of vertical motion ( from hereon: vertical RDM task ) , while being presented with the same auditory stimuli – all features of which were thus fully irrelevant for task performance . Behavioral responses on this visual task were orthogonal to the response tendencies potentially triggered by the auditory features , excluding any task- or response-related interference ( Figure 1B ) . Under this manipulation , all auditory features are task-irrelevant and are orthogonal to the response-mapping . To maximize the possibility of observing conflict detection when conflicting features are task-irrelevant and explore the effect of task automatization on conflict processing , participants performed the tasks both before and after extensive training , which may increase the efficiency of cognitive control ( Figure 1C; van Gaal et al . , 2008 ) . For content discrimination task I , mean error rates ( ERs ) were 2 . 6% ( SD = 2 . 7% ) and mean reaction times ( RTs ) 477 . 2 ms ( SD = 76 . 1 ms ) , averaged over all four sessions . For the vertical RDM , mean ERs were 19 . 2% ( SD = 6 . 6% ) and mean RTs were 711 . 4 ms ( SD = 151 . 3 ms ) . The mean ER of vertical RDM indicates that our staircasing procedure was effective ( see Materials and methods for details on staircasing performance on the RDM ) . To investigate whether our experimental design was apt to induce conflict effects for task-relevant sensory input and to test whether conflict effects were still present when sensory input was task-irrelevant , we performed repeated measures ( rm- ) ANOVAs ( 2 × 2 × 2 factorial ) on mean RTs and ERs gathered during the EEG recording sessions ( session 1 , ‘before training’; session 4 , ‘after training’ ) . This allowed us to include ( 1 ) task relevance ( yes/no ) , ( 2 ) training ( before/after ) , and ( 3 ) congruency of auditory content with location of auditory source ( congruent/incongruent ) . Note that congruency is always defined based on the relationship between two features of the auditorily presented stimuli , also when participants performed the visual task ( and therefore the auditory features were task-irrelevant ) . Detection of conflict is typically associated with behavioral slowing and increased ERs . Indeed , we observed that , across both tasks , participants were slower and made more errors on incongruent trials as compared to congruent trials ( the conflict effect , RT: F ( 1 , 23 ) = 52 . 83 , p<0 . 001 , ηp2 = 0 . 70; ER: F ( 1 , 23 ) = 9 . 13 , p=0 . 01 , ηp2 = 0 . 28 ) . This conflict effect was modulated by task relevance of the auditory features ( RT: F ( 1 , 23 ) = 152 . 76 , p<0 . 001 , ηp2 = 0 . 87; ER: F ( 1 , 23 ) = 11 . 15 , p=0 . 01 , ηp2 = 0 . 33 ) and post-hoc ANOVAs ( see Materials and methods ) showed that the conflict effect was present when the auditory feature content was task-relevant ( RTcont ( I ) : F ( 1 , 23 ) = 285 . 00 , p<0 . 001 , ηp2 = 0 . 93; ERcont ( I ) : F ( 1 , 23 ) = 23 . 85 , p<0 . 001 , ηp2 = 0 . 51; Figure 2A , left panel ) , but not when all auditory features were task-irrelevant ( RTVRDM: F ( 1 , 23 ) = 1 . 96 , p=0 . 18 , ηp2 = 0 . 08 , BF01 = 5 . 41; ERVRDM: F ( 1 , 23 ) = 0 . 26 , p=0 . 62 , ηp2 = 0 . 01 , BF01 = 4 . 55; Figure 2A , right panel ) . Because responses in the vertical RDM were made with the right hand only , we subsequently tested whether the auditory features in isolation affected the speed and accuracy of right-hand responses . For example , the spoken word ‘left’ may slow down responses made with the right hand more so than the spoken word ‘right’ ( the same logic holds for stimulus location ) . However , this was not the case . A 2 × 2 × 2 factorial rm-ANOVA on mean RTs with session ( before/after training ) , stimulus content ( 'left'/'right' ) , and stimulus location ( left/right ) showed that RTs were unaffected by sound content ( F ( 1 , 23 ) = 0 . 01 , p=0 . 92 , ηp2 = 0 . 00 , BF01 = 6 . 16 ) and sound location ( F ( 1 , 23 ) = 0 . 49 , p=0 . 49 , ηp2 = 0 . 02 , BF01 = 6 . 36 ) . Participants performed both behavioral tasks before and after extensive training of the content discrimination task to be able to investigate the role of training on conflict processing ( Figure 1C ) . RTs and ERs in the vertical RDM task were not modulated by behavioral training ( RTVRDM: F ( 1 , 23 ) = 2 . 07 , p=0 . 16 , ηp2 = 0 . 08 , BF01 = 0 . 32; ERVRDM: F ( 1 , 23 ) = 0 . 24 , p=0 . 63 , ηp2 = 0 . 01 , BF01 = 3 . 79 ) . Training did result in a decrease of overall RT on content discrimination task I , although ERs were not affected ( RTcont ( I ) : F ( 1 , 23 ) = 45 . 05 , p<0 . 001 , ηp2 = 0 . 66; ERcont ( I ) : F ( 1 , 23 ) = 1 . 77 , p=0 . 20 , ηp2 = 0 . 07 , BF01 = 0 . 89 ) . Moreover , the effect of conflict on RTs and ERs in this task decreased after behavioral training ( RTcont ( I ) : F ( 1 , 23 ) = 29 . 86 , p<0 . 001 , ηp2 = 0 . 57; ERcont ( I ) : F ( 1 , 23 ) = 9 . 76 , p=0 . 005 , ηp2 = 0 . 30; Figure 2—figure supplement 1A , B ) , suggesting increased efficiency of within-trial conflict resolution mechanisms . All other effects were not reliable ( p>0 . 05 ) . The observation that conflicting task-irrelevant stimuli had no effect on RTs and ERs , even after substantial training , whereas task-relevant conflicting stimuli did , may not come as a surprise because manual responses on the visual task ( motion up/down with index and middle finger of right hand ) were fully orthogonal to the potential conflicting nature of the auditory features ( i . e . , left/right ) . Further , content discrimination task I and the vertical RDM were independent tasks , requiring different cognitive processes . For example , mean RTs on the vertical RDM were on average 267 ms longer than mean RTs for content discrimination task I . However , caution is required in concluding that conflict detection is absent for task-irrelevant stimuli based on these behavioral results alone as neural and/or behavioral effects can sometimes be observed in isolation ( one is observed but not the other , e . g . , Canales-Johnson et al . , 2020; van Gaal et al . , 2014 ) . Therefore , in order to test whether unattended conflict is detected by the brain we turn to the multivariate pattern analysis ( MVPA ) of our neural data . Plausibly , the neural dynamics of conflict processing for task-irrelevant sensory input are different – in physical ( across electrodes ) and frequency space – from those related to the processing of conflict when sensory input is task-relevant . Therefore , we applied multivariate decoding techniques in the frequency domain to inspect whether and – if so – to what extent certain stimulus features were processed . These multivariate approaches have some advantages over traditional univariate approaches , for example , they are less sensitive to individual differences in spatial topography , because decoding accuracies are derived at a single participant level ( Fahrenfort et al . , 2018; Grootswagers et al . , 2017; Haxby et al . , 2001 ) . Therefore , group statistics do not critically depend on the presence of effects in specific electrodes or clusters of electrodes . Further , although a wealth of studies have shown that conflict processing is related to an increase in power of theta-band neural oscillations ( ~4–8 Hz ) after stimulus presentation ( Cohen and Cavanagh , 2011; Jiang et al . , 2015a; Nigbur et al . , 2012 ) , it is unknown whether this is also the case for task-irrelevant conflict . By performing our MVPA in frequency space , we could potentially find neural signatures in non-theta frequency bands related to the processing of task-irrelevant conflict . However , due to the temporal and frequency space that has to be covered , strict multiple comparison corrections have to be performed ( across time and frequency , see Materials and methods ) . Therefore , we adopted an additional hypothesis-driven analysis , which also allowed us to obtain evidence for the absence of effects . Throughout this paper , we will discuss our neural data in the following order . First , the MVPAs in the frequency domain are presented for all critical features of the task ( congruency , content , location , corrected for multiple comparisons ) . Then , we report results from the additional hypothesis-driven analysis , where we extracted classifier accuracies from a predefined time-frequency region of interest ( ROI ) ( 100–700 ms , 2–8 Hz ) on which we performed ( Bayesian ) tests ( see Materials and methods ) . This ROI was selected based on previous observations of conflict-related theta-band activity ( Cohen and Cavanagh , 2011; Cohen and van Gaal , 2014; Jiang et al . , 2015b; Nigbur et al . , 2012 ) . Specifically , for every task and every stimulus feature ( i . e . , congruency , content , location ) , we extracted average decoding accuracies from the ROI per participant and performed analyses on these values . First , we trained a classifier on data from all EEG electrodes to distinguish between congruent versus incongruent trials , for both content discrimination task I and the vertical RDM task . Above-chance classification accuracies imply that relevant information about the decoded stimulus feature is present in the neural data , meaning that some processing of that feature occurred ( Hebart and Baker , 2018 ) . We performed our main analysis on the combined data from both EEG sessions , thereby maximizing power to establish effects in our crucial comparisons . We also performed similar analyses on the session-specific data to investigate the role of behavioral training on processing of conflict . These results are discussed more in depth below and are shown in Figure 2—figure supplement 1C , D . Congruency decoding reveals that stimulus congruency was represented in neural data only when conflict was task-relevant ( p<0 . 001 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 2–12 Hz , peak frequency: 4 Hz , time range: 234–609 ms , peak time: 438 ms; Figure 2B , left panel ) . The conflict effect roughly falls in the theta-band ( 4–8 Hz ) , which confirms a firm body of literature linking conflict detection to post-conflict modulations in theta-band oscillatory dynamics ( Cavanagh and Frank , 2014; Cohen and Cavanagh , 2011; Cohen and van Gaal , 2014; Nigbur et al . , 2012 ) . Activation patterns that were calculated from classifier weights within the predefined time-frequency theta-band ROI ( 2–8 Hz , 100–700 ms ) revealed a clear midfrontal distribution of conflict-related activity ( Figure 5—figure supplement 1A ) . No significant time-frequency cluster was found for the vertical RDM task ( Figure 2B , right panel ) . To quantify the absence of this effect , we followed up this hypothesis-free ( with respect to frequency and time ) MVPA with a hypothesis-driven analysis focused on the post-stimulus theta-band . This more restricted analysis showed no significant effect ( t ( 23 ) = −0 . 50 , p=0 . 69 , d = −0 . 10 ) and an additional Bayesian analysis revealed moderate evidence in favor of the null hypothesis ( i . e . , no effect of conflict on theta-band power ) than the alternative hypothesis ( BF01 = 6 . 53 ) . Similar to our observation of decreased behavioral effects of conflict after behavioral training ( Figure 2—figure supplement 1A , B ) , decoding accuracies in the content discrimination task I were also lower after training ( t ( 23 ) = −3 . 01 , p=0 . 01 , d = −0 . 63; Figure 2—figure supplement 1C ) , suggesting more efficient conflict resolution , as reflected in neural theta oscillations as well . In the vertical RDM , behavioral training did not affect decoding accuracies of sound congruency ( t ( 23 ) = −1 . 24 , p=0 . 23 , d = −0 . 25 , BF01 = 2 . 36; Figure 2—figure supplement 1D ) . Thus , cognitive control networks − or possible substitute networks − are seemingly not capable of detecting conflict when sensory features are task-irrelevant . However , the question remains whether this observation is specific to the conflicting nature of the auditory stimuli or whether the auditory stimuli are not processed whatsoever when attention is reallocated to the visually demanding task . To address this question , we trained classifiers on two other features of the auditory stimuli , that is , location and content , to test whether these features were processed by the brain regardless of task relevance . Indeed , the content of auditory stimuli was processed both when the stimuli were task-relevant ( p<0 . 001 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 2–30 Hz , peak frequency: 4 Hz , time range: 47–1000 ms , peak time: 422 ms; Figure 2B , left panel ) and task-irrelevant ( p<0 . 001 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 2–20 Hz , peak frequency: 6 Hz , time range: 78–547 ms , peak time: 297 ms; Figure 2B , right panel ) . Similarly , the location of auditory stimuli could also be decoded from neural data for both content discrimination task I ( p<0 . 001 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 2–18 Hz , peak frequency: 6 Hz , time range: 63–-672 ms , peak time: 203 ms; Figure 2B , left panel ) and the vertical RDM task ( p<0 . 001 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 2–12 Hz , peak frequency: 6 Hz , time range: 156–484 ms , peak time: 281 ms; Figure 2B , right panel ) . The above chance performance of the classifiers for the auditory stimulus features demonstrates that location and content information were processed , even when these features were task-irrelevant . Processing of task-irrelevant stimulus features was , however , more transient in time and more narrowband in frequency as compared to processing of the same features in a task-relevant setting . Further , content decoding revealed a much broader frequency spectrum than any of the other comparisons in content discrimination task I . In the next experiment , we show that this is related to the fact that this feature was response-relevant and that this effect therefore partially reflects response preparation and response execution processes . Summarizing , we show that when ( conflicting ) features of an auditory stimulus are truly and consistently task-irrelevant , the conflict between them is no longer detected by – nor relevant to – the conflict monitoring system , but the features ( content and location ) are still processed in isolation . To investigate if , and how , behavioral training affects processing of sound content and location , we tested whether decoding accuracies for these features were different between the two experimental sessions . Decoding accuracies for the task-relevant feature of content discrimination task I ( i . e . , sound content ) were significantly increased after behavioral training in a delta- to theta-band cluster ( p<0 . 001 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 2–10 Hz , peak frequency: 2 Hz , time range: 234–484 ms , peak time: 344 ms; Figure 2—figure supplement 1C ) . This suggests that processing of task-relevant information ( i . e . , sound content ) is improved as a result of training . Decoding accuracies for sound location in the content discrimination task were not different before and after behavioral training ( no significant clusters; predefined ROI: t ( 23 ) = 0 . 12 , p=0 . 91 , d = 0 . 02 , BF01 = 4 . 63 ) . In the vertical RDM task , behavioral training did not affect the decoding accuracies within the predefined ROI for sound content ( t ( 23 ) = 0 . 75 , p=0 . 46 , d = 0 . 15 , BF01 = 3 . 61 ) and location ( t ( 23 ) = 0 . 04 , p=0 . 97 , d = 0 . 01 , BF01 = 4 . 66 , also no other significant clusters; Figure 2—figure supplement 1D ) . This suggests that processing of sound content and location ( Figure 2B ) , both task-irrelevant auditory features in the vertical RDM task , is automatic and not dependent on training . In conclusion , we observed neural signatures of the processing of sensory stimulus features ( i . e . , location and content of an auditory stimulus ) regardless of task relevance of these features , but a lack of integration of these features to form conflict when the auditory stimulus was fully task-irrelevant . Considerable training in content discrimination task I resulted in more efficient conflict processing ( i . e . , decreased behavioral conflict effects and theta-band activity after training; Figure 2—figure supplement 1 ) when the auditory stimulus was task-relevant , but this increased automaticity did not lead to detection of conflict when the auditory stimulus was fully task irrelevant . The experimental design of the first experiment rendered the auditory features to be located at the extreme ends of the scale of task relevance , that is , either the conflicting features were task-relevant and the conflicting features were consistently mapped to specific responses , or the conflicting features were task-irrelevant and the conflicting features were not mapped to responses . However , to further understand the relationship between the relevance of the conflicting features and the overlap with responses , we performed a second experiment containing four behavioral tasks . For this second experiment , we recruited 24 new participants . We included two auditory conflicting tasks , similar to content discrimination task I . In one of the auditory tasks ( from hereon: content discrimination task II , Figure 3A ) , participants again had to respond according to the content of the auditory stimulus , whereas in the other auditory task ( from hereon: location discrimination task , Figure 3B ) they were instructed to report from which side the auditory stimulus was presented ( i . e . , left or right speaker ) . Furthermore , we included two new tasks in which the conflicting features ( location and content ) were not task-relevant and participants responded to a non-conflicting feature that was part of the conflicting stimulus ( from hereon: volume oddball detection task , Figure 3C ) or the auditory stimulus was task-irrelevant but its features - location and content- overlapped with the responses to be given ( from hereon: horizontal RDM task , Figure 3D ) . The horizontal RDM task was similar to vertical RDM task of experiment 1; however , the dots were now moving on a horizontal plane . In other words , participants were instructed to classify the overall movement of moving dots to either the left or the right . As this is a visual paradigm , the simultaneously presented auditory stimuli are fully task-irrelevant . However , both features of conflict , the content ( i . e . , ‘left’ and ‘right’ ) and the location ( i . e . , left and right speaker ) , of the auditory stimuli could potentially interfere with participants’ responses on the visual task , thereby inducing a crossmodal Stroop-like type of conflict ( Stroop , 1935 ) . In the volume oddball detection task , participants were presented with the same auditory stimuli as before; however , one out of eight stimuli ( 12 . 5% ) was presented at a lower volume . Participants were instructed to detect these volume oddballs by pressing the spacebar with their right hand as fast as possible . If they did not hear an oddball , they were instructed to withhold from responding . In this task , theoretically , the selection of an object’s feature ( e . g . , volume ) could lead to the selection of all of its features ( e . g . , sound content , location ) , as suggested by theories of object-based attention ( Chen , 2012 ) . This in turn may lead to conflict detection , even if the conflicting features are task-irrelevant . Similar to experiment 1 , we included behavioral training in conflict-inducing tasks to inspect if enhanced automaticity of conflict processing would affect conflict detection under task-irrelevant sensory input . Participants performed 500 trials of the volume oddball detection task twice at the very beginning of a session and at the end of a session ( Figure 3E ) . During the first run of the task , neither sound content nor sound location was related to any behavioral responses , whereas during the second run these features might have acquired some intrinsic relevance through training on the other tasks . Furthermore , repeated exposure to conflict may prime the conflict monitoring system to exert more control over sensory inputs necessary for more efficient conflict detection , even when these sensory inputs are not task-relevant within the context of the task the participant is performing at that time . In order to keep sensory input similar , moving dots ( coherence: 0 ) were presented on the monitor during content discrimination task II , the location discrimination task , and the volume oddball detection task , but these could be ignored . Again , EEG was recorded while participants performed these tasks in order to see if auditory conflict was detected when the auditory stimulus or its conflicting features ( i . e . , location and content ) were task-irrelevant . We performed the same artifact rejection procedure as in experiment 1 . For one participant , on average 64 . 5% ( SD = 9 . 9% ) of all epochs within each task were removed in this procedure , which is 3 . 9 standard deviations from the average ratio of removed epochs in this experiment ( M = 10 . 3% , SD = 13 . 9% ) . Therefore , this participant was excluded from the EEG analysis of experiment 2 , resulting in N = 23 for the analysis of EEG data . Mean RT in the location discrimination task was 338 . 2 ms ( SD = 112 . 7 ms ) and mean ER was 4 . 5% ( SD = 3 . 2% ) . For content discrimination task II , RTs were on average 364 . 0 ms ( SD = 127 . 4 ms ) and ERs were 5 . 5% ( SD = 5 . 8% ) . For the horizontal RDM , RTs were on average 362 . 5 ms ( SD = 116 . 5 ms ) and ERs were 27 . 7% ( SD = 5 . 2% ) . Mean RTs of the volume oddball task were calculated for correct trials in which a response was made ( i . e . , hit trials ) and was 504 . 5 ms ( SD = 178 . 7 ms , on hit trials ) . On average , participants had 40 . 5 hits ( SD = 12 . 8 ) out of 61 . 8 oddball trials ( SD = 8 . 6 ) , per run of 500 trials . We will first discuss the behavioral results of the content discrimination , location discrimination , and horizontal RDM tasks as behavioral performance for the volume oddball task is represented in perceptual sensitivity ( d’ ) , rather than ER . rm-ANOVAs ( 3 × 2 factorial ) were performed on mean RTs and ERs from these three tasks , with the factors ( 1 ) task and ( 2 ) congruency of the auditory features . Again , congruency always relates to the combination of the auditory stimulus features sound content ( 'left' vs . 'right' ) and sound location ( left speaker vs . right speaker ) . We observed that participants were slower and made more errors on incongruent trials as compared to congruent trials ( RT: F ( 1 , 23 ) = 75 . 41 , p<0 . 001 , ηp2 = 0 . 77; ER: F ( 1 , 23 ) = 68 . 00 , p <0 . 001 , ηp2 = 0 . 75; Figure 4A ) . This conflict effect was modulated by task ( RT: F ( 1 . 55 , 35 . 71 ) = 22 . 80 , p<0 . 001 , ηp2 = 0 . 50; ER: F ( 1 . 58 , 36 . 36 ) = 10 . 18 , p<0 . 001 , ηp2=0 . 31 ) and post-hoc paired samples t-tests ( incongruent – congruent ) showed that conflict effects were only present in tasks where one of the conflicting features was task-relevant ( location discrimination task: RTloc: t ( 23 ) = 5 . 03 , p<0 . 001 , d = 1 . 03; ERloc: t ( 23 ) = 6 . 25 , p<0 . 001 , d = 1 . 28; content discrimination task II: RTcont ( II ) : t ( 23 ) = 8 . 95 , p<0 . 001 , d = 1 . 83; ERcont ( II ) : t ( 23 ) = 5 . 93 , p<0 . 001 , d = 1 . 21; horizontal RDM task: RTHRDM: t ( 23 ) = 1 . 44 , p=0 . 16 , d = 0 . 29 , BF01 = 1 . 88; ERHRDM: t ( 23 ) = 1 . 65 , p=0 . 11 , d = 0 . 34 , BF01 = 1 . 44 ) . Although in the horizontal RDM task conflict between sound content and location did not affect the speed of responses , stimulus content and location in isolation could have potentially interfered with behavioral performance given the overlap of these features with both the plane of dot direction ( left/right ) and the response scheme ( left/right hand ) . Indeed , trials containing conflict between sound location and dot direction resulted in slower RTs and increased ERs ( RT: t ( 23 ) = 2 . 12 , p=0 . 045 , d = 0 . 44; ER: t ( 23 ) = 5 . 94 , p<0 . 001 , d = 1 . 21 ) . Similar effects were observed for trials where sound content conflicted with the dot direction , but onlyin ERs ( RT: t ( 23 ) = 1 . 72 , p=0 . 10 , d = 0 . 35 , BF01 = 2 . 85; ER: t ( 23 ) = 5 . 55 , p<0 . 001 , d = 1 . 13 ) . This shows that sound content and location in isolation , even though both features were task-irrelevant , interfered with task performance . For the volume oddball task , we tested the effect of auditory congruency on RTs of trials which , by virtue of task instruction , only covers oddball trials in which a correct response was made ( i . e . , hits ) . rm-ANOVAs ( 2 × 2 factorial ) with the factors run number and feature congruency revealed no effects of auditory conflict on RT ( F ( 1 , 23 ) = 2 . 78 , p=0 . 11 , ηp2 = 0 . 10 , BF01 = 3 . 34; Figure 4A , no effects of training , see Figure 4—figure supplement 1A ) . To test whether individual features of the auditory stimulus interfered with right-hand responses , we performed an additional 2 × 2 factorial rm-ANOVA with sound content and location as factors . Auditory content significantly affected RTs , whereas sound location did not ( content: F ( 1 , 23 ) = 25 . 41 , p<0 . 001 , ηp2 = 0 . 53; location: F ( 1 , 23 ) = 0 . 99 , p=0 . 33 , ηp2 = 0 . 04 , BF01 = 3 . 90 ) . Specifically , RTs were slower for the spoken word 'left' ( incongruent with the responding hand , M = 528 . 6 ms , SD = 193 . 1 ms ) as compared to the spoken word 'right' ( congruent with the responding hand , M = 493 . 3 ms , SD = 174 . 9 ms ) , revealing interference of sound content in isolation on right-hand responses , similar to the horizontal RDM task . Although conflict between the auditory features was not present in RTs , we did observe that sensitivity ( d’ ) increased for incongruent ( M = 2 . 66 , SD = 0 . 77 ) compared to congruent trials ( M = 2 . 11 , SD = 0 . 81 , F ( 1 , 23 ) = 45 . 62 , p<0 . 001 , ηp2 = 0 . 67; Figure 4A ) . These results show that volume oddball detection performance increases on trials that contain conflict between sound content and location . This effect of conflict on behavioral performance can already be found in the first run , when sound content and location had not yet been related to any responses/task and were thus fully task-irrelevant ( t ( 23 ) = 5 . 71 , p<0 . 001 , d = 1 . 17 ) . There was no significant interaction between run number and auditory stimulus congruency ( F ( 1 , 23 ) = 1 . 40 , p=0 . 25 , ηp2 = 0 . 06 , BF01 = 2 . 13; Figure 4—figure supplement 1B; hit rates and false alarms are plotted in this figure supplement as well ) . We again trained multivariate classifiers on single-trial time-frequency data to test whether the auditory stimulus features ( i . e . , content , location , and congruency ) were processed when ( 1 ) the auditory conflicting features were task-relevant and overlapped with the response-mapping ( content and location discrimination tasks ) , ( 2 ) the auditory conflicting features were task-irrelevant and another feature of the conflicting stimulus was task-relevant ( volume oddball task ) , or when ( 3 ) the auditory conflicting features were task-irrelevant , but its conflicting features overlapped with the response-mapping in the task ( horizontal RDM task ) . Cluster-based analyses across the entire T-F range revealed neural signatures of conflict processing in the theta-band when the content of the auditory stimulus was task-relevant ( content discrimination task II: p<0 . 001 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 4–10 Hz , peak frequency: 4 Hz , time range: 328–953 ms , peak time: 438 ms; Figure 4B ) and when the volume of the auditory stimulus was task-relevant ( volume oddball task: p=0 . 03 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 2–6 Hz , peak frequency: 2 Hz , time range: 234–516 ms , peak time: 438 ms; Figure 4B , see Figure 4—figure supplement 1E: no training effects for the volume oddball task ) . Both observations of congruency decoding are in line with the presence of conflict-related behavioral effects in these tasks ( Figure 4A ) . No significant clusters of above-chance classifier accuracy were found after correcting for multiple comparisons in the location discrimination task and the horizontal RDM task ( Figure 4B ) . However , a hypothesis-driven analysis focused on the post-stimulus theta-band ( 2–8 Hz , 100–700 ms ) revealed that congruency decoding accuracies within this ROI were significantly above chance for both tasks as well ( location discrimination: t ( 22 ) = 2 . 00 , p=0 . 03 , d = 0 . 42; horizontal RDM: t ( 22 ) = 2 . 89 , p=0 . 004 , d = 0 . 60 ) . Thus , we observed that conflict of the auditory stimulus is detected when one of the auditory conflicting features is task-relevant ( content and location discrimination tasks ) , when one of its non-conflicting features is task-relevant ( volume oddball task ) , and when none of the auditory features is task-relevant but these features overlap with the response-mapping of the task ( horizontal RDM task ) . To qualify the differences between tasks , we combine the data from all experiments and compare effect sizes across tasks at the end of this section ( Figure 5 ) . Next , we trained classifiers to distinguish trials based on sound location and content in order to inspect sensory processing . We found neural signatures of the processing of sound content for all four tasks: content discrimination II ( p<0 . 001 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 2–30 Hz , peak frequency: 4 Hz , time range: 203–1000 ms , peak time: 469; Figure 4B ) , location discrimination ( p=0 . 01 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 2–6 Hz , peak frequency: 2 Hz , time range: 313–641 ms , peak time: 563 ms; Figure 4B ) , horizontal RDM task ( p<0 . 001 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 4–16 Hz , peak frequency: 4 Hz , time range: 78–328 ms , peak time: 281 ms; Figure 4B ) . For the volume oddball task , we observed a delta/theta-band cluster and a late beta-band cluster ( delta/theta: p<0 . 001 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 2–8 Hz , peak frequency: 4 Hz , time range: 94–797 ms , peak time: 281 ms; late beta: p=0 . 01 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 12–20 Hz , peak frequency: 20 Hz , time range: 672–953 ms , peak time: 828 ms; Figure 4B ) . Furthermore , sound location could be decoded from the content discrimination task II ( delta/theta: p=0 . 02 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 2–6 Hz , peak frequency: 2 Hz , time range: 453–688 ms , peak time: 609 ms; alpha: p=0 . 03 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 10–12 Hz , peak frequency: 12 Hz , time range: 531–750 ms , peak time: 578 ms; Figure 4B ) , the location discrimination task ( p<0 . 001 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 2–30 Hz , peak frequency: 2 Hz , time range: 109–1000 ms , peak time: 453 ms; Figure 4B ) , and the volume oddball task ( p<0 . 001 , one-sided: X¯>0 . 5 , cluster-corrected; frequency range: 2–22 Hz , peak frequency: 10 Hz , time range: −47 ms to 891 ms , peak time: 469 ms; Figure 4B ) . Initially , we did not observe a significant cluster of location decoding in the horizontal RDM , although the hypothesis-driven analysis revealed that accuracies within the predefined theta-band ROI were significantly above chance level as well ( t ( 23 ) = 2 . 47 , p=0 . 01 , d = 0 . 51 ) . One aspect of these results is worth highlighting . When participants responded to the location of the auditory stimulus , location decoding revealed a broadband power spectrum , similar to sound content decoding when sound content was task-relevant ( content discrimination tasks ) . This broad frequency decoding may be due to the fact that these features were task-relevant , but these results may also partially reflect response preparation and response execution processes as these auditory features were directly associated with a specific response . In order to test whether the earliest sensory responses were already modulated by task relevance and to link this to previous event-related potential ( ERP ) studies ( Alilović et al . , 2019; Molloy et al . , 2015; Woldorff et al . , 1993 ) , we performed an additional time-domain multivariate analysis on these sensory features ( T-F analyses are not well suited to address questions about the timing of processes ) . Because we were interested in the earliest sensory responses , we performed this analysis on data from experiment 2 only as all task parameters were best matched ( e . g . , in all tasks , a visual stimulus was presented , no training , etc . ) . We observed increased decoding for task-relevant sensory features compared to task-irrelevant features , starting ~250 ms ( sound location RT: M = 338 ms ) and ~330 ms ( sound content RT: M = 364 ms ) after stimulus presentation ( Figure 4—figure supplement 2 ) . The onset of these differences starts before a response is made , which may suggest that sensory processing of these features is indeed affected by task relevance; however , processes building up towards motor execution , such as decision-making and response preparation processes , cannot be excluded as potential factors driving the higher decoding accuracies in tasks where specific features are task-relevant and hence correlated with decision and motor processes . These results are elaborated upon in the Discussion . In conclusion , in line with the behavioral results , we observed that the processing of conflict between two stimulus features ( i . e . , location and content of an auditory stimulus ) was present in all tasks of experiment 2 . This indicates that conflict can be detected when one of the auditory conflicting features is task-relevant ( content and location discrimination tasks ) , when one of its non-conflicting features is task-relevant ( volume oddball task ) , and when there is overlap in the response-mapping with any of its task-irrelevant conflicting features ( horizontal RDM task ) . Overall , this reveals that when the conflicting stimulus itself is attended or when its conflicting features overlap with the response scheme , all of its features seem to be processed and integrated to elicit conflict . The neural data from all six different tasks over two experiments suggests that if sensory input is task-irrelevant , processing of that information is preserved , while cognitive control operations are strongly hampered ( Figures 2B , 4B , and 5B ) . To quantify this observation , we calculated Cohen's dz for all tasks and features ( based on the preselected ROI ) , sorted tasks according to the effect sizes of congruency decoding , and plotted Cohen's dz across tasks for all features ( conflict , content , and location; Figure 5 ) . For each feature and task , we extracted individual classifier area under the curve ( AUC ) values and performed an analysis of covariance ( ANCOVA ) on these accuracies , with fixed effects being task and stimulus feature . We found main effects for behavioral task and stimulus feature ( task: F ( 5 , 402 ) = 17 . 25 , p<0 . 001 , ηp2 = 0 . 18; stimulus feature: F ( 2 , 402 ) = 44 . 61 , p<0 . 001 , ηp2 = 0 . 18 ) . Crucially , the interaction between task and stimulus feature was also significant ( F ( 10 , 402 ) = 18 . 80 , p<0 . 001 , ηp2 = 0 . 32 ) , showing that the accuracy of conflict decoding decreased more across tasks as compared to content and location decoding . We next performed t-tests ( one-sided , X¯ > 0 . 5 ) on ROI accuracies from every task/feature combination to assess decoding performance for all stimulus features for the different task-relevance manipulations ( see Figure 5—source data 1 ) . We observed that congruency decoding accuracies were strongly influenced by task , whereas this was not the case for decoding accuracies of stimulus content and location ( Figure 5B ) . Note that these results were robust and not dependent on the specific ROI that was selected because using other ROI windows led to similar patterns of results , that is , decreased congruency decoding under task-irrelevant input , but relatively stable sensory feature decoding ( Figure 5—figure supplement 1C , D ) . Classifier weights were extracted from the ROI for all tasks and features , transformed to activation patterns and plotted in topomaps , to show the patterns of activity underlying the decoding results ( Figure 5—figure supplement 1A ) .
So why are auditory content and location not integrated to form conflict when all of the auditory features are task-irrelevant ? It has been suggested that the MFC is crucial for monitoring the presence of conflict , through the detection of coinciding inputs ( Cohen , 2014 ) . In our paradigm , it thus seems crucial that information related to the auditory features content and location reaches the MFC to be able to detect conflict , although control networks can undergo reconfiguration under certain circumstances ( Canales-Johnson et al . , 2020 ) . Previous studies have shown that task-irrelevant stimuli can still undergo elaborate processing ( Li et al . , 2002; Peelen et al . , 2009 ) . Our decoding results show that task-irrelevant features are indeed processed by the brain ( Figures 2B , 4B , and 5 ) . Interestingly , conflict between two task-irrelevant features was detected when another feature of the conflicting stimulus was task-relevant ( volume oddball task ) or when the conflicting features had overlap with the overall response scheme ( horizontal RDM task ) , but remained undetected when none of the auditory features was task-relevant and there was no overlap with the response scheme ( vertical RDM task ) . We argue that this difference is due to the fact that in the volume oddball and horizontal RDM tasks , the task-irrelevant conflicting features were selected through object-based attention . Theories of object-based attention have suggested that when one stimulus feature of an object is task-relevant and selected , attention ‘spreads’ to all other features of the attended stimulus , even when these features are task-irrelevant or part of a different stimulus or modality ( Chen , 2012; Chen and Cave , 2006; O'Craven et al . , 1999; Turatto et al . , 2005; Wegener et al . , 2014; Xu , 2010 ) . In the volume oddball task , a non-conflicting feature of the auditory stimulus ( volume ) was task-relevant , but this allowed for the selection of the other task-irrelevant features through object-based attention . In the horizontal RDM task , on the other hand , the conflicting features of the task-irrelevant auditory stimulus overlapped with the overall response scheme or task-set of the participant , namely discriminating rightward- versus leftward-moving dots . This may have led to the automatic classification of all sensory input according to this task-set ( as either coding for ‘left’ or ‘right’ ) , even when that input was not relevant for the task at hand . Possibly , through this classification , attentional resources could be exploited for the processing of these task-irrelevant features . This is especially interesting because in all conflict analyses incongruency was defined as the mismatch between the two features of the auditory stimulus ( location and sound ) and not between a visual feature ( leftward-moving dots ) and one feature of the auditory stimulus ( e . g . , the word ‘right’ ) . Note that we report additional behavioral results that show clear indications of conflict when the task-relevant feature of the visual stimulus interferes directly with a single task-irrelevant feature of the auditory task ( e . g . , auditory content-dot-motion conflict ) . When inspecting the T-F maps for the vertical RDM task , the relatively fleeting temporal characteristics of the processing of the task-irrelevant stimulus features ( sound content and location ) might suggest that the integration of these features may not be possible due to a lack of time as proposed in the incremental grouping model of attention ( Roelfsema , 2006; Roelfsema and Houtkamp , 2011 ) . However , the time window in which conflict was decodable when the auditory conflicting features were task-relevant coincides with the time range in which these features could be decoded when the auditory conflicting features were task-irrelevant ( Figure 5A ) . Therefore , it seems unlikely that the more temporally constrained processing of task-irrelevant stimulus features is the cause of hampered conflict detection . Besides time being a factor , the processing of task-irrelevant features in the vertical RDM task may have also been too constrained to ( early ) sensory cortices and therefore could not progress to integration networks , including the MFC , necessary for the detection of conflict . Speculatively , the processing of task-irrelevant auditory features was relatively superficial due to the relatively few remaining resources ( Lavie et al . , 2004; Sigman and Dehaene , 2006; Zylberberg et al . , 2010; Zylberberg et al . , 2011 ) , and combined with a lack of object-based attention , this may have prevented the propagation of the information to the MFC . It has been hypothesized that unattended ( sometimes referred to as ‘preconscious’; Dehaene et al . , 2006; Dehaene and Changeux , 2011 ) stimuli are not propagated deeply in the brain , but still allow for shallow recurrent interactions in sensory cortices . The poor spatial resolution of EEG measurements and the specifics of our experimental setup , however , do not allow to test these ideas regarding the involvement of spatially distinct cortices . Yet , previous work of our group suggests that task-irrelevant nonconscious information does not propagate to the frontal cortices , whereas task-relevant nonconscious information does . We demonstrated that masked task-irrelevant conflicting cues induced similar early processing in sensory cortices as compared to masked task-relevant cues , but prohibit activation of frontal cortices ( van Gaal et al . , 2008 ) . These findings are not conclusive , and so we believe that uncovering the role of task relevance in processing of ( nonconscious ) information deserves more attention in future work ( see also van Gaal et al . , 2012 for a discussion on this issue ) . We show that conflict processing is absent when conflicting features are fully task-irrelevant , while evidence of sensory processing is still present in neural data ( Figures 2B , 4B , and 5 ) . Even though sensory processing of auditory features seems relatively preserved under various levels of task relevance of these features , it appears that sensory processing may in fact also be affected when the feature is task-irrelevant ( Figure 4—figure supplement 2 ) , in line with previous studies ( e . g . , Alilović et al . , 2019; Jehee et al . , 2011; Kok et al . , 2012; Kouider et al . , 2016 ) , although to a lesser extent than conflict processing ( Figure 5B ) . For example , when sound location is the task-relevant feature ( i . e . , in the location discrimination task ) , decoding accuracies for that feature are more broadband in the frequency domain ( Figure 4B ) and higher in the time domain ( Figure 4—figure supplement 2 ) , compared to location decoding performance in other tasks . This increased decoding accuracy is present even before a response has been made , suggesting decreased early stage sensory processing in tasks where the decoded feature is not task-relevant . However , despite sensory processing being weakened under decreasing levels of task relevance , it is not diminished , in line with previous findings of ongoing processing in the ( near ) absence of attention ( Fahrenfort et al . , 2017; Li et al . , 2002; Peelen et al . , 2009 ) . Processing of conflict between the two interfering auditory features , on the contrary , is hampered when the features are fully task-irrelevant . This is further supported by the significant interaction between task and feature in terms of decoding performance within the predefined ROI ( Figure 5B ) . Summarizing , although processing of sensory features is degraded under decreasing levels of task relevance it is present regardless of attention , whereas detection of conflict between these features is no longer possible when the features are fully task-irrelevant . Besides object-based attention , the process through which attentional resources are allocated to the processing of task-irrelevant features of a task-relevant object , other mechanisms might also play a role in the extent to which sensory information is processed , such as the active suppression of task-irrelevant information . It has been shown that task-irrelevant information that is response-relevant , and can thus potentially interfere with performance on the primary task , can be suppressed to minimize interference ( Appelbaum et al . , 2011; Janssens et al . , 2018; Polk et al . , 2008; but see Egner and Hirsch , 2005 ) . This would result in more reduced sensory processing , indexed by lower decoding performance , for task-irrelevant features that are response-relevant than for task-irrelevant features that are not . Disentangling the effects of such mechanisms , object-based attention and their possible interactions on the processing of sensory and cognitive information , however , falls outside the scope of this work . For our main analysis , we trained a multivariate classifier on congruent versus incongruent trials and observed effects of task relevance of the performance of the classifier , that is , decoding performance was hampered when conflicting features were fully task-irrelevant ( Figures 2B , 4B , and 5B ) . Moreover , we report behavioral effects of conflict in all auditory tasks as well ( Figures 2A and 4A ) . Given that behavioral performance on the auditory tasks is worse for incongruent trials as compared to congruent trials , one may wonder whether our multivariate decoder is in fact picking up information related to conflict detection or processes related to task difficulty . Whether medial frontal theta-band oscillations are a reflection of conflict detection or task difficulty , and whether these factors can be dissociated in principle , has been the topic of debate in the literature ( Grinband et al . , 2011a; Grinband et al . , 2011b; McKay et al . , 2017; Ruggeri et al . , 2019; Yeung et al . , 2011 ) . On the one hand , it has been shown that activity in the dorsal medial prefrontal cortex is related to RT , suggesting that neural markers of conflict may in fact reflect time on task ( Grinband et al . , 2011b; Grinband et al . , 2011b; Ruggeri et al . , 2019 ) . On the contrary , other research has shown that enhanced prefrontal theta-band oscillations are found in conflicting trials even when controlling for RT ( Cohen and van Gaal , 2014 ) or task difficulty ( McKay et al . , 2017 ) . The decoding results presented in this work likely reflect conflict processing , and not just task difficulty , for two reasons . First , the spatial distribution and time-frequency dynamics of the congruency decoding results are comparable to those more commonly found in the literature on conflict processing , even in a study where conflicting signals were matched for RT ( Cohen and van Gaal , 2014 ) . Specifically , using the content discrimination task of experiment 1 as example , we observe effects of conflict centered on the theta-band and ~230–610 ms post-conflict presentation , with a clear medial frontal spatial profile ( Figure 2B , Figure 5—figure supplement 1A ) . Second , auditory stimulus conflict was decodable from neural data for two tasks in which there were either no effects of conflict – or task difficulty – on behavioral performance ( i . e . , horizontal RDM task ) , or even increased behavioral performance on conflicting trials ( i . e . , volume oddball task ) . Therefore , we believe that the observed congruency decoding results presented here are mainly driven by the detection of conflicting sensory inputs and are not , or much less so , driven by task difficulty . Contrary to the current study , previous studies using a variety of conflict-inducing paradigms and attentional manipulations reported conflict effects in behavior and electrophysiological recordings induced by unattended stimuli or stimulus features ( Mao and Wang , 2008; Padrão et al . , 2015; Zimmer et al . , 2010 ) . However , our study deviates from those studies in several crucial aspects . First , we explicitly separate task-relevant stimulus features that cause conflict and task-relevant features that do not , parsing the cognitive components that induce cognitive control in this context . Furthermore , in the RDM and volume oddball tasks we tested whether conflict between two task-irrelevant features could be detected by the brain . Specifically , we investigated if conflict between two task-irrelevant features would be detected in the presence or absence of object-based attention ( e . g . , volume oddball task vs . vertical RDM task ) , also manipulating whether task-irrelevant conflicting features mapped onto the response or not ( horizontal RDM task vs . vertical RDM task ) . This approach is crucially different from previous studies that exclusively tested whether a task-irrelevant or unattended stimulus ( feature ) could interfere with processing of a task-relevant feature ( Mao and Wang , 2008; Padrão et al . , 2015; Zimmer et al . , 2010 ) . Under such conditions , at least one source contributing to the generation of conflict ( i . e . , the task-relevant stimulus ) is fully attended , and therefore , one cannot claim that under those circumstances conflict detection occurs outside the scope of attention . It can be argued that in our horizontal RDM task the task-irrelevant auditory features ( location and content ) that mapped onto the response of the primary task could interfere with the processing of horizontal dot-motion , that is , the task-relevant feature . This is in fact true , as we found effects of auditory content-dot-motion and auditory location-dot-motion conflict in behavior ( both on RTs and ERs ) . This highlights that a single feature of a task-irrelevant stimulus can interfere with the response to a task-relevant stimulus when there are overlapping feature-response-mappings . This is different from two features of a task-irrelevant stimulus to produce inherent conflict ( e . g . , between auditory content and location ) , which is what we specifically investigated by always testing the presence of auditory content-location conflict only . A similar argument might be made for our vertical RDM and volume oddball tasks because in those cases the auditorily presented stimuli could potentially conflict with responses that were exclusively made with the right hand , for example , the spoken word ‘left’ or the sound from left location may conflict generally more with a right-hand response ( independent of the up/down classification or oddball detection ) than the spoken word ‘right’ or the sound from right location . In the vertical RDM task , the auditorily presented stimuli were truly task-irrelevant as both stimulus content and location in isolation did not affect behavior . In the volume oddball task , sound content and location were task-irrelevant features , but these features were part of the attended stimulus and hence selected through object-based attention . In this task , the content of the auditory stimuli ( e . g . , ‘left’ ) did interfere with right-hand responses to the volume oddball task , resulting in longer RTs ( compared to ‘right’ ) . Moreover , in this task we did find behavioral and neural effects of conflict between two auditory features ( Figure 4 ) . The absence of conflict effects in the vertical RDM and presence of such effects in the volume oddball task and horizontal RDM indicates that at least one feature of the stimulus containing the conflicting features should be task-relevant or associated with a response in order for conflict to be detected . Summarizing , we show that the brain is not able to detect conflict that emerges between two features of a task-irrelevant stimulus in the absence of object-based attention . Lastly , in other studies , conflicting stimuli were often task-irrelevant on one trial ( e . g . , because they were presented at an unattended location ) but task-relevant on the next ( e . g . , because they were presented at the attended location ) ( e . g . , Padrão et al . , 2015; Zimmer et al . , 2010 ) . Such trial-by-trial fluctuations of task relevance allow for across-trial modulations to confound any current trial effects ( e . g . , conflict-adaptation effect ) and also induce a ‘stand-by attentional mode’ where participants never truly disengage to be able to determine if a stimulus is task-relevant . We prevented such confounding effects in the present study , where the ( potentially ) conflicting features or the auditory stimulus were task-irrelevant on every single trial in the vertical RDM , horizontal RDM , and volume oddball task . One difference between the content and location discrimination tasks , on the one hand , and the volume oddball and RDM tasks , on the other , was the task relevance of the ( conflicting ) auditory features . Another major difference between these groups of tasks was , consequently , the origin of the conflict . When the auditory stimuli were task-relevant , the origin of conflict was found in the interference of a task-irrelevant feature on behavioral performance , whereas for the other tasks this was not the case . We argued that in the volume oddball and RDM tasks salient auditory stimuli could be intrinsically conflicting . Intrinsic conflict is often referred to as perceptual conflict , as opposed to the aforementioned behavioral conflict ( Kornblum , 1994 ) . Although perceptual conflict effects are usually weaker than response conflict effects , both in behavior and electrophysiology ( Frühholz et al . , 2011; van Veen et al . , 2001; Wang et al . , 2014 ) , this difference in the origin of the conflict is unlikely to explain why we did not observe effects of conflict under task-irrelevant sensory input , as opposed to earlier studies . First , several neurophysiological studies have previously reported electrophysiological modulations by perceptual conflict centered on the MFC ( Jiang et al . , 2015a; Nigbur et al . , 2012; van Veen et al . , 2001; Wang et al . , 2014; Zhao et al . , 2015 ) . Second , an earlier study using a task similar to ours ( but including only task-relevant stimuli ) showed effects of perceptual conflict , that is , unrelated to response-mapping , in both behavior and neural measures ( Buzzell et al . , 2013 ) . Third , the prefrontal monitoring system has previously been observed to respond when participants view other people making errors ( Jääskeläinen et al . , 2016; van Schie et al . , 2004 ) , suggesting that cognitive control can be triggered without the need to make a response . Fourth , in our volume oddball task , where conflict was perceptual in nature as well , we did observe conflict effects , both in behavior and neural data . The lack of conflict effects in the vertical RDM task might suggest a case of inattentional deafness , a phenomenon known to be induced by demanding visual tasks , which manifests itself in weakened early ( ~100 ms ) auditory evoked responses ( Molloy et al . , 2015 ) . Interestingly , human speech seems to escape such load modulations and is still processed when unattended and task-irrelevant ( Olguin et al . , 2018; Röer et al . , 2017; Zäske et al . , 2016 ) , potentially because of its inherent relevance , similar to ( other ) evolutionary relevant stimuli such as faces ( Finkbeiner and Palermo , 2017; Lavie et al . , 2003 ) . Indeed , the results of our multivariate analyses demonstrate that spoken words are processed ( at least to some extent ) when they are task-irrelevant as stimulus content ( the words ‘left’ and ‘right’ , middle row , Figures 2B and 4B , and Figure 5B ) and stimulus location ( whether the word was presented on the left or the right side’ , bottom row , Figures 2B and 4B , and Figure 5B ) could be decoded from time-frequency data for all behavioral tasks . For all tasks , classification of stimulus content was present in the theta-band ( 4–8 Hz ) , which is in line with a previously proposed theoretical role for theta oscillations in speech processing , namely that they track the acoustic envelope of speech ( Giraud and Poeppel , 2012 ) . After this initial processing stage , further processing of stimulus content is reflected in more durable broadband activity for the content discrimination tasks , possibly related to higher-order processes ( e . g . , semantic ) and response preparation ( middle row and left column , Figures 2B and Figure 4B ) . Similarly to processing of stimulus content , processing of stimulus location was most strongly reflected in the delta- to theta-range for all tasks ( Figure 4B ) , which may relate to the auditory N1 ERP component , an ERP signal that is modulated by stimulus location ( Fuentemilla et al . , 2006; Lewald and Getzmann , 2011; Salminen et al . , 2015 ) . We also observed above-chance location decoding in the alpha-band for task-relevant auditory stimuli , convergent with the previously reported role of alpha-band oscillations in orienting and allocating audio-spatial attention ( Weisz et al . , 2014 ) . Thus , the characteristics of the early sensory processing of ( task-irrelevant ) auditory stimulus features in our study are in line with recent findings of auditory and speech processing . Moreover , our observations are in line with recent empirical findings that suggest a dominant role for late control operations , as opposed to early selection processes , in resolving conflict ( Itthipuripat et al . , 2019 ) . Specifically , this work showed that in a Stroop-like paradigm both target and distractor information is analyzed fully , after which the conflicting input is resolved . Extending on this , we show preserved initial processing of task-irrelevant sensory input , but hampered late control operations necessary to detect conflict , at least in the current setup . Previous studies investigating conflict through masking procedures concluded that conflict detection by the MFC may happen automatically and is still operational under strongly reduced levels of stimulus visibility ( D'Ostilio and Garraux , 2012b; Jiang et al . , 2015b; van Gaal et al . , 2008 ) . Such automaticity can often be enhanced through training of the task . For example , training in a stop-signal paradigm in which stop-signals were rendered unconscious through masking led to an increase in the strength of behavioral modulations of these stimuli ( van Gaal et al . , 2009 ) . In order to see whether enhancing such automaticity could hypothetically increase the likelihood of conflict detection , we included extensive training sessions in the first experiment and had measurements of the volume oddball task before and after exposure to conflicting tasks in the second experiment . In experiment 1 , we found no neural effects of conflict detection in the vertical RDM task , even when participants had been trained on the auditory task for 3600 trials ( Figure 2B , Figure 2—figure supplement 1D ) . Training did result in a decrease of behavioral and neural conflict effects in content discrimination task I of experiment 1 , indicating that our training procedure was successful ( Figure 2—figure supplement 1A–C ) and suggesting more efficient functioning of conflict resolution mechanisms . In experiment 2 , participants performed the volume oddball task twice , once before and once after sound location and content had been mapped to responses . Again , we aimed to see if training on conflict tasks would enhance automaticity of conflict processing in a paradigm where the auditory conflicting features were task-irrelevant . We did not find any statistically reliable differences in behavioral conflict effects or accuracy of congruency decoding between the two runs ( Figure 4—figure supplement 1 ) . Therefore , it seems that the automaticity of conflict detection by the MFC and associated networks does not hold when the auditory stimulus is task-irrelevant ( at least after the extent of training and exposure as studied here ) . Remarkably , we report increased behavioral performance on the volume oddball task for incongruent trials as compared to congruent trials ( Figure 4A , Figure 4—figure supplement 1B–D ) . Speculatively , this increased behavioral performance ( d’ ) on incongruent trials may be due to attentional capture of conflicting stimuli . Attentional capture is the involuntary shift of attention towards salient stimuli ( Awh et al . , 2012; Theeuwes , 2010 ) . Possibly , the detection of conflict between sound content and location is a salient event causing ( re- ) capture of attention towards the auditory stimulus , resulting in better processing of the stimulus information and ultimately better oddball detection performance . Thus , following the detection of conflict , frontal networks would have to exert control over attentional resources and direct them towards the source of the conflict . Interestingly , cases of frontal control over attentional processes have been demonstrated in the past , for example , showing that task-irrelevant distractors that have been related to reward induce stronger attentional capture ( Anderson et al . , 2011 ) and that high working memory load increases the strength of attentional capture by distractors ( Lavie and Fockert , 2006 ) . The present study was however not optimized to test directly which underlying mechanisms are associated with increased sensitivity of conflicting sensory input , and this issue merits further experimentation . Summarizing , high-level cognitive processes that require the integration of conflict inducing stimulus features are strongly hampered when none of the stimulus features of the conflict inducing stimulus are task-relevant and hence object-based attention is absent . This work nicely extends previous findings of perceptual processing outside the scope of attention ( Peelen et al . , 2009; Sand and Wiens , 2011; Schnuerch et al . , 2016; Tusche et al . , 2013 ) , but suggests crucial limitations of the brain’s capacity to process task-irrelevant ‘complex’ cognitive control-initiating stimuli , indicative of an attentional bottleneck to detect conflict at high levels of information analysis . In contrast , the processing of more basic physical features of sensory input appears to be less deteriorated when input is task-irrelevant ( Lachter et al . , 2004 ) .
We performed two separate experiments , each containing multiple behavioral tasks . For each of these experiments , we recruited 24 healthy human participants from the University of Amsterdam . None of the participants took part in both experiments . So , in total 48 participants ( 37 females ) aged 18–30 participated in this experiment for monetary compensation or participant credits . All participants had normal or corrected-to-normal vision and had no history of head injury or physical and mental illness . This study was approved by the local ethics committee of the University of Amsterdam , and written informed consent was obtained from all participants after explanation of the experimental protocol . We will describe the experimental design and procedures for the two experiments separately . Data analyses and statistics were similar across experiments and will be discussed in the same section . Participants performed two tasks in which conflicting auditory stimuli were either task-relevant or task-irrelevant . In both tasks , conflict was elicited through a paradigm adapted from Buzzell et al . , 2013 , in which spatial information and content of auditory stimuli could interfere . In content discrimination task I , participants had to respond to the auditory stimuli , whereas in vertical RDM task participants had to perform a demanding RDM task , while the auditory conflicting stimuli were still presented ( Figure 1A ) . Participants performed both tasks on two experimental sessions of approximately 2 . 5 hr . In between these two experimental sessions , participants had two training sessions of 1 hr during which they only performed the task-relevant task ( Figure 1B ) . On each experimental session , participants first performed a shortened version of the RDM task to determine the appropriate coherency of the moving dots ( 73–77% correct ) , followed by the actual task-irrelevant auditory task , and finally the task-relevant auditory task . Participants were seated in a darkened , sound-isolated room , 50 cm from a 69 × 39 cm screen ( frequency: 120 Hz , resolution: 1920 × 1080 , RGB: 128 , 128 , 128 ) . Both tasks were programmed in MATLAB ( R2012b , The MathWorks , Inc ) , using functionalities from Psychtoolbox ( Kleiner et al . , 2007 ) . In the task-irrelevant auditory task , participants performed an RDM task in which they had to discriminate the motion ( up vs . down ) of white dots ( n = 603 ) presented on a black circle ( RGB: 0 , 0 , 0; ~14° visual angle; Figure 1A ) . Onset of the visual stimulus was paired with the presentation of the auditory conflicting stimulus . Participants were instructed to respond according to the direction of the dots by pressing the ‘up’ or ‘down’ key on a keyboard with their right hand as fast and accurate as possible . Again , participants could respond in a 2 s time interval , which was terminated after responses , and followed by an inter-trial interval of 850–1250 ms . Task difficulty , in terms of dot-motion coherency ( i . e . , proportion of dots moving in the same direction ) , was titrated between blocks to 73–77% correct of all trials within that block . Similar to content discrimination task I , the vertical RDM was divided into 12 blocks containing 100 trials each , separated by short breaks . Again , congruent and incongruent trials , with respect to the auditory stimuli , occurred equally often . In the second experiment , we wanted to investigate whether it is task irrelevance of the auditory stimulus itself or task irrelevance of the auditory features ( i . e . , content and location ) that determine whether prefrontal control processes are hampered . Participants performed two tasks in which auditory stimuli were fully task-relevant ( location discrimination and content discrimination ) , one task in which the auditory stimulus was relevant but the features auditory location and content were not ( volume oddball ) and one task in which the auditory stimulus itself was task-irrelevant but its features location and content could potentially interfere with behavior ( horizontal RDM ) . Participants came to the lab for one session , lasting 3 hr . Each session started with the volume oddball task , followed by ( in a counterbalanced order ) the location discrimination , content discrimination , and horizontal RDM tasks , and ended with a block of the volume oddball task again . We included the location discrimination and content discrimination tasks both to replicate the results of experiment 1 and also to see if there would be differences in these results between the two tasks . Specifically , we investigated whether processing of auditory stimulus features – as indicated by multivariate classification accuracies – would differ between the two tasks . Participants were seated in a darkened , sound-isolated room , 50 cm from a 69 × 39 cm screen ( frequency: 120 Hz , resolution: 1920 × 1080 , RGB: 128 , 128 , 128 ) . All four experiments were programmed in Python 2 . 7 using functionalities from PsychoPy ( Peirce et al . , 2019 ) . The auditory location discrimination task was identical to the auditory content discrimination task II , with the exception that participants were now instructed to respond according to the location of the auditory stimulus . Thus , participants had to press a left button ( ‘A’ ) for sounds coming from a left speaker and right button ( ‘L’ ) for sounds coming from a right speaker . Again , participants performed six blocks of 100 trials . In the volume oddball task , the same auditory stimuli were presented . Again , on every trial , a black disk with randomly moving dots ( coherence: 0 ) was presented to keep sensory input similar between tasks . Occasionally an auditory stimulus would be presented at a lower volume . The initial volume of the oddballs was set to 70% , but was staircased in between blocks to yield 83–87% correct answers . If participants' performance on the previous block was below or above this range , the volume increased or decreased with 5% , respectively . The odds of a trial being an oddball trial were 1/8 ( drawn from a uniform distribution ) . Participants were instructed to detect these oddballs by pressing the spacebar as fast as possible whenever they thought they heard a volume oddball . If they thought that the stimulus was presented at a normal volume , they were instructed to refrain from responding . The response interval was 800 ms , which was terminated at response . A variable inter-trial interval of 150–350 ms was initiated after this response interval . Participants performed two runs of this task , at the beginning of each session and at the end of each session . Each run contained five blocks of 100 trials each . In the horizontal RDM task , participants had to discriminate the motion ( left vs . right ) of white dots ( n = 603 ) presented on a black circle ( RGB: 0 , 0 , 0; ~14° visual angle ) . Onset of the visual stimulus was paired with the presentation of the auditory stimulus . Participants were instructed to respond according to the direction of the dots by pressing a left key ( ‘A’ ) or right key ( ‘L’ ) on a keyboard with left and right index fingers , respectively , as fast and accurate as possible . Participants could respond in an 800 ms time interval , which was terminated after responses , and followed by an inter-trial interval of 250–450 ms . Task difficulty , in terms of dot-motion coherency ( i . e . , proportion of dots moving in the same direction ) , was set to 0 . 3 in the first block . This value indicated an intermediate coherence as every trial in that block could be the intermediate coherence , but also half ( 0 . 15 ) or twice this intermediate coherence ( 0 . 6 ) . The intermediate coherence was titrated between blocks to 73–77% correct . If behavioral performance fell outside that range , 0 . 01 was added to or subtracted from the intermediate coherence . The horizontal RDM consisted of 10 blocks of 60 trials . All stimulus features ( i . e . , sound content , location , and congruency ) were presented in a balanced manner ( e . g . , 50% congruent , 50% incongruent ) . We were primarily interested in the effects of congruency of the auditory stimuli on both behavioral and neural data . Therefore , we defined trial congruency on the basis of these auditory stimuli in all behavioral tasks of the two experiments . All behavioral analyses were programmed in MATLAB ( R2017b , The MathWorks , Inc ) . In all tasks , trials with an RT <100 ms or >1500 ms were excluded from behavioral analyses . Missed trials were excluded in all tasks ( except the volume oddball task ) as well . In order to investigate whether current trial conflict effects were present under varying levels of task relevance and to inspect if training on/exposure to conflict-inducing tasks modulated such conflict effects , we performed rm-ANOVAs on different behavioral measures . For all tasks , excluding the volume oddball task , the rm-ANOVAs were performed on the ER over all trials and RTs on correct trials . For the volume oddball task , perceptual sensitivity ( d’; Green and Swets , 1966 ) and RTs of correct trials ( i . e . , ‘hit’ trials ) were analyzed with rm-ANOVAs . If the assumption of sphericity was violated , we applied a Greenhouse–Geisser correction . For the tasks from experiment 1 , we performed these ANOVAs with task relevance , training ( before vs . after ) and current trial congruency as factors ( 2 × 2 × 2 factorial design ) . Additional post-hoc rm-ANOVAs , for content discrimination task I and vertical RDM task separately ( 2 × 2 factorial design ) , were used to inspect the origin of significant factors that were modulated by task relevance . For content discrimination task I , the location discrimination task , and horizontal RDM task , we performed a rm-ANOVA with task and congruency as factors ( 3 × 2 factorial design ) . For the volume oddball task , we performed a rm-ANOVA with congruency and run number as factors ( 2 × 2 factorial design ) on RTs , d’ scores , hit rates , and false alarm rates . We also applied paired sample t-tests comparing the difference in these variables between incongruent and congruent for all tasks , within each experimental session ( vertical RDM , content discrimination task I ) and run ( volume oddball task ) . To test interference of the individual auditory features , sound content and sound location , on performance on the vertical RDM task and volume oddball task , we performed rm-ANOVAs on RTs with sound location and sound content as factors ( 2 × 2 factorial design ) . Additionally , for the horizontal RDM , we tested for auditory-visual conflict effects ( i . e . , conflict between sound content/location and dot direction ) in RT and ER with paired sample t-tests comparing incongruent and congruent trials . In case of null findings , we performed a Bayesian analysis ( rm-ANOVA or paired sample t-test ) with identical parameters and settings on the same data to test if there was actual support of the null hypothesis ( JASP Team , 2018 ) . EEG data were analyzed using custom-made software written in MATLAB , with support from the toolboxes EEGLAB ( Delorme and Makeig , 2004 ) and ADAM ( Fahrenfort et al . , 2018 ) . EEG data were recorded with a 64-channel BioSemi apparatus ( BioSemi B . V . , Amsterdam , The Netherlands ) at 512 Hz . Vertical eye movements were recorded with electrodes located above and below the left eye , and horizontal eye movements were recorded with electrodes located at the outer canthi of the left and the right eye . All EEG traces were re-referenced to the average of two electrodes located on the left and right earlobes ( mastoidal reference for one participant in experiment 1 ) . The data were band-pass filtered offline , with cutoff frequencies of 0 . 01–50 Hz . Next , epochs were created by taking data from −1 s to 2 s around onset of stimulus presentation . We then rejected epochs containing blink artifacts and high-voltage artifacts . Blinks were defined as VEOG data exceeding a threshold of ±100 mV in a time window of 0–800 ms post-stimulus . This procedure resulted in the removal of 5 . 03% ( SD = 9 . 71% ) of epochs in experiment 1% and 6 . 10% ( SD = 7 . 54% ) of epochs in experiment 2 . Subsequently , high-voltage artifacts were defined as events where voltage exceeded a threshold of ±300 mV in a time window of 0–800 ms post-stimulus on any EEG channel . With this second round of artifact rejection , 3 . 65% ( SD = 7 . 96% ) of all epochs were removed in experiment 1% and 4 . 16% ( SD = 8 . 86% ) in experiment 2 . In total , 8 . 68% ( SD = 12 . 09% ) of all trials were removed in experiment 1% and 10 . 26% ( SD = 13 . 92% ) were removed in experiment 2 . Note that this procedure ensures the absence of blinks and high-amplitude artifacts within the predefined ROI time window of 100–700 ms . The data of one participant in experiment 2 contained many artifacts . Specifically , across all five tasks performed in experiment 2 , 64 . 46% ( SD = 9 . 88% ) of all epochs were rejected for this participant . This was more than three standard deviations from the average number of rejected trials across participants and files and left too few trials for the decoding analysis . Therefore , this participant was removed from all EEG analyses . We applied a multivariate backwards decoding model to EEG data that was transformed to the time-frequency domain . We used multivariate analyses both because its higher sensitivity in comparison with univariate analyses and to inspect if and to what extent different stimulus features ( i . e . , location and content ) were processed in both tasks , without having to preselect spatial or time-frequency ROIs . The ADAM toolbox was used for time-frequency decomposition and decoding ( Fahrenfort et al . , 2018 ) . Single-trial power spectra were computed by convolving the EEG data with a complex wavelet ( wavelet size of 0 . 5 s ) after the application of a Hann taper ( epochs: −100 ms to 1000 ms , 2–30 Hz in linear steps of 2 Hz ) . Raw time-frequency data contained both induced and evoked power . Trials were classified according to current trial stimulus features ( i . e . , location and content ) , resulting in four trial types . As decoding algorithms are known to be time-consuming , epochs were resampled to 64 Hz . Then , we applied a decoding algorithm to the data according to a 20-fold cross-validation scheme using either stimulus location , stimulus content , or congruency as stimulus class . Specifically , a linear discriminant analysis ( LDA ) was trained to discriminate between stimulus classes ( e . g . , left vs . right speaker location , etc . ) . Classification accuracy was computed as the AUC , a measure derived from Signal Detection Theory ( Green and Swets , 1966 ) . The multivariate classifiers were on different subsets of trials , depending on the behavioral task . For the auditory tasks ( content discrimination I and II , location discrimination , and volume oddball detection ) , only correct trials were included in the analysis as errors tend to elicit a similar , albeit not identical , neural response as cognitive conflict and errors are more likely on incongruent trials ( Cohen and van Gaal , 2014 ) . For the volume oddball detection task , we additionally excluded all oddball trials , thus only testing correct rejections , in order to prevent conflict arising between responses made exclusively with the right hand and sound content and location . For the two visual tasks ( horizontal and vertical RDM ) , we trained the classifier on all trials . Topographical maps were created in order to investigate the spatial sources of activity related to the processing of the auditory features ( content , location , and congruency ) . We first extracted classifier weights for each task and feature from the predefined ROI ( 100–700 ms , 2–8 Hz ) , allowing us to directly compare the spatial distributions between features and tasks . However , raw classifier weights are not interpretable as neural sources of activity and therefore have to be reconstructed ( Haufe et al . , 2014 ) . Thus , classifier weights were transformed to activation patterns by multiplying them with the covariance in the EEG data . The topographical activity maps of tasks and features with low decoding performance should be interpreted with caution as activation patterns reconstructed from classifier weights may be unreliable when decoding performance is low ( Haufe et al . , 2014 ) . We applied a time-domain decoding analysis on EEG data to inspect the possible effect of task relevance of a stimulus feature on sensory processing . For this analysis , only EEG data from the tasks of experiment 2 were used as its parameters were comparable between tasks in this experiment ( e . g . , always visual stimulus present , no extensive training ) . For the analysis , we trained linear classifiers ( LDA ) to discriminate sound content ( ‘left’ vs . ‘right’ ) or sound location ( left speaker vs . right speaker ) . First , epochs ( from −100 ms to 1000 ms around stimulus presentation ) were resampled to 64 Hz similar to the time-frequency decoding analyses . Then , the models were trained and tested according to a 20-fold cross-validation scheme . The AUC scores we obtained via multivariate analyses of our EEG data were tested per timepoint with one-sided t-tests ( X¯>0 . 5 ) across participants against chance level ( 50% ) . These t-tests were corrected for multiple comparisons over time using cluster-based permutation tests ( p<0 . 05 , 1000 iterations ) . For each decoded stimulus feature , we then compared the decoding accuracies of the behavioral task in which the feature was task-relevant to all other tasks in a pairwise fashion ( e . g . , location decoding under location discrimination task vs . horizontal RDM task ) , with cluster-corrected two-sided t-tests against 0 . The AUC scores we obtained via multivariate analyses of our EEG data were tested per timepoint and frequency with one-sided t-tests ( X¯>0 . 5 ) across participants against chance level ( 50% ) . These t-tests were corrected for multiple comparisons over time and frequency using cluster-based permutation tests ( p<0 . 05 , 1000 iterations ) . This procedure yields time-frequency clusters of significant above-chance classifier accuracy , indicative of information processing . Note that this procedure yields results that should be interpreted as fixed effects ( Allefeld et al . , 2016 ) , but is nonetheless standard in the scientific community . In addition to the cluster analysis , we performed hypothesis-driven analyses on classifier accuracies that were extracted from a predefined time-frequency ROI . All these analyses were performed in JASP ( JASP Team , 2018 ) . We then applied an ANCOVA on these accuracies with fixed effects being task and stimulus feature . Next , one-sample t-tests ( one-sided , X¯>0 . 5 ) were performed on every task/feature combination to determine whether decoding accuracy of a specific feature within our preselected ROI was above chance during the various behavioral tasks . Additional Bayesian one-sample t-tests ( one-sided , X¯>0 . 5 , Cauchy scale = 0 . 71 ) were performed to inspect evidence in favor of the null hypothesis that decoding accuracy was not above chance . We performed the same analysis on different ROIs . The results of those analyses can be found in Figure 5—figure supplement 1B–D . | Focusing your attention on one thing can leave you surprisingly unaware of what goes on around you . A classic experiment known as ‘the invisible gorilla’ highlights this phenomenon . Volunteers were asked to watch a clip featuring basketball players , and count how often those wearing white shirts passed the ball: around half of participants failed to spot that someone wearing a gorilla costume wandered into the game and spent nine seconds on screen . Yet , things that you are not focusing on can sometimes grab your attention anyway . Take for example , the ‘cocktail party effect’ , the ability to hear your name among the murmur of a crowded room . So why can we react to our own names , but fail to spot the gorilla ? To help answer this question , Nuiten et al . examined how paying attention affects the way the brain processes input . Healthy volunteers were asked to perform various tasks while the words ‘left’ or ‘right’ played through speakers . The content of the word was sometimes consistent with its location ( ‘left’ being played on the left speaker ) , and sometimes opposite ( ‘left’ being played on the right speaker ) . Processing either the content or the location of the word is relatively simple for the brain; however detecting a discrepancy between these two properties is challenging , requiring the information to be processed in a brain region that monitors conflict in sensory input . To manipulate whether the volunteers needed to pay attention to the words , Nuiten et al . made their content or location either relevant or irrelevant for a task . By analyzing brain activity and task performance , they were able to study the effects of attention on how the word properties were processed . The results showed that the volunteers’ brains were capable of dealing with basic information , such as location or content , even when their attention was directed elsewhere . But discrepancies between content and location could only be detected when the volunteers were focusing on the words , or when their content or location was directly relevant to the task . The findings by Nuiten et al . suggest that while performing a difficult task , our brains continue to react to basic input but often fail to process more complex information . This , in turn , has implications for a range of human activities such as driving . New technology could potentially help to counteract this phenomenon , aiming to direct attention towards complex information that might otherwise be missed . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2021 | Preserved sensory processing but hampered conflict detection when stimulus input is task-irrelevant |
Outer hair cells ( OHCs ) in the mammalian ear exhibit electromotility , electrically driven somatic length changes that are thought to mechanically amplify sound-evoked vibrations . For this amplification to work , OHCs must respond to sounds on a cycle-by-cycle basis even at frequencies that exceed the low-pass corner frequency of their cell membranes . Using in vivo optical vibrometry we tested this theory by measuring sound-evoked motility in the 13–25 kHz region of the gerbil cochlea . OHC vibrations were strongly rectified , and motility exhibited first-order low-pass characteristics with corner frequencies around 3 kHz– more than 2 . 5 octaves below the frequencies the OHCs are expected to amplify . These observations lead us to suggest that the OHCs operate more like the envelope detectors in a classical gain-control scheme than like high-frequency sound amplifiers . These findings call for a fundamental reconsideration of the role of the OHCs in cochlear function and the causes of cochlear hearing loss .
The hair bundles of auditory sensory cells are deflected by sound-driven vibrations , causing mechano-electric transduction channels to open and close . The resulting receptor current modulates the cell’s membrane potential . The mammalian cochlea contains two distinct types of hair cells . The vast majority of nerve fibers that carry the acoustic information to the brain innervate the inner hair cells ( IHC ) . Up to a few kilohertz , IHC synapses can ‘phase-lock , ’ that is , code the individual cycles of tones . At higher frequencies ( >3 kHz ) , phase-locking rapidly declines and neural coding relies on the DC component of the IHC receptor potential generated by the asymmetric , or rectifying , nature of the IHC receptor current ( Russell and Sellick , 1978 ) . Outer hair cells ( OHC ) modify the mechanical vibrations inside the organ of Corti ( OoC ) , enabling frequency tuning and dynamic range compression . Dysfunctional and missing OHCs strongly reduce sensitivity , and this is a major cause of sensorineural hearing loss ( Ryan and Dallos , 1975 ) . The discovery of electromotility , length changes of isolated OHCs ( Brownell et al . , 1985 ) driven by variations in the membrane potential ( Santos-Sacchi and Dilger , 1988 ) , has intensified the study of OHCs and their functional significance . The membrane protein responsible for electromotility has been identified ( Zheng et al . , 2000 ) , and prestin knockout mice have profound hearing loss ( Liberman et al . , 2002 ) . The dominant view is that OHC electromotility drives vibrations within the OoC in a cycle-by-cycle manner ( Ashmore , 2011 ) over the entire audible range , which extends up to 150 kHz in some species ( Vater and Kössl , 2011 ) . If this view is correct , the AC receptor potentials of OHCs must be large enough to be effective up to high frequencies , even though the membrane capacitance is expected to reduce the AC receptor potentials ( and hence the OHCs' motility ) at a rate of 6 dB per octave ( Dallos , 1984 ) . The functional implication of this electrical low-pass filtering is a limitation in the OHC’s ability to provide cycle-by-cycle mechanical feedback , known as the RC problem ( Ashmore , 2011; Housley and Ashmore , 1992 ) . The electrical corner frequency of OHCs has been measured electrophysiologically in vitro , with highest values ranging from 480 Hz ( Mammano and Ashmore , 1996 ) to 1250 Hz ( Johnson et al . , 2011 ) , but no systematic in vivo data exist due to technical limitations and the cochlea’s extreme vulnerability . In addition to the electrical filtering , the motile process itself may also be too slow to provide high-frequency mechanical feedback . An early in vitro report claiming a bandwidth for electromotility of at least 79 kHz ( Frank et al . , 1999 ) was recently challenged ( Santos-Sacchi and Tan , 2018 ) . Again , in vivo estimates of the corner frequency of motility are missing . Here we use optical vibrometry to measure non-linear components of the OHCs' motile response and determine the corner frequency of OHCs in the high-frequency region of the intact gerbil cochlea .
In response to a tone pair ( Figure 1A ) , vibrations in the OHC/Deiters’ cell region showed a strong envelope-following component ( Figure 1B ) . This reveals a significant degree of rectification in OHC motion , producing 2nd-order distortions ( DP2s ) such as the ‘quadratic difference tone’ at f2-f1 . Using multitone stimuli , we mapped the spatial profile of the DP2s inside the OoC by cross-section measurements . DP2s were concentrated in the OHC/Deiters’ cell ‘hotspot area’ ( Cooper et al . , 2018 ) ( Figure 1C and D ) . They were observed at stimulus levels as low as 25 dB SPL ( Figure 1—figure supplement 1 ) , and disappeared post mortem ( Figure 1—figure supplement 2 ) . These observations confirm that OHCs are the source of the DP2s long known to exist from psychophysical ( Zwicker , 1979 ) , electrophysiological ( Kim et al . , 1980; Nuttall and Dolan , 1993 ) , and cochlear-mechanical ( Cooper and Rhode , 1997 ) studies . The OHC origin of DP2s is consistent with the significant rectified ( ‘DC’ ) component and 2nd harmonics observed in in vivo recordings of OHC receptor potentials ( Dallos , 1986; Cody and Russell , 1987 ) and cochlear microphonics ( Gibian and Kim , 1982 ) . Because rectification by OHC bundles produces DP2s in the receptor current , DP2s are an inevitable part of the electromotile response . More importantly , this part can be isolated through spectral analysis from recorded cochlear vibrations . This makes the DP2 spectrum ideally suited to studying the RC problem . To assess spectral trends , we employed a zwuis tone complex ( Figure 1E ) , a stimulus designed to produce a rich DP2 spectrum upon rectification ( van der Heijden and Joris , 2003; Victor , 1979 ) . Rectifying an N-component zwuis stimulus generates N2 distinct DP2 components at frequencies fk ±fm , each of which can be traced back to a pair of interacting primary frequencies ( fk , fm ) . The vibration spectrum obtained in the OHC region ( Figure 1F ) evoked by this stimulus reveals a rich family of DP2s having a systematic 6-dB/octave roll-off . This roll-off confirms the action of a low-pass filter between the hair bundle’s rectification and the motile response . Accurate estimation of the corner frequency , however , is hampered by the ~10-dB scatter of DP2 magnitudes within each frequency band . We identified three causes of this scatter ( Figure 2 ) . Their elimination reduces the scatter substantially , paving the way to accurate estimates of OHC corner frequencies . The first cause of scatter is combinatorial: when supplying an equal-amplitude input to a rectifier , the resulting second harmonics 2fk are 6 dB below the remaining DP2s ( fk ±fm , k > m ) ( Figure 2A ) . This reflects the binomial coefficients occurring in the second-order terms of the power series describing the nonlinearity ( Meenderink and van der Heijden , 2010 ) . Since each DP2 component can be uniquely traced back to its ‘parent primaries , ’ this is readily corrected . The second cause of scatter is the spatially distributed nature of DP2 generation . The primaries and DP2s propagate as traveling waves ( Kim et al . , 1980; Gibian and Kim , 1982 ) , ( Figure 2—figure supplement 1 ) , so the recorded DP2s are a vector sum of contributions along the path from stapes to recording place ( Schroeder , 1969 ) . The magnitude of slowly propagating components is affected by interference across generation loci ( Figure 2B ) , and the growth and subsequent decay of components entering their peak region further obfuscates their original magnitude . These confounding effects of propagation are eliminated by setting an upper frequency limit to the primaries and DP2s used for stimulation , analysis and iterated adjustment of the stimulus as described below . For this frequency limit we choose half the characteristic frequency ( CF/2 ) . Wave propagation below CF/2 is too fast to cause interference and magnitudes change little during fast propagation ( Ren et al . , 2011 ) ( Figure 2—figure supplement 1 ) . The third cause of scatter in the DP2 spectrum is the unequal amplitude of the primary components entering the rectifier , that is , the effective input that deflects the OHC bundles . Possible causes of unequal amplitudes at the OHC input include imperfections in the sound calibration as well as non-flatness of middle-ear transfer and intracochlear propagation . Even a perfectly regular trend in the input spectrum such as a roll-off creates a scattered effect in the DP2 spectrum ( see Appendix 1 ) . The amplitude of a DP2 component is proportional to the product of its parents’ amplitudes ( van der Heijden and Joris , 2003 ) . This bilinearity causes a scatter in DP2 magnitude ( expressed in decibels ) equal to twice the range of the primary input magnitudes ( Figure 2C and Appendix 1—figure 1 ) . If the OHC input were known , it would take a simple adjustment of the stimulus spectrum to equalize the primary amplitudes at the OHC input , and thereby regularize the DP2 spectrum as in Figure 2A . Current in vivo measurement techniques lack the spatial resolution to determine OHC bundle deflection , but the effective OHC input can be retrieved from the rich DP2 spectrum by exploiting the bilinear relationship between primary and DP2 amplitudes . This computational method was previously used to retrieve the effective IHC input from auditory-nerve responses ( van der Heijden and Joris , 2003 ) . Here we use it to compute the effective OHC input ( see Appendix 1 ) and to adjust the stimulus accordingly . The OHC input thus obtained differs from the linear component of OHC motion , and in fact resembles basilar membrane motion more closely ( Figure 2—figure supplement 2 ) . This means that motion recorded in the OHC region may not be used as a proxy for OHC input . The resemblance between basilar membrane motion and computed OHC input may shed light on the mechanisms underlying the deflection of OHC hair bundles . Within the current study , however , the OHC input is primarily of methodological interest . Experimental equalization of the effective OHC input and compensation for the combinatorial effect had the predicted effect of markedly reducing the scatter in the DP2 spectrum ( Figure 3 ) . After equalization , the DP2 spectrum recorded from the OHCs closely resembles that of the simple nonlinear circuit of Figure 2A with its first-order low-pass characteristics . This holds for both the magnitude and the phase , having a minus 6-dB/octave high-frequency slope and a 0 . 25-cycle high-frequency asymptote ( Figure 4C ) respectively . The corner frequency was 2 . 5 kHz , 2 . 7 octaves below the 16-kHz CF . We obtained data from different cochlear regions in different animals , with CFs ranging from 13 to 25 kHz . The adjustment of the relative stimulus amplitudes always reduced the scatter of the DP2 magnitudes . First-order low-pass characteristics were consistently found across CFs ( Figure 4 ) . The corner frequencies obtained ranged from 2 . 1 to 3 . 3 kHz , showing a weak trend of increasing with CF ( Figure 4D ) . In summary , the rectification displayed in vivo by OHC motility provides a unique opportunity to directly measure OHC corner frequencies without opening the cochlea . When equalizing the primary amplitudes at the OHC input , the DP2 spectra reveal an unmistakable first-order low-pass character , both in terms of magnitude and phase . In the frequency range probed here ( below CF/2 ) stiffness dominates OoC impedance rendering displacement proportional to force ( Dong and Olson , 2009 ) . Thus within the framework of models in which OHCs directly push the basilar membrane ( e . g . , Ramamoorthy et al . , 2007 ) , electromotile force itself suffers from the 6-dB/octave roll-off . The corner frequencies of 2 . 1–3 . 3 kHz that we measured in vivo were 2 . 8 ± 0 . 2 octaves below the CFs of our recording locations . These corner frequencies are higher than values of membrane corner frequency from in vitro studies at lower CF: 480 Hz ( guinea pig , CF ~7 kHz ) ( Mammano and Ashmore , 1996 ) ; 300–1250 Hz ( gerbil , CF , ~350–2500 Hz ) ( Johnson et al . , 2011 ) . The in vivo data fall considerably short of the extrapolations to higher CFs made in the in vitro gerbil study ( Johnson et al . , 2011 ) ( which predict electrical corner frequencies of 6 . 5–11 kHz for the CFs tested here ) . Our observation of a simple 6-dB/octave roll-off and minus 0 . 25-cycle phase asymptote indicates the dominance of a single low-pass mechanism in the entire frequency range tested . Comparison with the in vitro data suggests that this dominant factor is the RC time of the cell membrane , which is fundamental to the operation of all biological cells . The somewhat higher corner frequencies of the present study ( compared to the in vitro data ) may be attributed the more basal location of the OHCs of the present study . A corner frequency at 2 . 8 octaves below CF implies a 17-dB attenuation of the CF component . The shallow increase of OHC corner frequency with increasing CF suggests an even stronger attenuation at higher CFs than studied here . When driven by a sufficiently large electrical input , OHC motility can generate vibrations up to very high frequencies , both in vitro ( Frank et al . , 1999 ) and in vivo ( Ren et al . , 2016 ) , but for acoustic stimulation the low-pass filtering of the receptor potential will limit the frequency range . Various schemes ( reviewed in Johnson et al . , 2011 ) have been proposed to push the frequency limit of electromotility beyond the corner frequency of the OHC cell membrane into the CF range . Our findings do not support such schemes , as the ~2 . 5-kHz corner frequency is evident in the motile response itself . We assessed the quantitative effect of low-pass filtering by the OHCs ( Appendix 2 ) . A 16-kHz tone at the behavioral threshold of the gerbil is estimated to evoke an AC component of the OHC receptor potential of 5 . 7 μV at the peak of traveling wave . At the slightly more basal location where cycle-by-cycle amplification is assumed to start , it is ~1 μV . Inspection of the in vivo OHC recordings in guinea pig of Cody and Russell ( 1987 ) yield an 3 . 6-μV AC component at CF for a 17-kHz tone near the behavioral threshold , corresponding to ~0 . 6 μV at the spatial onset of the putative amplification . Even if these minute variations in the membrane potential could evoke a significant electromotile response , such a motile feedback is unlikely to improve sensitivity because of its expected poor signal-to-noise ratio ( van der Heijden and Versteegh , 2015a ) . Overall our data suggest that OHCs and IHCs have similar properties , namely , considerable rectification ( Pappa et al . , 2019 ) and a corner frequency not exceeding a few kilohertz . Thus , just like in high-frequency IHCs , the receptor potential of high-frequency OHCs is expected to mainly follow the envelope of the waveform that stimulates their hair bundles . In this sense both IHC and OHCs operate as envelope detectors . We therefore propose that OHC motility does not provide cycle-by-cycle feedback , but rather modulates sound-evoked vibrations ( Cooper et al . , 2018; van der Heijden and Versteegh , 2015b ) . In this scenario the dynamic range compression in the cochlea is based on an automatic gain control system ( van der Heijden , 2005 ) in which the degree of OHC depolarization determines the gain . The spatial confinement of the motile response to the OHC/Deiters’ cell region presents another challenge to the prevailing theory that OHCs drive basilar membrane motion directly . It rather suggests that electromotility controls the local coupling between OHCs and Deiters’ cells in a parametric fashion , perhaps dynamically adjusting the amount of dissipation in the Deiters’ cell layer . This fundamentally different view of the function of OHCs has great consequences for the experimental study of their role in hearing loss and the origins of the vulnerability of cochlear sensitivity . As to theoretical work , it is important that models of cochlear function , whether invoking cycle-by-cycle feedback or not , incorporate the findings of the present study .
The materials and methods employed in this study are summarized below . More extensive details are provided elsewhere ( Cooper et al . , 2018 ) . Sound evoked vibrations were recorded from the ossicles and cochlear partitions of deeply anesthetized female gerbils ( n = 27 , weight = 53–75 g ) . Spectral-domain optical coherence tomography ( SD-OCT ) measurements were made from the first turn of the intact cochlea , under open-bulla conditions – optical access to the partition being provided through the transparent round window membrane . The hearing thresholds of the animals were assayed using tone-evoked compound action potential ( CAP ) measurements from a silver electrode placed on the wall of the basal turn of the cochlea . Animals were anesthetized using intraperitoneally injected mixtures of ketamine ( 80 mg/kg ) and xylazine ( 12 mg/kg ) . Supplementary ( 1/4 ) doses of the same mixture were administered at intervals of 10–60 min to maintain the anesthesia at surgical levels throughout subsequent procedures . All experiments were performed in accordance with the guidelines of the Animal Care and Use Committee at Erasmus MC ( protocol AVD101002015304 ) . An SD-OCT system ( Thorlabs Telesto TEL320C1 ) was used for interferometric imaging and vibration measurements . Cross-sectional ( B-scan ) and axial images ( A-scans and M-scans ) were triggered externally using TTL pulses phase-locked to the acoustic stimulation system ( Tucker Davies Technologies system III ) at a sampling rate 111 . 6 kHz . The theoretical resolution of the OCT system was ~3 . 5 µm across a 3 . 5 mm depth-of-field ( i . e . , z-range ) , but the optics of our recording system ( a Mitotoyu IR imaging lens with an NA of 0 . 055 ) introduced an axial point spread function of ~6 µm FWHM and a lateral resolution ( in the xy plane ) of 13 µm ( all assessed in air , with a refractive index of 1; corresponding intracochlear measurements should scale inversely with the refractive index of perilymph , which we assumed to be 1 . 3 ) . The linear operating range of the OCT system was >500 µm . The amount of light incident on the cochlea was ~3 . 7 mW . The sensitivity of the A-scan’s phase-spectra to vibration permitted measurement noise-floors that ranged from ~30 pm/√Hz in the cochlea down to ~3 pm/√Hz in the middle ear . The OCT’s measurement beam was not aligned with any of the cochlea’s principal anatomical axes . The vibration measurements that we made should therefore be sensitive to structural movements in all three of the cochlea’s principal dimensions ( radial , transverse , and longitudinal ) . Specifically , in all recordings used for this study , the measurement beam pointed toward scala vestibuli , toward the apex of the cochlea , and away from the modiolus . When mapping vibrations across the width of the cochlea partition ( cf . Figure 1C , Figure 1—figure supplement 2 ) , measurements were spaced at intervals of between 6 and 12 µm in the xy-plane . Responses were analyzed by Fourier transformation of the vibration waveforms derived from contiguous groups of 3 pixels in each M-scan , where each pixel covers a depth of ~2 . 7 µm in the fluid-filled spaces of the cochlea , and ~3 . 5 µm in the air-filled spaces of the middle-ear . The statistical significance of each response component was assessed using Rayleigh tests of the component’s phase stability across time ( Cooper et al . , 2018 ) . Acoustic stimuli were tailored to fit the nature of each experiment , as described below . Each stimulus was coupled into the exposed ear-canal using a pre-calibrated , closed field sound-system . Stimuli were generally presented for 12 s , with inter-stimulus intervals ~ 1 min . Broad-band multi-tone ‘zwuis’ complexes ( van der Heijden and Joris , 2003 ) were used to determine the characteristic frequency and sensitivity functions of each recording site ( e . g . see Figure 4—figure supplement 1 ) . Each broad-band stimulus had 43 spectral components , spanning from 0 . 4 to 30 kHz with an average spacing of 705 Hz . The components all had equal amplitudes , with levels expressed in decibels re: 20 μPa ( i . e . , dB SPL ) , but stimulus phase was randomized across frequency . The unique property of a zwuis stimulus is that the frequencies of all of its primary components , and all of its potential inter-modulation distortion products up to the third order , are unambiguous . This means that all of the second-order distortion products ( i . e . DP2s ) studied in this paper can readily be attributed to a unique pair of spectral ‘parents’ ( see Appendix 1 ) . Narrow-band zwuis stimuli were used to simplify the analysis and interpretation of DP2 spectra . They consisted from 10 to 15 components , ranging from few hundred hertz to at least one octave below the characteristic frequency of the recording side . The first presentation of each narrow-band stimulus had equal primary amplitudes , but their relative amplitudes were adjusted during subsequent presentations ( fixing the average magnitude in dB SPL ) in order to equalize the input to the OHCs . This procedure is described in the Appendix 1 . | Our ears give us our sense of hearing . Their job is to collect sounds and pass this information on to the brain . Hair cells , a special group of cells in the ear , are responsible for detecting sound vibrations and turning them into the electrical signals that our brains can understand . The ear contains two populations of hair cells: inner hair cells that send signals to the brain , and outer hair cells that act as a protective ‘buffer’ by modulating sound vibrations entering the innermost part of the ear . When outer hair cells are damaged , the vibrations picked up by inner hair cells are much smaller than in a healthy ear . This has led to the idea that outer hair cells actively amplify sounds before passing them on . That is , outer hair cells simultaneously act like microphones ( by receiving sound from the environment ) and loudspeakers ( by re-emitting magnified vibrations ) . One problem with this amplifier theory is that it cannot explain how some animals are able to hear extremely high-pitched sounds . If the theory is true , outer hair cells should be able to re-emit ultrasonic vibrations . However , some observations suggest that they may not vibrate fast enough to do so . To test the amplifier theory , Vavakou et al . measured how outer hair cells in the ear of Mongolian gerbils responded to different sounds . This revealed that the motion of these cells could keep up with moderately high sounds ( around the upper end of a piano’s range ) , but were too sluggish to amplify ultrasound despite gerbils having good ultrasonic hearing . Further experiments showed that instead of acting like amplifiers , outer hair cells seem to monitor the loudness of sound and adjust the level accordingly before passing the vibrations on to the inner hair cells . These results shed new light on how outer hair cells help our ears work . Since damage to these cells can cause hearing loss , understanding how they work could one day guide new methods of protecting or even restoring hearing in vulnerable patients . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"physics",
"of",
"living",
"systems",
"neuroscience"
] | 2019 | The frequency limit of outer hair cell motility measured in vivo |
We report that a major subpopulation of monocyte-derived macrophages ( MDMs ) contains high levels of dUTP , which is incorporated into HIV-1 DNA during reverse transcription ( U/A pairs ) , resulting in pre-integration restriction and post-integration mutagenesis . After entering the nucleus , uracilated viral DNA products are degraded by the uracil base excision repair ( UBER ) machinery with less than 1% of the uracilated DNA successfully integrating . Although uracilated proviral DNA showed few mutations , the viral genomic RNA was highly mutated , suggesting that errors occur during transcription . Viral DNA isolated from blood monocytes and alveolar macrophages ( but not T cells ) of drug-suppressed HIV-infected individuals also contained abundant uracils . The presence of viral uracils in short-lived monocytes suggests their recent infection through contact with virus producing cells in a tissue reservoir . These findings reveal new elements of a viral defense mechanism involving host UBER that may be relevant to the establishment and persistence of HIV-1 infection .
The uracil nucleobase plays a central role in adaptive and innate immunity against HIV-1 when it is found in DNA rather than RNA ( Priet et al . , 2006; Sire et al . , 2008 ) . The most well-characterized uracil-centric innate immune response involves host cell DNA cytidine deaminase enzymes ( APOBECs ) , which selectively deaminate cytosine residues during ( - ) strand DNA synthesis thereby rendering the viral genome nonfunctional by hypermutation ( C/G to T/A ) . Work from our lab and others has suggested the presence of another uracil-mediated HIV-1 restriction pathway involving the incorporation of dUTP into viral DNA by reverse transcriptase to produce U/A base pairs ( 'uracilation' ) ( Weil et al . , 2013 ) . The dUTP-dependent pathway is thought to be restricted to non-dividing immune cells such as macrophages , monocytes and resting CD4+ T cells because only non-dividing cells have the requisite low levels of canonical dNTPs and elevated ratios of dUTP/ TTP ( Yan et al . , 2011; Kennedy et al . , 2011 ) . Notably , U/A pairs resulting from dUTP incorporation are 'invisible' to normal DNA sequencing methods and they retain the coding potential of normal T/A base pairs . Despite the similarity of U/A and T/A pairs , the presence of uracil in DNA has the potential to introduce diverse effects on viral infection including transcriptional silencing and engagement of the host uracil base excision repair ( UBER ) pathway ( Brégeon et al . , 2003; Krokan et al . , 2013a , 2014b ) . The presence , persistence , and ultimate fate of U/A pairs in HIV-1 proviral DNA is unexplored and is important for understanding the potential of tissue macrophages to establish and maintain a long-term HIV reservoir . A role for dUTP and the UBER enzyme uracil DNA glycosylase ( hUNG2 ) in HIV-1 infection have been long debated [see reference ( Weil et al . 2013 ) for a succinct review] . A dUTP-mediated host defense pathway was first suggested from inspection of the genomes of β-retroviruses , non-primate lentiviruses , and endogenous retroviruses , which have all captured a host dUTPase gene during viral evolution . Additional support was provided by the observation that dUTPase-deficient mutants of non-primate lentiviruses cannot infect non-dividing cells that contain high-dUTP levels ( Lichtenstein et al . , 1995; Turelli et al . , 1996; Threadgill et al . , 1993; Steagall et al . , 1995 ) . Intriguingly , HIV-1 does not encode a dUTPase yet can infect human macrophages with a reported dUTP:TTP ratio of 60:1 ( Kennedy et al . , 2011 ) . There is an equal level of intrigue concerning the role of hUNG2 in HIV infection . A pro-infective role for hUNG2 has been suggested by reports that hUNG2 suppresses mutations in the viral genome upon infection of macrophages ( Mansky et al . , 2000; Chen et al . , 2004; Priet et al . , 2005; Guenzel et al . , 2012 ) , but is completely dispensable for HIV-1 replication of cells with low-dUTP levels ( Kaiseer and Emerman , 2006 ) . In contrast , a modest restrictive role for hUNG2 has been suggested from the decreased infectivity of HIV virions lacking viral protein R ( Vpr ) . This restriction is attributed to a Vpr-dependent ubiquitin-mediated hUNG2 degradation pathway or through Vpr-induced transcriptional silencing of hUNG2 expression ( Schrofelbauer et al . , 2005; Ahn et al . , 2010; Langevin et al . , 2009 ) . These intriguing prior observations have motivated our further studies into the role of UBER in HIV infection , which now establish a profoundly restrictive role and unexpected effects on viral mutagenesis .
We hypothesized that viral uracilation and restriction in resting immune cells would require enzyme activities that support a high dUTP/TTP ratio and uracil base excision . Using sensitive and specific in vitro enzymatic assays ( Figure 1—figure supplement 1A–D ) ( Weil et al . , 2013; Hansen et al . , 2014; Seiple et al . , 2008 ) , we found that monocytes and monocyte-derived macrophages ( MDMs ) expressed high levels of SAMHD1 dNTP triphosphohydrolase to reduce the canonical dNTP pools ( Hansen et al . , 2014; Goldstone et al . , 2011 ) , undetectable dUTPase activity that allowed dUTP accumulation , and modest expression of the UBER enzymes uracil DNA glycosylase ( hUNG ) and abasic site endonuclease ( APE1 ) ( Figure 1—figure supplement 1E–H ) . Although resting CD4+ T cells also possessed high SAMHD1 , hUNG and APE activities , their dUTPase activity was at least seven-fold greater than MDMs . LC-MS analyses of the dUTP and canonical dNTP levels in resting and activated CD4+ T cells and MDMs revealed that the dUTP/TTP ratio was ~20 for MDMs , 1 . 1 for resting CD4+ T cells , and <0 . 05 for activated CD4+ T cells ( Figure 1—figure supplement 1I , J ) ( Gavegnano et al . , 2012; Hollenbaugh et al . , 2014 ) . Since reverse transcriptase has a nearly identical Km for dUTP and TTP ( Kennedy et al . , 2011 ) , the high ratio of dUTP/TTP indicates that uracil incorporation is a very frequent event during reverse transcription in the cell cytoplasm , which contains no UBER activity . To detect and map uracil in selected regions of the HIV genome , we developed uracil excision-droplet digital PCR ( Ex-ddPCR ) ( Figure 1A ) . Briefly , Ex-ddPCR involves isolation of total DNA from HIV-infected MDMs , after which half of the sample is treated with UNG to destroy template strands that contain uracil ( Figure 1A ) . Thus , any PCR amplicon that contains one or more uracils on each DNA strand is not amplified in the UNG-treated DNA sample . After performing ddPCR ( no UNG treatment ) and Ex-ddPCR ( UNG pre-treatment ) , the fraction of the amplicons containing at least one uracil on each strand was calculated from the counts of positive droplets for the ddPCR and Ex-ddPCR samples ( Figure 1A ) . A complete description of Ex-ddPCR is found in Materials and methods and Figure 1—figure supplement 2A–E . 10 . 7554/eLife . 18447 . 003Figure 1 . Ex-ddPCR and Ex-Seq determine the uracil content of the HIV DNA copies . ( A ) After infection of immune cells with the single-round VSV-G pseudo-typed virus HIVNL4 . 3 ( VSVG ) , total DNA is isolated and either digested by UNG to degrade uracil containing DNA or mock digested to provide a measure of total amplifiable DNA in the sample . The output ( Frac UDNA ) represents the fraction of amplified DNA copies containing at least one uracil on each DNA strand . Signal is normalized to a genomic reference copy standard ( RPP30 ) that does not contain uracil . ( B ) Activated , resting CD4+ T cells and MDMs were infected in vitro with HIVNL4 . 3 ( VSVG ) virus and the uracil content was measured 3 days post-infection ( dpi ) using primers that targeted the gag region . The data ( ± UNG digestion ) are shown as scatter plots and histograms . ( C ) Normalized coverage of the HIVNL4 . 3 ( VSVG ) -genome-positive strand in Excision-Seq ( Ex-Seq ) libraries prepared from total cellular DNA at 7 days post-HIV infection . ( D ) Fraction of the reads in panel C that contained uracil ( Frac U ) . ( E ) Discordant read pairs between HIV and human DNA present in Ex-Seq libraries prepared from total cellular DNA at 7 days post-infection with HIVNL4 . 3 ( VSVG ) virus . The number of discordant reads obtained by Ex-Seq in the absence and presence of UNG digestion are shown as white and black bars . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 00310 . 7554/eLife . 18447 . 004Figure 1—figure supplement 1 . Profiling enzyme activities and dNTP pool levels in immune target cells of HIV . Extracts from each indicated cell type were obtained as described in Methods . ( A ) Deoxyuridinetriphosphate hydrolase ( dUTPase ) activity was measured by monitoring the hydrolysis of dUTP to dUMP via PEI-cellulose TLC . Specificity was determined using a potent dUTPase inhibitor [compound 26 in Priet et al . ( 2005 ) ] . ( B ) SAMHD1 triphosphohydrolase activity was determined by C18 RP-TLC-based assay using 8-3H-labeled dGTP as the substrate . Specificity for SAMHD1 was determined using the inhibitor pppCH2dU . The mobilities of the substrate ( dGTP ) and product nucleoside ( dG ) are marked . ( C ) Endogenous uracil DNA glycosylase ( hUNG ) activity ( combined hUNG1 and hUNG2 ) was determined using a fluorescein-labeled DNA substrate that shows an increase in fluorescence upon uracil excision . Specificity for hUNG activity was determined by addition of the uracil DNA glycosylase inhibitor protein ( UGI ) . ( D ) Apyrimidinic endonuclease ( APE1 or 2 ) activity was measured using a fluorescein-labeled duplex DNA containing a single abasic site that increases in fluorescence upon endonuclease cleavage . ( E ) dUTPase activity . ( F ) SAMHD1 activity . ( G ) hUNG activity . ( H ) APE activity . ( I ) Measurement of deoxyribonucleotide ( dNTP ) pool levels were determined by an LC-MS method . ( J ) MDMs contain a high ratio of dUTP/TTP ratio compared to resting and activated CD4+ T cells . Abbreviations: rCD4+ , resting CD4+ T cell; aCD4+ , PHA activated CD4+ T cell; MDM , monocyte-derived macrophage . See Supplemental Methods for further details and references describing these assays . Number of experimental replicates ( n = 3 ) and errors are reported as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 00410 . 7554/eLife . 18447 . 005Figure 1—figure supplement 2 . In vitro generated calibration curves for evaluating uracil content in DNA amplicons and single-round HIV infections of cultured MDMs . ( A ) ddPCR primer and probe locations relative HIV genome ( HIVNL4 . 3 ( VSVG ) ) . ( B ) Generation of uracil-containing duplex DNA of increasing length and ratio of dUTP/TTP was achieved by in vitro DNA polymerization using Taq polymerase in the presence of various ratios of dUTP/TTP . These DNA standards ( 79 to 471 bp ) were analyzed by the Ex-ddPCR method as indicated . ( C ) The fraction of the DNA amplicons that contained detectable uracil ( Frac U ) increased with DNA amplicon length as well as the dUTP/TTP ratio used in the initial DNA synthesis . ( D ) In single-round infections of a mixed population of MDMs with HIVNL4 . 3 ( VSVG ) , the copy number of early , middle and late viral cDNAs were measured over a 30-day culture period using primers specific to different regions of the viral genome ( see above ) . The copy number is normalized to one million MDM target cells . ( E ) Uracil content was measured in each HIV DNA population using Ex-ALU-gag nested qPCR . Number of experimental replicates ( n = 1–3 ) and errors are reported as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 005 Ex-ddPCR analysis of HIV DNA isolated from single-round infections of activated and resting CD4+ T cells and MDMs using a VSV-G pseudo-typed replication-deficient HIV strain ( HIVNL4 . 3 ( VSVG ) ) established that uracilation of viral DNA occurs in MDMs , but not T cells ( Figure 1B ) . The scatter plots and histograms in Figure 1B show that the copy number for viral gag DNA isolated from MDMs at 3 days post-infection was 3 . 5-fold lower in the Ex-ddPCR experiment , showing that ~70% of gag amplicons contained uracil . Importantly , the genomic reference standard RNase P ( RPP30 ) contained no measurable uracil , indicating that uracil incorporation is specific to HIV DNA and occurs during reverse transcription . In contrast , the gag copy number for viral DNA collected from activated and resting T cells was the same for both the ddPCR and Ex-ddPCR reactions indicating that uracil was absent in viral DNA isolated from infected T cells ( Figure 1B ) . To augment Ex-ddPCR , we also applied the next-generation sequencing technology uracil Ex-Seq to globally map the frequency of U/A pairs across the entire HIV genome ( Bryan et al . , 2014 ) . Ex-Seq is similar to standard IIlumina sequencing ( Seq ) , except that UNG-mediated uracil excision is used to destroy uracil-containing templates prior to PCR amplification . To specifically enrich HIV-1 sequences , we used 5’-biotin-conjugated DNA probes that tiled both strands of the entire viral genome , yielding a 103-fold increase in HIV-derived fragments . Sequencing of viral DNA isolated from MDMs 7 days after infection with HIVNL4 . 3 ( VSVG ) showed uniform coverage across the genome except for notably increased reads at the 5´ and 3´-LTR regions , which was equally evident for both the Seq and Ex-Seq samples ( Figure 1C ) . We speculate that the elevated signal in the LTRs arises from non-uniform hybridization of the lock-down probes , which is normalized when converting the reads to Frac U . The ratio of the normalized sequencing reads ( Ex-Seq/Seq ) thus quantifies the fraction that contained at least one uracil on each strand ( Figure 1D ) . Thus , about 60% of the 100 bp reads contained uracil , leading us to conclude that uracilation was uniform across the viral genome . The above analyses report on both integrated and unintegrated viral DNA . However , the Ex-Seq experiment also provides specific information on the uracilation status of integrated proviruses and their genomic sites of integration ( Figure 1E ) . Sequence reads from integrated proviruses are unambiguously assigned by the presence of viral LTR and human genomic sequences ( 'discordant pairs' ) . Over 4000 discordant pairs were identified per million HIV reads for the Seq sample , which was reduced by about 60% for the Ex-Seq sample ( Figure 1E ) . This level of uracilation is similar to that of the total HIV DNA and supports our previous finding that uracils can persist in proviruses when nuclear hUNG2 expression is low ( Weil et al . , 2013 ) . For T cell targets of HIV-1 , expression of APOBEC3G ( A3G ) and APOBEC3F ( A3F ) has been shown to result in cytosine deamination at preferred sequence motifs present in the HIV minus-strand DNA , giving rise to mutagenic uracils ( C→T transition mutations ) ( Harris et al , 2003a , 2003b ) . We excluded APOBEC-catalyzed cytidine deamination as a major source of viral uracils in infected MDMs using several criteria . First , Ex-Seq data showed that the uracils appeared on both strands of the viral cDNA rather than the ( - ) strand that is exclusively targeted by APOBECs ( Yu et al . , 2004 ) . Second , the mutational spectrum of proviral DNA did not match the signature C/G→T/A hotspots for APOBEC cytosine deamination as determined using the program Hypermut 2 . 0 ( see sequencing studies below ) ( Rose and Korber , 2000 ) . Finally , we obtained direct evidence that viral uracils originated from the high dUTP/TTP ratio by increasing the intracellular TTP levels through the addition of 5 mM thymidine to the culture media . Addition of thymidine to the culture medium prior to infection led to a 14-fold decrease in the uracilation level of proviral DNA ( Figure 2A ) and a five-fold increase in proviral copies as compared to no thymidine supplementation ( Figure 2B ) . The combined results strongly support a mechanism where the viral uracils originate primarily from dUTP incorporation during reverse transcription . Since reverse transcriptase utilizes dUTP with the same efficiency and fidelity as TTP ( Kennedy et al . , 2011 ) , the uracils must be in the form of coding U/A base pairs . In Figure 2—figure supplement 1 , we present immunoblots showing that MDMs express very low levels of A3G and A3A , which is consistent with previous reports and the absence of tell-tale marks of enzymatic deamination described above ( Harris and Liddament , 2004 ) . 10 . 7554/eLife . 18447 . 006Figure 2 . Uracils arise from high dUTP/TTP not APOBEC-catalyzed cytosine deamination . ( A ) Culturing a mixed GFP+/GFP− MDM cell population with media supplemented with 5 mM thymidine resulted in a 14-fold decrease in the fraction of proviral copies that contained uracil . Uracil content was measured using Ex-ALU-gag nested qPCR with their statistical significance level ( *p<0 . 05 ) . ( B ) Mixed MDM cultures supplemented with 5 mM thymidine showed a five-fold increase in provirus copy number as measured using ALU-gag nested qPCR . Number of experimental replicates ( n = 4–5 ) and errors are reported as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 00610 . 7554/eLife . 18447 . 007Figure 2—source data 1 . Ex-ALU-gag qPCR for provirus and uracil detection in MDMs cultured with standard media or media supplemented with 5 mM Thymidine ( Figure 2A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 00710 . 7554/eLife . 18447 . 008Figure 2—figure supplement 1 . APOBEC3G ( A3G ) and APOBEC3A ( A3A ) are poorly expressed in a mixed population of MDMs . Western blots established that the expression levels of APOBEC3A ( A3A ) and A3G in a mixed population of MDMs over a 30-day culture period after infection with HIVSF162 ( CCR5 ) were consistently low or undetectable . The detection of A3G and A3A in PBMCs and their absence in HEK 293T ( 293 ) cells indicates the antibody is specific for each protein . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 008 During HIV infection of MDMs ~10% of the infected MDMs were positive for GFP fluorescence after infection with either replication deficient HIVNL4 . 3 ( VSVG ) or replication competent HIVSF162 ( CCR5 ) ( Figure 3A ) . This level of infection based on GFP fluorescence remained constant from about seven to 30 days after infection and did not change when the MOI was varied in the range 0 . 1–10 ( Figure 3—figure supplement 1 ) . This result led us to suspect that the MDM population might consist of a phenotypic mixture with different susceptibilities to HIV infection . ( We note that higher MOIs are typically used in our experiments to obtain reasonable viral copy numbers in the GFP− population , but the results are independent of MOI . ) Accordingly , we used GFP fluorescence to sort HIVNL4 . 3 ( VSVG ) infected MDMs into 99% pure GFP positive and negative populations ( Figure 3—figure supplement 2 ) . Droplet digital PCR measurements showed that the copy number of early viral gag-containing DNA products in the GFP− population was about seven-fold lower than the GFP+ population at 1 day post-infection ( ~2 versus 14 copies/cell , Figure 3B ) . This result suggested that viral infection was hindered at the entry and/or early reverse transcription steps in the GFP− population . Using ALU-gag nested qPCR to measure proviral copies ( O'Doherty et al . , 2002 ) , we found that the GFP- population contained only 1% of the copies seen in the GFP+ MDMs ( Figure 3C ) . The large decrease in copy number between the early DNA intermediates and the provirus stage suggested that a potent pre-integration restriction mechanism was also present in the GFP− population . 10 . 7554/eLife . 18447 . 009Figure 3 . MDMs consist of two distinct cell populations with respect to viral infection . ( A ) Flow cytometry analysis of MDMs infected with replication-deficient HIVNL4 . 3 ( VSVG ) and replication comeptent HIVSF162 ( CCR5 ) viruses . Expression of virally encoded GFP was only observed in 12–15% cells over a 30-day time-period . ( B ) After cell sorting according to GFP fluorescence , ddPCR was used to measure the copy number of HIVNL4 . 3 reverse transcription intermediates ( RTIs ) in GFP− ( white bars ) and GFP+ ( green bars ) MDMs . Level of statistical significance ( *p<0 . 05 ) . ( C ) ALU-gag nested qPCR measurement of the HIVNL4 . 3 ( VSVG ) provirus copy number in sorted GFP− and GFP+ MDMs . ( D ) Measurement of HIVSF162 ( CCR5 ) provirus copy number and uracil content over the course of a 30-day multi-round infection . ( E ) ELISA measurements of viral p24 protein levels in GFP− and GFP+ MDMs over the course of a 30-day multi-round infection . Number of experimental replicates ( n = 3–5 ) and errors are reported as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 00910 . 7554/eLife . 18447 . 010Figure 3—source data 1 . Ex-ALU-gag qPCR for measurement of reverse transcription intermediates ( RTIs ) and provirus content in GFP sorted MDMs ( Figure 3B , C ) . Provirus and uracil content in GFP sorted MDMs ( Figure 3D ) and virus output , p24 ( Figure 3E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 01010 . 7554/eLife . 18447 . 011Figure 3—figure supplement 1 . Uracilation is independent of multiplicity of infection ( MOI ) . MDMs were infected with HIVNL4 . 3 ( VSVG ) using MOIs of 0 . 1 , 1 and 10 . ( A ) The viral gag copy number increases with MOI . ( B ) The MOI does not significantly affect the fraction of viral gag copies that contained uracil . Number of experimental repetitions ( n = 3 ) and errors are reported as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 01110 . 7554/eLife . 18447 . 012Figure 3—figure supplement 2 . The sorted populations of in vitro infected GFP− and GFP+ MDMs are highly pure . MDMs were infected with HIVNL4 . 3 ( VSVG ) by spinoculation and then sorted using GFP fluorescence after 7 days . The purity of each sorted population ( GFP− and GFP+ ) was assessed by flow cytometry . Each population was greater than 99% pure . ( A ) Flow cytometry analysis of purified GFP− population . ( B ) Flow cytometry analysis of purified GFP+ population . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 01210 . 7554/eLife . 18447 . 013Figure 3—figure supplement 3 . Three different viral strains show a similar uracilation profile with in vitro infected MDMs independent of the differentiation regimen . ( A ) The kinetics and absolute levels of virus output as measured by p24 ELISA is indistinguishable for MDMs infected with HIVSF162 ( CCR5 ) and HIVBAL ( CCR5 ) . ( B ) Provirus copy number and uracil content at dpi = 3 are very similar in MDMs infected with HIVSF162 ( CCR5 ) , HIVNL4 . 3 ( VSVG ) or HIVBAL ( CCR5 ) ( MOI = 0 . 1 ) . Proviral copy numbers were measured using ALU-gag nested qPCR and the uracil content by Ex ALU-gag nested qPCR . ( C ) Infection of MDMs obtained from peripheral blood monocytes of three donors produces equivalent levels of uracilated HIVNL4 . 3 ( VSVG ) DNA as judged by ddPCR analysis of the GFP copies . ( D ) The fraction of viral GFP copies that contain uracil is similar for all three donors as determined using Ex-ddPCR ( ns; not significant ) . ( E ) Purified monocytes isolated from bulk PBMCs were differentiated to MDMs in vitro using the adherence method , or by addition of M-CSF or GM-CSF . The differentiation regimen did not significantly affect viral GFP expression in MDMs infected with HIVNL4 . 3 ( VSVG ) , or ( F ) proviral DNA copy number as determined by ALU-gag nested qPCR , or ( G ) the fraction of viral DNA copies that contained uracil as determined by Ex ALU-gag nested qPCR . Number of experimental replicates ( n = 2–3 ) and errors are reported as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 01310 . 7554/eLife . 18447 . 014Figure 3—figure supplement 4 . GFP expression from HIVNL4 . 3 infected MDM populations does not strongly correlate with CD14 and CD16 expression . MDMs ( dpi = 3 ) were initially distinguished by their characteristic FSC and SSC and then assessed for their expression levels of CD14 and CD16 markers . ( A ) MDMs were sorted according to GFP expression using FACS and then double stained for CD14 and CD16 expression before analysis by flow cytometry . ( B ) The relative CD14 and CD16 marker expression levels in mixed and GFP-sorted MDM populations were quantified using flow cytometry . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 014 We used infections with a replication-competent and macrophage-tropic virus ( HIVSF162 ( CCR5 ) ) to further elucidate the characteristics of the two MDM populations . First , the infected MDMs were sorted into GFP positive and negative populations at 1 day post-infection . The cells were then cultured at days 1 , 7 , 14 , and 30 post-sorting where we measured proviral DNA copy number , the fraction of proviruses containing uracil , and the HIV p24 levels in the culture supernatants for each cell population ( Figure 3D , E ) . Both MDM populations showed high levels of viral gag DNA products at day 1 ( red circle and blue square , Figure 3D ) . The GFP+ population was characterized by efficient integration and a high proviral DNA copy number that remained stable for 14 days . By day 30 , the copy number increased by ~four-fold , which we attribute to secondary infection by released viruses ( white bars , Figure 3D ) . In contrast , the GFP− population had a low efficiency of integration exemplified by a 50-fold reduction in copy number between the early gag DNA products and proviruses ( black bars , Figure 3D ) . Unlike the GFP+ cells , the proviral copy number in the GFP− population decreased by 20-fold between days 7 and 14 before returning to the day 7 level at 30 days . Once again , we attributed the later increase to secondary infection by released virus particles because it was not observed in single-round infections . We established that HIVNL4 . 3 ( VSVG ) , HIVSF162 ( CCR5 ) , and HIVBAL ( CCR5 ) viral strains and MDMs isolated from different donors behaved similarly ( Figure 3—figure supplement 3A–E ) , and that the results were independent of whether differentiated MDMs were obtained using adherence , M-CSF or GM-CSF protocols ( Figure 3—figure supplement 3F–H ) . The most striking difference between the two MDM populations was that the minor GFP+ population had no detectable proviral uracils , while ~90% of the proviral copies in the major GFP− fraction contained uracil at day one ( Figure 3D ) . The fraction of proviral copies that were uracilated ( Frac U ) decreased from 0 . 77 to 0 . 23 between days 7 and 14 before rising to 0 . 55 at day 30 ( presumably due to continuing infection ) . We explored the mechanistic basis for the differences in viral uracilation between the two MDM populations by measuring the dUTP/TTP ratio in the cell extracts obtained from each . The GFP+ MDMs had a 20-fold lower dUTP/TTP ratio , which stemmed in part from a five-fold higher TTP concentration compared to the GFP− MDMs . This difference between the GFP+ and negative MDMs confirms the key mechanistic requirement for viral uracilation: a high dUTP/TTP ratio . Consistent with their larger proviral copy numbers , the GFP+ MDMs produced 100 to 1000 times more p24 than the GFP− MDMs and the levels were fairly constant over a 30 day period ( Figure 3E ) . In contrast , the GFP− MDMs showed a 10-fold decrease in p24 levels between day seven and 14 ( Figure 3E ) , which coincides with the 20-fold reduction in proviral copy number in the same time period ( Figure 3D ) . The expression of detectable levels of p24 , but not GFP , in the GFP− MDMs suggests that truncated viral RNAs are being expressed that extend through the p24 coding sequence , but not GFP . We were quite surprised that the uracilation phenotype was restricted solely to MDMs that were incapable of expressing virally encoded GFP ( Figure 3A ) and were curious whether the minor , but highly infectable GFP+ MDM population might be derived from CD16+ monocyte precursors . Our reasoning is based on the reports that CD16+ monocytes are enriched in the CCR5 co-receptor and are the preferred target of HIV as compared to the more prevalent CD14+ monocytes which express lower levels of CCR5 and appreciable levels of the active low-molecular-weight forms of APOBEC3A and APOBEC3G ( Ellery et al . , 2007 ) . In addition , CD16+ monocytes typically comprise around 10% of the monocyte pool in uninfected donors ( Ellery et al . , 2007 ) , which curiously matches the level of GFP+ MDMs in our studies . Despite these suggestive relationships , multicolor flow cytometry established that the GFP fluorescence did not track with the CD14+ or CD16+ status of the MDMs ( Figure 3—figure supplement 4 ) . Aside from the 20-fold higher dUTP levels in the GFP− MDMs ( see above ) , we have no comprehensive understanding of what additional phenotypic differences exist between the two macrophage populations that result in the combined GFP−/uracilation phenotype . It is possible that the dominant GFP− phenotype arises from a complex combination of effects related to the metabolic state , or activation state of these cells . We previously established that hUNG2 activity was the primary determinant of whether uracilated proviral DNA survived or persisted in the HT29 model cell system ( Weil et al . , 2013 ) . To confirm this result in MDMs , we took advantage of three approaches: ( i ) knockdown of hUNG2 activity by transfection with a plasmid vector that overexpressed the potent uracil DNA glycosylase inhibitor protein ( Ugi ) ( Bennett et al . , 1993 ) , ( ii ) overexpression of hUNG2 by transfection with a plasmid expression vector , and ( iii ) infections with viral constructs where virus protein R ( Vpr ) was deleted . Viral protein R is known to interact with hUNG2 and induce its degradation through the assembly with the DDB1-CUL4 ubiquitin ligase complex and would be expected to recapitulate the phenotype produced by Ugi overexpression ( Eldin et al . , 2014 ) . We surmised that reduced hUNG2 levels would lead to greater infection efficiency , higher levels of uracil in viral DNA products , and that greater expression of hUNG2 would correlate with increased restriction and reduced uracil content in proviral DNA due to uracil excision by hUNG2 . It is important to note that plasmid DNA transfection methodology is essential in these experiments because the introduction of Ugi and hUNG2 through lentiviral packaging and transfection would be compromised by uracilation during reverse transcription . Primary macrophages derived from blood monocytes were cultured in the presence of M-CSF for 7 days prior to transfection with the pIRES-Ugi-eGFP and pIRES-hUNG2-eGFP expression plasmids ( Figure 4A ) . Because MDMs are exceedingly difficult to transfect , the cells expressing eGFP were sorted by FACS 48 hr post-transfection and the GFP+ cells ( indicating successful plasmid transfection; not expression of HIV-encoded GFP ) were allowed to adhere for 24 hr and then spin-infected with HIVSF162 ( CCR5 ) virus . Cells were harvested 3 days post-infection and total DNA was extracted for qPCR analyses . 10 . 7554/eLife . 18447 . 015Figure 4 . hUNG2 uracil excision activity are required for efficient pre-integration restriction . Effect of transient overexpression of nuclear hUNG2 and Ugi on copies of viral early reverse transcription intermediates ( RTIs ) and proviruses . ( A ) Experimental scheme for hUNG2 and Ugi overexpression in MDMs . ( B ) Copies were determined by qPCR using specific primers for early RTIs and provirus . ( C ) Fraction of viral RTIs and provirus that contained detectable uracil under conditions of hUNG2 depletion ( Ugi ) and overexpression ( hUNG2 ) . ND; Not detectable . ( D ) Western blot analysis of the vpr deletion virus HIVNL4 . 3 ( Δvpr ) revealed no detectable Vpr . Western blot analysis to detect nuclear ( hUNG2 ) and mitochondrial ( hUNG1 ) hUNG isoforms detected oscillation in the hUNG2 levels over the course of a 7-day multi-round infection . At day 7 , the HIVNL4 . 3 ( Δvpr ) shows abundant hUNG levels while hUNG2 is depleted in the wild-type infection . ( E ) Measurements of provirus copy number in a mixed population of GFP− and GFP+ MDMs infected with HIVNL4 . 3 ( VSVG ) ( wt ) and HIVNL4 . 3 ( Δvpr ) . ( F ) Ex-ALU-gag nested qPCR measurements of the fraction of uracilated proviral copies in a mixed population of GFP− and GFP+ MDMs infected with HIVNL4 . 3 ( VSVG ) ( wt ) and HIVNL4 . 3 ( Δvpr ) at 3 days post-infection . Level of statistical significance ( *p<0 . 05 ) . Number of experimental replicates ( n = 4–5 ) and errors are reported as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 01510 . 7554/eLife . 18447 . 016Figure 4—source data 1 . Ex-ALU-gag qPCR for measurement of reverse transcription intermediates ( RTIs ) , provirus and uracil content in MDMs transfected with plasmids encoding hUNG2 , Ugi or empty control ( Figure 4B , C ) . Provirus and uracil content in MDMs infected with wild-type ( wt ) or mutant vpr ( Δvpr ) virus ( Figure 3E , F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 016 For the cells that were transfected with the Ugi expression plasmid , the copy number of early reverse transcription ( RTI ) products and proviral DNA was increased about five-fold compared to transfection with the empty control plasmid ( Figure 4B ) . The Ugi-expressing cells had dramatically increased uracil content , consistent with efficient knockdown of hUNG2 activity and a prominent role for hUNG2 in uracil excision ( Figure 4C ) . In contrast , the cells transfected with the hUNG2 expression plasmid showed a 10-fold reduction in RTIs and proviral DNA relative to control ( Figure 4B ) , and we were unable to detect any uracil in viral DNA using Ex-qPCR ( Figure 4C ) . These results closely match our previous findings where high expression levels of hUNG2 served to effectively destroy uracilated viral DNA ( Weil et al . , 2013 ) . In a complementary approach to establish a role for hUNG2 in uracil excision , we infected the mixed population of MDMs with HIVNL4 . 3 ( VSVG ) , HIVSF162 ( CCR5 ) or HIVNL4 . 3 ( Δvpr ) viruses ( Figure 4D–F ) . Infections with HIVSF162 ( CCR5 ) and HIVNL4 . 3 ( Δvpr ) showed that the hUNG2 nuclear isoform ( but not the mitochondrial hUNG1 isoform ) selectively disappeared between 1 and 7 days post-infection with HIVSF162 ( vpr+ ) , but not with HIVNL4 . 3 ( Δvpr ) ( Figure 4D ) ( Eldin et al . , 2014 ) . As expected , infection with HIVNL4 . 3 ( Δvpr ) reduced the overall proviral copies present in the mixed population of GFP+ and GFP− MDMs ( Figure 4E ) , which is a non-specific effect that can be attributed to any of the known pro-infective functions of vpr ( Eldin et al . , 2014 ) . In the absence of vpr , two-fold fewer proviruses contained uracil , which is a uracilation-specific effect that can be attributed specifically to vpr-induced depletion of hUNG2 activity in GFP− MDMs ( Figure 4F ) ( Grogan et al . , 2011 ) . The larger effects observed above with Ugi and hUNG2 overexpression are expected because the hUNG2 level is manipulated before viral infection . We investigated whether the presence of uracil in proviral DNA affected the efficiency of producing viral RNA genome copies . In these experiments , we use reverse transcriptase quantitative PCR ( RT-qPCR ) to measure the copy number of extracellular viral genomic RNAs ( EVRs ) present in supernatants from infected GFP+ and negative MDMs that were cultured in standard growth media or media supplemented with IFNγ or IL-4 ( Figure 5A ) . These two cytokines were of interest because they stimulate macrophage differentiation into classical pro-inflammatory ( M1 ) and anti-inflammatory ( M2 ) polarization states , respectively , and could modulate the outcome of infection ( Xue et al . , 2014 ) . The EVR copies present in the supernatant of GFP− MDMs at three days post-cytokine stimulation was about 100-fold lower than the GFP+ population , independent of cytokine stimulation . The lower copies of genomic RNA produced from the uracilated proviruses in GFP− MDMs is comparable to the reduced proviral copy number in this population . This result indicates that uracilation does not significantly alter the output of viral RNA genomes when normalized for proviral DNA levels . We also found that inflammatory cytokine IFNγ stimulation reduced the EVR copy number in both MDM populations by almost 10-fold ( Figure 5B ) , which is therefore not correlated with the uracilation phenotype . 10 . 7554/eLife . 18447 . 017Figure 5 . Impact of uracilation and cytokines on viral DNA and RNA sequences . ( A ) Experimental protocol . ( B ) Quantitative reverse transcriptase-PCR ( qRT-PCR ) was used to determine the copy numbers of extracellular viral RNA in cell supernatants from sorted GFP+ ( green bars ) and GFP− ( white bars ) MDMs with and without cytokine stimulation . Number of experimental replicates ( n = 4 ) and errors are reported as mean ± SD . ( C ) Summary of mutations from limiting–dilution clonal sequencing of extracellular viral RNAs obtained from infected GFP− MDMs under different growth conditions . The mutation frequencies ( point mutations per total nucleotides sequenced ) were obtained from a ~500 bp amplicon of the env gene . ND; not detected . ( D ) Representative env sequences of extracellular viral RNAs produced by GFP− sorted MDMs . The top sequence is for the HIVSF162 ( CCR5 ) strain used to infect the MDMs ( HIVSF162_env_ref ) . Boxed regions at the protein and nucleotide sequence level show the mutation spectrum within the CD4-associated and co-receptor binding sites . ( E ) Breakdown of point mutations for viral RNA produced from infected MDMs and proviral DNA from infected T cells ( see text ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 01710 . 7554/eLife . 18447 . 018Figure 5—source data 1 . qPCR measurement of gag copies ( Figure 6B ) , provirus copies ( Figure 6C ) and uracil content ( Figure 6D ) in GFP sorted and cytokine-treated MDMs . Measurement of virus output , p24 ( Figure 6E ) . Provirus content was evaluated by Ex-ALU-gag qPCR in MDM ( Figure 6F ) producer cells and CEMx174 ( Figure 6G ) target cells . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 018 Limiting dilution clonal sequencing was performed on EVRs produced from sorted GFP− and GFP+ MDMs at 3 days post-stimulation ( Laird et al . , 2013; Ho et al . , 2013 ) , and revealed surprising differences in viral single-point mutation frequencies for these populations ( Figure 5C ) . Regardless of cytokine stimulation , the GFP+ population showed no detectable mutations in the roughly 12 , 000 total bases of the V3/V4 variable region of env that were sequenced . In contrast , the GFP− population showed a 1 . 2% basal mutation frequency , which responded differently to IFNγ and IL-4 stimulation: IFNγ stimulation increased the frequency by two-fold and IL-4 decreased it significantly ( Figure 5C ) . These results suggest that the cytokine environment of an in vivo infected macrophage could influence the evolution of uracilated proviruses . In contrast to the mutated env sequences , the LTR region of EVRs produced from the GFP− cells was devoid of detectable mutations after sequencing 7392 total bases ( Supplementary file 1 ) , which may reflect selection for expression of transcription competent LTR sequences . The absence of mutations in the env and LTR regions of EVRs isolated from the GFP+ MDM cultures , as well as the LTR region of EVRs obtained from the GFP− culture supernatants , virtually eliminates the possibility that the observed mutations in env arise from PCR or sequencing errors . Targeted sequencing of the env amplicon in uracilated viral DNA isolated from GFP− MDMs at day 14 post-infection in the absence of cytokine stimulation revealed a <0 . 007% mutation frequency in the V3/V4 variable region , which is significantly lower than the 1 . 2% frequency in EVRs produced from the same cell culture ( most viral DNA is integrated at day fourteen ) . This result suggests that the EVR mutations arise from transcriptional errors by RNA polymerase II ( RNAP II ) . Such mutations may derive from encounter of the polymerase with either uracils or repair intermediates resulting from uracil excision ( see Discussion ) . Illumina next-generation sequencing data covering the entire viral DNA genome ( obtained from total HIVNL4 . 3 viral DNA isolated from a mixed population of MDMs at day 7 post-infection in the absence of cytokines ) confirmed the low mutation frequency determined from targeted sequencing of the env V3/V4 variable region ( see Data Access information ) . The frequencies of single-base transition , transversion , and insertion/deletion mutations within the env region of EVRs produced from unstimulated and IFNγ or IL-4- stimulated GFP− cells were calculated from the sequencing data ( Figure 5C ) . The RNA ( + ) strand mutations were primarily single base changes mostly comprised of A→G , A→C , and T→C transitions and transversions ( Figure 5D , E ) , but also a few short 2–3 nucleotide insertions or deletions . A detailed analysis of the sequence dependence of the observed G→A ( + ) strand mutations using Hypermut 2 . 0 ( Rose and Korber , 2000 ) indicated that the sequences and frequencies were not consistent with A3G , A3F or A3A deaminase activity on the viral ( - ) strand cDNA ( Yu et al . , 2004 ) . To explore whether the mutation signature of uracilated viral DNA products and viral genomic RNA sequences derived from infected MDMs were distinct from those obtained from T cells , we infected activated CD4+ T cells with HIVBAL ( CCR5 ) virions produced from donor cells expressing endogenous levels of A3G ( Materials and methods ) . The infection was kept to a single round by introducing the entry inhibitor enfuvirtide after the initial infection ( Hildinger et al . , 2001 ) , and the V3/V4 env region of 14 proviruses were sequenced . The proviral sequences derived from infected T cells had a much higher mutation frequency ( ~1% ) than uracilated viral DNA in MDMs ( <0 . 007% ) ( Supplementary file 2 ) . In addition , the mutation spectrum of proviral DNA isolated from T cells differed from the EVRs produced from MDMs ( Figure 5E ) . The prominent proviral mutations found in T cells were elevated C→T transitions and T→G transversions , while the viral RNA sequences derived from MDMs were distinguished by elevated A→C transversions and T→C transitions ( Figure 5E ) . We note that previous studies have established that host RNAP II contributes little to HIV sequence evolution in T cell infections and that errors during reverse transcription are the predominant diversification mechanism ( Menendez-Arias , 2009 ) . In contrast , for in vitro infection of MDMs a majority of the mutations occur during transcription . To measure the levels of functional progeny viruses that emerged from infected MDMs in the absence and presence of cytokine stimulation , we followed the approach outlined in Figure 6A . MDMs infected with HIVSF162 ( CCR5 ) were first sorted based on GFP fluorescence at one day post-infection . The sorted GFP positive and negative cells were divided into three cultures that were stimulated with IFNγ , IL-4 , or no treatment . After 3 days , the MDM producer cultures were analyzed with respect to ( i ) gag DNA copy number as a surrogate for total viral DNA present ( Figure 6B ) , ( ii ) integrated proviral DNA copy number ( Figure 6C ) , ( iii ) the fraction of proviruses containing uracil ( Figure 6D ) , and ( iv ) p24 levels in culture supernatants ( Figure 6E ) . The virus-containing MDM culture supernatants were removed at three days post-stimulation ( dps ) , normalized to p24 and added to cultures of CEMx174 target cells . After 3 days , the proviral copy number per 106 target cells was measured ( Figure 6F ) . 10 . 7554/eLife . 18447 . 019Figure 6 . Effects of cytokine stimulation on viral transmission in GFP sorted MDM populations . ( A ) Experimental approach . ( B ) MDMs infected with HIVSF162 ( CCR5 ) were sorted into GFP+ ( green bars ) and GFP− ( white bars ) populations and analyzed with respect to gag copy number ( dps; days post-stimulation ) . ( C ) Provirus copy numbers for each MDM population were measured using ALU-gag nested qPCR at 3 dps . ( D ) The fraction of proviral DNA copies containing uracil was measured using Ex-ALU-gag nested qPCR . ND; Not Detected . ( E ) Viral growth kinetics of sorted MDM populations . ( F ) Levels of virus in culture supernatants of GFP+ and GFP− MDM producer cells were measured using p24 ELISA . ( G ) Viral supernatants from each MDM treatment were normalized to p24 levels and used to infect naïve CEMx174 target cells . Provirus copies were measured using ALU-gag qPCR . The results are normalized to one million target cells . In all cases , the number of experimental replicates ( n = 4–5 ) and errors are reported as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 01910 . 7554/eLife . 18447 . 020Figure 6—source data 1 . qPCR measurement of gag copies ( Figure 6B ) , provirus copies ( Figure 6C ) and uracil content ( Figure 6D ) in GFP sorted and cytokine treated MDMs . Viral growth kinetics ( Figure 6E ) and total virus ( Figure 6F ) content in culture supernatants as monitored by p24 ELISA . Ex-ALU-gag qPCR measurement of provirus content in CEMx174 target cells . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 020 Independent of stimulation , the GFP− MDM producer cells showed lower copy numbers of gag and proviral DNA ( ~20 and 104-fold ) and p24 levels ( 100-fold ) , as compared to GFP+ MDMs ( Cassetta et al . , 2013 ) . IFNγ or IL-4 stimulation had little effect on gag and proviral DNA copy numbers in both MDM producer populations ( Figure 6B , C ) , but IL-4 induced a modest decrease in the fraction of proviruses that contained uracil at 3 days post-stimulation ( Figure 6D ) . For both MDM types , the addition of either cytokine resulted in a reduction in p24 production , with IFNγ showing the largest effect relative to no stimulation ( ~10-fold decrease ) ( Figure 6E , F ) . Despite the very restrictive environment of GFP− MDM producer cells prior to integration ( Figure 3D ) , the harvested viruses derived from uracilated proviruses were highly competent for infection of CEMx174 target cells ( Figure 5G ) . To test whether HIV-1 DNA isolated from infected individuals contained detectable levels of uracil , we purified resting CD4+ T cells and monocytes from bulk peripheral blood mononuclear cells ( PBMCs ) of six participants suppressed on antiretroviral therapy ( ART ) . Highly purified populations of resting CD4+ T ( >95% ) cells and monocytes ( >98% ) were obtained and quantified by flow cytometry ( Figure 7—figure supplement 1A–C ) . Both ddPCR and Ex-ddPCR were used to quantify HIV DNA and measure the fraction of the viral pol amplicons that contained uracil ( Figure 7A , B ) Strain , 2013 . HIV DNA levels in resting CD4+ T cells were generally higher and more variable than those measured in monocytes ( geometric mean values of 526 copies/106 T cells and 47 copies/106 monocytes ) ( Figure 7A ) ( Wang et al . , 2013; Eriksson et al . , 2013 ) . Consistent with in vitro measurements ( Figure 1B ) , we found no detectable uracil ( Frac U ) in HIV DNA isolated from resting CD4+ T cells , but five of the six donor monocyte samples contained detectable uracils . The fraction of uracil-positive pol amplicons was in the range 20 to 80% for the five patient samples ( Figure 7B ) . The observation that viral DNA uracilation is specific to monocytes and macrophages provides an unambiguous marker of its origins and excludes contaminating T cell DNA as a possible source . 10 . 7554/eLife . 18447 . 021Figure 7 . Peripheral blood monocytes and alveolar macrophages contain high levels of uracil in HIV DNA . ( A ) Resting CD4+ ( rCD4+ ) T cells and monocytes were purified by negative bead selection from bulk PBMCs obtained from six ART-suppressed individuals . Total HIVpol DNA was quantified using Ex-ddPCR . Copy numbers were normalized to the human RPP30 gene to give an estimate of HIV copies/106 cells . Limit of detection ( LOD ) = 5 copies/106 cells . ( B ) The uracilated fraction of HIV pol DNA copies derived from monocytes and rCD4+ T cells was measured using Ex-ddPCR . Total genomic DNA was also isolated from matched , cryopreserved PBMCs and bronchial alveolar macrophages ( AM ) obtained from a single donor both pre-ART ( green circle ) and post-ART ( yellow circle ) , ( ***p<0 . 001 ) . Donor 4 had no detectable uracil ( pol ) and was excluded from this plot . ( C ) Infection of MDMs by HIV-1 and possible fates of uracilated viral DNA products ( see text ) . Level of statistical significance ( ***p<0 . 001 ) . Each experimental replicate is shown . Number of experimental replicates ( n = 1–3 ) and errors are reported as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 02110 . 7554/eLife . 18447 . 022Figure 7—figure supplement 1 . Purity and detection of HIV DNA in isolated cell populations from ART suppressed individuals . Resting CD4+ ( rCD4+ ) T cells and monocytes were isolated from six ART suppressed individuals by negative selection . The results for donor 1 are shown in the figure panels . ( A ) Monocytes were stained for the specific monocyte markers CD14 and CD16 . ( B ) The purity of the monocyte preparations was determined by negative staining for the T cell marker CD3 . ( C ) To determine if the purified resting CD4+ T cells contained contaminating activated CD4+ T cells , we stained for CD4 and the activation markers CD25/CD69/HLA-DR . ( D ) The purity of the purified monocyte and rCD4 cells from each patient is indicated . The HIVpolcopies were determined by ddPCR and are expressed as an average and standard deviation ( SD ) . When possible , two or three independent ddPCR measurements were taken . ( E ) NL4 . 3 plasmid was serially diluted over 100 , 000-fold to evaluate the detection limit of the ddPCR assay for HIV pol . ( F ) Dilution series of human genomic DNA with measurement of the RPP30 copy number by ddPCR . This series was used to determine the load limit for total DNA that can be accurately measured in a ddPCR reaction . ( G ) Measurement by ddPCR of HIV pol copy number in a background of genomic DNA obtained from 250 , 000 uninfected PBMCs . ( H ) Uracils are not detected in genomic DNA ( RPP30 gene ) of uninfected monocytes or T cells . Using RPP30 primers , Ex-ddPCR was performed on genomic DNA isolated from monocytes of healthy donors ( open circles ) or ART patients ( gray circle ) as well as rCD4+ T cells from ART patients . ( I ) Ex-ddPCR was used to determine that uracil is only found in HIV DNA isolated from monocytes and not rCD4+ T cells of HIV-infected patients . The shown data are for donor ID 2 ( Supplementary file 4 ) . For visualization , the histograms for the RPP30 Ex-ddPCR data have been down scaled by a factor of ~104 to facilitate comparison with the low abundance pol positive droplets . The insets show the raw scatter plots of the pol positive droplets that were used to generate corresponding histograms for each isolated cell population . The pol-positive droplets for the ddPCR and Ex-ddPCR experiments are shown in red and blue , respectively . The histograms for the pol-positive droplets follow the same color key . Number of experimental repetitions ( n = 2–9 ) and errors are reported as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 18447 . 022 We also obtained access to cryopreserved bronchial alveolar lavage ( BAL ) and PBMC isolates from a single HIV-1 infected donor that were collected prior to initiation of ART and six months after ART therapy and performed a similar analysis ( Figure 7B ) . Viral DNA from highly purified monocytes was tested , as well as bronchial alveolar lavage ( BAL ) samples that consisted of >95% alveolar macrophages ( AMs ) . Using the Ex-ALU-gag PCR method for detecting proviral DNA , we found that ~40% ( monocytes ) and 80% ( alveolar macrophages ) of the ALU-gag amplicons were positive for uracils . Similar levels of uracils were detected in pre- and post-ART BAL isolates ( Figure 7B ) . Although a stable reservoir for HIV is primarily located in resting CD4+ T cells , these results suggest that HIV DNA residing in monocytes/alveolar macrophages comprises another potential source for viral rebound after discontinued therapy , consistent with other reports ( Cribbs et al . , 2015 ) .
Previous literature contains diverse and sometimes conflicting data on the role of dUTP and host cell uracil DNA glycosylase in HIV infection [reviewed in Weil et al . ( 2013 ) ] . Our key finding that MDMs consist of two distinct populations with respect to the uracilation phenotype likely explains why this potent restriction pathway has remained largely elusive . The established elements of this pathway are depicted in Figure 7C . Upon entry into the macrophage , reverse transcriptase ( RT ) encounters a nucleotide pool environment that favors the incorporation of dUTP into HIV DNA products predominantly in the form of U/A base pairs . The high ratio of dUTP/TTP is maintained , at least in part , by the low levels of dUTPase expression combined with the high expression levels of SAMHD1 in macrophages . Uracilation of the viral DNA can proceed to high levels in the cytoplasmic compartment because the UBER machinery is sequestered in the nucleus . However , when the heavily uracilated HIV DNA enters the nucleus it is attacked and fragmented by nuclear UBER enzymes . Fragmentation is initiated by hUNG2 uracil excision , and is likely followed by abasic site excision by human abasic site endonuclease ( APE1 or APE2 ) ( Figure 1—figure supplement 1 ) ( Schrader et al . , 2009; Stavnezer et al . , 2014 ) . We estimate , based on comparing the measured levels of early and proviral DNA levels in the GFP− MDMs ( Figure 3 ) , that ~99% of the viral DNA is destroyed before the integration step and further destruction at the proviral stage is observed over the next 14 days ( Figure 3D ) . This pathway differs from original models of APOBEC3G-mediated virus restriction in that uracils are introduced on both DNA strands and they are not inherently mutagenic ( Priet et al . , 2006 ) . We speculate that lethal mutagenesis , functional genome variation , or faithful repair could result from host enzyme processing of uracilated proviruses ( Figure 7C ) . In non-dividing MDMs , we found that uracilated proviruses contained few mutations yet gave rise to mutated viral genomic RNAs ( Figure 5D ) . Although it is tempting to attribute these mutations to RNAP II transcriptional errors as it encounters uracil on the template DNA strand , RNAP II shows high-fidelity incorporation of A opposite to U during in vitro transcription Menendez-Arias , 2009 . Moreover , many aspects of the mutation spectrum in Figure 5E are not reconcilable with RNAP II incorporating incorrect bases opposite a template U ( although the frequent A→G transitions may be explained by the potential of U to pair with A or G ) ( Kuraoka , 2002 ) . The actual error mechanism is likely to be much more complex given that transcription coupled repair is expected to generate abasic sites , strand breaks , and even gaps when densely spaced uracils are excised . In this regard , RNAP II is known to stall at isolated abasic sites during in vitro transcription and could also misincorporate various ribonucleotides at such sites in vivo ( Yu et al . , 2003 ) . The observation that proviral DNA is relatively mutation free , while the RNA is not , strongly suggests that repair intermediates generated during transcription by the host UBER pathway are most often repaired to restore the proviral coding sequence . Faithful repair could involve the reincorporation of U opposite to A due to the perturbed nucleotide pools , or the introduction of T . The observation that RNA mutations were increased upon stimulation with the inflammatory cytokine IFNγ , and nearly absent in the presence of the anti-inflammatory cytokine IL-4 , indicates that the transcriptional programs induced by these cytokines are highly relevant to the outcome; a further understanding of this complex problem is of considerable interest . Although errors introduced by IFNγ stimulation may frequently lead to non-functional viral genomes , they also allow the viral DNA sequence to evolve rapidly in the absence of reverse transcription . Replication-competent viruses that emerge would then undergo fitness selection in the host . These potential outcomes make UBER a double-edged sword with respect to propagation of viral infection . Uracil has never been considered an epigenetic base like 5-substituted cytosines because of its low abundance in genomes and because the detection of U/A base pairs is not possible with standard sequencing methods . Nevertheless , the densely incorporated uracils detected here have the potential to alter DNA structure and dynamics ( Ye et al . , 2012 ) , which could affect site recognition by transcription factors or other DNA binding proteins . There is a growing body of evidence indicating that uracil can exert significant effects on protein binding to DNA . These findings include ( i ) our previous report that multiple uracils silenced transcription from both HIV-1 LTR and CMV promoters in a cell line ( Weil et al . , 2013 ) , ( ii ) uracils within the origin of replication in HSV-1 perturb the binding of HSV-1 origin binding protein ( Focher et al . , 1992 ) , ( iii ) U/A pairs disrupt AP-1 transcription factor DNA binding ( Rogstad et al . , 2002 ) , ( iv ) singly uracilated DNA disrupts RNase H splicing specificity during reverse transcription Klarmann , 2003 , ( v ) U/A pairs perturb maintenance of telomere length in B cells by disruption of sheltrin binding ( Vallabhaneni et al . , 2015 ) , and ( vi ) one or two U/A base pairs within the specific cleavage site of some restriction enzymes prevents DNA strand cleavage ( Roberts et al . , 2015 ) . In addition , the abasic site product of uracil excision is known to exert a large negative effect on transcription ( Luhnsdorf et al . , 2014 ) as would any mutations in transcription factor recognition sequences arising from error prone repair of excised uracils ( Shah et al . , 2015; Emiliani , 1998 ) . Combined , these potential effects of U/A pairs could profoundly silence HIV gene expression . Historically , macrophages have been controversial targets for HIV infection ( Sattentau and Stevenson , 2016 ) . Macrophages have unique cellular phenotypes that are conducive for a significant impact on infection because they persist for months after initial infection ( Gerngross and Fischer , 2015 ) , continue to produce infectious virus particles for their entire lifetime ( Crowe and Sonza , 2000 ) , and provide a drug tolerant environment for HIV propagation ( Gavegnano and Schinazi , 2009; Gavegnano et al . , 2013 ) . Moreover , their residence , within tissues that have poor drug penetrance ( lung , brain , gut ) ( Cribbs et al . , 2015 ) , has led to an increased interest in their potential role in the establishment and persistence of HIV infection . The detection of abundant viral uracils in monocytes isolated from HIV-1 infected individuals who showed full viral suppression from ART , in vivo The fairly short several day half-life of blood monocytes suggests that uracilated viral DNA detected in these cells arose from a recent infection , perhaps by passage of the monocyte through a tissue reservoir containing virus producing cells ( Wang et al . , 2013 ) . Such an infection pathway could involve a canonical CCR5-dependent viral entry mechanism or a recently described pathway involving phagocytosis of infected T cells by macrophages ( Baxter , 2014 ) . However , the high efficiency of ART and the low levels of infected T cells using current treatment regimes make the phagocytosis pathway for monocyte infection seem less plausible for our aviremic patient samples . Finally , the requirement for dUTP incorporation during reverse transcription virtually excludes the possibility that the infected monocytes containing uracilated DNA arose from latently infected myeloid stem cells , because monocyte differentiation does not involve reverse transcription and genomic DNA replication does not involve accumulation of detectable uracils ( Figure 7—figure supplement 1I ) . A further ramification of these data is that some uracilated proviruses may persist for the entire lifetime of a macrophage due to protection within highly compacted chromatin , which is highly resistant to UBER ( Ye et al . , 2012 ) . Under such circumstances , the potential time scale for uracil persistence could be years based on the findings that replication-competent viruses have been detected in brain microglial cells of patients even after years of viral suppression using ART ( Rappaport and Volsky , 2015; Robillard et al . , 2014 ) . If U/A pairs are found to down regulate viral gene expression through chromatin stabilization or by weakening the binding of transcription factors , some proviruses could remain transcriptionally silenced for years and then become activated when appropriate stimulatory conditions are present .
This study was approved by the Johns Hopkins Institutional Review Board and written informed consent was provided by both HIV-infected and healthy individuals ( IRB00038590 ) . Peripheral blood was obtained from healthy volunteers and ART suppressed patients with viral loads <20 copies HIV-1 RNA/mL . Monocytes and resting CD4+ T cells were obtained from donor PBMCs as described in Isolation of primary cells from donor PBMCs and the purity of the cell types was determined by flow cytometry . Vesicular stomatitis virus G protein ( VSV-G ) pseudo-typed HIV-1 virions ( HIVNL4 . 3 ( VSVG ) ) were generated as previously described ( Weil et al . , 2013 ) by co-transfection of HEK 293T ( RRID:CVCL_0063 ) cells with pNL4-3-ΔE-eGFP and pVSV-G . Virions containing the CCR5 co-receptor ( HIVSF162 ( CCR5 ) ) were obtained by transfection of HEK 293T cells with pSF162R3-GFP-Nef+ . HIVBAL ( CCR5 ) virus was obtained from NIH AIDS Research and Reference Reagent program ( cat #510 ) . Viral supernatants were collected 48 hr after transfection , DNaseI-treated to remove any plasmid carryover and virions were purified over a 20% ( wt/vol ) sucrose cushion . HIVNL4 . 3 ( Δvpr ) mutant virus construct was generated by site-directed mutagenesis of pNL4-3-ΔE-eGFP . Primers used for site-directed mutagenesis can be found in Supplementary file 3 . Peripheral blood mononuclear cells ( PBMCs ) were isolated from HIV-1 positive or negative donors using density centrifugation on a Ficoll-Hypaque gradient . Monocytes were isolated from the PBMC population using the Pan monocyte Isolation Kit ( Miltenyi Biotec ) . Monocytes were differentiated into macrophages over 7 days using MDM-20 media containing RPMI 1640 , 20% ( vol/vol ) autologous plasma , 10 ng/mL GM-CSF or M-CSF ( BD Biosciences , San Jose , CA ) , 1 × HEPES , and 1 × Glutamine ( Gibco , Gaithersburg , MD ) . Cultured MDMs were maintained in media containing RPMI 1640 + 10% ( vol/vol ) dialyzed FBS , 1% Pen/Strep . CD4+ T cells were isolated from PBMC population using the CD4+ isolation kit ( CD4+ T cell Isolation kit II , Miltenyi Biotec , San Diego , CA ) . CD4+ cells were activated with 0 . 5 μg/mL phytohemagglutinin ( PHA ) for 3 days in IL-2 containing media . Resting CD4+ cells were further enriched from the bulk CD4+ cell population by negative selection against biotinylated activation markers; CD25 , CD69 and HLA-DR ( Miltenyi Biotec ) . The purity of monocytes , activated and resting CD4+ cells was determined by flow cytometry with fluorescent labeling of anti-human; Fluorescein isothiocyanate ( FITC ) - and allophycocyanin ( APC ) -conjugated mAbs CD4 ( RRID:AB_314074 ) , CD25 ( RRID:AB_2280228 ) , CD69 ( RRID:AB_492844 ) and HLA-DR ( RRID:AB_10839413 ) . Isotype-matched control mAbs ( BD Biosciences ) were used for gating and quantification . Monocyte and MDM populations were stained using brilliant violet 421-conjugated ( BV421 ) anti-CD14 ( RRID:AB_10899407 ) , CD16-APC ( RRID:AB_2616904 ) and CD3-FITC ( RRID:AB_2616618 ) . The purity of these cells were analyzed by flow cytometry using a FACS Canto II ( BD Biosciences ) and FlowJo software ( Treestar ) . Using this method , we obtained monocytes with a purity of >99 . 9% with undetectable T cell contamination . Cells were then pelleted and used for protein extraction , dNTP extraction , or infection as described above; for cell infections , cells were analyzed for GFP expression by FACS analysis . In preparation for cell sorting studies , MDMs infected with GFP-reporter virus ( HIVNL4 . 3 ( VSVG ) or HIVSF162 ( CCR5 ) ) were washed twice with Hank’s balanced salt solution ( HBSS ) and then detached by incubating with Accutase ( Cell Technologies ) for 30 min at 37°C . MDMs were resuspended at 5 × 106 cells/mL in 1 × HBSS ( pH 7 . 2 ) , 5 mM EDTA and 0 . 5% BSA . Prior to sorting , cell suspensions were passed through a 35 mm nylon mesh ( BD Biosciences ) and propidium iodide added to gate out dead cells . Sorting was performed on a MoFlo Sorter using a 100 μm nozzle and the sort purify 1 mode . pIRES-Ugi-eGFP and pIRES-hUNG2-eGFP reporter plasmids were constructed by replacing the neoR cassette for eGFP using SmaI and XbaI cloning sites of pIRESneo3 ( Clonetech , Mountainview , CA ) . The empty pIRES-eGFP plasmid was used as a transfection control ( shown ) . Primary macrophages derived from blood monocytes were cultured in the presence of M-CSF for 7 days prior to transfection using jetPEI-Macrophage in vitro DNA transfection reagent ( Polyplus ) . Cells expressing GFP were sorted by FACS 48 hr post-transfection . GFP+ cells ( indicating the presence of hUNG2 or its inhibitor protein ) were allowed to adhere for 24 hr and then spin-infected with HIVSF162 ( CCR5 ) virus . Cells were harvested 3 days post-infection , and DNA was extracted using QIamp DNA kit ( Qiagen , Valencia , CA ) for PCR and Ex-PCR analyses . Due to the low efficiency of transfection using MDM target cells , it was not possible to obtain immunoblots for Ugi or hUNG2 expression levels in these samples . However , the specific and opposite effects of the Ugi and hUNG2 transfection , the absence of an effect upon transfection with pIRES-eGFP plasmid , and the alternative approach of deleting vpr are strongly supportive of our interpretations . Protein lysates were obtained using CelLytic M ( Sigma , St . Louis , MO ) reagent according to the manufacturer’s instructions . SAMHD1 activity was determined as previously described with several modifications ( Hansen et al . , 2014 ) . Protein lysates ( 5 μg ) were incubated in buffer containing; 5 mM [8-3H] dGTP ( activator/substrate ) ( Moravek Biochemicals , Brea , CA ) , 5 mM ATP , and 10 mM paranitrophenyl phosphate ( p-NPP ) in a total volume of 50 μL . ATP was included to inhibit nonspecific phosphatases and p-NPP was added to prevent the degradation of dUTP by alkaline phosphatase ( Williams and Parris , 1987 ) . The small molecule inhibitor ( pppCH2dU ) of SAMHD1 was used to determine specificity of the assay . Reactions were incubated at 37°C at varying time points , at which 2-μL samples were removed and quenched by spotting onto a C18-reversed phase thin layer chromatography ( TLC ) plate . The TLC plate was developed in 50 mM KH2PO4 ( pH 4 . 0 ) to separate substrate [dGTP ( Rf = 0 . 80 ) ] from products [dG ( Rf = 0 . 20 ) ] . Plates were exposed to a tritium-sensitive screen for 5 hr and then scanned on a Typhoon phosphoimager ( GE Healthcare , Pittsburgh , PA ) and the counts present in the substrate and product were quantified using the program Quantity One ( Bio-Rad , Hercules , CA ) . dUTPase activity was assessed as previously described ( Weil et al . , 2013 ) . Briefly , protein lysates were incubated with [5-3H] dUTP ( Moravek Biochemicals , Brea , CA ) and the fraction of substrate hydrolyzed to dUMP product was monitored . Reactions were incubated at 37°C for 1 hr , after which 2-μL samples was spotted onto a PEI-cellulose TLC plate . The TLC plates were developed in 0 . 5 M LiCl to separate substrate [dUTP ( Rf = 0 . 1 ) ] from the product [dUMP ( Rf = 0 . 6 ) ] . Plates were exposed to a tritium-sensitive screen for 5 hr and scanned on a Typhoon phosphoimager ( GE Healthcare ) . A molecular beacon hairpin reporter assay was used to determine endogenous UNG activity in crude cell extracts ( Weil et al . , 2013 ) . Reactions were performed using 5 μg of protein lysate and 50 nM S-pin-18 in 10 mM Tris-HCl ( pH 7 . 1 ) , 100 mM NaCl , 1 mM EDTA , and 0 . 2% Triton X-100 . Supplementing the reaction with the potent Uracil DNA glycosylase inhibitor ( Ugi ) showed no measurable activity . Additionally , a molecular beacon assay was used to monitor Apurinic/Apyrimidinic endonuclease ( APE ) activity ( Seiple et al . , 2008 ) . Two FAM and dabcyl quenched DNA duplex substrates were used . A specific substrate ( SS ) containing a single abasic site ( Φ ) and a non-specific substrate ( NS ) with the abasic site replaced with the canonical DNA base Thymine ( T ) . The sequence of the specific molecular beacon substrate; SS , 5’-FAM-GAGAAΦATAGTCGCG3’ and 5’-CGCGACTATGTTCTC-dabsyl-3’ ( where Φ is a tetrahydrofuran abasic site analog ) . Independent reactions were performed using 5 μg of protein lysate and 50 nM in 10 mM Tris-HCl ( pH 7 . 1 ) , 100 mM NaCl , 1 mM EDTA , and 0 . 2% Triton X-100 . To determine APE activity , the difference in the initial rate of the specific substrate from the non-specific rate was measured . Rates were measured on a Fluoro-Max 3 ( Horiba Jobin Yvon , Edison , NJ ) . The quantification of dNTPs from primary immune cells was performed by quantitative LC-MS . To remove the signal depression by the abundant rNTPs ( Kennedy et al . , 2010 ) , a boronic acid chromatography step was performed prior to analysis . Briefly , NTPs were methanol extracted for 1 hr at 4°C from 5 million cells . Samples were supplemented with isotopically labeled ( Mansky et al . , 2000 ) C ( Priet et al . , 2005 ) N-dNTPs ( Sigma-Aldrich ) for accurate determination of each metabolite . Supernatants were dried under vacuum and then resuspended in 50 μL ( 200 mM Ammonium acetate , pH 8 . 8 ) . This was directly applied to 100 μL of pelleted boronic acid resin and incubated at 4°C for 1 hr . Resin was pelleted by centrifugation , supernatant removed and dried under vacuum . Sample was resuspended in mobile phase A [2 mM NH4PO4H2 , 3 mM Hexylamine ( HMA ) in dd H2O] . Chromatographic separation was performed with an analytical Hypersil Gold C18 column ( 100 × 1 mm , 3 μm particle size , Thermo Scientific , Waltham , MA ) , using a ( Agilent , Santa CLara , CA ) LC-MS system equipped with a binary pump . Mobile phase A was [2 mM NH4PO4H2 , 3 mM Hexylamine ( HMA ) in dd H2O] and mobile phase B [Acetonitrile] . The flow rate was maintained at 50 µL/min and an injection volume of 20 µL . The autosampler was held constant at 4°C and the column at 30°C . The chromatographic gradient began at 10% mobile phase , followed by a linear gradient that reached 60% in 15 min . At the end of each run , the column was equilibrated for 10 min . Analyte detection was performed with the same LC-MS system with an electrospray ionization source , using multiple-reaction monitoring ( MRM ) analysis in positive ionization mode . The following MRM transitions ( parent → product ) were detected as previously described with the addition of dUTP: dATP ( 492 m/z → 136 m/z ) , dGTP ( 508 m/z → 152 m/z ) , dCTP ( 468 m/z → 112 m/z ) , TTP ( 483 m/z → 81 m/z ) and dUTP ( 469 m/z → 81 m/z ) ( Fromentin et al . , 2010 ) . A standard curve was constructed using equimolar amounts of dNTPs . Infections with HIV viral strains ( HIVNL4 . 3 ( VSVG ) , HIVSF162 ( CCR5 ) , HIVBAL ( CCR5 ) ) were performed as previously described ( Jordan et al . , 2003 ) . Briefly , primary cells were plated at a density of ~100 , 000 cells per well in a 96-well plate . Virus was added to each well , and the plate was spin infected for 2 hr at 1200 × g and 30°C and then incubated at 37°C until the given time points . Following incubation , cells were collected and stored at −80°C . We determined that A3G and A3A were poorly expressed or undetectable in MDMs prior to and after infection with HIVSF162 ( CCR5 ) ( Figure 2—figure supplement 1 ) . To validate the antibody activities and specificities , A3G and to a lesser extent A3A , were both detected in PBMCs , but not HEK 293T cells which are known not to express either enzyme ( Figure 2—figure supplement 1 ) . Protein extraction was performed using CelLytic M reagent ( Sigma ) according to the manufacturer’s instructions . Protein concentration was determined by the Bradford assay ( BioRad ) . Twenty-five micrograms of cell lysate was run on an SDS/PAGE gel and transferred to a PVDF membrane . Primary antibody ( 1:1000 ) was used for detection of APOBEC3G and is cross-reactive with APOBEC3A . Membrane was incubated overnight with anti-APOBEC3G obtained from the NIH AIDS Reagent Program , Division of AIDS , NIAID , NIH ( cat . no . 10201; from Immunodiagnostics: Murine Human APOBEC3G ( CEM15 ) Monoclonal Antibody ( RRID:AB_2617158 ) . The secondary HRP-conjugated anti-rabbit IgG ( 1:2000 ) was used for detection ( ab97080; Abcam , Cambridge , MA , RRID:AB_10679808 ) . Global protein expression was assessed by blotting MDM extracts for GAPDH ( see below for further information ) . The primary antibody for GAPDH ( 1:2000 ) was a rabbit polyclonal antibody ( ab15246; Abcam , RRID:AB_613387 ) , and the secondary antibody ( 1:5000 ) was HRP-conjugated anti-rabbit IgG ( ab97080; Abcam ) . Ten micrograms of each sample was run on an SDS/PAGE gel and transferred to a PVDF membrane . Detection of virion-associated viral protein R ( vpr ) was achieved using a rabbit polyclonal antibody ( 1:2000 ) obtained through the NIH AIDS Reagent Program , Division of AIDS , NIAID , NIH ( cat . no . 11836; from Dr . Jeffrey Kopp , RRID:AB_2617159 ) and secondary antibody ( 1:5000 ) was HRP-conjugated goat anti-mouse IgG ( ab97040; Abcam , RRID:AB_10698223 ) . Virion associated proteins were probed by re-suspending purified virus directly in buffer containing SDS . For an internal control , the same membrane was also blotted for p24 using the mouse monoclonal anti-p24 ( 1:2000 ) primary antibody obtained from the NIH AIDS Reagent Program , Division of AIDS , NIAID , NIH ( cat . no . 24–2; from Dr . Michael H . Malim , RRID:AB_1124906 ) . The secondary antibody ( 1:5000 ) used was an HRP-conjugated goat anti-mouse IgG ( ab97040; Abcam ) . To determine the uracil status of viral DNA , a modified droplet digital PCR ( Ex-ddPCR ) method was developed by inserting a pre-digestion step using UNG ( Figure 1 ) . Total genomic DNA was isolated using a QIamp DNA extraction kit ( Qiagen ) according to manufacturer’s protocol . Genomic DNA was first fragmented using the endonuclease BSAJ-1 ( 1 U ) in Cutsmart buffer ( NEB ) for 1 hr at 60°C . Uracil containing DNA sequences were digested using 50 nM UNG in 1x TMNB+ buffer ( 10 mM Tris-HCl ( pH 8 . 0 ) 20 mM NaCl , 11 mM MgCl2 and 0 . 002% Brij-35 ) . Reactions were cleaned up using a MinElute PCR Purification kit ( Qiagen ) . UNG or mock digested DNA were diluted in to a PCR master mix containing ( 1x Platinum taq buffer , 1x droplet stabilizer ( RainDance Technologies , Billerica , MA ) , 1 mM dNTPs , 0 . 9 μM forward and reverse primers , 0 . 15 μM probe , 2 mM MgCl2 , 0 . 5 μM carboxy-X-rhodamine ( ROX , Sigma ) as a passive reference dye , 1 U platinum taq polymerase and varying amounts of template DNA and loaded in to a RainDance source chip ( RainDance Technologies ) . Different primer/probe sets were designed to tile the HIV genome to report on heterogeneity of uracil incorporation . Amplification was performed with the following thermal program: 95°C for 10 min , followed by 44 cycles of: 95°C for 15 s and then 60°C for 1 min , followed by a final step of heating to 95°C for 10 min . The final high-temperature cycle cures the droplets . For each step , the ramp cycle time was decreased to 0 . 5°C/s . Amplification samples were transferred to a RainDance Sense instrument . Copy numbers of the human RNase P ( RPP30 ) gene were measured in the same reaction mixture to determine cell number . DNA isolated from the Jurkat-based cell line ( J-Lat ) ( Jordan et al . , 2003 ) containing a full-length integrated HIV genome was used to establish gating parameters for HIV and RPP30-positive droplets . J-Lat cells were obtained through the NIH AIDS Reagent Program , Division of AIDS , NIAID , NIH: J-Lat Full Length Clone ( clone 9846 ) from Dr . Eric Verdin . Ex-ddPCR data were analyzed using the RainDrop Analyst software package ( RainDance Technologies ) . Drop-size gating was performed to remove drops of atypical size using a lower and upper size gate boundary for each PMT of 800 and 1150 , respectively . Intact drops for each of the three clusters ( one negative; quenched drops , and two positive; PMT1+ and PMT2+ ) were gated independently using the ellipse-gating tool . Gated clusters were spectrally compensated using the following definitions: ( intact drops/quenched drops ) as the negative control for both PMT1+ and PMT2+ , while ( intact drops/PMT1+ ) and ( intact drops/PMT2+ ) were selected for PMT1+ and PMT2+ , respectively . Positive cluster gating on spectrally compensated data was again achieved by using the ellipse-gating tool . A Poisson correction statistic was applied to both positive targets ( HIV and RPP30 ) to remove sampling error ( Sanders et al . , 2011 ) . Determination of uracil-free DNA transcripts was determined using Equation 1 . ( 1 ) Frac UDNA=\ {positive droplets ( no UNG ) −positive droplets ( +UNG ) }{positive droplets ( no UNG ) } Ex-seq libraries were constructed from genomic DNA by shearing for 12 min on BioRuptor ( Diagenode , Denville , NJ ) , end repair , dA-tailing , and adapter ligation ( NEB , Ipswich , MA ) with dual indexed adapters ( GSE76091 ) . Library concentrations were quantitated with a Qubit fluorometer and mixed at equal proportions to yield 500 ng of library DNA . HIV enrichment was performed using lockdown probes designed against the forward and reverse strand of the HIV construct used for infection , HIVNL4 . 3 ( GSE76091 ) . Briefly , libraries were mixed with 5 μg Cot-1 and XGen Blocking oligos and hybridized with the lockdown probes at 3 pmol total for 4 hr at 65°C . Probes were isolated on Streptavidin Dynabeads ( M-270; Invitrogen ) to capture HIV DNA using Nimblegen SeqCap EZ Hybridization and Wash kit as indicated in the XGen lockdown probe protocol . DNA was eluted in water and samples were split in half , followed by uracil removal in one sample with UNG ( 5 U ) and T4 Endonuclease IV ( 10 U ) digestion for 2 hr at 37°C on beads . On-bead PCR was performed for 22 cycles with short Illumina primers ( GSE76091 ) and Maxima Taq DNA polymerase ( Invitrogen ) . PCR products were purified with 1 . 9 × volumes of Ampure XP beads ( Beckman Coulter , Brea , CA ) and quantified with a Qubit fluorometer . Forward and reverse lockdown reactions were pooled for UNG- and UNG+ independently and two lanes of Illumina HiSeq were performed . Data files from this study have been deposited to the NCBI Gene Expression Omnibus ( GEO; [Edgar et al . , 2002] ) under accession GSE76091 . Demultiplexed FASTQ records were trimmed to remove low-quality cycles ( cycles 1 and 2 ) , and processed records were aligned to the human ( hg19 ) and HIV reference ( HIVNL4 . 3 ( VSVG ) ) with bwa mem . Alignments in BAM format were converted to bedGraph coverage with BEDTools . Samples were deduplicated with samtools rmdup and Piccard markduplicates . VCF files were generated for mutation analysis from the rmdup files with freebayes and filtered with vcffilter for quality scores greater than 40 ( Garrison and Marth , 2012 ) . Mutations were further filtered for a depth of greater than 10 reads and 10 or more unique sequencing start sites . Chimeric and discordant reads were identified with samtools and were normalized to counts per million reads that mapped to HIV . A reproducible software pipeline for analysis of Excision-seq data is available at https://github . com/hesselberthlab/stivers-hiv . To specifically measure fractional uracilation of proviral DNA , we used a modification of the ALU-gag nested qPCR approach , in which an UNG digestion step is inserted before the initial PCR amplification ( Ex-ALU-gag nested qPCR ) ( Weil et al . , 2013; O'Doherty et al . , 2002 ) . Ex-Alu-gag nested PCR begins with the first amplification ( 40 cycles ) used a primer complementary to genomic ALU sequences and a gag primer and ~100 ng genomic DNA . In addition to samples , reactions containing: gag primer only , no-template , and genomic DNA isolated from uninfected matched cells were used as controls . A standard curve for Ex-ALU-gag nested qPCR was generated using DNA isolated from J-Lat cells . Although there is some debate concerning the measurement of absolute proviral copy numbers using ALU-gag nested qPCR , Ex- ALU-gag nested qPCR is unambiguous with respect to determining the change in copy number resulting from UNG digestion . This is due to the fact that identical DNA samples are used for the PCR amplification steps . Thus , the fraction of the proviral population that contains uracil within the gag amplicon may be reliably measured . FACS sorted ( GFP−/+ ) MDM populations were maintained in culture media for indicated times prior to cytokine stimulation . MDMs were either stimulated by co-culture with the IFNγ ( 50 ng/mL , R&D Systems , Minneapolis , CA ) and 200 ng HIVtat peptide ( ProSpec Ltd . , East Brunswick , NJ ) or IL-4 ( 50 ng/mL , R&D Systems ) . At 3 days post-stimulation ( dps ) , the cytokine containing media is removed and replaced with fresh culture media . Viral supernatants were collected at varying time points from each well and assayed for HIV-1 p24 production using a sensitive ELISA ( Alliance HIV-1 p24 antigen ELISA kit , Perkin Elmer , Hopkinton , MA ) . T cells were infected with replication-competent HIVBAL ( CCR5 ) as previously described above . The HIVBAL ( CCR5 ) virus was obtained through NIH AIDS Reagent Program , Division of AIDS , NIAID , NIH ( cat . no . 510; contributors: Dr . Suzanne Gartner , Dr . Mikulas Popovic and Dr . Robert Gallo ) . Briefly , CD4+ T cells isolated from PBMCs of an HIV-1-negative donor were purified using a negative selection method ( CD4+ T cell Isolation Kit II , Miltenyi Biotec ) and activated with PHA for 3 days in IL-2-containing medium . A spinoculation method ( 2 hr , 37°C , 1200 × g ) was used to infect activated CD4+ T cells with replication-competent HIVBaL ( CCR5 ) virus ( 300 ng p24/106 cells ) . Following incubation for 6 days in IL-2 containing media , supernatants were collected and stored at −80°C . Freshly isolated CD4+ T cells were activated by αCD3/αCD28 stimulation for 3 days in IL-2 containing medium and then plated ( 100 μL/well ) into a 96-well v-bottom plate . HIVBaL ( CCR5 ) ( 500 ng p24/106 cells ) was added to each well and cells were infected by spinoculation . Following infection , cells were suspended in 40 mL of IL-2 containing medium supplemented with enfuvirtide ( 10 μM ) to prevent additional rounds of viral replication . Following a 2-day incubation at 37°C , genomic DNA was extracted from 107 CD4+ cells using the Qiagen Gentra Purgene Cell Kit A . DNA was subjected to a nested limiting dilution PCR protocol adapted from Ho et al . with the following modifications: 30 cycles were run for the outer PCR , 40 cycles were run for each inner PCR , and all clonal outer wells were subjected to the 4 inner PCRs ( A-D; corresponding to gag , pol , rev and env regions ) regardless if positive or negative for the gag inner PCR ( Ho et al . , 2013 ) . HIV infected ( HIVSF162 ( CCR5 ) ) MDMs were sorted into GFP– and GFP+ populations and diluted serially in five-fold increments ( 1× 105–100 cells ) using 12-well plates ( Costar ) . Cells were maintained in culture for 14 days prior to cytokine simulation and supernatant collection . Extracellular viral RNA was collected from the supernatants of individual wells that contained the fewest number of input cells and were p24-positive in an attempt to obtain virus released from a cell containing a single provirus . Extracellular viral RNA was purified using a ZR-96 Viral RNA Kit ( Zymo Research ) . Isolated viral RNAs were DNase-treated ( Life Technologies ) and reverse transcribed using a qScript cDNA synthesis kit ( Quanta Biosciences ) according to the manufacturer’s protocol . Using the products of RT-PCR , nested PCR reactions were used to amplify the LTR and env regions of interest ( primer sequences are listed in Supplementary file 3 ) . Limiting dilution PCR was performed using Platinum Taq High-Fidelity polymerase ( Invitrogen ) . For LTR amplification , outer primers 5LTROut and mod_VQA_R and inner primers 5LTRIn and mod_VQA_R were used . Thermocycler settings were the same for both outer and inner LTR PCR reactions: 94°C for 2 min , followed by 29 cycles of: 94°C for 15 s , 55°C for 30 s and then 68°C for 1 min , followed by a final extension at 68°C for 7 min . For env amplification , outer primers ES7 and ES8 and inner primers Nesty8 and DLoop were used . Thermocycler settings were the same for both outer and inner env PCR reactions: 94°C for 3 min , followed by 25 cycles of: 94°C for 30 s , 55°C for 30 s and then 68°C for 1 min , followed by a final extension at 68°C for 5 min . cDNA was diluted ( 1/20–1/1000 ) and used as input for the outer PCR reaction . Aliquots ( 1 μL ) from each outer PCR product were used as input for inner PCR reactions and subjected to 1% agarose gel electrophoresis . Inner clonal PCR reactions from selected dilutions were used to identify dilutions where <20% of the PCR reactions were positive , where the corresponding outer PCR dilution contained one template with >90% of possibility by Poisson statistics . PCR products were gel extracted using the QIAquick Gel Extraction Kit ( Qiagen ) and directly sequenced ( Sanger ) without cloning ( Genewiz , South Plainfield , NJ ) . Forward and reverse sequences for each sample were aligned into one consensus contig per sample using default assembly parameters with CodonCode Sequence assembly and Alignment software and aligned to HIVSF162 ( CCR5 ) reference sequence . Chromatograms showing double peaks were taken as evidence for more than one template present in the initial PCR reaction and were discarded from further analysis . Bronchoscopy and broncho-alveolar lavage was performed on a single HIV-1 infected patient pre and post-ART treatment . At the time of the first bronchoscopy , the patient was ART naive with recent plasma viral load 24 , 162 copies/mL and CD4 459 cells/mL . Second bronchoscopy occurred 9 . 2 months later , after 6 months of ART treatment ( efaviernz/emtricitabine/tenofovir ) with plasma viral load <20 copies/mL and CD4 413 cells/mL . To obtain alveolar macrophages ( AMs ) , bronchoscopy and lavage were done , as described elsewhere ( Popesu , 2014 ) . Briefly , lavage fluid was collected , filtered and BAL cells were pelleted by centrifugation . BAL cells were washed with HBSS and counted . After centrifugation , cells were re-suspended in Recovery Cell Culture freezing media according to the BAL count ( 106 cells/mL ) and stored in a −140°C freezer . PBMCs were isolated on a ficoll-hypaque gradient and were processed in the same manner as BAL cells . Monocytes and resting CD4+ T cells purified via negative selection ( Miltenyi Biotec ) from blood collected from six HIV infected patients on antiretroviral therapy ( ART ) . Cell purity was determined as described above in Purity of isolated cell populations . Cellular DNA was extracted using a QIamp DNA Midi Kit ( Qiagen ) following the manufacturer’s protocol . The frequency of total HIV DNA ( pol copies/106 cells ) was determined by ddPCR using published primers to conserved regions of HIV pol ( Strain , 2013 ) and the reference cellular gene RNase P ( RPP30 ) for genomic quantification . Total cellular DNA was estimated by halving the number of RPP30 copies , and HIV copy numbers per diploid cell were calculated as the ratio of template pol copies per diploid cell ( pol copies/106 cells ) . Due to the low frequency of HIV DNA in a large background of host cellular DNA , it was desirable to load as much total DNA ( ~106 cells ) as possible to maximize assay sensitivity . See Figure 7—figure supplement 1 for further details . The limit of detection and quantification of the ddPCR assay was evaluated by establishing the loading limit and intrinsic limit of detection . A dilution series of uninfected PBMC DNA was used to determine the maximum amount of total DNA that could be accurately measured in a single ddPCR reaction . Cellular RPP30 was used to monitor input DNA and showed a linear response across a wide range ( 103–106 ) of RPP30 copies . The intrinsic limit of detection of ( 5 copies/106 cells ) was determined by serially diluting DNA from infected CD4+ T cells into a background of healthy donor PBMC DNA ( 250 , 000 cells ) . For patient samples , replicate measurements were taken to improve the accuracy . Data were analyzed for statistical significance ( HIV copy number and Frac UDNA ) by a two-tailed Student’s t test for independent samples using GraphPad Prism ( La Jolla , CA ) using significance levels: ***p<0 . 001 , **p<0 . 01 , *p<0 . 05 , not significant [ns]; n indicates the number of independent experimental replicates . Raw and processed sequencing data files ( FASTAQ ) from this study have been deposited to the NCBI Gene Expression Omnibus ( GEO; http://www . ncbi . nlm . nih . gov/geo/ ) ( Barrett et al . , 2011 ) under accession number GSE76091 . A reproducible software pipeline for analysis of Excision-seq data is available at https://github . com/hesselberthlab/stivers-hiv . | Human immunodeficiency virus type 1 ( HIV-1 ) infects and kills immune cells known as CD4+ T cells , leading to the disease AIDS . Current drug treatments enable HIV-1 infected patients to live relatively long and healthy lives . However , no cure for HIV-1 exists because the virus lives indefinitely in a resting state within the genetic material – or genome – of the infected cell , where it is not susceptible to drug treatments . Most HIV-1 research focuses on T cells , but another type of immune cell – the macrophage – may also harbor resting HIV-1 in its genome . Compared to other cells , macrophages are unusual because they produce large amounts of a molecule called deoxyuridine triphosphate ( dUTP ) . Most cells , including T cells , keep dUTP levels very low because it closely resembles molecules that are used to make DNA and so it can be accidentally incorporated into the cell’s DNA . When this happens , the cell removes the dUTP from the DNA using enzymes in a process called uracil base excision repair ( UBER ) . To hide inside the cell’s genome , HIV-1 needs to produce a DNA copy of its own genome , but it was not known what happens when HIV-1 tries to do this within a macrophage that contains high levels of dUTP and UBER enzymes . Here , Hansen et al . reveal that about 90% of macrophages have exceptionally high levels of dUTP and are poorly infected by HIV-1 . The high levels of dUTP result in the virus incorporating dUTP into its DNA , which is then attacked and fragmented by UBER enzymes . However , about one in a hundred viral DNA molecules do manage to successfully integrate into the genome of the macrophage . This viral DNA later gives rise to new virus particles through an error-prone process that , by introducing new mutations into the virus genome , may help HIV-1 to evolve and persist . Further experiments examined cells that give rise to macrophages from infected patients who had been on anti-HIV drug therapy for several years . Hansen et al . found that there was lots of dUTP in the DNA sequences of HIV-1 viruses found in these “precursor” cells . These precursor cells only live for several days before being eliminated , so the presence of viruses containing dUTP suggests these cells were infected recently . A future challenge will be to identify new anti-HIV drugs that specifically target macrophages and to understand the role of error-prone production of new viral genomes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"immunology",
"and",
"inflammation"
] | 2016 | Diverse fates of uracilated HIV-1 DNA during infection of myeloid lineage cells |
Cationic antimicrobial peptides ( CAPs ) such as defensins are ubiquitously found innate immune molecules that often exhibit broad activity against microbial pathogens and mammalian tumor cells . Many CAPs act at the plasma membrane of cells leading to membrane destabilization and permeabilization . In this study , we describe a novel cell lysis mechanism for fungal and tumor cells by the plant defensin NaD1 that acts via direct binding to the plasma membrane phospholipid phosphatidylinositol 4 , 5-bisphosphate ( PIP2 ) . We determined the crystal structure of a NaD1:PIP2 complex , revealing a striking oligomeric arrangement comprising seven dimers of NaD1 that cooperatively bind the anionic headgroups of 14 PIP2 molecules through a unique ‘cationic grip’ configuration . Site-directed mutagenesis of NaD1 confirms that PIP2-mediated oligomerization is important for fungal and tumor cell permeabilization . These observations identify an innate recognition system by NaD1 for direct binding of PIP2 that permeabilizes cells via a novel membrane disrupting mechanism .
Host defense peptides , which include cationic antimicrobial peptides ( CAPs ) , are a group of innate immune molecules produced by essentially all plant and animal species that act as a first line of defense against microbial invasion . In common with most innate immunity peptides , they are relatively small ( typically <100 amino acid residues ) , are predominantly cationic , and typically harbor a substantial number of hydrophobic amino acids ( Hancock and Lehrer , 1998; Brogden , 2005; Lai and Gallo , 2009 ) . Although originally identified due to their potent activity against microbial pathogens , several CAPs also exhibit cytolytic activity against a range of mammalian tumor cells ( Lichtenstein et al . , 1986; Cruciani et al . , 1991; Hancock and Sahl , 2006; Schweizer , 2009 ) . The defensins are a family of CAPs that are ubiquitously expressed in plants , animals , insects , and fungi that play an important role in innate immune defense against microbial threats ( Brogden , 2005; Lay and Anderson , 2005; Hancock and Sahl , 2006; Lai and Gallo , 2009 ) . The plant defensins belong to a large family of molecules that are highly variable in sequence but have a conserved structure . The sequence variability leads to several biological functions including antimicrobial activity , regulation of plant development , and pollen tube guidance ( Carvalho and Gomes , 2009; De Coninck et al . , 2013 ) . Even those plant defensins that have been ascribed antifungal activity have large differences in sequence and are likely to act by different mechanisms ( van der Weerden and Anderson , 2013 ) . The plant defensins are small ( ∼5 kDa , 45–54 amino acids ) , basic , cysteine-rich proteins that display a family-defining disulfide bond array ( in a CI–CVIII , CII–CV , CIII–CVI , and CIV–CVII configuration ) known as the cysteine-stabilized αβ ( CSαβ ) motif . This motif consists of a triple-stranded antiparallel β-sheet , which is cross-braced via three disulfide bonds at the core of the molecule to an α-helix ( in a βαββ arrangement ) . The fourth conserved disulfide bond further rigidifies the protein by linking together the N- and C-terminal regions of the molecule , effectively generating a highly stable pseudocyclic molecule ( Janssen et al . , 2003; Lay et al . , 2003b , 2012; Lay and Anderson , 2005 ) . This CSαβ fold is also conserved in defensins found in other organisms , including insects and fungi ( Lay and Anderson , 2005 ) . NaD1 , a plant defensin isolated from the flowers of the ornamental tobacco ( Nicotiana alata ) , exhibits potent antifungal activity against pathogenic fungi , including Fusarium oxysporum , Botrytis cinerea , Aspergillus niger , Cryptococcus species , as well as the yeasts Saccharomyces cerevisiae and Candida albicans ( Lay et al . , 2003a , 2003b , 2012; van der Weerden et al . , 2008 , 2010; Hayes et al . , 2013 ) . NaD1 inhibits fungal growth in a three-stage process that involves specific interaction with the cell wall and entry into the cytoplasm before cell death ( van der Weerden et al . , 2008 , 2010 ) . Interaction with NaD1 also leads to hyper-production of reactive oxygen species , inducing oxidative damage that contributes to its fungicidal activity on Candida albicans ( Hayes et al . , 2013 ) . Many CAPs have been postulated to act at the level of the plasma membrane of target cells . Suggested mechanisms of action for membrane permeabilization are based on the ( 1 ) carpet , ( 2 ) barrel-stave , and ( 3 ) toroidal-pore models ( reviewed in Brogden , 2005 ) . In the carpet model , the CAPs act like classic detergents , accumulating and forming a carpet layer on the membrane outer surface , leading to local disintegration ( including membrane micellization or fragmentation ) upon reaching a critical concentration . Other CAPs are suggested to aggregate on the membrane surface before inserting into the bilayer forming a ‘barrel-stave’ pore where the hydrophobic peptide regions align with the lipid core and the hydrophilic peptide regions form the interior of the pore . Alternatively , in the toroidal pore model , the CAPs induce the lipid monolayers to bend continuously through the pore , with the polar peptide faces associating with the polar lipid head groups ( Brogden , 2005 ) . Although these models have been useful for describing potential mechanisms underlying the antimicrobial activity of various CAPs , it is not clear how well they represent the actual configuration of CAPs at the membrane . Furthermore , the oligomeric state of CAPs required for their activity based on the postulated models remains unknown . Indeed , it has long been hypothesized that the molecules could form proteinaceous pores and function through insertion into membranes ( Brogden , 2005 ) . However , to date , the structural basis of CAP activity at the target membrane has not been defined . In addition to the uncertainty about the configuration of CAPs at the membrane , the role of ligands in modulating the recognition of target surfaces by CAPs remains unclear . One class of ligands that has been linked to plant defensin antifungal activity are sphingolipids ( Wilmes et al . , 2011 ) , a key component of fungal cell walls and membranes . Plant defensins that bind sphingolipids include RsAFP2 from radish ( binds glucosylceramide , GlcCer ) ( Thomma et al . , 2003; Thevissen et al . , 2004 ) , DmAMP1 from dahlia ( binds mannose- ( inositol-phosphate ) 2-ceramide , M ( IP ) 2C ) ( Thevissen et al . , 2000 , 2003 ) , as well as the pea defensin Psd1 ( Goncalves et al . , 2012 ) and sugarcane defensin Sd5 ( de Paula et al . , 2011 ) that both bind membranes enriched for specific glycosphingolipids . MsDef1 , a defensin from Medicago sativa , has also been implicated in binding sphingolipids , as a mutant of the fungus Fusarium graminearum that is depleted in glucosylceramide , is highly resistant to MsDef1 ( Ramamoorthy et al . , 2007 ) . In this report , we have identified the cellular phospholipid phosphatidylinositol 4 , 5-bisphosphate ( PIP2 ) as a key ligand that is recognized during membrane permeabilization of fungal and mammalian plasma membranes . Using X-ray crystallography , we have defined the molecular interaction of NaD1 with PIP2 and demonstrate that NaD1 forms oligomeric complexes with PIP2 . Structure-guided mutagenesis revealed a critical arginine residue ( R40 ) that is pivotal for NaD1:PIP2 oligomer formation and that oligomerization is required for plasma membrane permeabilization . Engagement of PIP2 is mediated by NaD1 dimers that form a distinctive PIP2-binding ‘cationic grip’ that interacts with the head groups of two PIP2 molecules . Functional assays using NaD1 mutants reveal that the mechanism of membrane permeabilization by NaD1 is likely to be conserved between fungal and mammalian tumor cells . Together , these data lead to a new perspective on the role of ligand binding and oligomer formation of defensins during membrane permeabilization .
To define the molecular basis of NaD1 target cell membrane permeabilization activity , we set out to identify potential ligands for NaD1 . Membrane lipids represent an attractive target for NaD1; therefore , we investigated whether NaD1 interacts with cellular lipids using protein–lipid overlay assays based on lipid strips immobilized with 100 pmoles of various biologically active lipids ( Poon et al . , 2010; Patel et al . , 2013 ) . NaD1 specifically bound to certain phospholipids , including several phosphatidylinositol mono-/bis-/tri-phosphates , phosphatidylserine , phosphatidic acid , cardiolipin , and sulfatide ( Figure 1A ) . Interestingly , NaD1 bound the functionally important plasma membrane phospholipid PIP2 ( Figure 1A ) but did not bind to a panel of other membrane lipids or sphingolipids . To confirm that the ability of NaD1 to engage PIP2 was not a result of immobilization on the lipid strip , we confirmed that NaD1 also bound PIP2 in the context of a membrane bilayer using a liposome pull-down assay ( Figure 1B ) . 10 . 7554/eLife . 01808 . 003Figure 1 . Interaction of NaD1 with lipids . ( A ) Detection of NaD1 binding to cellular lipids by protein-lipid overlay assay . Blots are representative of at least two independent experiments for each strip . ( B ) Binding of NaD1 to PIP2-containing liposomes . NaD1 in A and B was detected using a rabbit anti-NaD1 antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 00310 . 7554/eLife . 01808 . 004Figure 1—figure supplement 1 . Relative binding of NaD1 to lipids . Quantitation by densitometry of the relative binding of NaD1 to lipids ( normalized to PtdIns ( 4 , 5 ) P2 in A and B , sulfatide in C ) on a ( A ) Membrane Lipid Strip , ( B ) PIP Strip , and ( C ) SphingoStrip . Data shown as mean ± SD ( n = 2 ) for A and C , and mean ± SEM ( n = 3 ) for B . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 004 To gain insight into the NaD1:PIP2 interaction at the atomic level , we determined the crystal structure of NaD1 in complex with PIP2 . The structure of monomeric NaD1 ( Lay et al . , 2012 ) was used to solve the structure of a NaD1:PIP2 complex by molecular replacement and refined to a resolution of 1 . 6 Å with values of Rwork/Rfree of 0 . 155/0 . 184 ( Table 1 ) . Upon PIP2 binding , NaD1 forms an arch composed of 14 NaD1 molecules ( Figure 2A ) , with a final arch diameter of 90 Å and a width of 35 Å . The asymmetric unit contains all 14 NaD1 molecules that form the final arch , with the symmetry of the arch being entirely non-crystallographic . Fourteen PIP2 molecules are bound in an extended binding groove ( Figure 2B ) on the inside of the arch ( Figure 2A ) . The entire oligomeric complex is held together by a complex network of interactions , which include numerous NaD1:NaD1 ( Figure 3A , B ) and NaD1:PIP2 interactions ( Figure 3C , D ) . Notably , the arch-shaped oligomer displays a small degree of pitch , which although noticeable is not sufficient to allow the formation of an extended coil in the crystal ( Figure 2B ) . 10 . 7554/eLife . 01808 . 005Table 1 . Data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 005NaD1:PIP2 nativeData collection Space groupC2221 Cell dimensions a , b , c ( Å ) 79 . 64 , 132 . 04 , 153 . 01 α , β , γ ( ° ) 90 . 00 , 90 . 00 , 90 . 00 Wavelength ( Å ) 0 . 9537 Resolution ( Å ) *40 . 84–1 . 6 ( 1 . 69–1 . 60 ) Rsym or Rmerge*0 . 092 ( 0 . 617 ) I/σI*11 . 6 ( 2 . 2 ) Completeness ( % ) *99 . 7 ( 94 . 7 ) Redundancy*6 . 7 ( 5 . 4 ) Refinement Resolution ( Å ) 40 . 37–1 . 6 No . reflections105745 Rwork/Rfree0 . 155/0 . 184 No . atoms Protein10326 Ligand/ion845 Water816 B-factors Protein21 . 5 Ligand/ion28 . 9 Water31 . 2 R . m . s . deviations Bond lengths ( Å ) 0 . 010 Bond angles ( ° ) 1 . 651*Values in parentheses are for highest resolution shell . 10 . 7554/eLife . 01808 . 006Figure 2 . Crystal structure of the NaD1:PIP2 complex . ( A ) Two orthogonal views of a cartoon representation of the NaD1:PIP2 oligomer comprising 14 NaD1 monomers ( shown as ribbons ) and 14 PIP2 molecules ( shown as green sticks ) . The surface of the NaD1 oligomer is shown in gray . ( B ) Surface representation of the NaD1 14-mer , displaying the extended binding groove on the inside of the arch . Coloring is by atom type ( N in blue , O in red , S in yellow , and C in gray ) . For clarity the 14 bound PIP2 molecules were omitted . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 00610 . 7554/eLife . 01808 . 007Figure 3 . Detailed view of the crystal structure of the NaD1:PIP2 complex . In all panels , hydrogen bonds and salt bridges are shown as black dotted lines . ( A ) View of the interface of two NaD1 monomers revealing the hydrogen bonding pattern , with monomer I shown in cyan and monomer II in magenta . Secondary structure elements are labeled in black . For clarity bound PIP2 molecules are omitted . ( B ) Cartoon diagram of four molecules of NaD1 forming a dimer of dimers . ( C ) PIP2 binding site on monomer I . Cartoon diagram of the PIP2 binding site in monomer I on dimeric NaD1 . ( D ) PIP2 binding site on monomer II . Cartoon diagram of the PIP2 binding site on monomer II on dimeric NaD1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 00710 . 7554/eLife . 01808 . 008Figure 3—figure supplement 1 . Cartoon of two NaD1 dimers with four bound PIP2 molecules . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 00810 . 7554/eLife . 01808 . 009Figure 3—figure supplement 2 . Simulated anneal omit map of a single PIP2 molecule bound to an NaD1 dimer , contoured at 1σ . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 009 The observed NaD1:PIP2 oligomer can be described as an assembly of seven NaD1 dimers , which comprise two distinct NaD1:NaD1 interfaces . The first interface is formed by an antiparallel alignment of the β1-strand from each of two NaD1 molecules ( monomers I and II ) and exhibits two-fold symmetry between the associated monomers ( Figure 3A ) . It comprises an average buried surface area of 430 Å2 and is formed by a network of six hydrogen bonds involving R1 , K4 , E6 , E27 , K45 , and C47 . This dimeric arrangement leads to the formation of a ‘cationic grip’ ( Figure 4A , B ) , which is able to accommodate two PIP2 head groups simultaneously ( Figure 3—figure supplement 1 ) . A second interface is formed by the dimeric NaD1 ( comprising monomers I and II ) and adjacent NaD1 monomers III and IV ( Figure 3B ) . This interface is formed by hydrogen bonds involving N8 of monomer I , R1 , E2 , K17; D31 of monomer II , R1 , K17; D31 of monomer III; and N8 of monomer IV , effectively forming a dimer of dimers ( Figure 3—figure supplement 1 ) . The interactions between two dimers are repeated seven times to allow formation of the observed 14-mer . The full 14-mer is thus constructed using two different interfaces . 10 . 7554/eLife . 01808 . 010Figure 4 . The dimeric NaD1 ‘cationic grip‘ with two bound PIP2 molecules . ( A ) Surface view in two orientations of a NaD1 dimer ( monomer I in cyan and monomer II in magenta ) with two bound PIP2 molecules ( yellow and green ) . ( B ) The same as in A except that the surface shows a qualitative electrostatic representation ( blue is positive , red in negative , and white is uncharged or hydrophobic ) . Figure generated using Pymol . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 010 In addition to NaD1:NaD1 interactions , oligomer formation requires the presence of PIP2 . NaD1 binds PIP2 primarily via a ‘cationic grip’ that is created by a NaD1 dimer , which results in the formation of a distinct binding site ( Figure 3—figure supplement 2 ) formed by K4 together with residues 33–40 , which comprise a characteristic ‘KILRR’ motif ( Figure 3C , D ) . PIP2 forms a dense network of hydrogen bonds involving K4 , H33 , K36 , I37 , L38 , and R40 of a single NaD1 monomer . In oligomeric NaD1:PIP2 , a single PIP2 binding site also contains interactions with neighboring NaD1 monomers ( Figure 3C , D; Figure 3—figure supplement 1 ) . Bound PIP2 forms additional hydrogen bonds with R40 from monomer II and K36 from monomer IV′ , with the full PIP2 binding site in the oligomer comprising contributions from three different NaD1 molecules ( Figure 3C , D ) . Consequently , oligomer formation appears to be highly cooperative , with multiple interactions between adjacent NaD1 and PIP2 molecules required to form the observed 14-mer ( Figure 3 ) . To confirm that oligomer formation is not a crystallization artifact , we treated mixtures of NaD1 and PIP2 in aqueous solution with the crosslinker BS3 , which resulted in covalent cross-linking of multiple NaD1 molecules that occurred only in the presence of PIP2 ( Figure 5A ) , whereas NaD1 on its own only formed a dimer as reported previously ( Lay et al . , 2012 ) . 10 . 7554/eLife . 01808 . 011Figure 5 . NaD1 forms oligomers with PIP2 . ( A ) Ability of NaD1 to form multimers in the presence of PIP2 as determined by protein–protein cross-linking with BS3 followed by SDS-PAGE and Coomassie Brilliant Blue staining . ( B ) TEM of NaD1:PIP2 complexes . TEM micrographs of NaD1 alone , PIP2 alone , or NaD1 in complex with PIP2 . Data in A and B are representative of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 011 We next imaged NaD1:PIP2 oligomers using transmission electron microscopy ( TEM ) . Complexes of NaD1:PIP2 ( 1:1 . 2 molar ratio ) were applied to a carbon-coated copper grid and imaged . Strikingly , long string-like fibrillar structures were observed when both NaD1 and PIP2 were present , whereas they were absent on grids bearing either NaD1 or PIP2 in isolation ( Figure 5B ) . Although the NaD1:PIP2 oligomer we observed by crystallography displays a subtle pitch , it is not sufficient to allow continuous addition or concatenation of 14-mers to form the fibrils observed by TEM , with the ends of two 14-mers running into each other . However , given that the crystal structure of the oligomer reveal an outer diameter of 90 Å , with a corresponding diameter of the fibrils under TEM of 10 nM , additional twisting of the 14-mer could allow for the formation of continuous coils with a diameter to match the fibrils observed under TEM . Based on our oligomeric NaD1:PIP2 structure , we performed site-directed mutagenesis on NaD1 to confirm the role of proposed key amino acid residues in PIP2 binding , oligomerization , and fungal cell killing . Examination of the PIP2 binding pockets in the NaD1:PIP2 oligomer suggests that R40 , which contacts two adjacent PIP2 molecules simultaneously and interacts with the phosphate moiety at position 4 , is critical for cooperative binding of PIP2 and therefore formation of the NaD1:PIP2 oligomer ( Figure 6A ) . Mutation of R40 should not lead to loss of PIP2 binding , since PIP2 would still form five hydrogen bonds and ionic interactions with NaD1 and should only impact oligomerization . In contrast , I37 contributes to PIP2 binding but not oligomerization . We generated recombinant proteins of NaD1 ( rNaD1 ) and NaD1 mutants ( rNaD1 ( R40E ) and rNaD1 ( I37F ) ) and confirmed their correct folding by CD spectroscopy ( data not shown ) and evaluated the ability of the mutant NaD1 to bind phospholipids , undergo PIP2-induced oligomerization , and kill the filamentous fungus F . oxysporum f . sp . vasinfectum . 10 . 7554/eLife . 01808 . 012Figure 6 . Multimerization of the NaD1:PIP2 complex . ( A ) Schematic representation of residues from neighboring NaD1 monomers involved in binding two PIP2 molecules . Ability of rNaD1 , rNaD1 ( R40E ) , and rNaD1 ( I37F ) to ( B ) bind cellular lipids by protein-lipid overlay assay , ( C ) form multimers in the presence of PIP2 as determined by protein–protein cross-linking with BS3 followed by SDS-PAGE and Coomassie Brilliant Blue staining , and ( D ) to inhibit fungal cell growth . Error bars in D indicate SEM ( n = 3 ) . Data in B–D are representative of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 01210 . 7554/eLife . 01808 . 013Figure 6—figure supplement 1 . Relative binding of rNaD1 , rNaD1 ( I37F ) , and rNaD1 ( R40E ) to lipids . Quantitation by densitometry of the relative binding ( normalized to PtdIns ( 4 , 5 ) P2 ) of ( A ) rNaD1 , ( B ) rNaD1 ( I37F ) , and ( C ) rNaD1 ( R40E ) to lipids on PIP Strips . Data shown as mean ± SD ( n = 2 ) for A and B , and mean ± SEM ( n = 3 ) for C . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 013 As predicted , mutation of R40 to glutamic acid led to a largely unchanged binding to PIP2 , with the remaining five hydrogen bonds and ionic interactions formed between PIP2 and rNaD1 ( R40E ) compensating for the loss of two ionic interactions as well as the charge repulsion . However , it did result in reduced binding to PI ( 4 ) P ( Figure 6B ) and oligomerization ( Figure 6C ) that correlated with substantially reduced fungal cell killing ( Figure 6D ) . It is important to note that although PIP2 binding was maintained , the loss of interaction with the 4-phosphate moiety of PIP2 results in loss of cooperative binding and therefore ablation of oligomerization , which is critically dependent on R40 forming ‘bridging’ interactions between two neighboring PIP2 molecules . In contrast , mutation of I37 to phenylalanine had little effect on PIP2 binding specificity , oligomerization , and fungal cell killing ( Figure 6B–D ) . These data support our defined NaD1-PIP2 structure and demonstrate that the coordinated oligomerization of NaD1 by interaction with PIP2 is an important event during fungal cell killing . Since PIP2 is a critical component of mammalian plasma membranes , we investigated whether NaD1 also harbored permeabilization activity against mammalian cells . To this end , we performed a flow cytometry-based cell permeabilization assay on U937 monocytic lymphoma cells to measure uptake of the membrane-impermeable nucleic acid dye propidium iodide ( PI ) . NaD1 permeabilized the plasma membrane of the U937 cells and induced a change in cell morphology in a concentration-dependent manner ( Figure 7A ) . Furthermore , rapid leakage of intracellular ATP from U937 cells was observed within the first 200 s following exposure to NaD1 ( Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 01808 . 014Figure 7 . NaD1 kills mammalian tumor cells by membrane permeabilization . ( A ) Forward scatter , side scatter , and PI uptake analysis of U937 cells treated with NaD1 . ( B ) Binding of FITC-dextran and ( C ) LDH release by NaD1-treated U937 cells . Error bars in C indicate SEM ( n = 3 ) . Data in A–C are representative of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 01410 . 7554/eLife . 01808 . 015Figure 7—figure supplement 1 . NaD1 rapidly permeabilizes U937 cells . Ability of NaD1 ( 0 . 5–20 μM ) to mediate the release of ATP from U937 cells was examined using the ATP bioluminescence assay . The energy-dependency of the light-emitting luciferase-catalyzed oxidation of luciferin was used to indirectly measure the amount of ATP released by permeabilized cells over 30 min . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 01510 . 7554/eLife . 01808 . 016Figure 7—figure supplement 2 . Reduced and alkylated NaD1 ( NaD1R&A ) does not permeabilize U937 cells . Ability of NaD1R&A to permeabilize U937 cells was investigated using the ( A ) PI uptake , ( B ) LDH release , and ( C ) ATP bioluminescence assays , with native NaD1 being included as a positive control . Error bars in A and B indicate SEM ( n = 3 ) . Data in A–C are representative of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 01610 . 7554/eLife . 01808 . 017Figure 7—figure supplement 3 . Tumor/transformed cells are more susceptible to killing by NaD1 than normal/primary cells . MTT cell viability assays were performed on the indicated cell lines . The IC50 values are shown as mean ± SD ( n≥ 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 017 NaD1-mediated lysis of U937 cells was confirmed by the uptake of FITC-dextran ( up to 40 kDa ) and the release of lactate dehydrogenase ( 140 kDa ) into the supernatant after NaD1 treatment ( Figure 7B , C ) . In contrast , reduced and alkylated NaD1 ( NaD1R&A ) showed no cytotoxic activities against U937 cells ( Figure 7—figure supplement 2 ) , confirming the importance of the NaD1 tertiary structure for the ability to induce membrane permeabilization . It should be noted that NaD1 also permeabilized a diverse range of normal primary human cells and tumor cell lines ( Figure 7—figure supplement 3 ) , with the highest levels of activity exhibited against tumor cell lines . Collectively , these data suggest that , in addition to antifungal activity , NaD1 also exhibits antiproliferative properties against mammalian cells . We then sought to examine changes in cell morphology upon NaD1 treatment . Live confocal laser scanning microscopy ( CLSM ) revealed rapid changes on the cell surface of NaD1-permeabilized tumor cells and showed the formation of large plasma membrane blebs , with adherent cells ( HeLa and PC3 ) forming multiple blebs of different sizes ( Video 1 ) and non-adherent cells ( U937 ) forming typically one to two large blebs ( Video 2; Figure 8A ) . Moreover , bleb size was frequently larger than the actual cell ( diameter >20 μm ) and did not retract over a period of 20 min ( Figure 8—figure supplement 1 ) . 10 . 7554/eLife . 01808 . 018Video 1 . NaD1 rapidly induces membrane blebbing and permeabilization of HeLa cells . Live CLSM of PKH67-stained HeLa cells in the presence of PI . Cells were imaged over a period of 10 min ( 5 s/frame ) , with NaD1 ( 10 μM ) being added to cells at 1 min . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 01810 . 7554/eLife . 01808 . 019Video 2 . Formation of a single large membrane bleb on a U937 cell following NaD1 treatment . Three-dimensional reconstruction of CLSM images of a NaD1-treated ( 10 μM ) PKH67-/PI-stained U937 cell . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 01910 . 7554/eLife . 01808 . 020Figure 8 . NaD1 induces membrane blebbing of tumor cells . ( A ) CLSM of PKH67-stained NaD1 ( 10 μM ) permeabilized HeLa and U937 cells . ( B ) CLSM of U937 cells treated with NaD1 ( 20 μM ) in the presence of PI and 4 kDa FITC-dextran . Arrows indicate entry of PI . It should be noted that in this experiment the detector gain on the helium–neon laser ( red channel ) was increased compared to that used in A to enable visualization of the cell lysis events . Scale bars represent 10 μm . Data in A and B are representative of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 02010 . 7554/eLife . 01808 . 021Figure 8—figure supplement 1 . NaD1-induced membrane blebs do not retract once U937 cells are permeabilized . Live confocal laser scanning microscopy ( CLSM ) of NaD1-treated U937 cells in the presence of PI . Cells were imaged over a period of 40 min , with NaD1 ( 10 μM ) being added at 7 min . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 02110 . 7554/eLife . 01808 . 022Figure 8—figure supplement 2 . NaD1-mediated membrane permeabilization occurs at the blebs of PC3 cells . Live CLSM of PC3 cells treated with NaD1 in the presence of PI and 4 kDa FITC-dextran . Cells were imaged over a period of 10 min ( 5 s/frame ) , with NaD1 ( 20 μM ) , and 4 kDa FITC-dextran ( 100 μg/ml ) being added to cells at 30 s . It should be noted that individual PC3 cells show a variable rate of permeabilization by NaD1 reflecting a kinetic effect but eventually all cells are permeabilized under the experimental conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 022 In our CLSM studies , we noticed that NaD1-induced membrane blebbing typically coincided with PI uptake ( Video 1 ) . To determine whether membrane blebbing occurs prior to , during , or following membrane permeabilization , we treated U937 cells with NaD1 in the presence of PI and 4 kDa FITC-dextran to monitor the entry of these molecules into NaD1-sensitive cells ( Figure 8B; Video 3 ) . FITC-dextran and NaD1 were added at 00:35 min , with FITC-dextran being excluded from cells with an intact membrane . Bleb formation was first observed for the cell located at the center of the panel at 03:25 min , with PI staining appearing at a specific point at the edge of the bleb . From 03:25 to 04:15 min , PI staining was observed in the bleb and the cytoplasm with FITC-dextran also entering the cell from the bleb site . At 04:20 min , PI-stained molecules were ‘expelled’ out of the cell , possibly at the same region that PI first entered the bleb . Similar results were also observed for PC3 cells ( Figure 8—figure supplement 2 ) . These data suggest that ( i ) small molecules such as PI can enter the cell initially at a ‘weakened’ point at the membrane bleb , ( ii ) the bleb continues to enlarge while PI and 4 kDa FITC-dextran enters , and ( iii ) intracellular contents are released at the bleb site , representing cytolysis . 10 . 7554/eLife . 01808 . 023Video 3 . NaD1-mediated membrane permeabilization occurs at the blebs of U937 cells . Live CLSM of U937 cells treated with NaD1 in the presence of PI and 4 kDa FITC-dextran . Cells were imaged over a period of 10 min ( 5 s/frame ) , with NaD1 ( 20 μM ) and 4 kDa FITC-dextran ( 100 μg/ml ) being added to cells at 30 s . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 023 We next determined the specific mechanism by which NaD1 permeabilizes mammalian cells . Firstly , we tested the binding of BODIPY-labeled NaD1 to U937 cells . BODIPY-NaD1 permeabilized U937 cells at a level comparable to unlabeled NaD1 and bound to both viable ( 7AAD-negative ) and permeabilized ( 7AAD-positive ) cells , with more BODIPY-NaD1 bound to membrane-damaged cells ( Figure 9A ) . These data suggest that NaD1 can interact with U937 cells prior to membrane permeabilization and accumulates on/within NaD1-sensitive cells . 10 . 7554/eLife . 01808 . 024Figure 9 . Subcellular localisation of BODIPY-NaD1 in tumor cells . ( A ) Detection of BODIPY-NaD1 binding to viable and permeabilized U937 cells by flow cytometry . ( B ) CLSM of subcellular localization of BODIPY-NaD1 ( 10 μM ) on permeabilized U937 , PC3 , and HeLa cells . Scale bars represent 10 μm . Data in A and B are representative of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 024 We then determined the subcellular localization of BODIPY-NaD1 on permeabilized tumor cells . BODIPY-NaD1 accumulated at membrane bleb ( s ) , in the cytoplasm and nucleolus and possibly at certain cytoplasmic organelles in U937 , PC3 , and HeLa cells ( Figure 9B; Videos 4 and 5 ) . It is worth noting that the formation of large plasma membrane blebs has been reported previously in mammalian cells in which physically- or chemically-induced detachment of the plasma membrane from the actin cortex had occurred , including through enzymatic modification or sequestration of the inner membrane phospholipid , PIP2 ( Niebuhr et al . , 2002; Sheetz et al . , 2006; Keller et al . , 2009 ) . 10 . 7554/eLife . 01808 . 025Video 4 . BODIPY-NaD1 accumulates at the plasma membrane and certain intracellular organelles . Three-dimensional reconstruction of CLSM images of BODIPY-NaD1-treated ( 10 μM ) PI-stained PC3 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 02510 . 7554/eLife . 01808 . 026Video 5 . BODIPY-NaD1 accumulates at the plasma membrane and certain intracellular organelles . Three-dimensional reconstruction of CLSM images of BODIPY-NaD1-treated ( 10 μM ) PI-stained HeLa cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 026 We then asked whether the binding of NaD1 to PIP2 at the inner leaflet of the plasma membrane could lead to the formation of blebs , as PIP2 is a key mediator of cytoskeleton-membrane interactions ( Raucher et al . , 2000; Sheetz , 2001 ) . HeLa cells expressing GFP-PH ( PLCδ ) , which binds specifically to PIP2 , were treated with NaD1 and showed a marked delay from the initiation of blebbing ( rapid small membrane blebbing ) to membrane permeabilization compared with cells expressing free GFP ( Figure 10; Videos 6 and 7 ) . Quantitation of the kinetics of cell permeabilization indicated that GFP-PH ( PLCδ ) -expressing cells took approximately 2 . 5 times as long as GFP only cells to permeabilize in response to NaD1 ( 235 ± 40 s vs 90 ± 20 s , respectively ) ( Figure 10B ) . These results suggest that the expression of GFP-PH ( PLCδ ) may compete with NaD1 for PIP2 binding at the inner leaflet of the plasma membrane and interfere with NaD1-induced cell permeabilization . 10 . 7554/eLife . 01808 . 027Figure 10 . Expression of GFP-PH ( PLCδ ) in HeLa cells significantly delays NaD1-mediated cell permeabilization compared with cells expressing free GFP . ( A ) CLSM of NaD1 ( 10 μM ) treated HeLa cells expressing GFP-PH ( PLCδ ) . Scale bars represent 10 μm . ( B ) The average length of time taken for NaD1 ( 10 μM ) to permeabilize ( PI-positive ) GFP-PH ( PLCδ ) -expressing vs free GFP-expressing HeLa cells were analyzed over a period of 15 min . For GFP-PH ( PLCδ ) -expressing cells , n = 21; for free GFP-expressing cells , n = 29 . Error bars indicate SEM , * = p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 02710 . 7554/eLife . 01808 . 028Video 6 . Expression of GFP-PH ( PLCδ ) in HeLa cells delays NaD1-mediated cell permeabilization . Live CLSM of GFP-PH ( PLCδ ) transfected HeLa cells treated with NaD1 in the presence of PI . Cells were imaged over a period of 15 min ( 5 s/frame ) , with NaD1 ( 10 μM ) being added to cells at 30 s . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 02810 . 7554/eLife . 01808 . 029Video 7 . Expression of free GFP in HeLa cells does not delay NaD1-mediated cell permeabilization . Live CLSM of free GFP transfected HeLa cells treated with NaD1 in the presence of PI . Cells were imaged over a period of 20 min ( 5 s/frame ) , with NaD1 ( 10 μM ) being added to cells at 30 s . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 029 We then evaluated the effect of our loss-of-function mutant rNaD1 ( R40E ) , which exhibited reduced killing activity of fungal cells , on U937 cells . rNaD1 ( R40E ) showed dramatically reduced ability to permeabilize these tumor cells , in contrast to the control rNaD1 and rNaD1 ( I37F ) proteins as demonstrated by the PI uptake assay ( Figure 11A ) . Similarly , rNaD1 ( R40E ) -treated U937 cells displayed no LDH release ( Figure 11B ) or FITC-dextran uptake ( Figure 11C ) compared to the control proteins . 10 . 7554/eLife . 01808 . 030Figure 11 . Permeabilization of U937 cells is impaired in rNaD1 ( R40E ) . Ability of rNaD1 , rNaD1 ( R40E ) and rNaD1 ( I37F ) to permeabilize U937 cells as assessed by ( A ) PI uptake , ( B ) LDH release , and ( C ) FITC-dextran binding assays . Error bars in A and B indicate SEM ( n = 3 ) . Data in A–C are representative of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 030 In their totality , these data support our notion that the coordinated oligomerization of NaD1 by interaction with PIP2 is a critical event for fungal and tumor cell killing .
CAPs , of which defensins are a major family , are weapons within the armory of host defense peptides that are utilized by animals and plants in their fight against pathogenic threats . NaD1 is a defensin from the ornamental tobacco that has potent activity against fungi and yeast ( Lay et al . , 2003a , 2012; van der Weerden et al . , 2008 , 2010; Hayes et al . , 2013 ) . This defensin operates by a three-step mechanism—specific interaction with the cell wall followed by permeabilization of the plasma membrane and entry into the cytoplasm ( van der Weerden et al . , 2010 ) . However , the precise molecular basis of membrane permeabilization and passage through the membrane is poorly defined , particularly in terms of the specific lipid targets on the membrane and the structural definition of defensin-lipid interactions . In this study , we identified the lipid targets of NaD1 as phosphoinositides ( PIPs ) and show that the binding of particular PIPs such as PIP2 , mediates NaD1 oligomerization and membrane permeabilization . This expands on our previous study revealing that the ability of NaD1 to homo-dimerize enhances its antifungal activity ( Lay et al . , 2012 ) . It has been postulated that the ability of human defensins and other CAPs to form higher oligomeric states at the plasma membranes of target cells is a contributing factor in membrane disruption and/or permeabilization ( Hill et al . , 1991; Wimley et al . , 1994; Hoover et al . , 2000; Mader and Hoskin , 2006; Wei et al . , 2010 ) , but to date no structural explanations have been reported . Our crystal structure of a NaD1:PIP2 oligomeric complex reveals the first detailed molecular description of a plant defensin-lipid interaction and identifies a new mechanism of membrane permeabilization . The proposed antifungal mechanisms of action for plant defensins are diverse and includes membrane permeabilization , generation of reactive oxygen species with induction of apoptosis , and dysregulation of Ca2+ influx and K+ efflux ( reviewed in De Coninck et al . , 2013; van der Weerden and Anderson , 2013 ) . The structural basis of the interaction of these defensins with lipid has not been defined . However , it is interesting to note that the equivalent region to the ‘KILRR’ loop of NaD1 ( the loop between the β2 and β3 strands ) , that is critical in forming the lipid-binding ‘cationic grip’ , has been implicated as functionally important for the antifungal activity of a number of plant defensins . For example , mutations in the equivalent region of RsAFP2 abolished antifungal activity ( De Samblanx et al . , 1997 ) , and a chimeric protein generated by replacing this region of MsDef1 with the equivalent region of MtDef4 ( a functionally distinct defensin that does not bind sphingolipid ) resulted in functional conversion into a defensin able to inhibit the growth of a glucosylceramide-deficient , MsDef1-resistant F . graminearum strain ( Sagaram et al . , 2011 ) . Based on our observation that NaD1 binds phosphoinositides through the same loop region , it is tempting to speculate that plants have evolved a suite of defensins that specifically interact through their β2–β3 loop regions with different membrane lipids to mediate fungal cell killing . The β2–β3 region has also been implicated in the α-amylase activity of the plant defensin , VrD1 , suggesting that this region is also functionally important in defensins with different activities ( Lin et al . , 2007 ) . Not surprisingly , the β2–β3 region of plant defensins exhibits considerable sequence divergence which is likely to reflect their ability to bind different ligands and therefore the various mechanisms of antifungal activity ( van der Weerden and Anderson , 2013 ) . The antifungal plant defensins , DmAMP1 and RsAFP2 , interact with different sphingolipids . DmAMP1 binds M ( IP ) 2C ( Thevissen et al . , 2000 , 2003 ) and S . cerevisiae strains that have gene disruptions to encoded proteins within the M ( IP ) 2C biosynthetic pathway are rendered DmAMP1-resistant ( Thevissen et al . , 2005 ) . In contrast , RsAFP2 binds to GlcCer , present in fungi such as Pichia pastoris and C . albicans and does not cause permeabilization or growth inhibition of strains that do not express GlcCer ( Thevissen et al . , 2004 ) . The fact that DmAMP1 is able to act on these strains supports the notion that DmAMP1 and RsAFP2 have different lipid targets . It is worthwhile noting that RsAFP2 does not bind to GlcCer derived from human or soybean ( Thevissen et al . , 2004 ) , suggesting that there is some level of selectivity , while its inability to permeabilize artificial liposomes containing GlcCer indicates that binding alone is insufficient for its permeabilization action ( Thevissen et al . , 2004 ) . This is the first report of a defensin from any species targeting a phosphoinositide , such as PIP2 . Interestingly , a number of toxins have been reported to directly or indirectly target PIP2 , resulting in plasma membrane reorganization and permeabilization . The marine bacterium Vibrio parahaemolyticus causes gastroenteritis in humans by acting as an inositol polyphospholipid 5′ phosphatase . It targets PIP2 by catalyzing the removal of the 5′ phosphate moiety , resulting in the disruption of PIP2-mediated cytoskeletal interactions leading to membrane blebbing and cell lysis ( Broberg et al . , 2010 ) . A similar mode of action has also been described for the effector protein ipgD of the bacillary dysentery causing Gram-negative pathogen Shigella flexneri ( Niebuhr et al . , 2002 ) . The equinatoxin from the sea anemone Actinia equina also causes membrane blebbing and cell lysis via a mechanism involving Ca2+-mediated PIP2 hydrolysis at the inner membrane and the formation of membrane pores in target cells ( Garcia-Saez et al . , 2011 ) . In this study , we show that NaD1 also causes plasma membrane blebbing and permeabilization of human cells . However , in contrast to the enzymatic or pore-forming mechanism ( s ) of the above toxins , it does so through the direct binding of PIP2 . The ability of NaD1 to form an oligomeric complex with PIP2 suggests it could potentially sequester PIP2 from the plasma membrane , leading to membrane destabilization , blebbing , and ultimately cell lysis . Such an oligomerization process would be predicted to efficiently displace any PIP2-binding molecules . Indeed , the ability of lipid-binding proteins to oligomerize on membranes has been proposed as an important mechanism in mediating high-avidity membrane interactions ( Lemmon , 2008 ) . The increased sensitivity of tumor cells to NaD1 compared with its effects on healthy primary cells may be attributed to a number of differences in the physical properties of the plasma membranes of these two cell types . These include an increase in the expression of negatively charged outer membrane components , such as O-glycosylated mucins ( Yoon et al . , 1996; Kufe , 2009 ) and phosphatidylserine ( Utsugi et al . , 1991; Ran et al . , 2002 ) , which could allow stronger initial electrostatic interactions between the cationic NaD1 and the cell surface . The increased levels of microvilli ( Chaudhary and Munshi , 1995; Ino et al . , 2002 ) and higher degree of membrane fluidity ( Sok et al . , 1999; Zeisig et al . , 2007 ) in tumor cells may also facilitate the aggregation of a greater number of NaD1 molecules to the cell surface and assist in penetration and/or destabilization of the membrane , respectively . The precise mechanism of action for the increased sensitivity of mammalian tumor cells over normal cells to defensins and whether this approach can be harnessed for selective tumor cell killing remains to be determined . The ability of NaD1 to permeabilize both fungal and mammalian cell membranes suggests a common mode of interaction with membranes and our findings described herein implicate PIP2 in both settings . Although the plasma membranes on mammalian and fungal cells are different in overall lipid compositions ( typically mammalian cells are rich in zwitterionic phospholipids whereas fungal membrane have higher levels of anionic phospholipids ) , PIP2 is important in both species ( Di Paolo and De Camilli , 2006; van Meer et al . , 2008 ) . PIP2 is normally found on the inner leaflet of mammalian cells and although it is a minor species comprising only 0 . 5–1% of phospholipids , it plays a major regulatory role in a number of important membrane-related processes , including signal transduction , ion channel function , and cytoskeletal attachment ( McLaughlin and Murray , 2005 ) . Similar functions for PIP2 have been reported or suggested in the plasma membrane of fungi ( van Meer et al . , 2008 ) . The importance of PIP2 in many fundamental cellular processes would certainly make it an attractive target for defense against pathogens . In addition to binding PIP2 , we show that NaD1 is also able to bind to a number of other phospholipids . The functional consequences of the promiscuous binding by NaD1 remains to be determined . The various phospholipids recognized by NaD1 have different subcellular distributions and functions . In its normal physiological setting as a plant antifungal molecule , the ability of NaD1 to bind a number of different phospholipids may enable defense against a wide array of different fungal pathogens . How NaD1 is able to enter cells remains to be defined . It is possible that the binding of PIPs such as PI ( 3 ) P may mediate entry of NaD1 into cells . Indeed , cell surface-expressed PI ( 3 ) P has been suggested to mediate entry of eukaryotic pathogen effector molecules , such as Avr from oomycetes and fungi , into plant cells ( Kale et al . , 2010 ) . Structurally , one can envisage that PIPs other than PIP2 can fit into the cationic grip and be able to induce NaD1 oligomerization . Inspection of the cationic grip indicates that an additional phosphate group at the 3-inositol position can be accommodated and would contact R40 in a manner similar to the phosphate in the 4-inositol position ( Figure 12 ) and may induce oligomerization in a manner analogous to the 4-phosphate . In contrast , a phosphate at the 6-inositol position is unlikely to be tolerated without reorganization of the binding site due to a steric clash with S35 . Overall , a combination of phosphate groups on the inositol ring in position 5 , together with an additional phosphate in position 3 or 4 , appears to be capable of oligomerization . Furthermore , PIPs harboring phosphate groups in positions 3 , 4 , and 5 should be well tolerated in the grip . 10 . 7554/eLife . 01808 . 031Figure 12 . PIP2 fit into the NaD1 ‘cationic grip’ . Cut-away of PIP2 bound in the NaD1 cationic grip . NaD1 dimer surface is shown in gray , PIP2 in green and select NaD1 residues are shown as stick representation . NaD1 monomeric chains are colored in gray and salmon . Carbon atoms in the inositol ring are numbered ( 1–6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 031 Our structure of the NaD1:PIP2 complex suggests that formation of the cationic grip via a NaD1 dimer is important for the formation of a functional PIP2 binding site . Notably , this dimeric arrangement is different to the dimers we previously observed for NaD1 alone ( Lay et al . , 2012 ) . Although one of the two dimer configurations observed was based on an interface formed by two opposing β1 strands , the overall configuration places the β2–β3 loops at opposite ends of the dimer ( Lay et al . , 2012 ) , so that the ‘cationic grip’ is absent . Consequently , a substantial reorientation is required to convert the NaD1 dimer observed in the absence of PIP2 into a configuration that can engage PIP2 and oligomerize . Although CAP:lipid interactions ( e . g . , nisin:lipid II , plectasin:lipid II ) have been investigated using structural methods ( Hsu et al . , 2004; Schneider et al . , 2010 ) , the formation of defensin:ligand oligomers as observed for NaD1:PIP2 is unique . NaD1 engages PIP2 in a cooperative manner where three NaD1 monomers form a complete PIP2 binding site , with each NaD1 monomer participating in the formation of three distinct PIP2 binding sites . This ability to participate in the engagement of multiple PIP2 molecules is mediated by two key basic residues , K4 and R40 . The lantibiotic peptide nisin binds specifically to the membrane-bound cell wall precursor lipid II [undecaprenyl-pyrophosphoryl-MurNAc- ( pentapeptide ) -GlcNAc] , leading to pore formation and permeabilization of the bacterial cytoplasmic membrane ( Sahl and Bierbaum , 1998; Wiedemann et al . , 2001 ) . Structural studies have revealed that nisin forms a 1:1 complex with lipid II ( Hsu et al . , 2004 ) , however the precise contribution of lipid II binding to plasma membrane permeabilization is not fully established . Interestingly , the report on the nisin:lipid II structure indicated that nisin precipitated upon lipid II addition and the structure could only be determined by dissolving the nisin:lipid II complex in DMSO ( Hsu et al . , 2004 ) . It is conceivable that nisin also forms oligomers that were present in the precipitated sample that was discarded . The fungal defensin plectasin is another antimicrobial peptide that requires lipid II binding for its biological activity ( Schneider et al . , 2010 ) . NMR spectroscopy was used to map a putative lipid II binding site on plectasin , however the interaction appears again to be a 1:1 complex , with no suggestion of oligomer formation . Furthermore , plectasin does not permeabilize target microbial membranes and acts by directly targeting bacterial cell wall biosynthesis ( Schneider et al . , 2010 ) . Amongst the many described phospholipid-binding proteins , a number of different domain structures have been defined that exhibit stereospecific recognition of specific phosphoinositide head groups in the context of cellular membrane surfaces . These include the pleckstrin homology ( PH ) , ‘Fab1 , YOTB , Vac1 , EEA1’ ( FYVE ) , and Phox-homology ( PX ) domains ( Bravo et al . , 2001; Lemmon , 2008 ) , with all three showing structural similarities to that we describe for PIP2 binding by NaD1 . In each of these domains , phosphoinositide binding is mediated through pockets that are strategically lined with basic residues formed by two distal β loops . Pleckstrin homology domains consist of two perpendicular antiparallel β-sheets followed by a C-terminal amphipathic helix , with the canonical phosphoinositide binding pocket comprising basic residues derived from two distal β-loops on the two β-sheets ( Lietzke et al . , 2000 ) . PX domains contain a triple-stranded antiparallel β-sheet followed by a helical subdomain made up of four α-helices , with the β1 and β2 loops together with α-helix 3 forming a positively-charged pocket responsible for binding phosphoinositides ( Bravo et al . , 2001 ) . FYVE domains are small cysteine-rich Zn2+ binding domains , consisting of two β-strands followed by a small C-terminal α-helix , with the binding pocket for the inositol head group of PI ( 3 ) P comprising basic clusters at either end of the β1 strand , one of which possesses a common ( R/K ) ( R/K ) HHCR motif ( Kutateladze and Overduin , 2001 ) . The topology of the phosphoinositide binding pocket formed by the NaD1 dimer is similar to that of the PH domains , however the formation of the pocket is unique and involves the symmetrical juxtaposition of identical β2–β3 loops ( comprised of KILRR ) by dimerization of two NaD1 monomers . Thus , despite the very different overall folds of NaD1 and the other phosphoinositide binding domains , key structural features that govern how they recognize phosphoinositides are generally conserved . It is of interest that a recent report describes the ability of human α-defensin 6 to self-assemble into high-order oligomers termed nanonets ( Chu et al . , 2012 ) . In contrast to NaD1:PIP2 oligomers , these fibril-like structures appear to form without involvement of defined extrinsic ligands and rely on stochastic binding to bacterial surface proteins to initiate self-assembly . Although composed of defensins , these nanonets do not harbor direct antibacterial activity per se , but rather act by trapping bacteria to prevent cellular adhesion and invasion ( Chu et al . , 2012 ) . Together with our findings , these studies indicate that defensins are able to form different fibrils or oligomers for diverse functions in innate immunity . Certain CAPs and cell penetrating peptides have been reported to penetrate biological membranes with or without membrane permeabilization ( Henriques et al . , 2006; Ashida et al . , 2011 ) . Although it is yet to be elucidated how NaD1 could pass through the plasma membrane to interact with phospholipids at the inner leaflet , our data provide several significant insights . NaD1 permeabilizes mammalian cells by forming a novel phosphoinositide recognition complex associated with the formation of membrane blebs and membrane rupture , possibly involving disruption of cytoskeleton-membrane interactions through the binding of PIP2 at the inner leaflet ( Figure 13 ) . This novel mechanism of cell lysis is distinct from that proposed for other CAPs that act via pore formation or a non-specific charge-based interaction with the plasma membrane ( Brogden , 2005 ) and the well-defined pore-forming ability of cholesterol-dependent cytolysins ( Rossjohn et al . , 1997 ) , perforin ( Law et al . , 2010 ) or complement membrane attack complex ( Hadders et al . , 2007 ) . These findings not only reveal a new mechanism of cell lysis but also uncover a potential evolutionarily conserved innate defense mechanism that can target cell membranes through the recognition of a ‘phospholipid pattern/code’ . 10 . 7554/eLife . 01808 . 032Figure 13 . Proposed molecular mechanism of NaD1-mediated tumor cell lysis . Schematic representation of NaD1-induced membrane blebbing and permeabilization . The assembly of NaD1:PIP2 oligomer can potentially be formed by ( i ) sequential recruitment of a NaD1 monomer followed by a PIP2 molecule or ( ii ) dimerization of two single NaD1:PIP2 complex followed by the recruitment of NaD1:PIP2 dimers . DOI: http://dx . doi . org/10 . 7554/eLife . 01808 . 032
NaD1 was isolated from its natural source , whole Nicotiana alata flowers , as described previously ( van der Weerden et al . , 2008 ) . Reduction and alkylation of NaD1 was performed as described previously ( van der Weerden et al . , 2008 ) . BODIPY labeling of NaD1 was performed as described previously ( van der Weerden et al . , 2010 ) . NaD1 binding to lipids spotted on Membrane Lipid Strips , PIP Strips or SphingoStrips was performed according to manufacturer’s instructions ( Echelon Biosciences , Salt Lake City , UT ) . Briefly , lipid strips were incubated with PBS/3% fatty acid-free BSA for 120 min at RT to block non-specific binding . The lipid strips were then incubated with NaD1 or NaD1 mutants ( 1 μg/ml ) diluted in PBS/1% fatty acid-free BSA for 60 min at 4°C and washed thoroughly for 60 min at RT with PBS/0 . 1% Tween-20 . Membrane-bound NaD1 was detected by probing the lipid strips with 2 μg/ml of a protein A purified rabbit anti-NaD1 antibody ( van der Weerden et al . , 2008 ) diluted in PBS/1% fatty acid-free BSA for 60 min at 4°C , followed by a HRP-conjugated donkey anti-rabbit Ig antibody ( GE Healthcare , Buckinghamshire , United Kingdom ) diluted to 1:2000 in PBS/1% fatty acid-free BSA for 60 min at 4°C . After each antibody incubation , the lipid strips were washed for 30 min at RT with PBS/0 . 1% Tween-20 . Chemiluminescence was detected using the enhanced chemiluminescence reagent ( GE Healthcare ) and developed using Hyperfilm ( GE Healthcare ) . Relative binding of proteins to the lipids was quantitated by densitometry using ImageJ software ( National Institutes of Health , Bethesda , MD; http://rsb . info . nih . gov/ij ) . The data were normalized to the PtdIns ( 4 , 5 ) P2 ( for PIP Strips and Membrane Strips ) or sulfatide ( for SphingoStrips ) . Liposomes were prepared as described previously ( Zhang et al . , 2001 ) using lipids purchased from Avanti Polar Lipids ( Alabaster , AL ) ; L-α-phosphatidylcholine ( PC , chicken egg ) , L-α-phosphatidyl-DL-glycerol ( PG , chicken egg ) , L-α-phosphatidylethanolamine ( PE , chicken egg ) , L-α-phosphatidylinositol ( PI , bovine liver ) , L-α-phosphatidylserine ( PS , porcine brain ) , and L-α-phosphatidylinositol-4 , 5 bisphosphate ( PIP2 , porcine brain ) . Lipids ( dissolved in chloroform ) and PIP2 ( dissolved in chloroform , methanol and water ) were combined with the desired ratio of lipid components ( PC:PE:PS:PI 50:30:10:10 , PC:PE:PS:PI:PIP2 50:30:10:8:2 , PC:PG 75:25 , PC:PG:PIP2 75:20:5 ) . The lipid mixture was dried under a stream of nitrogen gas followed by further drying under a vacuum for 3 hr . The lipid films were rehydrated in 500 µl of 50 mM HEPES ( pH 7 . 0 ) to a concentration of 14 mg/ml for 2 hr with occasional vortexing . Lipid mixtures were freeze-thawed five times before sonicating for 8 min until the mixture cleared . Liposomes were washed twice in 50 mM HEPES ( pH 7 . 0 ) prior to liposome binding assay . 50 µl of 14 mg/ml liposomes were incubated with 0 . 5 µg of NaD1 for 30 min at 25°C . Sample was pelleted by centrifugation at 16000×g and 30 µl of supernatant collected . The pellet was washed twice with 100 µl of 50 mM HEPES pH 7 . 0 . Supernatant and pellet samples were analyzed for the presence of protein by SDS-PAGE and immunoblotting using a rabbit anti-NaD1 antibody as described for the protein-lipid overlay assay . The NaD1:PIP2 complexes were generated by mixing NaD1 at 10 mg/ml and PIP2 at a molar ratio of 1:1 . 2 . Crystals were grown in sitting drops at 20°C in 0 . 2 M ammonium sulfate , 7% PEG 3350 , 32% MPD , and 0 . 1 M imidazole pH 7 . Diffraction data were collected from crystals flash cooled in mother liquor at 100 K at the Australian Synchrotron ( beamline MX2 ) and processed with Xds ( Kabsch , 2010 ) . The structure was solved by molecular replacement with PHASER ( Storoni et al . , 2004 ) using the structure of NaD1 ( Lay et al . , 2012 ) as a search model . The final model was built with Coot ( Emsley and Cowtan , 2004 ) and refined with Phenix ( Adams et al . , 2010 ) to a resolution of 1 . 6 Å . All data collection and refinement statistics are summarized in Table 1 . Refinement yielded Rwork and Rfree values of 15 . 5% and 18 . 4% , respectively . All programs were accessed via the SBGrid suite ( Morin et al . , 2013 ) . The coordinates have been deposited in the Protein Data Bank ( accession code 4CQK ) . Figures were prepared using PyMol . NaD1 at 1 mg/ml ( 5 μl ) was incubated with 2 . 3 , 0 . 46 , and 0 . 092 mM PIP2 ( 5 μl ) at room temperature for 30 min . Protein complexes were cross-linked through primary amino groups by the addition of 12 . 5 mM bis[sulfosuccinimidyl] suberate ( BS3; 10 μl ) in a buffer containing 20 mM sodium phosphate and 150 mM NaCl , pH 7 . 1 , at room temperature for 30 min . Samples were reduced and denatured , and subjected to SDS-PAGE prior to Coomassie Brilliant Blue staining . TEM imaging was performed according to the procedure described by Adda et al . ( 2009 ) . In brief , samples ( 10 μl ) were applied to 400-mesh copper grids coated with a thin layer of carbon for 2 min . Excess material was removed by blotting and samples were negatively stained twice with 10 μl of a 2% ( wt/vol ) uranyl acetate solution ( Electron Microscopy Services , Hatfield , PA ) . The grids were air-dried and viewed using a JEOL JEM-2010 transmission electron microscope operated at 80 kV . Recombinant NaD1 and the point mutants rNaD1 ( R40E ) and rNaD1 ( I37F ) were cloned , expressed , and purified from the methylotropic yeast Pichia pastoris as described in Lay et al . ( 2012 ) . The ability of rNaD1 , rNaD1 ( I37F ) , and rNaD1 ( R40E ) to inhibit the growth of F . oxysporum f . sp . vasinfectum was assessed as described in van der Weerden et al . ( 2008 ) , except that 6400 spores/well were used and growth was assessed after 48 hr . Each test was performed in triplicate . For data analysis , Prism 5 software ( GraphPad Software Inc . , San Diego , CA ) was used to plot 4-parameter sigmoidal curves through the data . Human epithelial cervical cancer ( HeLa ) cells , prostate cancer ( PC3 ) cells , and monocytic lymphoma ( U937 ) cells were cultured in RPMI-1640 medium ( Invitrogen , Carlsbad , CA ) . All culture media were supplemented with 5–10% fetal calf serum , 100 U/ml of penicillin , and 100 μg/ml of streptomycin ( Invitrogen ) . Cell lines were cultured at 37°C in a humidified atmosphere containing 5% CO2 . Adherent cell lines were detached from the flask by adding a mixture containing 0 . 25% trypsin and 0 . 5 μM EDTA ( Invitrogen ) . Flow cytometry-based PI uptake assay was performed to analyze the ability of NaD1 and related defensins to permeabilize tumor cells . Unless stated otherwise , U937 cells were suspended to 1 × 106 cells/ml in 0 . 1% BSA/RPMI-1640 and incubated with protein samples at 37°C for 30 min . Samples were added to PBS containing a final concentration of 1 μg/ml PI ( Sigma-Aldrich , St Louis , MO ) and placed on ice . Samples were then analyzed immediately using the BD FACSCanto II Flow Cytometer and BD FACSDiva software v6 . 1 . 1 ( BD Biosciences , St Jose , CA ) . The resultant flow cytometry data were analyzed using FlowJo software v8 . 8 . 6 ( Tree Star , Ashland , OR ) . Cells were gated appropriately based on forward scatter and side scatter and cell permeabilization was defined by PI-positive staining . U937 cells and protein samples were prepared as per the PI uptake assay , with the exception that 100 μg/ml of FITC-dextran ( 4 , 10 , 20 , or 40 kDa , Sigma-Aldrich ) was present during the incubation at 37°C for 30 min . Samples were washed twice with 0 . 1% BSA/PBS to remove unbound FITC-dextran and added to PBS containing a final concentration of 1 μg/ml 7-aminoactinomycin D ( 7AAD ) prior to analysis using the BD FACSCanto II Flow Cytometer and BD FACSDiva software . The resultant flow cytometry data were analyzed using FlowJo software . ATP bioluminescence assay ( Roche , Mannheim , Germany ) was used according to manufacturer’s instructions to measure the release of ATP from permeabilized tumor cells following treatment with defensins . Briefly , U937 cells were suspended to 1 × 106 cells/ml in 0 . 1% BSA/PBS and mixed with luciferase reagent at a ratio of 4:5 ( vol:vol ) . The mixture of cells and luciferase reagent were added simultaneously to each well containing protein samples and luciferase activity was measured immediately on a SpectraMax M5e plate reader ( Molecular Devices , Sunnyville , CA ) at RT for 30 min with readings taken at 30 s intervals . The resultant data were analyzed using SoftMaxPro 5 . 2 software ( Molecular Devices ) . LDH cytotoxicity assay kit II ( Abcam , Cambridge , United Kingdom ) was used according to manufacturer’s instructions to detect the release of the cytosolic enzyme , LDH , from U937 cells following treatment with defensins . Briefly , U937 cells were suspended at a cell concentration of 1 × 106 cells/ml in 0 . 1% BSA/RPMI-1640 and incubated with protein samples at 37°C for 30 min . Cells were then pelleted by centrifugation at 600×g and the supernatant was added to LDH reaction mix for 30 min at RT . The absorbance of the enzymatic product at 450 nm was measured using a SpectraMax M5e plate reader , with the resultant data analyzed using SoftMaxPro 5 . 2 software . Mammalian cells were seeded in quadruplicate into wells of a flat-bottomed 96-well microtitre plate ( 50 μl ) at various densities starting at 2 × 106 cells/ml . Four wells containing complete culture medium alone were included in each assay as a background control . The microtitre plate was incubated overnight at 37°C under a humidified atmosphere containing 5% CO2/95% air , prior to the addition of complete culture medium ( 100 μl ) to each well and further incubated at 37°C for 48 hr . Optimum cell densities ( 30–50% confluency ) for cell viability assays were determined for each cell line by light microscopy . Mammalian cells were seeded in a 96-well microtitre plate ( 50 μl/well ) at an optimum density determined in the cell optimization assay as above . Background control wells ( n = 8 ) containing the same volume of complete culture medium were included in the assay . The microtitre plate was incubated overnight at 37°C , prior to the addition of NaD1 at various concentrations and the plate was incubated for a further 48 hr . The cell viability 3- ( 4 , 5-dimethyl-2-thiazolyl ) -2 , 5-diphenyl-2H-tetrazolium bromide ( MTT , Sigma-Aldrich ) assay was performed as follows: the MTT solution ( 1 mg/ml ) was added to each well ( 100 μl ) and the plate incubated for 2–3 hr at 37°C under a humidified atmosphere containing 5% CO2/95% air . Subsequently , the media was removed and replaced with dimethyl sulfoxide ( 100 μl , DMSO , Sigma-Aldrich ) and placed on a shaker for 5 min to dissolve the tetrazolium salts . Absorbance of each well was measured at 570 nm and the IC50 values ( the protein concentration to inhibit 50% of cell growth ) were determined using OriginPro software v8 . 1 . 13 . 88 ( OriginLab Corporation , Northampton , MA ) . Live imaging was performed on a Zeiss LSM 510 or LSM 780 confocal microscope using a 40× or 63× oil immersion objective in a 37°C/5% CO2 atmosphere . Adherent cells were cultured on coverslips prior to imaging , while non-adherent cells were immobilized onto 10% poly-L-lysine-coated coverslips . All cell types were prepared for imaging in RPMI medium containing 0 . 1% BSA and 1–2 μg/ml PI . NaD1 , BODIPY-NaD1 , and FITC-Dextran ( 100 μg/ml ) was added directly to the imaging chamber via a capillary tube . In certain experiments , cells were either stained with PKH67 ( Sigma-Aldrich ) or transfected with a plasmid construct for free GFP or GFP-PH ( PLCδ ) using Lipofectamine 2000 Reagent ( Invitrogen ) as per manufacturer’s instructions prior to imaging . The images were analyzed using ImageJ software or Zen software ( Zeiss , Oberkochen , Germany ) . For quantification of CLSM experiments involving transfected cells , the effects of NaD1 on HeLa cells were observed over a 15 min timeframe from the time of NaD1 addition . Non-expressing cells were excluded from analysis . | It is often said that attack is the best form of defense; and the immune systems of plants and animals will often target the cell membranes of microbes and other pathogens in order to defend themselves . Disrupting the cell membrane causes essential contents to leak from the cell , and eventually , the cell will burst and die . Most plants and animals produce small proteins called defensins that kill microbes by attacking their cell membranes . These defensins are thought to either destabilize the cell membrane by coating its outer surface or to insert themselves into the membrane to form open pores that allow vital biomolecules to leak out of the cell . However , the exact mechanism by which defensins attack microbial membranes is not understood . In this study , Poon , Baxter , Lay et al . show that a defensin called NaD1—which was isolated from the ornamental tobacco Nicotiana alata—binds to a molecule from the cell membrane called phosphatidylinositol 4 , 5-bisphosphate , or PIP2 for short . By working out the three-dimensional structure of this complex , Poon , Baxter , Lay et al . show that it contains 14 PIP2 molecules and 14 NaD1 molecules in an arch-shaped structure and suggest that sequestering large numbers of PIP2 molecules in this way destabilizes the cell membrane of the microbe . These findings raise a number of questions: are there other small proteins that can destabilize cell membranes in a similar manner to defensins ? Do the immune systems of other organisms also recognize molecules from microbial cell membranes to trigger this kind of counterattack ? Furthermore , since defensins can also kill tumor cells , a better understanding of how they work might also lead to new treatments for cancer and other diseases in humans . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | Phosphoinositide-mediated oligomerization of a defensin induces cell lysis |
Type XVII collagen ( COL17 ) is a transmembrane protein located at the epidermal basement membrane zone . COL17 deficiency results in premature hair aging phenotypes and in junctional epidermolysis bullosa . Here , we show that COL17 plays a central role in regulating interfollicular epidermis ( IFE ) proliferation . Loss of COL17 leads to transient IFE hypertrophy in neonatal mice owing to aberrant Wnt signaling . The replenishment of COL17 in the neonatal epidermis of COL17-null mice reverses the proliferative IFE phenotype and the altered Wnt signaling . Physical aging abolishes membranous COL17 in IFE basal cells because of inactive atypical protein kinase C signaling and also induces epidermal hyperproliferation . The overexpression of human COL17 in aged mouse epidermis suppresses IFE hypertrophy . These findings demonstrate that COL17 governs IFE proliferation of neonatal and aged skin in distinct ways . Our study indicates that COL17 could be an important target of anti-aging strategies in the skin .
Skin is a highly structured organ in which stem cell self-renewal , cell proliferation and differentiation are coordinated to maintain homeostasis . In addition to hair follicles and other skin appendages , the interfollicular epidermis ( IFE; non-haired skin ) comprises distinct cellular populations . IFE stem cells reside in the basal cell layer; these cells both self-renew and generate the terminally differentiated outer cell layers which function as barriers to the external environment and prevent loss of body fluids ( Giangreco et al . , 2008; Hsu et al . , 2014; Jones et al . , 2007; Natsuga , 2014 ) . Organismal aging , or physical aging , is defined as tissue impairment arising from the accumulation of numerous intrinsic and extrinsic factors that induce cell damage chronologically . The relationship between organismal aging and stem cells is an inescapable bond , and the heterogeneity of stem cells in organs may be reduced with aging ( Goodell and Rando , 2015 ) . Human skin aging is exemplified by alterations in the dermis and skin appendages , such as the thinning of dermis , dryness , wrinkles , gray hair and hair loss ( Rittié and Fisher , 2015 ) . However , the influence of aging on IFE has been controversial . Although decreased epidermal proliferation has been reported in in vitro and in vivo studies using aged individuals and mice ( Giangreco et al . , 2008; Gilchrest , 1983; Grove and Kligman , 1983 ) , several recent studies have reported contradictory results , showing sustained and increased proliferation in the aged epidermis ( Charruyer et al . , 2009; Stern and Bickenbach , 2007 ) . Thus , how organismal aging affects the IFE and its stem cells has not been clarified ( Keyes et al . , 2013 ) . The extracellular matrix proteins of the basement membrane zone ( BMZ ) are important components of the IFE stem cell niche and connect the dermis and epidermis functionally . Type XVII collagen ( COL17 ) is a type II transmembrane protein that is located along the hemidesmosomes in the BMZ . The N-terminus of COL17 is localized in hemidesmosomes , and its extracellular domain reaches the lamina densa ( McMillan et al . , 2003 ) . Non-hemidesmosomal COL17 in keratinocytes and human skin has also been reported ( Hirako et al . , 1998 ) . COL17 has been characterized as a target protein in the autoimmune blistering disease bullous pemphigoid ( Nishie , 2014 ) and also as being the defective protein in junctional epidermolysis bullosa ( JEB ) , a congenital blistering disease ( Fine et al . , 2014 ) . Recently , COL17 has been shown to form a niche for hair follicle stem cells ( HFSCs ) , as mice lacking the protein and human JEB patients with mutations in COL17A1 , the open-reading frame encoding COL17 , exhibit a premature aged skin phenotype , including gray hair and hair loss ( Matsumura et al . , 2016; Nishie et al . , 2007; Tanimura et al . , 2011 ) . Additionally , reduced labeling of epidermal basement membrane proteins , including COL17 , in human skin has been associated with aging ( Langton et al . , 2016 ) . However , the role of COL17 in maintaining IFE and its stem cells is still unclear . In the present study , we explore the comprehensive role of COL17 in regulating IFE homeostasis and also characterize age-related IFE alterations associated with a modified BMZ , including COL17 . We show that COL17 is indispensable for regulating IFE proliferation in neonatal mice through activating the Wnt pathway . The IFE hyperproliferation is induced by organismal aging and can be reversed by COL17 replenishment .
We investigated the phenotype of paw skin to study IFE in Col17a1−/− mice ( Nishie et al . , 2007 ) in the absence of hair follicles . Neonatal IFE skin ( P1 , postnatal day 1 ) of Col17a1−/− mice showed transient epidermal hyperproliferation , as demonstrated by counting the epidermal layers and the numbers of epidermal cells and phospho-Histone H3 ( PH3 ) -positive cells ( Figure 1a–b ) . The numbers of proliferating cell nuclear antigen ( PCNA ) - and Bromodeoxyuridine ( BrdU ) -positive cells in the Col17a1−/− IFE basal cells at P1 were also increased compared with the controls ( Figure 1c ) , indicating that COL17 deletion affects both the S and M phases in the cell cycle of IFE neonatal keratinocytes . The proliferative IFE phenotype of Col17a1−/− mice gradually waned postnatally , and the epidermal thickness and the number of PH3-positive cells became comparable with those of the controls at P20 ( Figure 1a–b ) . 10 . 7554/eLife . 26635 . 003Figure 1 . COL17 deletion induces transient IFE hyperproliferation in neonates . ( a ) Hematoxylin and eosin ( H&E ) staining and E-cadherin ( E-cad ) labeling ( with PI nuclear counterstain ) of Col17a1−/− and control IFE skin samples from Col17a1+/- or Col17a1+/+ littermates ( Control ) at P1 ( n = 5 ) and P20 ( n = 4 ) . Scale bar: 20 μm . Quantitation of the number of epidermal layers and epidermal cell counts . The values are shown as relative ratios to the controls . ( b ) PH3 staining at P1 and P20 . Scale bar: 20 μm . The number of epidermal basal cells positively labeled for PH3 per mm epidermis ( n = 4 ) . BM , basement membrane . ( c ) PCNA and BrdU labeling at P1 . Scale bar: 20 μm . Quantitation of PCNA- ( n = 5 ) and BrdU-positive basal cells ( n = 4 ) . The values are shown as relative ratios to the controls . ( d ) Quantitative RT-PCR ( qRT-PCR ) of Itga6 , Itgb1 , Tgm1 , Ppl and Ivl mRNAs ( n = 5 ) . ( e ) Loricrin and cleaved caspase-3 staining ( representative images from 3 mice ) . Scale bar: 20 μm . BM , basement membrane . ( f ) An in silico model of the epidermal cell proliferation upon the reduced adhesion of committed progenitor cells to the BMZ . The details are described in the Material and Methods . The data in all of the histograms are the means ± SE . *0 . 01<p<0 . 05 , **0 . 001<p<0 . 01 , ****p<0 . 0001 . Student’s t-tests . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 00310 . 7554/eLife . 26635 . 004Figure 1—figure supplement 1 . Barrier function assay of Col17a1−/− and littermate controls ( E18 . 5 ) . Inflammatory cell infiltration in the dermis of Col17a1−/− and littermate controls ( P1 ) and ultrastructural findings of the basement membrane zone in the paw epidermis of Col17a1−/− and littermate controls ( P1 ) . ( a ) mRNA expression levels of Itgb4 , Lamb3 and Lamc3 ( n = 4 ) . ( b ) Dye permeabilization with Toluidine blue ( representative images from three Col17a1−/− mice and eight control mice ) . ( c ) Transepidermal water loss ( TEWL ) ( n = 3 for Col17a1−/− mice; n = 8 for control ) . ( d ) In the Col17a1−/− paw epidermis , hemidesmosomes ( HD ) comprising inner plaques ( IP ) , outer plaques ( OP ) and anchoring fibrils were blurred compared with those of the Col17a1-/- mice . LL: Lamina Lucida , LD: Lamina densa . Scale bar = 0 . 2 μm . Representative images from two mice . ( eg ) Quantification of immune cells in the paw skin dermis of Col17a1−/− and control mice ( n = 4 ) . The data are the means±SE . *0 . 01<p<0 . 05 , **0 . 001<p<0 . 01 , Student’s t-tests . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 00410 . 7554/eLife . 26635 . 005Figure 1—figure supplement 2 . Proliferative ability of the back skin IFE from Col17a1−/− mice and NHEKs treated with COL17A1 siRNAs . ( a ) PH3- and PCNA-positive cells in the Col17a1−/− back skin ( n = 4 for Col17a1−/−; n = 5 for control ) . ( b ) COL17A1 knockdown efficiency in NHEKs . The left panel shows COL17 immunoblotting of lysates from NHEKs treated with siRNAs . The right panel shows the qRT-PCR results of COL17A1 ( n = 3 ) . ( c–d ) Cell proliferation curve ( c ) and slope ( d ) of NHEKs treated with siRNAs ( n = 3 ) . ( e–f ) Colony formation assay of NHEKs treated with siRNAs . Gross appearance ( e ) , total colony number ( f-left ) and the percentage of colonies that were larger than 0 . 5 mm ( f-right ) ( n = 3 ) . The data are presented as the means±SE . *0 . 01<p<0 . 05 , **0 . 001<p<0 . 01 , Student’s t-tests . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 005 We investigated whether the expression levels of markers of basal cells and differentiated cells were altered in the hyperproliferative IFE of Col17a1−/− mice at P1 . The gene expression of Itga6 , which might compensate for COL17 deficiency , was increased in Col17a1−/− IFE skin , whereas the expression of Itgb1 , itgb4 , lamb3 and lamc2 was not altered ( Figure 1d , Figure 1—figure supplement 1a ) . The mRNA expression levels of Tgm1 , Ppl and Ivl were somewhat higher in Col17a1−/− IFE skin ( Figure 1d ) , while loricrin-labeled granular cell layers were not expanded in these mice ( Figure 1e ) . Dye-permeability and transepidermal water loss at day 18 . 5 of embryogenesis ( E18 . 5 ) were comparable between the Col17a1−/− mice and the controls ( Figure 1—figure supplement 1b–c ) , indicating that IFE keratinocyte differentiation was not greatly altered in Col17a1−/− skin . There was no increase in the number of cells positive for cleaved caspase-3 , a marker of apoptosis , in the Col17a1−/− epidermis compared with the controls ( Figure 1e ) . The electron microscopy results indicated that the dermo-epidermal junction of the Col17a1−/− IFE presented hypoplastic hemidesmosomes in accordance with previous observations on the back skin of Col17a1−/− mice ( Nishie et al . , 2007 ) ( Figure 1—figure supplement 1d ) . There were no significant differences in the number of inflammatory infiltrates , including CD3+ , F4/80+ and Ly-6G+ cells , in the dermis of Col17a1−/− mice and control mice , which excludes inflammation as a contributor to immature hemidesmosome formation ( Figure 1—figure supplement 1e–g ) . To explain the transient epidermal hyperproliferation of the Col17a1−/− neonatal IFE , we exploited an in silico model ( Kobayashi et al . , 2016 ) to recapitulate the epidermal development . As COL17 serves as a linker between the epidermis and dermis and is expressed in basal cells , most of which are committed progenitor cells in the epidermis ( Doupé and Jones , 2012; Lim et al . , 2013 ) , epidermal thickness was calculated upon the loosening of the attachment of committed progenitor cells to the BMZ . In the model with diminished adhesion , the epidermal thickness was transiently increased and gradually returned to the baseline ( Figure 1f ) , which was compatible with the experimental observations of Col17a1−/− neonates and with the transient hypertrophy of Itga6- and Itgb1-null epidermis ( Brakebusch et al . , 2000; Niculescu et al . , 2011 ) . These data suggest that COL17 deletion induces epidermal thickening by the hyperproliferation of IFE keratinocytes , at least partially through the loosening of dermo-epidermal adhesion at the neonatal stage . The site specificity of the hyperproliferative phenotype in the neonatal paw epidermis of Col17a1/-/ mice was confirmed by the comparable expression of proliferation markers in the back skin IFE of Col17a1−/− mice and control mice ( Figure 1—figure supplement 2a ) . The discordance between the paw epidermis and back skin IFE might be explained either by the influence of hair follicle development on the back skin IFE or by the distinct regulation of the IFE at each body site ( Rompolas et al . , 2016; Roy et al . , 2016; Sada et al . , 2016 ) . We also investigated cell-intrinsic properties due to COL17 defects using cultured normal human epidermal keratinocytes ( NHEKs ) . The cell proliferation rates of NHEKs treated with COL17A1 siRNAs were slightly decreased ( Figure 1—figure supplement 2b–d ) , which is compatible with reduced proliferation of cultured keratinocytes derived from Col17a1−/− mice ( Tanimura et al . , 2011 ) , and the colony-forming abilities of these cells were similar to those of control cells ( Figure 1—figure supplement 2e–f ) . These data indicate that the proliferation potential of Col17a1−/− IFE is dependent on in vivo conditions . Various signaling molecules are involved in controlling hair follicle stem cells and epidermal homeostasis ( Kretzschmar and Watt , 2014 ) ; the relationships between these signaling molecules and BMZ proteins are only partially understood ( Margadant et al . , 2010; Rognoni et al . , 2014; Tanimura et al . , 2011 ) . To further explore the underlying mechanisms of the transient epidermal hyperproliferation phenotype of Col17a1−/− IFE , we first screened the gene expression profiles of the receptors , co-receptors , transcription factors and cofactors of the major signaling pathways , including Wnt , TGF-β/BMP , Notch , Hedgehog and FGF in neonatal Col17a1−/− and control skin samples . The screening data showed that the expression levels of Wnt-related molecules ( Fzd4 , Nfatc2 , Nfatc4 and Tcf7 ) were significantly decreased in Col17a1−/− neonatal IFE skin compared with controls ( Figure 2—figure supplement 1a ) . Although the expression of some TGF-β−related genes was altered in Col17a1−/− IFE skin ( Figure 2—figure supplement 1a ) , the TGF-β and p-Smad2 immunostaining in Col17a1−/− IFE keratinocytes was comparable with that in IFE cells in the controls , in contrast to the TGF-β and p-Smad2 reduction in Col17a1−/− HFSCs ( Tanimura et al . , 2011 ) ( Figure 2—figure supplement 1b–c ) . The gene expression profiles for Notch , Hedgehog and the FGF pathway were not significantly affected upon COL17 deletion ( Figure 2—figure supplement 1a ) . To validate these findings , we analyzed the gene expression profiles of specific Wnt-related molecules in the independent replication samples . In line with the screening qPCR data , the expression levels of the Wnt target genes ( Tcf7 l1 , Tcf7 l2 , and Axin2 ) , the receptor gene ( Fzd4 ) , and stimulatory Wnt genes ( Wnt2 , Wnt2b , Wnt5a ) were significantly downregulated in Col17a1−/−mice , whereas expression of the inhibitory Wnt gene ( Wnt4 ) was increased ( Figure 2a ) ( Bernard et al . , 2008; Mikels and Nusse , 2006 ) . The changes in these genes were paralleled by a reduction of basal cells positive for LEF1 , a nuclear Wnt mediator of Wnt signaling , in neonatal Col17a1−/− mice IFE ( Figure 2b ) . This reduction was also transient , and the number of LEF1-positive basal cells was comparable between the Col17a1−/− mice and the control mice at P4 , which might account for the transient epidermal hyperproliferation of the paw IFE in Col17a1−/− mice . As β-catenin binds to LEF1 in the nucleus in the Wnt-canonical pathway ( Lien and Fuchs , 2014 ) , we performed β-catenin immunostaining . The number of nuclear β-catenin-positive cells in Col17a1−/− mice IFE was diminished compared to the controls ( Figure 2c ) . 10 . 7554/eLife . 26635 . 006Figure 2 . COL17 deficiency destabilizes Wnt-β-catenin signaling in neonates . ( a ) qRT-PCR of Wnt-related molecules in Col17a1−/− and control IFE skin samples at P1 ( n = 5 ) . Student’s t-test . ( b ) LEF1 staining of Col17a1−/− and control IFE skin at P1 ( n = 5 ) , P4 ( n = 4 ) and P10 ( n = 4 ) . Inlet: higher magnification of LEF1-positive basal cells at P1 . Scale bar: 20 μm . Quantitation of LEF1-positive basal cells as a percentage of all basal cells ( % ) . Student’s t-test . ( c ) β-catenin staining of Col17a1−/− and control IFE skin at P1 . Nuclear β-catenin accumulation is indicated with arrows . The quantification of nuclear β-catenin-positive cells ( n = 3 ) . Student’s t-test . Scale bar: 20 μm . ( d ) Wnt activity in STF293 cells expressing hCOL17 treated with Wnt3a CM ( n = 3 ) . One-way ANOVA test , followed by Tukey’s test . ( e ) Wnt activities in the hindpaw IFE from ins-Topgal+ ( Control: left ) and ins-Topgal+:Col17a1−/− ( Col17a1−/−: right ) mice . Calculated areas devoid of hair follicles or sweat glands are indicated with squares in the representative figures . The results are quantified as the Wnt-activated area per unit ( n = 4 ) . Scale bar: 100 μm . Mann-Whitney test . ( f ) H&E , E-cad , PH3 and PCNA staining of IFE skin samples from K14-ΔNLef and littermate controls at P1 . Scale bar: 20 μm . The numbers of epidermal layers , epidermal cell counts and PCNA- and PH3-positive basal cells ( n = 4 ) . Student’s t-test . ( g ) Quantification of BrdU- and PH3-positive cells in WT paw skin IFE treated with Wnt inhibitors ( IWP-2 ( n = 6 ) vs DMSO ( n = 5 ) or Wnt-C59 ( n = 6 ) vs DMSO ( n = 4 ) ) . The data are presented as the means ± SE . Student’s t-test . *0 . 01<p<0 . 05 , **0 . 001<p<0 . 01 , ***0 . 0001<p<0 . 001 , ****p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 00610 . 7554/eLife . 26635 . 007Figure 2—figure supplement 1 . mRNA profiles of signaling molecules and TGF-β staining . ( a ) qRT-PCR screening of genes involved in the Wnt , TGF-β/BMP , Notch , Hedgehog , and FGF signaling pathways . mRNA samples from IFE skin of Col17a1−/− and littermate controls at P1 were analyzed ( n = 5 ) . *0 . 01<p<0 . 05 , Student’s t-tests . ( b ) TGF-β-stained IFE of Col17a1−/− and littermate controls at P1 ( representative images from three mice ) . ( c ) p-Smad2 staining of IFE from Col17a1−/− and littermate controls at P1 ( representative images from three mice ) . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 00710 . 7554/eLife . 26635 . 008Figure 2—figure supplement 2 . The details of LacZ staining in ins-Topgal+ mice with COL17 deficiency . ( a ) Schematic model of murine hindpaw on LacZ staining . The dark blue area indicates sweat glands-abundant regions and the light blue area represents to regions with hair follicles . Sweat glands and hair follicles are major sources of Wnt activities as marked with LacZ . The white area , without sweat glands or hair follicles , shows the regions used for the quantification . ( b ) Gross appearance of LacZ staining on ins-Topgal+ ( Control: left ) and ins-Topgal+:Col17a1−/− ( Col17a1−/−: right ) mice hind paw . Scale bar: 500 μm . ( c ) LacZ staining on ins-Topgal+ ( Control: left ) and ins-Topgal+:Col17a1−/− ( Col17a1−/−: right ) mice in high magnification for quantification . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 00810 . 7554/eLife . 26635 . 009Figure 2—figure supplement 3 . Wnt signaling and proliferation profile in JEB patients epidermis with COL17 deficiency . LEF1 , β-catenin and PH3 labeling of skin specimens from two JEB patients and controls . LEF1-positive cells and nuclear β-catenin accumulation are indicated with arrows . Scale bar: 20 μm . Quantification of LEF1- and PH3-positive basal cells per mm epidermis . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 009 To further evaluate the relationship between Wnt signaling and COL17 , we utilized SuperTopFlash 293 ( STF293 ) reporter cells ( Tsukiyama et al . , 2015 ) . Overexpression of human COL17 significantly upregulated Wnt activity in this cell line ( Figure 2d ) . Additionally , to visualize Wnt signaling in vivo , we crossed ins-Topgal+ mice ( Moriyama et al . , 2007 ) with Col17a1−/− mice . The LacZ-positive area that was indicative of active Wnt signaling in the IFE was significantly diminished in the ins-Topgal+:Col17a1−/− mice ( Figure 2e , Figure 2—figure supplement 2 ) . These results suggest that COL17 expression stabilizes Wnt signaling . To examine whether these findings correlate with the phenotype of JEB patients with COL17 deficiency , we also performed immunostainings for LEF1 , β-catenin and PH3 in JEB skin . In the JEB epidermis , the numbers of LEF1-positive cells and cells with nuclear β-catenin were decreased , while the number of PH3-positive cells was elevated ( Figure 2—figure supplement 3 ) ; these findings were compatible with the data from the Col17a1−/− mice , although this result requires further investigation due to the small sample size . We next investigated whether a defect in Wnt signaling in the IFE could be responsible for the hyperproliferation phenotype and whether that phenotype could be reversed by the introduction of COL17 . K14-deltaNLef1 mice express a Lef1 transgene that lacks a β-catenin-binding site under the control of the keratin 14 ( K14 ) promoter and serve as a model of inactive Wnt signaling in the epidermis ( Niemann et al . , 2002 ) . Neonatal K14-deltaNLef1 mice IFE exhibited epidermal thickening and a larger number of PH3-positive basal cells than the controls ( Figure 2f ) . The number of PCNA-positive basal cells was also increased , albeit not significantly . To confirm that the abated Wnt activities were related to neonatal IFE proliferation in vivo , we intraperitoneally administered Wnt inhibitors ( IWP-2 or Wnt-C59 ) into wild-type mouse neonates ( Kuo et al . , 2016; Carotenuto et al . , 2017 ) . The number of BrdU- and PH3-positive epidermal cells was increased in mice treated with these inhibitors at P1 compared with untreated control mice ( Figure 2g ) . Transgenic rescue by the expression of human COL17 ( hCOL17 ) under the K14 promoter in Col17a1−/− mice ( Nishie et al . , 2007 ) abrogated IFE hyperproliferation and restored LEF1- and nuclear β-catenin-positive basal cells ( Figure 3a , c–d ) . The expression levels of Wnt-related genes that were altered in Col17a1−/− mice at P1 were restored by transgenic rescue with hCOL17 ( Figure 3b ) . These data demonstrate that COL17 unambiguously contributes to the maintenance of neonatal IFE proliferation via its effect on Wnt signaling . 10 . 7554/eLife . 26635 . 010Figure 3 . Induction of human COL17 abrogates epidermal hyperproliferation and the expression of Wnt-β-catenin signaling molecules in neonatal Col17a1−/− IFE . ( a ) H&E , E-cad , PH3 and PCNA staining of IFE skin specimens from Col17a1+/+ or Col17a1+/- ( as hCOL17-; CTL ( control ) ) and hCOL17+; Col17a1−/− littermate mice at P1 . Quantification of epidermal layers , epidermal cell counts , and PH3- and PCNA-positive cells ( n = 4 ) . Scale bar: 20 μm . ( b ) Gene expression of Wnt-related molecules in IFE skin samples from hCOL17-; CTL and hCOL17+; Col17a1−/− littermate mice at P1 ( n = 4 ) . ( c ) LEF1 staining of IFE skin samples from hCOL17-; CTL and hCOL17+; Col17a1−/− littermates at P1 ( n = 4 ) . Scale bar: 20 μm . ( d ) β-catenin staining of IFE skin samples from hCOL17-; CTL and hCOL17+; Col17a1−/− littermates at P1 ( n = 4 ) . Nuclear β-catenin is indicated with arrows . The number of nuclear β-catenin-positive cells . Scale bar: 20 μm . The data are the means ± SE . Student’s t-tests . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 010 Because Col17a1−/− mice ( 3-month-old ) clearly exhibited the premature aging phenotypes of gray hair and hair loss ( Figure 4—figure supplement 1 ) ( Nishie et al . , 2007; Tanimura et al . , 2011 ) , we examined the effects of physical aging on the IFE and the relationship between aging and COL17 . We compared wild-type ( WT ) young mice ( 6 to 10 weeks old ) with aged mice ( 19 to 27 months old ) . In the IFE of aged mice , the epidermis was thickened , and the numbers of PH3- , BrdU- and PCNA-positive cells were increased ( Figure 4a ) . This phenomenon was specific to paw skin and was not observed in the back skin IFE ( Figure 4—figure supplement 2 ) , as was the case for the neonatal Col17a1−/− IFE . These data indicate that physical aging leads to paw IFE hyperproliferation . 10 . 7554/eLife . 26635 . 011Figure 4 . Physical aging affects epidermal proliferation and COL17 distribution . ( a ) H&E , E-cad , PH3 , BrdU and PCNA staining of IFE skin from young ( 6–10 weeks old ) and aged ( 19–27 months old ) adult C57BL/6 wild-type ( WT ) mice . Scale bar: 20 μm . The numbers of epidermal layers , epidermal cell counts , and PH3- , BrdU- and PCNA-positive basal cells ( n = 5 ) . Student’s t-test . ( b ) The gene expression levels of Itga6 , Itgb1 , Tgm1 , Ppl , Evpl , and Col17a1 in IFE skin samples from young and aged WT mice ( n = 5 for Itga6 , Itgb1 , Tgm1 and Col17a1; n = 3 for Ppl and Evpl ) . Student’s t-test . ( c ) COL17 staining ( antibodies to the juxtamembranous portion ) in IFE skin samples from the following groups: young and aged WT mice ( n = 5 ) , young ( <15 years old ) and aged ( >85 years old ) normal human individuals ( representative images from three human samples ) , and Klotho+/- and Klotho−/− littermates at 6 weeks ( representative images from three mice ) . Scale bar: 20 μm . The quantitative fluorescent intensity of lateral membrane of IFE basal cells from young and aged WT mice ( n = 5 ) . Mann-Whitney test . ( d ) COL17 labeling following the Triton X-100 treatment of IFE skin from young/aged WT mice and human individuals ( representative images from three samples ) . Scale bar: 20 μm . ( e ) The optical sectioning of 3D reconstructed whole mount COL17-stained skin from young and aged murine WT IFE . The IFE cell membrane was visualized with wheat germ agglutinin ( WGA ) . DAPI ( 4' , 6-diamidino-2-phenylindole ) was used for nuclear staining . Scale bar: 10 μm . The quantitative fluorescent intensity of COL17 in lateral membrane of IFE basal cells from young and aged WT mice ( n = 6 ) . Mann-Whitney test . ( f ) The distributions of COL17 and desmogleins 1 and 2 in young murine WT IFE using N-SIM ( structured illumination microscopy ) image reconstruction ( representative images from two mice ) . Basal keratinocytes were depicted by white lines . Scale bar: 5 μm . BM , basement membrane . The data are the means ± SE . *0 . 01<p<0 . 05 , **0 . 001<p<0 . 01 , ***0 . 0001<p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 01110 . 7554/eLife . 26635 . 012Figure 4—figure supplement 1 . Gross appearance of Col17a1−/− and littermate controls ( 3 months old ) . ( a ) Gray and sparse hair was noted in Col17a1−/− skin . ( b ) Scaly paw skin was observed in Col17a1−/− skin ( representative images from four mice ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 01210 . 7554/eLife . 26635 . 013Figure 4—figure supplement 2 . Proliferation markers in back skin from young and aged mice . Quantification of PH3- and PCNA- positive cells in young and aged paw skin ( n = 5 ) . Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 01310 . 7554/eLife . 26635 . 014Figure 4—figure supplement 3 . Intracellular COL17 labeling of the IFE from young/aged WT mice and human individuals . Antibodies against the intracellular portions of COL17 , which recognize the full-length protein , were used . The antibodies do not react with the shed ectodomain of COL17 . Scale bar: 20 μm . Samples from Klotho+/− and Klotho−/− littermates were obtained at the age of 6 weeks . Representative images from three samples . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 01410 . 7554/eLife . 26635 . 015Figure 4—figure supplement 4 . Expression of BMZ and extracellular matrix proteins in the IFE skin of young and aged WT mice and healthy individuals . ( a ) Staining of integrin α6 and β1 in young/aged mouse and human IFE samples ( representative images from five mice ) . Scale bar: 20 μm . ( b ) Labeling of BP230 and plectin in the mice IFE ( representative images from five mice ) . Scale bar: 20 μm . ( c , d ) Expression data from the microarray ( four mice from young and aged groups , respectively ) of genes encoding collagens ( c ) and laminins ( d ) . The fold changes are based on the young group . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 01510 . 7554/eLife . 26635 . 016Figure 4—figure supplement 5 . Analysis of epidermal proteases in IFE skin samples from young and aged WT mice . ( a ) Expression levels from the microarray analysis of genes encoding proteases involved in COL17 digestion and degradation ( n = 4 ) . ( b ) COL17 staining in Serpine1−/− and littermate control mice IFE at 8 weeks ( representative images from four mice ) . The antibody against the juxtamembranous NC16A domain of COL17 was used . ( c ) ELANE labeling of young and aged IFE ( representative images from five mice ) . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 016 The gene expression levels of Itga6 and Itgb1 were decreased in the aged mouse IFE , whereas that of Col17a1 did not change ( Figure 4b ) . Furthermore , the expression levels of differentiation markers ( Tgm1 , Ppl , and Evpl ) were not greatly affected . Although Col17a1 gene expression was unaffected , the distribution of COL17 was dramatically altered in the IFE of aged mice and humans . The apico-lateral portion of COL17 ( non-hemidesmosomal COL17 ) was greatly reduced in the IFE of aged mice and humans , as detected with antibodies targeting the extracellular portion of COL17 ( Figure 4c ) and is in line with the recent publication on age-related alterations of COL17 in human skin ( Langton et al . , 2016 ) . This phenomenon was also observed in the IFE of Klotho−/− mice , a premature aging model ( Kuro-o , 2008 ) . The involvement of COL17 proteolytic cleavage , so-called ectodomain shedding ( Nishie et al . , 2012 , 2015 ) , in the altered COL17 distribution was excluded using several antibodies that recognize the intracellular portions of COL17 ( Figure 4—figure supplement 3 ) . To confirm the depletion of non-hemidesmosomal COL17 in the aged IFE , the sections were treated with Triton X-100 prior to immunostaining ( Hirako et al . , 1998 ) . In both mouse and human skin , Triton X-100-treated young IFE lost the apico-lateral portion of COL17 and resembled the aged epidermis ( Figure 4d ) . To further examine the reduction in the apico-lateral portion of COL17 in aged mice , we performed whole-mount staining of the IFE ( Figure 4e ) . In 3D-reconstructed images , COL17 was confined to the basement membrane in the aged IFE , while basal cells of young mice had COL17 that was associated with the apical , lateral and basal membranes . Using high-resolution structured illumination microscopy ( SIM ) imaging on young WT mouse IFE ( Figure 4f ) , COL17 was detected at the apico-lateral cell periphery of basal keratinocytes but did not co-localize with desmosomal proteins ( desmogleins 1 and 2 ) , indicating that the apico-lateral COL17 in basal keratinocytes was not incorporated into desmosomes , which are highly insoluble . In contrast to COL17 dynamics with aging , the distributions of integrins α6 and β1 were not modified by aging ( Figure 4—figure supplement 4a ) . Other hemidesmosomal proteins ( BP230 and Plectin ) were also unchanged ( Figure 4—figure supplement 4b ) . Sustained Col17a1 gene expression in IFE skin contrasted with the decreased expression of many collagens and laminins with aging ( Figure 4—figure supplement 4c–d ) . These results suggest that physical aging leads to paw IFE hypertrophy and modifies COL17 distribution in the epidermis in a posttranslational manner . To elucidate the mechanisms that underlie the altered COL17 distribution that occurs with physical aging , we reproduced the environment of aged IFE . Because calcium distribution is changed in the aged epidermis ( Denda et al . , 2003 ) and the calcium concentration in the epidermis is lower in aged individuals than in young individuals ( Rinnerthaler et al . , 2015 ) , whole IFE skin samples collected from young mice were treated with EDTA to simulate changes in calcium concentration in aged IFE . Whole mount staining showed that EDTA treatment eliminated apico-lateral COL17 in basal cells , while hemidesmosomal COL17 was present ( Figure 5a ) . This finding suggests that the calcium concentration ( or the concentrations of other molecules chelated by EDTA ) may affect the COL17 distribution in the IFE . 10 . 7554/eLife . 26635 . 017Figure 5 . COL17 distribution is modulated by aPKC . ( a ) COL17 staining of whole IFE skin treated with 5 mM EDTA . The quantitative fluorescent intensity of COL17 in basal cells of IFE from control ( PBS ) and 5 mM EDTA treated ( n = 4 ) . Mann-Whitney test . Samples were taken from young adult WT mice at 6–10 weeks . Scale bar: 20 μm . ( b ) Phospho-aPKC labeling ( indicated with arrows ) and quantitative fluorescent intensity results from young and aged WT IFE skin ( n = 4 ) . Mann-Whitney test . Scale bar: 20 μm . ( c ) Representative figures of asymmetric cell division ( ACD; scored as perpendicular to basement membrane ) and symmetric cell division ( SCD; in parallel to basement membrane ) in young IFE . Survivin staining indicates the direction of the cell division . Laminin β1 signifies basement membrane . Scale bar: 10 μm . Graph of percentage of ACD and SCD in young and aged IFE ( n = 4 ) . Student’s t-test . ( d–e ) The pharmacological inhibition of pan-aPKC ( d , 1 μM of Go6983; 0 . 00002% DMSO as control ) and aPKCλ/ζ ( e , 10 μM of myr PSI; water as control ) in whole IFE skin from young adult WT mice , followed by COL17 staining . The quantitative fluorescent intensity of COL17 in lateral membrane of basal cells from control and 1 μM Go6983 treated ( d ) and 1 μM myr PSI ( e ) ( n = 4 ) . Mann-Whitney test . Scale bar: 10 μm . BM , basement membrane . ( f–h ) EDTA treatment ( 5 mM; PBS as control , f ) and pharmacological inhibition of pan-aPKC ( g , 1 μM of Go6983; 0 . 00002% DMSO as control ) and aPKCλ/ζ ( h , 10 μM of myr PSI; water as control ) on 3D epidermis . The relative fluorescent intensity of COL17 in lateral membrane of basal cells was measured ( n = 4 ) . Mann-Whitney test . BM , basement membrane . Scale bar: 20 μm . The data are the means ± SE . *0 . 01<p<0 . 05 , **0 . 001<p<0 . 01 , ***0 . 0001<p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 017 Among the numerous cellular events influenced by calcium dynamics , we focused on atypical protein kinase C ( aPKC ) , a key regulator of epithelial polarity ( Niessen et al . , 2013 ) . The expression levels of aPKCζ and aPKCλ/ι , aPKC isoforms expressed by basal epidermal cells are dependent on the calcium concentration in cultured keratinocytes ( Helfrich et al . , 2007 ) , and the ablation of aPKCλ/ι in the epidermis leads to premature aging phenotypes , such as gray hair , hair loss and IFE thickening , closely resembling Col17a1−/− mice ( Niessen et al . , 2013; Osada et al . , 2015 ) . Phospho-aPKC , an active form of aPKC , was reduced in IFE with aging ( Figure 5b ) . Consistent with this finding , the axis of cell division in the aged IFE tended toward asymmetric cell division ( ACD ) ( Figure 5c ) . This phenotype emulates the adult IFE of aPKCλ conditional-null mice ( Niessen et al . , 2013 ) . To recapitulate aged skin with attenuated aPKC , we treated young mouse IFE samples with Go6983 , a pan-PKC inhibitor , and myristoylated pseudosubstrate inhibitor ( myr PSI ) , which is specific for aPKCζ and aPKCλ/ι ( Gschwendt et al . , 1996; Helfrich et al . , 2007; Standaert et al . , 1999 ) . According to the whole mount staining , the pharmacological inhibition of aPKC diminished the apico-lateral COL17 in basal cells ( Figure 5d–e ) . EDTA treatment and aPKC inhibition also diminished apico-lateral membranous COL17 in basal cells of reconstructed human 3D epidermis ( Figure 5f–h ) . These data indicate that an aging-induced alteration of aPKC contributes to spatial changes in COL17 in the IFE . COL17 undergoes posttranslational modifications , such as ectodomain shedding and degradation , in physiological and pathological settings through the action of several proteases , such as disintegrin and metalloproteinases 9 , 10 , and 17 ( ADAM9/10/17 ) , matrix metalloproteinase-9 ( MMP9 ) , neutrophil elastase ( ELANE ) and other serine proteases ( Hirako et al . , 1998; Nishie , 2014 ) . There were no discernible changes between aged and young IFE skin in the expression levels of the proteases genes known to degrade COL17 ( Figure 4—figure supplement 4a ) . The involvement of MMP9 in COL17 degradation in aged IFE was further excluded by the observation of the IFE in Serpine1−/− mice , which lack plasminogen activator inhibitor-1 , blocking MMP9 activity ( Figure 4—figure supplement 4b ) . ELANE , which degrades COL17 adjacent to HFSCs ( Matsumura et al . , 2016 ) , was located along the BMZ of the IFE , an expression pattern different from that is seen in hair follicles and back skin ( Figure 4—figure supplement 4c ) . Physical aging did not affect ELANE labeling in the IFE . These data suggest that proteases may not play major roles in the loss of apico-lateral COL17 in the IFE . Next , we tested whether hCOL17 overexpression could reverse the aged IFE phenotype . K14-hCOL17 transgenic mice , which overexpress hCOL17 under the K14 promoter , showed COL17 on all surfaces of IFE basal cells at an age >19 months ( Figure 6a ) . In association with this observation , the epidermis was thinner , and the numbers of PH3- and PCNA-positive cells were reduced in the K14-hCOL17 aged IFE compared to the wild type ( Figure 6b ) . However , the expression levels of Itga6 and Itgb1 were similar in K14-hCOL17 aged IFE and age-matched controls ( Figure 6c ) . These data suggest that COL17 might act downstream of these stem cell markers and can keep the IFE in a juvenile state when overexpressed . Premature aged Col17a1−/− IFE ( 3 months old , Figure 4—figure supplement 1 ) exhibited epidermal thickening with an increased number of BrdU-positive basal cells , while the numbers of PCNA- or PH3-positive cells were comparable with age-matched controls ( Figure 6d ) . This result was akin to wild-type aged IFE with a loss of apico-lateral COL17 ( Figure 4c ) . Unlike neonatal IFE hypertrophy via inactive Wnt signaling , there were no obvious alterations in Wnt-related molecules in the young adolescent IFE ( Figure 6—figure supplement 1 ) . These results demonstrate that COL17 overexpression suppresses epidermal hyperproliferation associated with physical aging in the IFE . 10 . 7554/eLife . 26635 . 018Figure 6 . Overexpression of human COL17 ablates hyperproliferation in aged IFE . ( a ) COL17 labeling and its quantitative fluorescent intensity in IFE skin from WT and K14-hCOL17 aged mice ( >19 months old ) ( n = 5 ) . The antibody used in this assay detects both human and murine COL17 . Scale bar: 20 μm . Mann-Whitney test . ( b ) H&E- , E-cad- , PH3- and PCNA-stained skin samples from WT and K14-hCOL17 aged IFE . The numbers of epidermal layers , total epidermal cell counts , and PH3- and PCNA-positive basal cell counts ( n = 5 for H&E , E-cad , and PCNA staining; n = 4 for PH3 in aged K14-hCOL17 mice; n = 3 for PH3 of aged WT mice ) . Student’s t-test . Scale bar: 20 μm . ( c ) The gene expression levels of Itga6 and Itgb1 in IFE skin samples from WT and K14-hCOL17 aged IFE ( n = 5 ) . Student’s t-test . ( d ) H&E- , E-cad- , PH3- and PCNA-staining; quantifications of epidermal layers; total epidermal cells; and PH3- , PCNA- , and BrdU-positive basal cells in the IFE of Col17a1−/− mice and littermate controls ( 3 months old ) ( n = 4 ) . Student’s t-test . Scale bar: 20 μm . The data are the means ± SE . *0 . 01<p<0 . 05 , **0 . 001<p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 01810 . 7554/eLife . 26635 . 019Figure 6—figure supplement 1 . qRT-PCR assessment of Wnt-related molecules in IFE skin samples from Col17a1−/− and littermates at the age of 12 weeks . The data are the means±SE of three mice per each group . Student’s t-tests . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 019
BMZ proteins , including COL17 , are key attachment structures that support epidermal cell homeostasis throughout life ( Watt and Fujiwara , 2011 ) . Previously , it has been suggested that COL17 serves as a niche for HFSCs ( Matsumura et al . , 2016 ) , which subsequently maintain melanocyte stem cells ( Tanimura et al . , 2011 ) . Here , we present the unidentified role of COL17 in regulating homeostasis in the paw IFE in terms of proliferation ( Figure 7 ) . COL17 deletion in the neonatal epidermis led to the transient hyperproliferation of the IFE , mediated via waned Wnt-β-catenin signaling . Physical aging led to increased IFE proliferation and the loss of the non-hemidesmosomal COL17 distribution associated with altered aPKC signaling . The restoration of COL17 in aged IFE reversed the hyperproliferative state . 10 . 7554/eLife . 26635 . 020Figure 7 . A model of the role of COL17 in maintaining IFE homeostasis . A graphical abstract of this study . COL17 regulates paw IFE homeostasis in coordination with Wnt signaling at the neonatal stage . Physical aging diminishes non-hemidesmosomal COL17 labeling in IFE keratinocytes , leading to IFE hyperproliferation , associated with altered aPKC activities . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 020 Wnt-β-catenin signaling is one of master regulators of skin development and homeostasis in several components , including hair follicles , other appendages and the IFE ( Lim and Nusse , 2013; Lu and Fuchs , 2014 ) . Activating and inactivating mutations of Wnt-β-catenin signaling molecules give rise to distinct phenotypes in the IFE of murine back skin in terms of proliferation and differentiation ( Lim and Nusse , 2013 ) . Recently , the relationship between the IFE of murine paw skin and Wnt-β-catenin signaling was described ( Lim et al . , 2013 ) . IFE basal cells require Wnt-β-catenin signaling to proliferate and express both Wnt ligands and inhibitors by an autocrine process in adolescent mice ( Lim et al . , 2013 ) . Our study uncovers the role of COL17-stabilized Wnt signaling in governing neonatal IFE proliferation , although the interaction between COL17 and Wnt-related molecules remains elusive . Our observation of IFE hyperproliferation in the neonatal paw under inactive Wnt signaling ( either by genetically modified mice or by inhibitors ) is in contrast to the observed IFE hypotrophy in 1-month-old mice with inducible β-catenin loss-of-function ( Lim et al . , 2013 ) . This discrepancy might be due to the difference of the observational time points ( neonate vs 1-month-old ) and/or of the model systems used in each study ( conventional vs inducible ) . Integrins , which are also BMZ components , have been implicated as having functional roles in regulating epidermal proliferation and differentiation ( Margadant et al . , 2010; Watt , 2014 ) . Interestingly , mice and human epidermis deficient in Kindlin-1 , an integrin co-activator , also show altered Wnt and TGF-β signaling ( Rognoni et al . , 2014 ) . In contrast to Col17a1−/− IFE , the absence of Kindlin-1 resulted in the activation of Wnt signaling in the IFE . The previous and current findings suggest that the regulation of Wnt signaling in the IFE is subject to complex regulation by the epidermal BMZ . For epidermal maintenance and proliferation , the presence of epidermal stem cells is essential . However , the theories on how these stem cells reside in the epidermis and how they regulate epidermal homeostasis remain controversial and may be dependent on the location on the body ( Mascré et al . , 2012; Rompolas et al . , 2016; Sada et al . , 2016; Sánchez-Danés et al . , 2016 ) . In the IFE , it is proposed that the majority of basal cells exist as plausible committed progenitors ( CPs ) , which undergo cell division and differentiation , in part through asymmetric cell divisions ( Alcolea and Jones , 2014; Doupé and Jones , 2012 ) . In the foot pad , the proliferative potentials of CPs are elevated compared to other epidermal sites , and the clones of foot pad CPs are also enlarged in aged skin ( Dunnwald et al . , 2003 , 2001; Stern and Bickenbach , 2007 ) . Other in vitro studies report the expansion of proliferative CPs in the aged epidermis ( Charruyer et al . , 2009 ) . Our data on hypertrophic IFE in aged skin reinforce the notion that physical aging abolishes the quiescent state of the IFE . This hypothesis requires clarification using conventional fluorescent lineage tracking and/or genetics-based cell fate mapping ( e . g . sequencing lineage barcodes ) ( Woodworth et al . , 2017 ) . Moreover , in our study , the introduction of hCOL17 into aged IFE ablates epidermal hyperproliferation , which is in line with the rejuvenation of HFSCs upon hCOL17 overexpression ( Matsumura et al . , 2016 ) . These findings highlight the role of COL17 as an anti-aging molecule in the skin . As the epidermis is a highly structured tissue , epithelial polarity is required for homeostasis ( Tellkamp et al . , 2014 ) . Although physiological aging is reported to affect aPKCs in neurons ( Pascale et al . , 2007 ) , the alteration of aPKC in the epidermis upon aging is poorly understood . Our study adds skin to the list of organs in which aPKC is modified with aging . It is noteworthy that in early adult aPKCλ-deficient mice , the IFE showed a similar proliferative phenotype to those of aged WT IFE ( Niessen et al . , 2013; Osada et al . , 2015 ) . Asymmetric cell divisions , rather than symmetric cell division , are increased in these mice and are thought to force the gradual loss of quiescence , contributing to premature aged phenotypes ( Tellkamp et al . , 2014 ) . Altered COL17 distribution with aging or with pharmacological inhibition of aPKC might result from improper cell polarity . To further confirm our results , the observation of COL17 in an aPKCλ-deficient epidermis would be critical as we cannot completely exclude the involvement of non-specific drug effects on the epidermis in COL17 dynamics . Recently , aging has been illuminated as a trigger for the clonal expansion of stem cells , which disturbs tissue health by means of oncogenesis ( Goodell and Rando , 2015 ) . The accumulation of Ki-67-positive cells and prolonged hyperplasia in the aged epidermis during recovery from carcinogenic treatments has been reported ( Golomb et al . , 2015 ) . It would be intriguing to observe skin carcinogenesis upon COL17 deficiency , as its relevance in colon and lung cancers has been described ( Liu et al . , 2016a , 2016b; Moilanen et al . , 2015 ) . To further elucidate the mechanisms and dynamics of COL17 in regulating IFE homeostasis , an investigation with either drug-inducible expression or loss of COL17 expression under specific promoters is warranted . The discrepancy in the epidermal growth rate with aging between hyperproliferation , as observed in our findings and the reports of others ( Charruyer et al . , 2009; Stern and Bickenbach , 2007 ) , and hypoproliferation ( Giangreco et al . , 2008 ) might be attributed to the following: ( 1 ) UV exposure , which inhibits cell growth , on human skin; ( 2 ) a wide divergence of epidermal thickness among the different locations on the body ( Porter et al . , 1998 ) and among individual people ( Waller and Maibach , 2005 ) ; ( 3 ) differences in hygiene status among the animal facilities and ( 4 ) the influence of the development and cyclical growth of hair on the IFE of haired skin . Our observations can exclude the involvement of most of these extrinsic and intrinsic confounding factors by restricting the monitoring to the paw IFE of congenic mice at a single animal facility . However , it is noteworthy that chronic stimulation via ambulation might affect paw IFE homeostasis with physical aging , which is a limitation of our study . Considering chronic stimulation and lack of UV exposure , it would be fair to say that our data on the murine paw IFE could be extrapolated into the human palmoplantar and buttock epidermis . In closing , our study has revealed an unrecognized link between COL17 and epidermal proliferation in neonatal and aged IFE . We propose that COL17 is a good candidate to target to prevent epidermal aging and oncogenesis .
Col17a1−/− ( RRID:MGI:3711939 ) , K14-hCOL17 ( a courtesy gift of Prof . Kim B Yancey ) , hCOL17+;Col17a1−/− ( RRID:MGI:3711948 ) , and K14-deltaNLef1 ( RRID:MGI:2667413 ) mice were generated as previously described ( Niemann et al . , 2002; Nishie et al . , 2007 ) . ins-TOPGAL+ mice ( RRID:IMSR_RBRC05918 ) ( Moriyama et al . , 2007 ) were obtained from the RIKEN BRC ( Tsukuba , Ibaraki , Japan ) and bred with Col17a1−/− mice to produce ins-Topgal+;Col17a1−/− mice . C57BL/6 strain mice and Klotho−/− ( RRID:MGI:2181617 ) mice were purchased from Clea ( Tokyo , Japan ) . Serpine1−/− mice ( RRID:IMSR_JAX:002507 ) were purchased from The Jackson Laboratory ( Bar Harbor , Maine , USA ) . For label analysis , 1-day-old mice were intraperitoneally administered 10 μg BrdU ( BD Pharmingen , New Jersey , USA ) per head , and adult mice were intraperitoneally administered 8 . 33 μg/g per head 4 hr before sacrifice . The following antibodies were used: anti-E-cadherin ( Cell Signaling Technology , Danvers , Massachusetts , USA; 24E10 , RRID: AB_10694492 ) , anti-phospho histone H3 ( Ser10 ) ( Merck Millipore , Billerica , Massachusetts , USA , RRID: AB_11210699 ) , anti-PCNA ( Dako , Santa Clara , California , USA; PC10 , RRID: AB_2160651 ) , anti-BrdU ( Dako; M0744 , RRID:AB_10013660 ) , anti-loricrin ( Covance , Princeton , New Jersey , USA , RRID:AB_10064155 ) , anti-cleaved caspase-3 ( Cell Signaling Technology , RRID:AB_2341188 ) , FITC-conjugated anti-CD3e ( BioLegend , San Diego , California; 145–2 C11 , RRID:AB_394595 ) , Alexa Fluor 488-conjugated anti-F4/80 ( Affymetrix , Santa Clara , California , USA; BM8 , RRID: AB_893479 ) , FITC ( fluorescein isothiocyanate ) -conjugated anti-Ly-6G ( Beckman coulter , Brea , California , USA; RB6-8C5 , RRID: AB_394643 ) , anti-TGFβ1 ( Santa Cruz Biotechnology , Dallas , Texas , USA , RRID: AB_632486 ) , anti-p-Smad2 ( Cell Signaling Technology; 138D4 , RRID:AB_490941 ) , anti-LEF1 ( Cell Signaling Technology; C12A5 , RRID: AB_823558 ) , anti-β-catenin ( BD; 14/Beta-catenin , RRID: AB_397554 ) , anti-hCOL17 NC16A domain ( TS39-3 , homemade ) ( Ujiie et al . , 2014 ) , anti-extracellular portion of hCOL17 ( mAb233 , homemade ) ( Nishizawa et al . , 1993 ) , anti-murine COL17 NC14A domain ( MoNC14A , homemade ) ( Nishie et al . , 2015 ) , anti-cytoplasmic portion of COL17 ( Abcam , Cambridge , UK; ab186415 , 1A8c , homemade ( Nishizawa et al . , 1993 ) ) , anti-plectin ( Abcam; ab32528 , RRID: AB_777339 ) , anti-BP230 ( Cosmo bio , Tokyo , Japan; 239 , RRID:AB_1961833 ) , anti-desmogleins 1 and 2 ( PROGEN , Wieblingen , Heidelberg , Germany; DG3 . 10 , RRID: AB_1284107 ) , anti-integrin β1 ( Chemicon International , Billerica , Massachusetts , USA; MB1 . 2 , RRID: AB_2128202 ) , anti-integrin α6 ( BD Pharmingen; GoH3 , RRID: AB_2296273 ) , anti-ELANE ( Abcam ) , anti-phospho aPKC ( Santa Cruz; sc-271962 , RRID:AB_10708397 ) , and anti-survivin ( Cell Signaling Technology; 71G4B7 , RRID:AB_10691694 ) , anti-laminin β1 ( Abcam; LT3 , RRID: AB_775971 ) . Specimens from mice paw skin and human skin were fixed in formalin and embedded in paraffin after dehydration or were frozen on dry ice in an optimal cutting temperature ( OCT ) compound . Frozen sections were fixed with 4% paraformaldehyde ( PFA ) or acetone or were stained without fixation . Antigen-retrieval with pH 6 . 0 ( citrate ) or pH 9 . 0 ( EDTA ) buffer was performed on deparaffinized sections . Sections were incubated with the primary antibodies overnight at 4°C . After washing in phosphate-buffered saline ( PBS ) , the sections were incubated with secondary antibodies conjugated with Alexa488 , Alexa647 or FITC for 1 hr at room temperature ( RT ) . The nucleus was stained with propidium iodide ( PI ) or 4' , 6-diamidino-2-phenylindole ( DAPI ) . All the stained immunofluorescent samples were observed using a confocal laser scanning microscope ( Olympus Fluoview FV1000; Olympus , Tokyo , Japan , RRID:SCR_014215 ) . To quantify the relative intensity of the staining , the signals for immunofluorescence were obtained and analyzed using FV1000 software ( RRID:SCR_014215 ) or Image J ( NIH , Bethesda , Maryland , USA , RRID:SCR_003070 ) . The values were normalized by the section length or by the tissue area and are indicated as arbitrary units ( A . U . ) . For immunohistochemistry , horseradish peroxidase ( HRP ) -tagged antibodies were used . Sections were blocked before antibody labeling and counterstained with hematoxylin . Images were captured with a bright field microscope ( Nikon , Tokyo , Japan; Keyence , Tokyo , Japan ) . For staining , skin samples of normal humans , both young ( <10 years old ) and aged ( >85 years old ) , were obtained from non-sun-exposed areas . To distinguish hemidesmosomal from non-hemidesmosomal COL17 in basal keratinocytes , frozen skin sections were treated with 0 . 5% Triton X-100 in PBS for 1 hr at RT before primary antibody incubation , as described previously ( Hirako et al . , 1998 ) . To evaluate the skin thickness , at least three footpad areas per each mouse were counted and averaged . Epidermal cells were quantified as the number of PI-positive cells per epidermal length . The number of epidermal cell layers was evaluated using E-cadherin labeling . The numbers of basal cells positive for proliferation markers and Wnt-related molecules were counted and normalized using the total epidermal length or total basal cell number . For general histological analysis , each immunostaining was repeated in at least three independent experiments , and sweat gland areas were excluded from observation . For high-resolution structured illumination microscopy ( SIM ) imaging , an N-SIM microscope ( Nikon ) with an electron-multiplying charged-coupled device camera ( DU-897; Andor Technology , Tokyo , Japan ) was used as described previously ( Hashimoto et al . , 2016 ) . Image reconstruction was performed using NIS-Elements software ( RRID:SCR_014329 ) . In division axis orientation determination , the direction of cell division was verified using survivin staining as described previously ( Niessen et al . , 2013 ) . The angle of division was confirmed by scaling the angle of the plane transecting two daughter cells relative to the plane of the basement membrane as labeled laminin β1 . The total number of cell divisions including ACD and symmetrical cell division ( SCD ) were set to 100% per sample . HEK293 Cells ( RRID:CVCL_0045 , authenticated by STR gene profiling ) which constitutively expresses the SuperTopFlash construct ( SuperTopFlash 293 ( STF293 ) ) ( Tsukiyama et al . , 2015 ) was cultured in Dulbecco’s Modified Eagle medium ( DMEM ) supplemented with 10% bovine serum ( Sigma-Aldrich , St . Louis , Missouri , USA ) and 1% penicillin-streptomycin-amphotericin B ( Wako , Osaka , Japan ) . The cells used in this assay were subject to regular mycoplasma testing . Wnt 3a-conditioned medium ( Wnt3a CM ) was kindly provided by S . Takada . Full-length human COL17A1 cDNA ( NM_0004940 ) was subcloned into the mammalian expression vector pcDNA3 . 1/Zeo ( ThermoFisher , Waltham , Massachusetts , USA ) and transfected into STF293 cells . The transfected cells were selected with Zeocin ( Thermo Fischer Scientific , Waltham , Massachusetts , USA ) . STF293 cells stably expressing either human COL17A1 or empty vector were seeded onto 96-well plates and co-transfected with the pGL 4 . 75 plasmid , which contains a Renilla reniformis luciferase gene ( Promega , Fitchburg , Wisconsin , USA ) , using Lipofectamine2000 ( Invitrogen , Waltham , Massachusetts , USA ) and Opti-MEM ( Thermo Fischer Scientific ) . At 24 hr after seeding , Wnt3a CM was added to the medium at at a final concentration of 30% of the total medium by volume . After 2 days of stimulation with the Wnt3a CM , the Firefly and Renilla luciferase activities were measured using a Dual-Luciferase reporter assay system ( Promega ) and a Spectra Max Paradigm ( Molecular device , Tokyo , Japan ) . All the values were normalized to Renilla luciferase activity . Either IWP-2 ( Sigma Aldrich ) or Wnt-C59 ( Cayman Chemical , Ann Arbor , Michigan , USA; both dissolved in DMSO ) were intraperitoneally injected into WT mice at P0 ( IWP-2: 10 mg/g body , Wnt-C59: 20 mg/g body ) ( Carotenuto et al . , 2017; Kuo et al . , 2016 ) . The paw skin samples were collected 24 hr after injection . For X-gal staining on ins-Topgal+ mice skin , a beta-galactosidase staining kit ( Takara-bio , Shiga , Japan ) was used according to the provider’s protocol . Briefly , hindpaw samples were fixed with 4%PFA for 2 hr at RT and soaked in staining solution overnight . Tissues were mounted with a Mowiol solution . Images for evaluation were collected from the heel of the hind paw to exclude the involvement of hair follicles and sweat glands . The individual light settings of each experiment were identical for both control and Col17a1−/− mice , when capturing images with a bright field microscopy ( Nikon ) . The LacZ-positive area was quantified by ImageJ . For epidermal whole mount staining , epidermal sheets were taken from whole paw mice skin as described previously with some modifications ( Kubo et al . , 2009 ) . For cell membrane detection , Alexa633-conjugated wheat germ agglutinin ( Invitrogen , Waltham , Massachusetts , USA ) was used . Three-dimensional ( 3D ) reconstruction images were generated , and analyses were performed using the Metamorph software ( Molecular Devices , Tokyo , Japan , RRID:SCR_002368 ) . Whole mice paw skin samples were treated with the following inhibitors for 1 hr at 4°C: EDTA , pan-PKC inhibitor Go6983 ( Tocris Bio , Bristol , UK ) and myristoylated pseudosubstrate Inhibitor ( myr PSI; Calbiochem , Billerica , Massachusetts , USA ) before staining . These inhibitors were used at the indicated concentrations as previously described ( Amagai et al . , 1995; Atwood et al . , 2013; Wu et al . , 2006 ) . Reconstituted 3D epidermis ( LabCyte EPI-MODEL ) derived from cultured human epidermal keratinocytes was purchased from J-TEC ( Aichi , Japan ) . Prior to pharmacological treatment , the reconstituted epidermis was cultured in the medium for 12 hr according to the manufacturer’s protocol . The treatment details were as described in ex-vivo inhibitor treatment . RNA was extracted from whole paw skin using the RNeasy Mini kit ( QIAGEN ) , and the cDNA was prepared using the SuperScript III First-Strand Synthesis System ( Thermo Fisher Scientific ) . The qRT-PCR was performed using the designated primers and fast SYBR Green ( Thermo Fisher Scientific ) in a STEP-One Plus sequence detection system ( Applied Biosystems , Waltham , Massachusetts , USA ) . All the primers used in this study are listed in Table 1 . 10 . 7554/eLife . 26635 . 021Table 1 . Primers used in qRT-PCR . DOI: http://dx . doi . org/10 . 7554/eLife . 26635 . 021Gene nameForward ( 5′→3′ ) Reverse ( 5′→3′ ) mTcf7l1TGGTCAACGAATCGGAGAATTCACTTCGGCGAAATAGTCGmTcf7l2CTCCACAGCTCAAAGCATCACACCACCTTCGCTCTCATCTmLef1CGCTAAAGGAGAGTGCAGCTAGCTGTCTCTCTTTCCGTGCTmTcf7GCCAGAAGCAAGGAGTTCACACTGGGCCAGCTCACAGTATmCtnnb1AAGGCTTTTCCCAGTCCTTCCCCTCATCTAGCGTCTCAGGmFzd7GACCAAGCCATTCCTCCGTGCAGGTAGGGAGCAGTAGGGTAmFzd4AACCTCGGCTACAACGTGACGGCACATAAACCGAACAAAGGAAmNfatc4CAAGCTGCGAGGATGAGGAGACAGCATGGAGGGGTATCCTmWnt4GTCAGGATGCTCGGACAACATCACGTCTTTACCTCGCAGGAmWnt5aGGACCACATGCAGTACATTGGCGTCTCTCGGCTGCCTATTTmFzd3TGATGAGCCATATCCCCGACTGCCTATGAAATAGCGAGCAAATGmFzd8CCGCTGGTGGAGATACAGTGCGGTTGTAGTCCATGCACAGmItga6ATGCCACCTATCACAAGGCTGCATGGTATCGGGGAATGCTmItgb1ATCATGCAGGTTGCGGTTTGTGGAAAACACCAGCAGTCGTmTgm1TTTGATGGGTGGCAGGTTGTGCCATTCTTGACGGACTCCAmCol17GATGGCACTGAAGTCACCGATATCCATTGCTGGTGCTCCCmPplGCATGCTGAGTGGAAGGAGTAAGTCTGAGTCCACCTTGCGmEvplTCCTACAAGCTGCAAGCACATCTAAGGAGCAGCGGTAGGTmIvlCTCCTGTGAGTTTGTTTGGTCTCACACAGTCTTGAGAGGTCCCmCyc1ATCGTTCGAGCTAGGCATGGGCCGGGAAAGTAAGGGTTGAmGusbCAGGGTTTCGAGCAGCAATGACCCAGCCAATAAAGTCCCGmGapdhTCCTGCACCACCAACTGCTTAGCTGGATGCAGGGATGATGTTCTGGhCOL17TCAACCAGAGGACGGAGTCATCGACTCCCCTTGAGCAAACh18SGGCGCCCCCTCGATGCTCTTAGGCTCGGGCCTGCTTTGAACACTCT An in situ skin permeability assay using Toluidine blue was performed as previously described ( Hardman et al . , 1998 ) . Transepidermal water loss ( TEWL ) of the embryo back skin was measured using an evaporimeter ( AS-VT100RS;Asahi Biomed , Tokyo , Japan ) ( Yanagi et al . , 2008 ) . Biopsy samples from paw skin specimens were fixed in 2% glutaraldehyde solution , post-fixed in 1% OsO4 , dehydrated , and embedded in Epon 812 . The embedded samples were sectioned at 1 µm thickness for light microscopy and thin-sectioned for electron microscopy ( 70 nm thick ) . The thin sections were stained with uranyl acetate and lead citrate and examined by transmission electron microscopy ( H-7100; Hitachi , Tokyo , Japan ) . Normal human epidermal keratinocytes ( NHEKs , Lonza ) in passages 3–4 were seeded with KGM-Gold ( Lonza ) in a 96-well plastic bottom plate . The cells were transfected with 10 μM of either human COL17A1 siRNAs or the control ( Mock ) ( Silencer Select siRNAs , Thermo Fisher , Waltham , Massachusetts , USA ) using RNAimax ( Thermo Fisher ) and Opti-MEM ( Thermo Fisher ) . Cell growth curves were generated using an xCELLigence system ( ACEA bioscience , San Diego , California , USA , RRID:SCR_014821 ) at 24 hr after the knockdown procedure . For the colony formation assay , 1000 cells/ml of NHEKs ( passages 3–4 ) were seeded in 6-well plates on mitomycin C ( Wako , Osaka , Japan ) treated 3T3-J2 feeder cells ( RRID:CVCL_W667 ) and were transfected with either COL17A1 siRNA or mock siRNA , as described above . Cells were maintained in KGM-Gold with 10% fetal bovine serum ( Sigma-Aldrich , St . Louis , Missouri , USA ) for 2 weeks . To detect the colonies , cells were fixed with 4% paraformaldehyde for 10 min and stained for 20 min with 1% rhodamine ( Sigma-Aldrich ) diluted in distilled water . The colony number and size were analyzed by ImageJ ( NIH , Bethesda , Maryland , USA ) . All the cells used in this assay were subject to regular mycoplasma testing . Cultured cells were collected and lysed in 1% NP-40-containing buffer . The samples were loaded on a NuPAGE 4–12% Bis-Tris gel ( Thermo Fisher ) and then transferred to a PVDF membrane . The membrane was incubated with primary antibodies ( anti-C-terminal region of human COL17 ( 09040 ) ( Ujiie et al . , 2014 ) and anti-beta tubulin ( Abcam , Cambridge , UK , RRID:AB_2210370 ) ) followed by secondary antibodies conjugated with horseradish peroxidase . The blots were detected using ECL-Plus ( GE Healthcare ) . Skin samples from two patients with the generalized other subtype of JEB were used in this study . The JEB patients harbored truncation mutations in COL17A1 , and their skin specimens were negative for COL17 expression ( Nakamura et al . , 2006 ) ( Masuda et al . , manuscript in preparation ) . The patients were a newborn Japanese male and a 15-year-old Japanese male . As controls for the immunofluorescence staining , skin samples were collected from age- , site- , and sex-matched healthy individuals . The total RNA extracted from young ( 6- to 10-week-old ) and aged ( 19- to 27-month-old ) mouse whole paw skin was hybridized to an Agilent SurePrint G3 Mouse v2 8 × 60K 1 color 8 microarray . N = 4 for each group . The microarray data were analyzed using GeneSpring software ( Agilent Technology , Santa Clara , California , USA , RRID:SCR_009196 ) . Microarray data are available in the ArrayExpress database ( www . ebi . ac . uk/arrayexpress , RRID:SCR_002964 ) under accession number E-MTAB-4916 . The authors Y . K . and M . N . previously introduced a mathematical model that could simulate the homeostasis of the epidermis ( Kobayashi et al . , 2016 ) . In this model , the four epidermal processes were taken into account: ( i ) the kinetic interactions of epidermal cells , ( ii ) the reproduction of stem cells and daughter cells , ( iii ) differentiation , and ( iv ) calcium dynamics . Migration of the cells after leaving the basal layer was naturally realized by the hard-core repulsive interactions among the cells , and the differentiation process was controlled by the calcium dynamics . Numerical investigation of the model revealed that because of the calcium dynamics , the thickness of the epidermis was maintained , and the lower boundary of the stratum corneum retained a flat structure . For the present purpose , we modified this model in such a way that we could investigate the effect of the binding strength of the stem cells to the basal membrane . The kinetic interactions and the reproduction process were modified as follows . Each cell was represented as a sphere characterized by its position xi= ( xi , yi , zi ) and radius R . The basal membrane was modeled as a monolayer of densely packed immobile particles , each having the radius Rm . The cells obeyed the following equations of motion:μ dxidt=−∂∂xi ( ∑j∈ΩLJVLJ ( |xi−xj| ) +∑j∈ΩmVm ( |xi−xj| ) ) , where μ is the coefficient of friction; VLJ is the repulsive interaction between cell i and cell j , which is either one of the adjacent cells or the nearest membrane particle; and Vm is the interaction with the membrane . The interactions were considered only for the particles that are in contact with cell i . The first summation was taken over the set ΩLJ , which consists of both cells and membrane particles , and the second summation was over Ωm , which contains only the membrane particles . The first potential was given byVLJ ( r ) =ε{ ( R+Rjr ) 6−12 ( R+Rjr ) 12 } , where ε is a positive constant . If cell j , interacting with i , represented the epidermal cell , then Rj=R , and if cell j is the membrane particle , then Rj=Rm . The second potential Vm was assumed to be cell-dependent . For the suprabasal cells , Vm=0; for the stem cells , the interaction was spring-like:Vm ( r ) =Ks2{ ( r− ( R+Rm ) }2 , where Ks is a positive constant . For the daughter cells , it was assumed to be a nonlinear spring:Vm ( r ) ={Kd2[a{ ( r− ( R+Rm ) }2−b2{ ( r− ( R+Rm ) }4]0 ( r > r∗ ) , ( r≤r∗ ) , where Kd , a , and d are positive constants and r*=R+Rm+a/b . That is , the daughter cells could leave the basal layer when the distance became greater than the threshold r* , while the stem cells were bound to the basal membrane and could not leave it . The cell reproduction process was modeled in the following way: each reproducible cell was assigned a deterministic cell cycle with the period T . After the period elapsed , the cell entered into the stochastic division process , which was the Poisson process with the rate γ . When the division of cell i started , the new particles j and k were created with the radius R at the same place as cell i , with cell i itself deleted: xj ( t0 ) =xk ( t0 ) =xi ( t0 ) . Note that the overlapping of j and k was allowed . The force exerted on j by k was given as the partial derivative −∂∂xj of the following elastic potential:U ( | xj−xk | ) =K'2 ( | xj−xk |−l ( t ) ) 2 , where K' is a positive constant and l ( t ) is the natural distance between j and k , which linearly increases as l ( t ) =α ( t−t0 ) with a constant α . When l ( t ) reached l ( t ) =2R , the division was considered to be completed , and the potential U was no longer computed . The differentiation processes and calcium dynamics were the same as in the manuscript ( Kobayashi et al . , 2016 ) , which were summarized as follows: each suprabasal cell was assigned a degree of differentiation S as an internal variable . When S reached the threshold S* , the cell underwent terminal differentiation and became a cornified cell . The advancement of S was accelerated when the calcium concentration of the cell was high , and calcium excitation was induced when the cell was in contact with the cornified cells , reflecting the fact that high-calcium concentration was observed beneath the stratum corneum . We numerically solved this model with the following initial conditions . According to previous results , we first obtained the steady states of the epidermis with an undulated basal membrane . We used these states as initial conditions . For each of five different initial conditions , we decreased the binding strength , Kd , to the basal membrane from Kd=5 . 0 to Kd=2 . 0 and ran simulations . For comparison , we also ran simulations for each of the initial conditions without changing Kd=5 . 0 . Other parameters were chosen: μ=1 . 0 , R=1 . 4 , Rm=1 . 0 , ε=1 . 0 , Ks=25 . 0 , a=0 . 0868 , b=0 . 376 , K'=5 . 0 , γ=0 . 00813 , and α=0 . 14 . By choosing these parameters and others used in the previous work [1] , we assumed that the average cell division and turnover periods were approximately 3 and 28 days , respectively . Figure 1F shows the ratio of the epidermis to its initial value . Each curve represents the ensemble average of five independent simulations with different shapes of the basal membrane . Statistical analysis was performed using GraphPad Prism ( GraphPad Software , La Jolla , California , USA , RRID:SCR_002798 ) . p-Values were determined using Student’s t-test , the Mann-Whitney test or one-way ANOVA followed by Tukey’s test . p-Values are indicated as *0 . 01<p<0 . 05 , **0 . 001<p<0 . 01 , ***0 . 0001<p<0 . 001 , ****p<0 . 0001 . The values are shown as the means ± standard errors ( SE ) . Sample size for animal experiments was determined on the basis of pilot experiments . | The skin is one of the largest organs of the body and is constantly confronted with a range of external stresses including germs , heat and scratches . The outermost part of the skin is called the epidermis and it acts as a barrier to the external environment and works to stop the body from losing water . An abnormally thin or thick epidermis can impair the skin’s ability to perform these roles . As such , the ability of epidermal cells to proliferate ( i . e . divide to make new cells ) is tightly regulated , both when the animal first develops and when it ages . However , most of the underlying mechanisms that regulate these processes are unknown . Watanabe et al . have now identified type XVII collagen ( called COL17 for short ) as a key molecule that controls how often epidermal cells in skin from mice and humans divide . COL17 is a protein that is made in the deepest layer of the epidermis , and it prevents the epidermis from thickening in newborn mice by coordinating with the Wnt signaling pathway . This signaling pathway , amongst other things , controls how often some cells divide . Older mice have a thicker epidermis than their younger counterparts . Watanabe et al . revealed that the distribution of COL17 in the epidermis also changes dramatically with age in mice and humans . Further experiments with mice showed that introducing COL17 back into the epidermis helped the tissue retain a more youthful state even in animals that had reached an old age . Together these findings give scientists a better understanding of how the ability of epidermal cells to divide is regulated at various stages in a mammal’s life . The new findings also point to COL17 as a promising component in future anti-aging strategies targeted at the skin . Yet first , further work will be needed to uncover how the production of COL17 is controlled in the epidermis . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2017 | Type XVII collagen coordinates proliferation in the interfollicular epidermis |
The intracellular signaling molecule Dishevelled ( Dvl ) mediates canonical and non-canonical Wnt signaling via its PDZ domain . Different pathways diverge at this point by a mechanism that remains unclear . Here we show that the peptide-binding pocket of the Dvl PDZ domain can be occupied by Dvl's own highly conserved C-terminus , inducing a closed conformation . In Xenopus , Wnt-regulated convergent extension ( CE ) is readily affected by Dvl mutants unable to form the closed conformation than by wild-type Dvl . We also demonstrate that while Dvl cooperates with other Wnt pathway elements to activate canonical Wnt signaling , the open conformation of Dvl more effectively activates Jun N-terminal kinase ( JNK ) . These results suggest that together with other players in the Wnt signaling pathway , the conformational change of Dvl regulates Wnt stimulated JNK activity in the non-canonical Wnt signaling .
The multiple Wnt signaling–related pathways are crucial to various developmental processes ( Logan and Nusse , 2004; Angers and Moon , 2009 ) . By regulating the cellular β-catenin level , canonical Wnt signaling controls cell fate , while non-canonical Wnt signaling plays a key role in controlling convergent extension ( CE ) and polarized cellular orientation . Dishevelled ( Dvl , or Dsh in Drosophila ) , a key component of both Wnt signaling pathways , relays Wnt signals downstream from the membrane-bound Wnt receptor Frizzled ( Fz ) ( Noordermeer et al . , 1994; Theisen et al . , 1994; Axelrod et al . , 1998; Wong et al . , 2003; Park et al . , 2005; Wang et al . , 2006; Schwarz-Romond et al . , 2007; Simons et al . , 2009 ) . While Dvl mediates both canonical and non-canonical Wnt signals , different pathways diverge at this point . Therefore , Dvl has been described as the ‘policeman’ at the intersection who directs different Wnt signals in different directions ( Boutros and Mlodzik , 1999 ) . However , the mechanism by which Dvl relays Wnt signals from Fz to different downstream components is not well understood . Dvl contains highly conserved DIX , PDZ , and DEP domains and a highly conserved extreme C-terminus ( Figure 1 ) ( Wharton , 2003; Wallingford and Habas , 2005 ) . The PDZ and DIX domains are reported to be involved in canonical Wnt signaling , while the PDZ and DEP domains play a crucial role in non-canonical Wnt signaling ( Sokol et al . , 1995; Axelrod et al . , 1998; Boutros et al . , 1998; Li et al . , 1999; Moriguchi et al . , 1999; Yamanaka et al . , 2002; Gao et al . , 2010 ) . The central PDZ domain not only participates in both pathways but also binds directly to the membrane-bound Wnt receptor Fz ( Wong et al . , 2003 ) . Many Wnt signaling regulators have been reported to mediate the different Wnt signaling pathways by interacting directly with the PDZ domain of Dvl ( Wharton , 2003; Wallingford and Habas , 2005; Gao and Chen , 2009 ) . Notably , in most species the extreme C-terminus of Dvl resembles a class III PDZ-binding motif ( E/D-X-Ф , where Ф represents hydrophobic residues such as F , I , L , M , or V ) , while that of Dsh resembles a class II PDZ-binding motif ( Ф-X-Ф ) ( Figure 1 ) ( Tonikian et al . , 2008; Lee and Zheng , 2010 ) . Because these two motifs suggest the possibility of intramolecular binding , we hypothesized that the C-terminus of Dvl/Dsh binds intrinsically to the Dvl PDZ domain . 10 . 7554/eLife . 08142 . 003Figure 1 . The C-terminal tail of Dishevelled ( Dvl ) is a PDZ domain binding motif . Sequence alignment of the C-terminus of Dvl/Dsh from selected species ( Wallingford and Habas , 2005 ) , showing residue numbers . DOI: http://dx . doi . org/10 . 7554/eLife . 08142 . 003 Here we use biophysical methods to investigate the interaction of the Dvl extreme C-terminus with the Dvl PDZ domain and demonstrate that Dvl adopts a closed conformation . We also show , in a Xenopus model , that disruption of this intramolecular interaction activates Jun N-terminal kinase ( JNK ) and enhances the CE phenotype associated with activation of non-canonical Wnt signaling . Further , we demonstrate that a Dvl PDZ-binding peptide or small molecule that inhibits canonical β-catenin signaling enhances JNK activity by releasing the Dvl C-terminus from its autoinhibitory closed conformation .
After our initial pull-down test indicated that Dvl C-terminus might interact with the Dvl PDZ domain , we decided to use two different biophysical assays to determine the binding affinity of the Dvl PDZ domain for the Dvl-C peptide and for a peptide derived from the C-terminus of Drosophila Dsh ( ‘Dsh-C peptide’ ) . We first used a competitive binding assay . Binding of the PDZ domain to a fluorescently-labeled peptide ( Rox-DprC ) derived from the C-terminus of Dapper ( Dpr ) , a known binding partner of the Dvl PDZ domain ( Cheyette et al . , 2002 ) , was monitored by fluorescence polarization . Both Dvl-C and Dsh-C peptides competitively inhibited the interaction of Rox-DprC with the Dvl-1 PDZ domain ( Figure 2 ) , indicating that the three peptides competed for the same binding site on the PDZ domain . The inhibition constants ( KI ) calculated from two independent experiments ( 12 . 3 ± 7 . 8 μM for the Dvl-C peptide and 26 . 8 ± 8 . 4 μM for the Dsh-C peptide ) were similar to the binding affinity of Rox-DprC to the PDZ domain ( KD ∼ 7 . 9 ± 0 . 9 μM ) ( Lee et al . , 2009a , 2009b ) . We then measured the binding between Dvl-C and the Dvl-1 PDZ domain by isothermal titration calorimetry ( ITC ) ( Figure 3 ) and obtained a KD of 7 . 0 ± 0 . 7 μM , which is consistent with the KI value observed in the fluorescence study . 10 . 7554/eLife . 08142 . 004Figure 2 . Competitive binding experiments . The KD value of the fluorescence-labeled Dapper ( Dpr ) -derived peptide Rox-DprC was obtained by plotting 1/ΔmP vs 1/[PDZ] , where ΔmP is the fluorescence polarization change ( ×1000 ) of Rox-DprC and [PDZ] is the concentration of the PDZ domain of Dvl . The KI values of the Dvl-C and Dsh-C peptides were obtained by using the equation KDapp = ( KD/ ( 1 + [I]/KI ) ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08142 . 00410 . 7554/eLife . 08142 . 005Figure 3 . Isothermal titration calorimetry experiment . The MicroCal Auto-iTC-200 was used to obtain the binding affinity of Dvl-C peptide and Dvl PDZ protein in 50 mM phosphate buffer . The concentration of Dvl-C peptide in the syringe was 1 . 05 mM and the concentration of Dvl PDZ domain in the cell was 0 . 114 mM . The KD value was averaged from two independent experiments at 25°C . DOI: http://dx . doi . org/10 . 7554/eLife . 08142 . 005 We next conducted NMR chemical shift perturbation studies ( Wong et al . , 2003 ) . Unlabeled Dvl-C and Dsh-C peptides were repetitively titrated into a solution of 15N-labeled Dvl-1 PDZ domain . During these titrations , the NMR resonance of the residues in the PDZ domain's peptide binding site ( Wong et al . , 2003; Lee et al . , 2009a , 2009b ) exhibited large chemical shift perturbations ( Figure 4 ) , further suggesting that both peptides interact with the Dvl PDZ domain . Moreover , during the Dvl-C peptide titration , a few resonances of the PDZ domain disappeared and reappeared ( Figure 4 ) , indicating that the Dvl-C peptide binds to the Dvl1 PDZ domain with higher binding affinity than the Dsh-C peptide does and that the complex is formed in the intermediate exchange range on the NMR time scale . 10 . 7554/eLife . 08142 . 006Figure 4 . Direct interaction of the Dvl C-terminus and PDZ domain . ( A ) Overlap of 1H-15N HSQC spectra of 15N-labeled PDZ domain without ( blue ) and with the unlabeled peptide ( SEFFVDVM ) derived from the extreme C-terminus of Dvl . Free: blue; final: red; ratio of peptide:protein = 20:1 . ( B ) Overlap of 1H-15N HSQC spectra of 15N-labeled PDZ domain without ( blue ) and with unlabeled peptide ( QDVSVSNYVL ) derived from the C-terminus of Drosophila Dsh ( Dsh-C ) . Blue: free; red: final; ratio of peptide:protein = 20:1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08142 . 006 To structurally analyze the binding of the Dvl-1 PDZ domain and the Dvl-C peptide , we determined the solution structure of the complex that formed ( Figure 5 ) . The Dvl PDZ domain contains six β-strands ( βA∼βF ) and two α-helix ( αA and αB ) structures ( Wong et al . , 2003; Lee et al . , 2009a , 2009b ) . As expected , we found that the Dvl-C peptide fits into the αB/βB peptide-binding groove of the Dvl PDZ domain and forms an additional β-strand with the αB-structure of the Dvl PDZ domain ( Figure 5A–C ) . Nuclear Overhauser effect ( NOE ) data indicated that six residues in the Dvl-C peptide participate in binding ( Figure 5—source data 1 , 2 ) . The side chain of Met ( 0 ) in the Dvl-C peptide is located within a hydrophobic pocket formed by the side chains of residues Leu262 , Ile264 , and Ile266 in the βB-structure and residues Val325 , Leu321 , and Val318 in the αB-helix structure ( Figure 5C , D ) . Notably , the side chain of residue Asp ( -2 ) in the Dvl-C peptide forms a hydrogen bond with the side chain of residue Arg322 in the Dvl PDZ domain ( Figure 5E ) . Consistent with this finding , the Arg322Ala substitution dramatically weakened the binding of Dvl PDZ to the Dvl-C peptide ( Figure 5—figure supplement 1 ) ( Lee et al . , 2009a ) . The side chains of two hydrophobic Dvl-C residues , Val ( -3 ) and Phe ( -5 ) , interact with the side chains of the Val318 and Ile266 residues in the Dvl PDZ domain ( Figure 5E ) . 10 . 7554/eLife . 08142 . 007Figure 5 . Solution NMR structure of Dvl PDZ domain in complex with Dvl-C peptide . ( A ) 2D plane of 3D 13C-F1-half-filtered F2-edited NOESY-HSQC spectrum ( mixing time , 300 ms ) at 15°C . The ratio of peptide:protein was 10:1 . [13C , 15N-PDZ] = ∼1 mM . ( B ) A stereo view of the backbone of 15 superimposed structures of the Dvl PDZ–Dvl-C peptide complex . ( C ) Ribbon diagram of the lowest-energy structure of the Dvl PDZ/Dvl-C peptide complex . ( D ) Surface of Dvl-1 PDZ bound to Dvl-C peptide ( carbon , green; nitrogen , blue; sulfur , yellow; oxygen , red; hydrogen atoms are omitted for clarity ) . ( E ) Structural details of the Dvl-C peptide–PDZ domain complex . The side chain of Asp ( -2 ) in the Dvl-C peptide forms a hydrogen bond with the side chain of Arg322 on the αB-structure . DOI: http://dx . doi . org/10 . 7554/eLife . 08142 . 00710 . 7554/eLife . 08142 . 008Figure 5—source data 1 . Intermolecular NOEs between the Dvl-C peptide and the PDZ domain obtained from 13C-half-filtered NOESY-HSQC spectraa . DOI: http://dx . doi . org/10 . 7554/eLife . 08142 . 00810 . 7554/eLife . 08142 . 009Figure 5—source data 2 . Structure statistics for the 15 lowest-energy peptide-PDZ complexes . DOI: http://dx . doi . org/10 . 7554/eLife . 08142 . 00910 . 7554/eLife . 08142 . 010Figure 5—figure supplement 1 . The mutant Dvl-1 PDZ ( R322A ) domain binds more weakly than wild-type Dvl-1 PDZ domain to the Dvl-C peptide . Overlap of the 1H-15N HSQC spectra of the labeled mutant Dvl PDZ ( R322A ) domain and the Dvl-C peptide . The mutation dramatically weakens interaction with the Dvl-C peptide ( free: blue; final titration: red; final peptide:protein ratio = 4:1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08142 . 010 To verify that the binding pocket of the Dvl PDZ domain is occupied by the intrinsic C-terminus , we generated two constructs: mC1 ( residues 251–695 of mouse Dvl-1 ) , containing the PDZ and DEP domains ( with an intact PDZ-binding motif ) and mC1-CΔ7 ( residues 251–688 ) , lacking the PDZ-binding motif . We then examined the binding of Rox-DprC to the two constructs . Little polarization change was observed when mC1 proteins were added to Rox-DprC solution , whereas large polarization changes were induced by adding mC1-CΔ7 ( Figure 6 ) , indicating that the binding pocket of the Dvl PDZ domain was occupied by its intrinsic C-terminus in the mC1 construct . The binding affinity between Rox-DprC and mC1-CΔ7 was 3 . 7 ± 0 . 5 μM , closely approximating the affinity between Rox-DprC and the Dvl-1 PDZ domain . In contrast , the estimated binding affinity between Rox-DrpC and mD1 was greater than 70 μM , suggesting not only that the Dvl PDZ domain binds intrinsically to the Dvl C-terminus but also that this binding interferes with interactions between the Dvl PDZ domain and its other binding partners in the Wnt signaling pathways . 10 . 7554/eLife . 08142 . 011Figure 6 . The binding pocket of the Dvl PDZ domain can be occupied by its intrinsic C-terminus . ( A ) Schematic representation of protein constructs mC1 ( residues 251–695 ) and mC1-CΔ7 ( residues 251–688 ) numbered according to the mouse Dvl-1 protein sequence . ( B ) Polarization change of the fluorescence-labeled peptide Rox-DprC ( Rox-SGSLKLMTTV , derived from the C-terminus of Dpr ) after addition of mC1-CΔ7 and mC1 proteins in 50 mM phosphate with 0 . 3 M NaCl and 6 mM β-mercaptoethanol . For the binding of Rox-DprC to mC1 , KD is 3 . 8 ± 0 . 5 μM the value was obtained by fitting the titration data with the equation: ΔmP = ΔmPmax × [P]/ ( [P] + KD ) , where ΔmP is the polarization change of Rox-DprC , [P] is the concentration of protein , and both KD and ΔmPmax are the fitting variables . For the binding of Rox-DprC to mC1-CΔ7 , KD was estimated as 68 ± 5 μM . Because of the limitation in the titration study , to estimate the KD value , although we used the same equation to fit the titration data , in the fitting , Kd was the only variable and the maximum polarization change , ΔmPmax , was fixed to the value that was obtained in the titration study of Rox-DprC binds to mC1-CΔ7 . DOI: http://dx . doi . org/10 . 7554/eLife . 08142 . 011 After establishing that Dvl can adopt a ‘closed’ conformation in which its C-terminus binds to its PDZ domain , we used a Xenopus model ( Xenopus Dvl-2 , XDsh ) ( Sokol et al . , 1995; Sokol , 1996; Umbhauer et al . , 2000 ) to investigate how this intrinsic interaction affects the role ( s ) of Dvl in the Wnt signaling pathways . We first generated two constructs , both myc-tagged at the N-terminus: wild-type XDsh ( residues 1–732 ) ( Sokol , 1996 ) and mutant XDsh-CΔ8 ( residues 1–724 , lacking the C-terminal PDZ-binding motif and therefore unable to form the intrinsic interaction between the its PDZ domain and the C-terminus ) . To avoid interfering the function of Dvl C-terminus , we placed the myc-tag at the N-terminus . However , the tag may affect the N-terminal DIX domain or potentially have some nonspecific effects . Therefore , in the Xenopus studies , we always paired the two constructs in the experiments to minimalize the potential of unexpected effects . To examine canonical Wnt signaling , we used a luciferase assay with a siamois promoter reporter ( Brannon et al . , 1997 ) . The Xenopus Wnt target gene construct with siamois promoter-driven luciferase reporter and equivalent quantities of XDsh-CΔ8 and wild-type XDsh mRNA , respectively , were coinjected into the animal pole region of 2-cell Xenopus embryos , and luciferase activity was then assayed in ectodermal explants dissected at the late blastula stage . Three different doses ( 80 pg , 200 pg , and 500 pg ) of XDsh and XDsh-CΔ8 mRNAs were used . At low and intermediate doses , neither XDsh nor XDsh-CΔ8 substantially affected canonical Wnt signaling ( not shown ) . At the high dose , both constructs enhanced canonical Wnt signaling , but wild-type XDsh was a stronger activator ( Figure 7 ) , consistent with a previous report showing involvement of the extreme C-terminal region of Dvl in activation of canonical Wnt signaling ( Tauriello et al . , 2012 ) . 10 . 7554/eLife . 08142 . 012Figure 7 . Effect of wild-type XDsh and XDsh-CΔ8 activity on canonical Wnt signaling . Luciferase assay using a Siamois promoter reporter ( Sialuc ) . Sialuc DNA ( 200 pg ) was injected alone or with myc-tagged Xdsh or Xdsh-CΔ8 mRNA ( 500 pg ) into the animal pole region of 2-cell Xenopus embryos . Ectodermal explants were dissected at the early gastrula stage for luciferase assay . Values are the means ±SD from four independent experiments ( p < 0 . 05 ) . Inset shows a representative western blot using anti-myc antibody ( 9E10 ) to control for XDsh and XDsh-CΔ8 protein expression in the four experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08142 . 012 To examine how the conformational change of Dvl affects the Wnt-regulated CE phenotype ( associated with the non-canonical Wnt signaling ) , we injected three different doses of wild-type XDsh and of XDsh-CΔ8 mRNA into the dorsal blastomeres of 4-cell Xenopus embryos . At all three doses ( 80 pg , 200 pg , and 500 pg ) , XDsh-CΔ8 mRNA caused greater body axis shortening and dorsal tail flexion than did wild-type XDsh ( Figure 8A ) , suggesting that XDsh-CΔ8 is more active than wild-type XDsh in inducing the CE phenotype . We then compared Xdsh-CΔ8 with another Dvl mutant that lacks the PDZ domain , the well-established Xdd1 ( Sokol , 1996 ) . Like XDsh-CΔ8 , Xdd1 is in ‘open’ conformation because it has no PDZ domain to interact with its C-terminus . We injected equal quantities ( 500 pg ) of N-terminal myc-tagged Xdd1 , XDsh-CΔ8 , and wild-type XDsh mRNA , respectively , into the dorsal blastomeres of 4-cell Xenopus embryos and found that the phenotypes were very similar after injection of XDsh-CΔ8 and after injection of Xdd1 ( Figure 8B ) , suggesting that the ‘open’ Dvl is more active than the ‘closed’ wild-type Dvl in inducing the Xenopus CE phenotype . 10 . 7554/eLife . 08142 . 013Figure 8 . The open conformation of Dvl significantly enhances gain-of-function planar cell polarity ( PCP ) signaling . ( A ) Xenopus embryonic abnormal convergent extension ( CE ) phenotypes induced by injection of wild-type XDsh and XDsh-CΔ8 mRNA at three increasing concentrations ( arrows at bottom represent 80 pg , 200 pg and 500 pg of injected mRNA; above are numbers of embryos injected from two independent experiments ) . Phenotypes are severe ( green ) , mild ( red ) , and normal ( blue ) . ( B ) Comparison of phenotypes induced by dose-equivalent injections ( 500 pg mRNA ) of XDsh , XDsh-CΔ8 , and Xdd1 ( a well-established dominant-negative XDsh mutant ) . XDsh-CΔ8 and Xdd1 induced similar phenotypes . The numbers of embryos injected from three independent experiments are listed on the top of each column . DOI: http://dx . doi . org/10 . 7554/eLife . 08142 . 013 In Xenopus embryos , JNK is known to regulate CE through non-canonical Wnt-PCP signaling ( Sokol , 1996 , 2000; Jones and Chen , 2007 ) , and either hyperactivation or JNK depletion dysregulates CE ( Yamanaka et al . , 2002 ) . We therefore investigated how the change of Dvl conformation affects JNK activity . We overexpressed wild-type XDsh , Xdd1 , and XDsh-CΔ8 in the ventral regions of 4-cell Xenopus embryos and dissected the ventral regions at the early gastrula stage to examine JNK phosphorylation by western blot . Overexpression of wild-type XDsh induced slightly greater JNK phosphorylation than observed in uninjected control cells . However , both Xdd1 and XDsh-CΔ8 induced JNK phosphorylation more potently ( Figure 9A ) , suggesting that deletion of the XDsh C-terminal region renders XDsh more active in the non-canonical Wnt pathway . This finding is consistent with an early report of more effective JNK activation in COS-7 cells expressing mutant Dvl lacking the PDZ domain than in cells expressing wild-type Dvl ( Sokol et al . , 1995; Axelrod et al . , 1998; Li et al . , 1999; Moriguchi et al . , 1999; Yamanaka et al . , 2002; Gao et al . , 2010 ) . 10 . 7554/eLife . 08142 . 014Figure 9 . The open conformation of Dvl disrupts CE by activating Jun N-terminal kinase ( JNK ) . ( A ) Western blot of phosphorylated JNK in ventral mesoderm cells overexpressing wild-type XDsh or its mutants . At equivalent protein level , Xdd1 and XDsh-CΔ8 more potently induce JNK phosphorylation than wild-type XDsh . ( B ) Xenopus 4-cell stage embryos were dorsally coinjected with equal quantities of wild-type XDsh mRNA or XDsh-CΔ8 mRNA ( 500 pg ) and the AP1-luciferase reporter DNA ( 200 pg ) ; luciferase activity was assayed at the late gastrula stage . Inset shows a representative Western blot using anti-myc antibody to control XDsh and XDsh-CΔ8 protein levels . Values are the mean and SD from three independent experiments ( XDsh vs XDsh-CΔ8 , p < 0 . 05 ) . ( C–H ) The dominant negative JNK mutant ( dnJNK ) rescues activin-induced explant elongation blocked by overexpression of XDsh-CΔ8 or Xdd1 . ( C ) Uninjected explants treated with activin show extensive elongation . ( D ) XDsh-injected explants treated with activin show moderate inhibition of explant elongation . ( E ) Injection of XDsh-CΔ8 strongly inhibits explant elongation . ( F ) Injection of Xdd1 similarly inhibits explant elongation as injection of XDsh-CΔ8 . ( G , H ) dnJNK rescues explant elongation inhibited by XDsh-CΔ8 or Xdd1 . ( I ) dnJNK also recues CE defects produced by overexpression of XDsh-CΔ8 or Xdd1 in whole embryos . Phenotypes are severe ( green ) , mild ( red ) , and normal ( blue ) . Numbers on the top indicate total embryos scored from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08142 . 014 To further demonstrate that the conformation of Dvl regulates its activation of JNK , we used the AP1-luciferase reporter ( Rui et al . , 2007 ) to monitor JNK activation in whole Xenopus embryos . When equal quantities ( 500 pg ) of XDsh-CΔ8 or wild-type XDsh mRNA were injected into the dorsal region of 4-cell Xenopus embryos and luciferase activity was assayed at the late gastrula stage , XDsh-CΔ8 clearly activated AP1-luciferase activity more strongly ( Figure 9B ) , indicating that Dvl in the open conformation activates JNK more potently than Dvl in the closed conformation . To clarify the role of the Dvl C-terminus in JNK activation , we analyzed activin-induced changes in the length and shape of Xenopus ectodermal explants; these changes represent the CE phenotype associated with non-canonical Wnt signaling ( Figure 9C–H ) . In Xenopus , both activation and inhibition of non-canonical Wnt signaling result in planar cell polarity ( PCP ) defects and produce CE phenotype ( Djiane et al . , 2000 ) . Ectodermal explants from control embryos and embryos injected with different mRNAs that encode Dvl and different mutants were dissected at the early blastula stage and treated with activin . The phenotypes were monitored at equivalent early neurula stages . As expected , XDsh-CΔ8 and the dominant-negative mutant Xdd1 strongly inhibited the activin-induced elongation of ectodermal explants ( CE outcomes ) as compared to controls , while wild-type XDsh had little effect ( Figure 9D–F ) . However , co-expression of a dominant-negative JNK mutant ( dnJNK ) , which inhibits non-canonical Wnt signaling ( Yamanaka et al . , 2002; Carron et al . , 2005 ) , with XDsh-CΔ8 or Xdd1 efficiently rescued explant elongation ( Figure 9G , H ) , indicating that the CE defects induced by XDsh-CΔ8 and Xdd1 are the result of JNK activation . In addition , we also confirmed that co-expression of dnJNK with XDsh could also rescue explant elongation ( not shown ) . Finally , to further confirm that the effect of Dvl's open conformation on the CE phenotype resulted from JNK activation , we coinjected dnJNK with XDsh-CΔ8 or Xdd1 , and analyzed the phenotypes in Xenopus whole embryo . Injection of XDsh-CΔ8 or Xdd1 alone resulted in embryos with short and bent axis , reflecting CE defects . However , the abnormal CE phenotype was substantially rescued when the embryos were coinjected with dnJNK ( Figure 9I ) , suggesting that JNK activation induced by the Dvl open conformation ( i . e . , use of XDsh-CΔ8 or Xdd1 ) had resulted in the CE phenotype . Many Wnt signaling regulators directly bind to and inhibit the Dvl PDZ domain ( Wharton , 2003; Wallingford and Habas , 2005 ) . To block Wnt signaling transduction at the Dvl level , we also developed a series of small-molecule inhibitors that disrupt Fz–Dvl interaction ( Shan et al . , 2005 , 2012; Grandy et al . , 2009; Shan and Zheng , 2009; Lee et al . , 2009b ) . Because all of the Wnt-regulating proteins and small molecule inhibitors target the site on the surface of the Dvl PDZ domain that binds to the molecule's own C-terminus , these agents should release the Dvl C-terminus from its intramolecular binding and open the closed conformation of Dvl . These molecules can be used to probe the effect of the conformational change of Dvl . For this study we chose a small molecule inhibitor of Fz–Dvl interaction , compound 3289–8625 ( Grandy et al . , 2009 ) and a protein inhibitor of Wnt signaling , TMEM88 ( Lee et al . , 2010 ) . We previously showed that the small molecule 3289–8625 penetrates the Xenopus embryo and binds to the PDZ domain of Dvl ( Grandy et al . , 2009 ) . Microinjection of embryos with XDsh mRNA and incubation in medium containing compound 3289–8625 increased the prevalence of the non-canonical Wnt signaling–associated CE phenotype ( Figure 10A ) . We also previously reported the protein TMEM88 to be a novel Wnt signaling inhibitor whose C-terminal region binds to the Dvl PDZ domain ( Lee et al . , 2010 ) . When we coinjected embryos with mRNAs encoding the C-terminal half of TMEM88 ( TMEM88-C ) and wild-type XDsh , abnormal CE phenotypes were again more prevalent than in embryos injected only with XDsh mRNA ( Figure 10A ) ; this result is consistent with the fact that gain-of-function of PCP signaling induced by activated Dvl affects gastrulation movements and disrupts axis elongation ( Djiane et al . , 2000 ) . CE defects were much less prevalent in control embryos incubated in 3289–8625 or injected with TMEM88-C mRNA alone ( Figure 10A ) . 10 . 7554/eLife . 08142 . 015Figure 10 . The open conformation of Dvl induced by targeting the Dvl PDZ domain potentiates Wnt/JNK signaling . ( A ) Regulation of XDsh-mediated PCP signaling by a PDZ-binding small molecule or peptide is shown by the gain-of-function CE phenotypes of whole embryos that were uninjected ( controls ) or injected with XDsh mRNA with or without treatment with the Dvl inhibitors 3209–8625 or coinjected with XDsh and TMEM88-C mRNAs . ( B ) Inhibiting the Dvl PDZ domain blocks canonical Wnt signalling induced by Dvl overexpression . Wild-type XDsh mRNA was injected alone or coinjected with an equal quantity of TMEM88-C mRNA in the animal pole region of two-cell stage Xenopus embryos , and ectodermal explants were dissected at the late blastula stage . TOPFLASH luciferase activity values are the mean and SD from three independent experiments ( p < 0 . 05 ) . ( C ) Inhibition of the Dvl PDZ domain by TMEM88 opens the conformation of Dvl and potentiates Wnt/JNK signaling induced by Dvl overexpression . Xenopus 4-cell stage embryos were injected dorsally with wild-type XDsh mRNA or coinjected with equal quantities of wild-type XDsh mRNA and TMEM88-C mRNA . AP1 luciferase activity was assayed at the late gastrula stage . Values are the mean and SD from three independent experiments ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08142 . 015 To further dissect how inhibition of the Dvl PDZ domain affects Dvl's role in the Wnt/β-catenin and Wnt/JNK signaling pathways , we used the TOPFLASH and AP1-luciferase reporter assays . To target the Dvl PDZ domain with TMEM88-C , we again coinjected the embryos with mRNAs encoding TMEM88-C and wild-type XDsh . As we reported previously ( Lee et al . , 2010 ) , binding of the PDZ domain by TMEM88-C antagonized Wnt/β-catenin activity induced by Dvl overexpression ( Figure 10B ) . Further , this opening of the Dvl conformation potentiated Wnt/JNK signaling induced by Dvl overexpression ( Figure 10C ) .
Here we showed that a C-terminal motif of Dvl can bind intrinsically to the Dvl PDZ domain , forming a ‘closed’ conformation . Although the binding affinity of this reaction is not notably high , the closed conformation is stable and may predominate among intracellular Dvl proteins ( Zhou et al . , 2006 ) . To compare the effects of the two Dvl conformations , we examined a C-terminal–truncated Dvl mutant , the established Xdd1 mutant ( neither of these forms the ‘closed’ conformation in solution ) , and wild-type Dvl in different Xenopus assays and found that both of the ‘open’ Dvl constructs significantly enhanced the CE phenotype mediated by Wnt-JNK signaling . To further support this observation we observed the competitive binding to the PDZ domain of the Dvl C-terminus and two agents known to bind the Dvl PDZ domain―the peptide TMEM88-C and the small-molecule inhibitor 3289–8625 ( Grandy et al . , 2009 ) ; we reasoned that by competing with the intrinsic binding of the C-terminus , these two agents should induce an open Dvl conformation . Indeed , the two molecules enhanced JNK activation by wild-type Dvl . Interestingly , consistent with these findings , several groups have reported that JNK is activated by sulindac , a nonsteroidal anti-inflammatory drug we previously demonstrated to bind to the Dvl PDZ domain ( Czibere et al . , 2005; Rice et al . , 2006; Lee et al . , 2009b; Singh et al . , 2011 ) . Therefore , we conclude that the open conformation of Dvl is likely to initiate JNK activation . Dvl has been suggested to exist in activated and inactivated states within the cell and to be activated by Wnt signals ( Lee et al . , 2003; Wharton , 2003; Wallingford and Habas , 2005; Gao and Chen , 2009 ) ; however , the form of the ‘active state’ has not been determined . Our findings suggest that the activation state of Dvl is determined by its conformation , such that in the absence of Wnt ligand , Dvl adopts a closed conformation that represents its inactive state . Wnt signaling opens the closed conformation of Dvl and thereby activates Dvl . For example , in a working model of the canonical Wnt signaling pathway , the simultaneous binding of Wnt ligand to both of its membrane-bound receptors , Fz and LRP5/6 , initiates canonical Wnt signaling by causing dimerization of the two receptors . Within the cell , the close proximity of the two receptors' cytoplasmic tails triggers the formation of signalosomes ( Bilic et al . , 2007 ) containing Dvl and Axin; Dvl binds to Fz through its PDZ domain ( Wong et al . , 2003 ) , thus acquiring the open conformation . This open conformation also allows the Dvl DEP domain to interact with the membrane through nonspecific charge–charge interactions ( Noordermeer et al . , 1994; Theisen et al . , 1994; Axelrod et al . , 1998; Wong et al . , 2003; Park et al . , 2005; Wang et al . , 2006; Schwarz-Romond et al . , 2007; Simons et al . , 2009 ) that in turn promote the Fz–Dvl interaction . Axin binds LRP5/6 ( Mao et al . , 2001; Tamai et al . , 2004 ) , and the DIX domains of Dvl and Axin interact to further stabilize the complex ( Schwarz-Romond et al . , 2007; Fiedler et al . , 2011 ) . The network of interactions in the super-complexes stabilizes the signalosomes remarkably , although the individual interactions are relatively weak . Energetically , the super-complex is more stable than the closed conformation of Dvl and therefore can capture the PDZ domain of Dvl . By ousting the Dvl C-terminus from its bond with the Dvl PDZ domain , the Fz binding motif also opens the conformation of Dvl as it is captured in the signalosome . Although other Wnt signaling pathways are less well defined , it is clear that most , if not all , non-canonical Wnt pathways are triggered by the interaction between Fz and Dvl , which is likely to open the conformation of Dvl as well ( Habas and Dawid , 2005; Gao and Chen , 2009 ) . Therefore , we propose that the active form of Dvl is its open conformation and that this active form initiates JNK-related Wnt signaling pathways . The Dvl DEP domain is essential to activate JNK cascades ( Boutros et al . , 1998 ) . The hypothesis that ‘opened’ Dvl stimulates JNK activity is consistent with a report that Daam1 , a key player that connects Dvl to JNK , exists in an autoinhibited state and is activated by binding to Dvl ( Liu et al . , 2008 ) . Daam1 binds to the Dvl DEP domain ( Habas et al . , 2001 ) , which is likely to be obstructed in the closed conformation . Therefore , opening of Dvl's conformation is a key step in the Wnt-stimulated cascade that activates JNK .
The cDNAs encoding the PDZ ( residues 251–340 ) , DEP ( residues 377–503 ) , mC1 ( residues 251–695 ) , and mD1-CΔ7 ( residues 251–688 ) domains of mouse Dvl-1 were sub-cloned into the pET28a vector . The N-terminally 6xHis-tagged proteins were expressed in BL21 ( DE3 ) Escherichia coli and purified by Ni-affinity chromatography followed by gel filtration chromatography as we described previously ( Wong et al . , 2003 ) and describe in the supplementary information . Peptides were synthesized by the Hartwell Center for Bioinformatics & Biotechnology at St . Jude Children's Research Hospital; they were purified by reverse-phase high-performance liquid chromatography and confirmed by MSI-MS as described in the supplementary information . For the fluorescence spectroscopy studies , a Fluorolog-3 spectrofluorometer ( HORUBA Instruments Inc , Edison , NJ ) with a 10 × 4 mm quartz cell ( Hellma Inc . ) with magnetic stirring was used . To confirm that the binding site of the Dvl-1 PDZ domain was occupied by intrinsic C-terminus , we generated two proteins , mC1 and mC1-CΔ7 , which were separately titrated into the fluorescence-labeled peptide Rox-DprC . The KI of both peptides was determined in two independent experiments by using the equation KDapp = KD ( 1 + [I]/KI ) , where KDapp is the apparent KD of Rox-Dpr-C with the Dvl-C or Dsh-C peptide and [I] is the concentration of both peptides ( Figure 2 ) . For the ITC studies , Auto-iTC-200 ( MicroCal ) was used to obtain the binding affinity of Dvl-C peptide and Dvl PDZ protein in 50 mM phosphate buffer . The concentration of Dvl-C peptide in the syringe was 1 . 05 mM and the concentration of Dvl PDZ domain in the cell was 0 . 114 mM ( Figure 3 ) . The KD value was averaged from two independent experiments at 25°C . All NMR experiments were performed at 15°C using Bruker Avance 800-MHz spectrometers equipped with triple-resonance , 5-mm triple axis–shielded gradient probes . For titration experiments we used the Varian Unity INOVA 600 MHz spectrometer equipped with a triple-resonance , 5-mm triple-axis shielded gradient probe at 25°C . We used NMR-derived data and a simulated annealing protocol using the program CNS within the HADDOCK software ( ver . 1 . 2 ) ( Dominguez et al . , 2003 ) . Coordinates of mouse Dvl PDZ were taken from the X-ray structure of the Xenopus PDZ domain ( 1L6O:A ) , whose amino acid residues were modified to fit the mouse Dvl PDZ domain ( Wong et al . , 2003 ) . Two different types of restraints were used to determine the complex structures: ( 1 ) Ambiguous interaction restraints were chosen on the basis of the chemical shift perturbation , intermolecular NOEs between the Dvl-C peptide/PDZ complex , and solvent accessibility ( calculated by using the program NACCESS [Hubbard and Thornton , 1993] ) ; ( 2 ) Unambiguous distant restraints were obtained from several different types of NOESY experiments , including 2D [F1 , F2]-double filtered NOESY experiments and 3D F1-half-filtered and F2-edited 13C-NOESY-HSQC experiments using the 13C/15N PDZ domain of Dvl with unlabeled Dvl-C peptide . NOE restraints were grouped into distance ranges according to their relative intensity: strong ( 1 . 8–2 . 5 and 1 . 8–3 . 0 Å ) , medium ( 1 . 8–4 . 0 Å ) , and weak ( 1 . 8–5 . 0 Å ) . A total of 45 unambiguous restraints ( 23 sequential intramolecular NOEs from the Dvl-C peptide bound to the PDZ domain and 22 intermolecular NOEs between the Dvl-C peptide and the Dvl-1 PDZ domain ) were used . Two types of restraints were combined to generate 2000 initial structures of the Dvl PDZ/Dvl-C peptide complex; 200 of these structures were selected by using NOE-derived restraints , and 100 structures were then obtained for final refinement . We ultimately selected the 15 lowest-energy conformers from the final 100 complex structures for further structural analysis . Xenopus eggs were obtained from females previously injected with 500 IU of human chorionic gonadotropin ( Sigma ) and artificially fertilized . Synthesis , microinjection of capped mRNAs , and treatment of ectodermal explants with activin were previously described ( Carron et al . , 2005 ) . After microinjection , some embryos were incubated in medium containing 2 μg/ml of the small-molecule compound 3289–8625 until the desired stage ( Grandy et al . , 2009 ) . To examine canonical Wnt signaling , both the siamois promoter–driven luciferase reporter ( Brannon et al . , 1997 ) and the TOPFLASH luciferase reporter were used . The siamois promoter reporter DNA construct ( Sialuc , 200 pg ) or the TOPFLASH reporter DNA construct ( 200 pg ) was injected , alone or with mRNA ( 500 pg ) encoding wild-type XDsh or a mutant XDsh-CΔ8 lacking the PDZ-binding motif , into the animal pole region of 2-cell Xenopus embryos . Ectodermal explants were dissected from injected embryos at the late blastula stage . To assess non-canonical Wnt signaling , we used the AP1-luciferase reporter ( Rui et al . , 2007 ) to monitor activation of JNK in Xenopus whole embryos . The reporter DNA ( 200 pg ) was injected alone or coinjected with equal quantities ( 500 pg ) of XDsh-CΔ8 or wild-type XDsh mRNA into the dorsal region of 4-cell Xenopus embryos , which were allowed to develop to late gastrula stage . A Lumat LB9507 luminometer ( Berthold Technologies GmbH & Co ) was used to perform the luciferase assays ( Promega ) . We used cell lysates from 10 explants or five whole embryos to measure luciferase activity . All experiments were performed at least in triplicate using different batches of embryos and the mean value was calculated using Student's t-test . Synthetic mRNAs ( 500 pg ) corresponding to myc-tagged wild-type XDsh and different mutants were injected into the ventral blastomeres of Xenopus embryos at the 4-cell stage . At the early gastrula stage , 10 ventral mesoderm explants were dissected and analyzed by western blot using anti-phospho-JNK ( Thr183/Tyr185 , Thr221/Tyr223 ) antibody ( Millipore ) and anti-myc 9E10 antibody ( Santa-Cruz Biotechnology ) . The two dorsal blastomeres of 4-cell Xenopus embryos were injected with mRNA ( 500 pg ) encoding XDsh , XDsh-CΔ8 , or Xdd1 , either alone or with mRNA ( 500 pg ) encoding dnJNK . For the dose-depending studies , three different amounts of XDsh and XDsh-CΔ8 mRNAs ( 80 pg , 200 pg and 500 pg ) were used . The embryos were cultured to the larval stage , then grouped and counted according to normal , mild , or severe CE abnormal ( JNK gain-of-function ) phenotype . To investigate the effect of XDsh , XDsh-CΔ8 , and Xdd1 on CE in vitro , the same mRNAs were injected , alone or with dnJNK mRNA , into the animal pole region of 2-cell Xenopus embryos . Ectodermal explants were dissected at the early blastula stage , incubated with activin for 1 hr , cultured to the early neurula-stage equivalent , and examined for the extent of explant elongation . To investigate how the C-terminal region of TMEM88 affects Dvl function , equal amounts of mRNAs ( 500 pg ) encoding the C-terminal half of TMEM88 ( TMEM88-C ) and wild-type XDsh were coinjected into the two dorsal blastomeres at the 4-cell stage , and the embryos were again grouped and counted as having normal , mild , or severe CE phenotypes . The above experiments were performed at least twice using different batches of embryos . | The development of an animal embryo depends on a number of signaling pathways that pass information from the outside of the cell to the inside . These pathways include Wnt signaling , which also regulates cell growth . The pathways must be precisely controlled; abnormal Wnt activity has been implicated in several human diseases , ranging from heart disease to cancer . Wnt signaling is complex , and actually comprises two major pathways: the canonical pathway ( which depends on a protein called β-catenin ) and the PCP pathway ( which doesn't depend on β-catenin ) . Both pathways are triggered when Wnt molecules bind to receptors on the outside of the cell . These receptors pass the signal into the cell and to a protein called ‘Dishevelled’ ( or ‘Dvl’ for short ) . This protein then passes the signal on through either the canonical or PCP pathway . Nevertheless it is not clear how the Dishevelled protein can direct the signal specifically down either one of these pathways . Lee et al . now show that the Dishevelled protein can take on at least two different shapes . When it is ‘closed’ , one end of the protein is tucked inside a pocket elsewhere on the protein's surface . But when Dishevelled is ‘open’ , this end of the protein moves out of this pocket . Further experiments using frogs ( called Xenopus , which are commonly used in research ) reveal that mutant versions of Dishevelled that were unable to take on the closed form strongly affected an aspect of the frog's development that involves the PCP pathway . Lee et al . then demonstrate that while Dishevelled cooperates with several other Wnt pathway components to activate the canonical pathway , the open form of Dishevelled activates the PCP pathway . The next challenge following on from this work is to find out how Wnt molecules binding to the receptor trigger the shape change in Dishevelled . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | Conformational change of Dishevelled plays a key regulatory role in the Wnt signaling pathways |
Yeast is a powerful model for systems genetics . We present a versatile , time- and labor-efficient method to functionally explore the Saccharomyces cerevisiae genome using saturated transposon mutagenesis coupled to high-throughput sequencing . SAturated Transposon Analysis in Yeast ( SATAY ) allows one-step mapping of all genetic loci in which transposons can insert without disrupting essential functions . SATAY is particularly suited to discover loci important for growth under various conditions . SATAY ( 1 ) reveals positive and negative genetic interactions in single and multiple mutant strains , ( 2 ) can identify drug targets , ( 3 ) detects not only essential genes , but also essential protein domains , ( 4 ) generates both null and other informative alleles . In a SATAY screen for rapamycin-resistant mutants , we identify Pib2 ( PhosphoInositide-Binding 2 ) as a master regulator of TORC1 . We describe two antagonistic TORC1-activating and -inhibiting activities located on opposite ends of Pib2 . Thus , SATAY allows to easily explore the yeast genome at unprecedented resolution and throughput .
Saccharomyces cerevisiae is an invaluable model for cell biology ( Weissman , 2010 ) . Despite the simplicity of its genome , its inner working mechanisms are similar to that of higher eukaryotes . Furthermore , its ease of handling allows large-scale screenings . Yeast genetic screens have classically been performed by random mutagenesis , followed by a selection process that identifies interesting mutants . However elegant the ‘tricks’ implemented to expose the sought-after mutants , this selection phase remains a tedious process of finding a needle-in-a-haystack ( Weissman , 2010 ) . The selection phase can limit the throughput and the saturation of classical yeast genetic screens . To circumvent these problems , a second-generation genetic screening procedure has been developed . Ordered deletion libraries for every non-essential gene have been generated ( Giaever et al . , 2002 ) . The growth of each individual deletion strain can be assessed , either by robot-mediated arraying , or by competitive growth of pooled deletion strains , followed by detection of ‘barcodes’ that identify each deletion strain . These second-generation approaches also have limitations . First , ordered libraries of complete deletions only cover non-essential genes . Second , deletion strains are prone to accumulate suppressor mutations ( Teng et al . , 2013 ) . To alleviate these problems , deletion libraries can be propagated in a diploid-heterozygous form . Additional steps are then required to make them haploid . In addition , while single genetic traits can be crossed into a pre-existing library , allowing for instance pairwise genetic-interaction analysis ( Costanzo et al . , 2010 ) , introducing multiple and/or sophisticated genetic perturbations becomes problematic , since crossing requires a selection marker for each important trait . Typically , deletion libraries are missing in most biotechnology-relevant backgrounds . Finally , manipulating ordered libraries requires non-standard equipment , such as arraying robots , limiting the pervasiveness of these approaches . Recently , an innovative approach called Transposon sequencing ( Tn-seq ) was developed in various bacterial models ( Christen et al . , 2011; Girgis et al . , 2007; van Opijnen et al . , 2009 ) , and in the fungus Schizosaccharomyces pombe ( Guo et al . , 2013 ) . By allowing en masse analysis of a pool of transposon mutants using next-generation sequencing , this strategy eliminates the drawbacks of previous genetic screens . Here , we describe an adaptation of the Tn-seq strategy for S . cerevisiae , that combines the advantages of both first and second generation screens , while alleviating their limitations . The method is based on the generation of libraries of millions of different clones by random transposon insertion ( Figure 1A ) . Transposons inserted in genes that are important for growth kill their hosts and are not subsequently detected . These genes therefore constitute transposon-free areas on the genomic map . Transposon-based libraries can be grown in any condition to reveal condition-specific genetic requirements . Unlike ordered deletion libraries , transposon-based libraries can easily be generated de novo from different strain backgrounds , are not limited to coding sequences and do not require the usage of robots . 10 . 7554/eLife . 23570 . 003Figure 1 . Principle of the method . ( A ) Outline of the experimental procedure . Left , the transposon ( green ) can insert either into non-essential DNA ( blue ) and give rise to a clone , or into essential DNA ( orange ) , in which case no clone is formed . Right , procedure to identify transposon insertion sites by deep-sequencing . ( B ) Profile of the transposon density across the whole genome , when the transposon original location is either a centromeric plasmid ( top ) or the endogenous ADE2 locus on chromosome XV ( bottom ) . The dashed lines indicate the chromosome centromeres . ( C ) Six examples of genomic regions and their corresponding transposon coverage in seven independent transposon libraries of indicated genotypes . Each vertical grey line represents one transposon insertion event . Genes annotated as essential are shown in orange , others in blue . Green arrowheads indicate the places where the absence of transposon coverage coincides with an essential gene . ( D ) Histogram of the number of transposons found in every annotated gene ( CDS ) . The vertical dashed line is the median of the distribution . ( E ) Same as D , with genes categorized as non-essential ( blue ) and essential ( orange ) according to previous annotations . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 00310 . 7554/eLife . 23570 . 004Figure 1—figure supplement 1 . Size distribution of the colonies appearing on SD +Galactose -Ade . ( A ) Histogram of colony area for a randomly chosen sample of 402 colonies . ( B ) Example of picture of colonies ( top ) and segmentation thereof ( bottom ) used to calculate histogram in ( A ) . Scale bar , 1 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 00410 . 7554/eLife . 23570 . 005Figure 1—figure supplement 2 . Genome-wide analysis of transposon insertion sites . ( A ) Frequency plot of nucleotide composition around transposon insertion site . The strand is determined according to the orientation of the transposon insertion . Plot was calculated with a random sample of 50 , 000 transposon in the wild-type library 1 . Note that GC-content of the yeast genome is 38% . ( B ) Preferential transposon insertion into internucleosomal DNA . Three genomic regions are shown . For each , the top panel shows the nucleosomal density as determined in ( Lee et al . , 2007 ) , and the bottom panels show transposon insertion as in Figures 2–5 . ( C ) Correlation analysis of nucleosome and transposon density . Transposon densities were calculated on the wild-type library 1 by averaging transposon number within a 40 bp moving average . Top , correlation coefficient calculated between the nucleosome and the transposon data offset by the indicated number of bp . The correlation is most negative when offset = 0 . The periodicity is ~160 bp . Middle , autocorrelation of transposon data . The periodicity is identical as above . Bottom , autocorrelation of the nucleosome density data . The periodicity is identical as above . ( D ) Analysis of transposon number relative to distance from centromere . For each chromosome , the number of transposons mapping within a certain distance from their respective centromere was computed and plotted ( grey lines ) . The number of transposons increases linearly with the distance , except in the vicinity of the centromere , where transposon enrichment can be observed . The intercept of the linear regression , computed on the linear part of the plot and multiplied by 16 chromosomes , gives a rough estimate of the numbers of transposons enriched at pericentromeric regions ( ~20% of the total transposon number ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 00510 . 7554/eLife . 23570 . 006Figure 1—figure supplement 3 . Transposon density in essential and non-essential genes . As in Figure 1D–E , except that , for each gene , the transposon density ( i . e . number of transposons divided by length of the gene ) is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 006 This method can successfully uncover sets of genes essential in given conditions , genome-wide genetic interactions and drug-targets . Transposon insertions can generate loss- and gain-of-function variants . Finally , our approach not only shows which protein is important for growth , but also which part of the protein is essential for function , allowing genome-wide mapping of structural protein domains and screening of phenotypes at a sub-gene resolution .
The detailed procedure can be found in the Materials and methods section . The method utilizes the Maize Ac/Ds transposase/transposon system in yeast ( Lazarow et al . , 2012; Weil and Kunze , 2000 ) . Briefly , cells in which the ADE2 gene is interrupted by the MiniDs transposon are induced to express the transposase Ac , on galactose-containing adenine-lacking synthetic defined medium ( SD +galactose -adenine ) . Transposase-induced MiniDs excision is followed by repair of the ADE2 gene . Cells with repaired ADE2 will be able to form colonies . The excised transposon then re-inserts at random genomic locations with a frequency of ~60% ( Lazarow et al . , 2012 ) . We have generated seven libraries , displayed together in all figures to illustrate the reproducibility of the approach . All libraries were generated in ade2Δ strains derived from BY4741 and BY4742 backgrounds . Additional mutations ( dpl1Δ , dpl1Δ psd2Δ , VPS13 ( D716H ) , mmm1Δ VPS13 ( D716H ) , YEN1on , Table 1 ) will be described in the following sections . The complete dataset is available ( Supplementary file 1 ) and searchable here: http://genome-euro . ucsc . edu/cgi-bin/hgTracks ? hgS_doOtherUser=submit&hgS_otherUserName=benjou&hgS_otherUserSessionName=23bDePuYrk10 . 7554/eLife . 23570 . 007Table 1 . Characteristics of the librariesDOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 007LibraryNumber of coloniesReads mappedTransposons mappedMedian read per transposonNumber of MiSeq runsOverlap between MiSeq runsWild-type 1~1 . 6×10631794831284162222*54% , 88%*Wild-type 2~2 . 4×10615303285258568121NAVPS13 ( D716H ) ~4 . 7×10624958456414114132†41% , 42%†Mmm1Δ VPS13 ( D716H ) ~1 . 9×10617799948303323121NAdpl1Δ~2 . 3×1061507715640112681NAdpl1Δ psd2Δ~2 . 9×1061164956136317991NAYEN1on~2 . 8×106951787749512561NAWild-type 2 + rapamycin~2 . 4×106966495616932291NA* The harvested library was grown in two flasks , one at 30°C and the other at 37°C . DNA was extracted separately from the two cultures and sequenced in two separate MiSeq runs† The library was harvested as ten subpools , which were grown in ten separate flasks . DNA was extracted separately . In one case , DNA from all ten subpools was pooled and processed to sequencing in one MiSeq run . In the other case , DNAs were kept separate and processed until the PCR step ( 1 × 100 µl PCR by subpool ) . PCR products were pooled and sequenced as another MiSeq run . Typically 7 , 000–10 , 000 colonies with a narrow size distribution ( Figure 1—figure supplement 1 ) can be generated on a 8 . 4 cm-Ø petri dish . In the case of wild-type library 1 , 240 plates yielded ~1 . 6E6 clones ( Table 1 ) . To detect transposon insertion sites , transposed cells were scraped off the 240 plates and pooled ( Figure 1A ) . This pool was used to reinocculate SD medium lacking adenine ( SD +Dextrose -Adenine ) , and the culture was grown to saturation . This step was used to dilute non-transposed ade- cells still present on the petri dishes . The culture was then harvested by centrifugation . Genomic DNA was extracted and digested with frequent-cutting restriction enzymes , followed by ligase-mediated intramolecular circularization . Circular DNA was PCR-amplified using transposon-specific outwards-facing primers . PCR products were then sequenced on a MiSeq machine ( Figure 1A ) . We aligned the sequencing reads of the wild-type library ( Table 1 ) to the reference yeast genome and counted the number of mapped transposons . To account for the fact that Illumina sequencing is imperfectly accurate , we considered that two reads of the same orientation mapping within two bp of each other originated from the same transposon ( see Source code 1 ) . In this analysis , 284 , 162 independent transposons could be mapped onto the genome , representing an average density of one transposon every 42 bp , and a median number of 22 reads per transposon . No large area of the genome was devoid of transposon ( Figure 1B ) . Consistent with analyses in Maize ( Vollbrecht et al . , 2010 ) , no strong sequence preference was detected in the insertion sites ( Figure 1—figure supplement 2A ) . In many instances , though , insertion frequency was modulated along the genome with a periodicity of ~160 bp . Superimposing nucleosome occupancy data ( Lee et al . , 2007 ) showed that this was due to favored transposon insertion in inter-nucleosomal DNA ( Figure 1—figure supplement 2B , Gangadharan et al . , 2010 ) . This effect can be appreciated at the genome-scale . Indeed , an autocorrelation analysis unraveled a ~160 bp periodic signal in the genome-wide transposon density ( measured using a 40 bp moving average , Figure 1—figure supplement 2C ) . This periodicity was comparable to that of the nucleosomal density data ( Lee et al . , 2007 ) . Finally , while no large region was devoid of transposon , some regions were actually preferentially targeted by transposons . These were the pericentromeric regions ( Figure 1B , top ) , which were specifically enriched by ~20% of the transposon insertions ( Figure 1—figure supplement 2D ) . The explanation for this phenomenon may pertain to nuclear architecture and to the propensity of our transposon to insert close to its excision site ( Lazarow et al . , 2012 ) . Because the transposon is initially excised from a centromeric plasmid , and because centromeres cluster in the nuclear periphery ( Jin et al . , 2000 ) , the transposon might tend to reinsert in the pericentromeric regions of other chromosomes . We confirmed this assumption by sequencing a small-scale library ( ~30 000 insertions mapped ) in which the MiniDS transposon was originally at the endogenous ADE2 locus , rather than on a centromeric plasmid . In this library , preferential targeting was not observed at pericentromeric regions , but rather in the vicinity of ADE2 , confirming our assumption ( Figure 1B , bottom ) . The transposon map clearly showed that a fraction of the coding DNA sequences ( CDS ) were devoid of insertions . These coincided almost exactly with genes annotated as essential ( Figure 1C , green arrowheads ) . The median number of transposons inserted in the CDSs of all genes was 18 per gene ( Figure 1D ) . This number raised to 21 for annotated non-essential genes , but dropped to three for annotated essential genes ( Figure 1E ) . This decrease was not due to a difference in the length between essential and non-essential genes , since normalizing the number of transposon insertions to the CDS length ( transposon density ) did not abrogate this effect ( Figure 1—figure supplement 3 ) . Thus our method distinguishes , in a single step , genes that are required for growth from those that are not . Several genes , although annotated as non-essential , harbored no or very few transposons ( Supplementary file 2 ) . This can be attributed to the following reasons . ( 1 ) Because sequencing reads mapping to repeated sequences were discarded during alignment , repeated genes appear as transposon-free . ( 2 ) Several annotated dubious ORFs overlap with essential genes , and thus appear as transposon-free . ( 3 ) Several genes are necessary for growth in particular conditions , and are therefore not annotated as essential , yet are required in our growth conditions ( SD +galactose -adenine ) . These include genes involved , for instance , in galactose metabolism and adenine biosynthesis . It came as a surprise that some genes annotated as essential were tolerant to transposon insertion . While some of these clearly corresponded to annotation inconsistencies , many reflected an unanticipated outcome of our approach . We observed many instances of ‘essential’ CDSs containing transposon-permissive regions . A striking example is GAL10 ( Figure 2A ) . GAL10 encodes a bifunctional enzyme with an N-terminal epimerase , and a C-terminal mutarotase domain ( Majumdar et al . , 2004 ) . While the epimerase activity is indispensable in our conditions , the mutarotase activity is dispensable , as cells were fed a mixture of α- and β-D-Galactose , thus not requiring the conversion of one into the other . In accordance , the 3’ end of GAL10 was permissive for transposon insertion . The junction between the permissive and non-permissive domains of GAL10 corresponds exactly to the junction of the two domains in the Gal10 protein . Several examples of essential genes with dispensable 3’-ends are shown in Figure 2A . We confirmed the dispensability of the 3' end for two genes , TAF3 and PRP45 ( Figure 3 ) . 10 . 7554/eLife . 23570 . 008Figure 2 . Examples of genes showing partial loss of transposon coverage . The grey level is proportional to the number of sequencing reads . Known functional domains are indicated . ( A ) Essential genes for which C-terminal truncations yield a viable phenotype . ( B ) Essential genes for which N-terminal truncations yield a viable phenotype . ( C ) Essential genes for which various truncations yield a viable phenotype . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 00810 . 7554/eLife . 23570 . 009Figure 2—figure supplement 1 . Detection of essential protein domains . Top , algorithm to detect essential protein domains . This algorithm is implemented in Source code 2 . For each gene , a score is computed as follows: the longest interval between transposon n and transposon n + 5 , multiplied by the total number of transposons mapping to that gene , divided by the gene length to the power of 1 . 5 . A score of 0 is assigned to genes targeted by less than 20 transposons , in which the longest interval is smaller than 300 bp , and/or in which the longest interval represents more than 90% or less than 10% of the CDS length . Bottom , yeast genes sorted according to their domain likelihood score . Vertical black bars above the graph indicate previously annotated essential genes . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 00910 . 7554/eLife . 23570 . 010Figure 2—figure supplement 2 . Transposon maps in the 100 highest scoring genes . Grey scale indicates the number of sequencing reads as in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 01010 . 7554/eLife . 23570 . 011Figure 2—figure supplement 3 . Transposon maps in the genes scoring 101 to 200 . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 01110 . 7554/eLife . 23570 . 012Figure 2—figure supplement 4 . Transposon maps in the genes scoring 201 to 300 . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 01210 . 7554/eLife . 23570 . 013Figure 2—figure supplement 5 . Transposon maps in the genes scoring 301 to 400 . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 01310 . 7554/eLife . 23570 . 014Figure 3 . TAF3 and PRP45 can be truncated without visible effects on cell growth . ( A ) A truncation of TAF3 was generated in a heterozygous diploid strain ( left ) by introduction of an HA tag and a G418-resistance cassette ( HA kanr ) . The strain was tetrad dissected ( middle ) . Tetrads 2 and 3 were further analyzed by PCR to confirm the Mendelian segregation of the truncated allele ( right ) . ( B ) A complete TAF3 deletion was generated in a heterozygous diploid strain ( left ) by introduction of a G418-resistance cassette ( kanr ) . Meiosis yields only two viable , G418-sensitive spores per tetrad , confirming that TAF3 complete deletion is lethal . ( C–D ) As in ( A–B ) but applied to PRP45 . Asterisks in the right panel designate PCR reactions that were inefficient at amplifying the large truncated allele . The genotype of these spores can nevertheless be inferred from the Mendelian segregation of the G418 resistance . ( E ) Top , cryo-EM structure of the S . cerevisiae spliceosome ( PDB accession 5GMK , Wan et al . , 2016 ) . Bottom , the same structure stripped of every protein except Prp45 . The essential portion of Prp45 as defined in ( C ) is in green and the non-essential part is in red and yellow . U2 , U6 , U5 and substrate RNAs are depicted in pale blue , pink , dark blue and orange , respectively . The red circle indicates the catalytic active site of the spliceosome . ( F ) Alignment of the Human , S . cerevisiae , and S . pombe Prp45 orthologs . The green , red and yellow boxes are colored as in ( E ) . The yellow box features the most conserved region of the protein . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 014 TAF3 encodes a 47 kDa central component of TFIID ( Sanders et al . , 2002 ) . Our data show that only the first 76 amino acids are necessary for growth . Using homologous recombination , we replaced all but the sequence coding for the first 90 amino acids with an HA tag and a G418-resistance cassette ( KanMX6 ) in a diploid strain . Sporulation and progeny segregation confirmed that strains expressing only the first 90 amino acids of Taf3 were viable ( Figure 3A ) . By contrast , complete replacement of TAF3 gave rise to only two viable , G418-sensitive spores per meiosis , confirming that TAF3 is essential ( Figure 3B ) . The essential region of Taf3 corresponds to a predicted histone-like bromo-associated domain ( Doerks et al . , 2002 ) . PRP45 encodes a central component of the spliceosome . From our data , only the first 140 amino-acids of Prp45 are necessary for growth , which we confirmed using the same strategy as for TAF3 ( Figure 3C–D ) . Prp45 is predicted to be intrinsically disordered , thus no clear domain boundaries are available to rationalize our data . However , a recent cryo-EM structure of the S . cerevisiae spliceosome offers clues on the structure of Prp45 ( Wan et al . , 2016 ) . Prp45 is centrally located within the spliceosome , has an extended conformation and makes several contacts with various proteins and snRNAs . In particular , the most conserved region of Prp45 is a loop that makes extensive contacts with U2 and U6 snRNAs close to the active site ( Figure 3E–F , yellow ) . This loop belongs to the dispensable part of Prp45 , surprisingly suggesting that splicing can occur without it . Our method thus offers insights that neither sequence conservation nor structural analysis could have predicted . We also observed CDSs in which the 5’ end is permissive for transposon insertion while the 3’ end is not ( Figure 2B–C ) . Again , our data show a general good coincidence between indispensable regions and annotated domains . It is surprising that transposons can be tolerated in the 5’ of such genes , since several stop codons in all frames of the transposon should interrupt translation and prevent the production of the essential C-terminus . We speculate that the production of the C-terminus is enabled by spurious transcription events . A remarkable example is SEC9 ( Figure 2C ) . SEC9 encodes a SNARE protein . The essential SNARE domain , located at the C-terminus ( Brennwald et al . , 1994 ) , is devoid of transposons . The N-terminus of the protein is known to be dispensable for growth ( Brennwald et al . , 1994 ) . We indeed observe several transposons inserted upstream of the SNARE domain . The extreme 5’ of the gene constitutes another transposon-free region even though it encodes a dispensable part of the protein ( Brennwald et al . , 1994 ) . It is possible that transposon insertion in this 5’ region generates an unexpressed , unstable or toxic protein . Other examples of genes showing various combinations of essential domains are shown in Figure 2C . We devised an algorithm to score genes according to their likelihood of bearing transposon-tolerant and -intolerant regions ( Source code 2 ) . In short , we computed for each CDS the longest interval between five adjacent transposons , multiplied it by the total number of transposons mapping in this CDS , and divided the result by the CDS length . We additionally imposed the following criteria: the interval must be at least 300 bp , must represent more than 10% and less than 90% of the CDS length , and a minimum of 20 transposons must map anywhere within the CDS . ~1200 genes fulfilled these requirements ( Figure 2—figure supplement 1 ) , of which the 400 best-scoring ones showed clear domain boundaries ( Figure 2—figure supplements 2–5 ) . Essential subdomains are only expected in essential genes and indeed , this gene set was overwhelmingly enriched for previously-annotated essential genes ( Figure 2—figure supplement 1 ) . Thus , our method allows to identify not only genes , but also subdomains that are important for growth , yielding valuable structure-function information about their cognate proteins . Our approach can easily identify essential genes and essential protein domains . In addition , its ease makes it a potential tool to uncover genes that are not essential in standard conditions but become important in specific conditions . Indeed , our approach yields two measures — the number of transposons mapping to a given gene , and the corresponding numbers of sequencing reads . Since in most cases , transposon insertion obliterates the gene function , both measures may be used as a proxy for fitness . We assessed the usefulness of these metrics in various genetic screens . Synthetic genetic interaction screening is an extremely powerful approach to establish networks of functional connections between genes and biological pathways , and to discover new protein complexes ( Costanzo et al . , 2010; Schuldiner et al . , 2005 ) . We have recently identified single amino-acid substitutions in the endosomal protein Vps13 that suppress the growth defect of mutants of the ER-mitochondria encounter structure ( ERMES ) ( Lang et al . , 2015b ) . Suppression is dependent on the proper function of the mitochondrial protein Mcp1 ( Tan et al . , 2013 ) and on the endosomal protein Vam6/Vps39 ( Elbaz-Alon et al . , 2014; Hönscher et al . , 2014 ) . We generated a transposon library from a strain bearing both the VPS13 suppressor allele VPS13 ( D716H ) and a deletion of the ERMES component MMM1 . In these conditions , we expected VPS13 , MCP1 and VAM6/VPS39 to become indispensable , while ERMES components ( MDM10 , MDM12 and MDM34 ) should become dispensable . Figure 4A shows , for each of the 6603 yeast CDSs , the number of transposon insertion sites mapped in the wild-type ( x-axis ) and in the mmm1Δ VPS13 ( D716H ) library ( y-axis ) . Most CDSs fall on a diagonal , meaning that they were equally transposon-tolerant in both libraries . Consistent with the ERMES suppressor phenotype of the VPS13 ( D716H ) mutation , ERMES components fell above the diagonal ( that is , they bore more transposons in the mmm1Δ VPS13 ( D716H ) than in the wild-type library , Figure 4A–B ) . By contrast , VPS13 , MCP1 and VAM6/VPS39 fell under the diagonal , as expected ( Lang et al . , 2015a , Figure 4A , C ) . Many other genes known to display synthetic sick or lethal interaction with mmm1Δ ( Hoppins et al . , 2011 ) were also found , including TOM70 , VPS41 , YPT7 , VMS1 and YME1 ( Figure 4A , C ) . 10 . 7554/eLife . 23570 . 015Figure 4 . Genetic interaction analyses . Libraries in panels B , C , E and G are displayed in the same order as in Figure 1C . ( A ) Comparison of the number of transposons inserted in each of the 6603 yeast CDSs in the wild-type ( x-axis ) and mmm1Δ VPS13 ( D716H ) ( y-axis ) libraries . ( B ) Transposon coverage of genes encoding ERMES components is increased in libraries from strains bearing the VPS13 ( D716H ) allele . ( C ) Examples of genes showing synthetic sick/lethal interaction with mmm1Δ VPS13 ( D716H ) . ( D ) Comparison of the number of transposons inserted in each of the 6603 yeast CDSs in the wild-type ( x-axis ) and dpl1Δ ( y-axis ) libraries . ( E ) Transposon coverage of the HIP1 locus in the dpl1Δ his3Δ library and in all the other libraries ( HIS3 ) . ( F ) Comparison of the number of transposons inserted in each of the 6603 yeast CDSs in the dpl1Δ ( x-axis ) and dpl1Δ psd2Δ ( y-axis ) libraries . ( G ) Transposon coverage of the PSD1 locus in the dpl1Δ psd2Δ and in all other libraries . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 01510 . 7554/eLife . 23570 . 016Figure 4—figure supplement 1 . Volcano plots comparing libraries or combinations of libraries as indicated . The calculated fold-change in transposon density between the two sets of libraries is plotted in log2 scale on the x-axis . The -log10 ( p-value ) ( computed using the Student’s t-test ) is plotted on the y-axis . ( 1 ) The VPS13 ( D716H ) and the mmm1Δ VPS13 ( D716H ) strains were generated in a MET17 background while all other libraries where generated in a met17Δ background . As a result MET17 and the overlapping ORF YLR302C appear as transposon free in the reference set . MET6 is more targeted by transposons in met17Δ libraries , likely because Met17 produces homocysteine , which needs to be converted to methionine by Met6 , or might otherwise accumulate to toxic levels . ( 2 ) The VPS13 ( D716H ) and mmm1Δ VPS13 ( D716H ) strains were generated in a Matα background , while the others were generated in a Mata background . ( 3 ) YGR190C overlaps with HIP1 . ( 4 ) GPP1 shows synthetic lethality with a recessive Mendelian variant present in the psd2Δ dpl1Δ strain . However , this variant is neither linked to DPL1 nor to PSD2 ( data not shown ) . ( 5 ) ERMES component genes scores very high with respect to fold change , because two of the five libraries in the reference set bear the VPS13 ( D716H ) allele . The p-value , by contrast , is not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 016 As a second proof-of-principle , we generated a library from a strain in which the dihydrosphingosine phosphate lyase gene DPL1 ( Saba et al . , 1997 ) was deleted , and another library from a strain in which both DPL1 and the phosphatidylserine decarboxylase 2 gene PSD2 ( Trotter and Voelker , 1995 ) were deleted ( Figure 4D , F ) . In this latter double-deleted strain , phosphatidylethanolamine can only be generated via the mitochondrial phosphatidylserine decarboxylase Psd1 , and thus any gene required for lipid shuttling to and from mitochondria should become indispensable ( Birner et al . , 2001 ) . LCB3 , which displays synthetic sick interaction with DPL1 ( Zhang et al . , 2001 ) , was less transposon-tolerant in both the dpl1Δ and the dpl1Δ psd2Δ libraries ( Figure 4D ) . By contrast PSD1 , which displays a synthetic lethality with PSD2 on media lacking choline and ethanolamine ( Birner et al . , 2001 ) , was transposon-intolerant in the dpl1Δ psd2Δ library only ( Figure 4F–G ) . Interestingly , we also found that VPS13 was less transposon-tolerant in the dpl1Δ psd2Δ , consistent with a role for Vps13 in interorganelle lipid exchange ( AhYoung et al . , 2015; Kornmann et al . , 2009; Lang et al . , 2015a , 2015b ) . Additionally , when comparing the dpl1Δ and wild-type libraries , one of the best hits was the histidine permease HIP1 ( Figure 4D–E ) . This did not , however , reflect a genetic interaction between HIP1 and DPL1 , but instead between HIP1 and HIS3; during the construction of the strains , ADE2 was replaced by a HIS3 cassette in the wild-type strain , while it was replaced by a NAT cassette in the dpl1Δ strain . The histidine-auxotroph dpl1Δ strain , therefore required the Hip1 permease to import histidine . Thus , synthetic genetic interactions are visible through pairwise comparison of the number of transposons per genes . However , this metrics leads to a significant spread of the diagonal ( Figure 4A , D , F ) . This spread is due to the intrinsic noise of the experiment . Indeed , the number of transposons per gene is expected , for each gene , to follow a binomial distribution . Sampling variability may thus mask biologically relevant differences . To overcome this limitation , we reasoned that comparing sets of one or more libraries against each other , rather than comparing two libraries in a pairwise fashion ( as in Figure 4 ) , would greatly improve the signal-to-noise ratio . We thus calculated an average fold-change of the number of transposons per gene between the experimental and reference sets , as well as a p-value ( based on a Student’s t-test ) associated with this change . The fold-change and p-values were then plotted as a volcano plot ( Figure 4—figure supplement 1 , Supplementary file 3 ) . In volcano plots , synthetic genes appeared well separated from the bulk of neutral genes , showing that parallel library comparison is a robust way to increase the signal-to-noise ratio . Synthetic lethality is one type of genetic interaction . Another type is rescue of lethal phenotype , where a gene deletion is lethal in an otherwise wild-type strain but can be rescued by a suppressor mutation . We describe two such phenomena observed in our libraries . The first concerns the septin gene CDC10 . Septin proteins are cytoskeletal GTPases that form a ring at the bud neck . This structure is necessary for vegetative growth in S . cerevisiae , and all septin genes are essential with the exception of CDC10 ( Bertin et al . , 2008; McMurray et al . , 2011 ) . Indeed , at low temperature , cdc10Δ cells are viable and able to assemble a septin ring . This Cdc10-less ring is based on a Cdc3-Cdc3 interaction , instead of the normal Cdc3-Cdc10 interaction ( McMurray et al . , 2011 ) . Because the propensity of Cdc3 to self-assemble is weak , low temperature is thought to be necessary to stabilize the interaction . Since we grew all libraries at moderately high temperature ( 30°C ) , CDC10 was , as expected , essentially transposon-free in most libraries ( Figure 5A ) . In the dpl1Δ psd2Δ library , however , the number of transposons mapping within CDC10 increased significantly , indicating that the absence of Psd2 and Dpl1 suppressed the cdc10Δ phenotype ( Figure 5A , Figure 4—figure supplement 1 , bottom left ) . Genetic analysis revealed that the dpl1Δ allele alone allowed cdc10Δ cells to grow at higher temperature , indicating that the Cdc10-less septin ring was stabilized in the absence of Dpl1 ( Figure 5B–C ) . Genetic analysis also demonstrated that the rescue of cdc10Δ by dpl1Δ was independent of PSD2 . It is unclear why the suppressive effect was detected in the dpl1Δ psd2Δ library , but not in the dpl1Δ library . We speculate that differences in growth conditions between experiments have obscured either the suppression in the dpl1Δ library or the involvement of Psd2 in our tetrad analyses . 10 . 7554/eLife . 23570 . 017Figure 5 . Synthetic rescue of lethal phenotypes . ( A ) Transposon coverage of CDC10 in the seven libraries . The coverage is increased in the dpl1Δ psd2Δ library . ( B ) Tetrad dissection of a PSD2/psd2Δ DPL1/dpl1Δ CDC10/cdc10Δ triple heterozygote at 30°C ( left ) and 25°C ( right ) . The cdc10Δ spores of ascertained genotype are circled with a color-coded solid line . cdc10Δ spores for which the genotype can be inferred from the other spores of the tetrad are circled with a color-coded dashed line . ( C ) Quantification of growing and non-growing cdc10Δ spores of the indicated genotype obtained from 48 tetrads ( three independent diploids ) . ( D ) Transposon coverage of DNA2 in the seven libraries . The coverage is increased in the YEN1on library . ( E ) Tetrad dissection of a DNA2/dna2Δ YEN1/YEN1 single heterozygote and of a DNA2/dna2Δ YEN1/YEN1on double heterozygote at 30°C ( left ) and 25°C ( right ) . All viable dna2Δ spores additionally carry the YEN1on allele ( red circle ) . ( F ) FACS profile of propidium-iodide-stained DNA content in DNA2 and dna2Δ YEN1on strains exponentially growing at 30°C ( left ) and 25°C ( right ) . For DNA2 panels , each profile is an overlay of two independent strains . For dna2Δ YEN1on panels , each profile is an overlay of four independent strains . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 017 Dpl1 is an ER protein that does not contact any of the septin subunits; its destabilizing effect on the septin ring must therefore be indirect . Since Dpl1 is a regulator of sphingolipid precursors ( Saba et al . , 1997 ) and since the septin ring assembles in contact with the plasma membrane ( Bertin et al . , 2010 ) , it is most likely the changing properties of the membrane in DPL1-deficient cells that allow or restrict the assembly of Cdc10-less septin rings . This hypothesis is particularly appealing because temperature has a profound effect on membrane fluidity and composition ( Ernst et al . , 2016 ) . Thus , the stabilizing effect of low temperature on Cdc10-less septin rings might not only be the result of a direct stabilization of Cdc3-Cdc3 interaction , but also of changes in plasma membrane properties , which can be mimicked by DPL1 ablation . The second example of rescue of a lethal phenotype was observed in a library made from a strain expressing a constitutively active version of the Holliday-junction resolvase Yen1 , a member of the Xeroderma Pigmentosum G ( XPG ) family of 5’-flap endonucleases ( Ip et al . , 2008 ) . In wild-type strains , Yen1 is kept inactive during the S-phase of the cell cycle via Cdk-mediated phosphorylation ( Matos et al . , 2011 ) . Rapid dephosphorylation at anaphase activates Yen1 for the last stages of mitosis . When nine Cdk1 target sites are mutated to alanine , Yen1 becomes refractory to inhibitory phosphorylation and active during S-phase ( Yen1on ) ( Blanco et al . , 2014 ) . To investigate the cellular consequence of a constitutively active Yen1 , we generated a library in a YEN1on background . We discovered that , under these conditions , the essential gene DNA2 became tolerant to transposon insertion ( Figure 5D ) . Further genetic analyses confirmed that the presence of the YEN1on allele led to a rescue of the lethal phenotype of dna2Δ strains; spores bearing the dna2 deletion failed to grow unless they additionally bore the YEN1on allele ( Figure 5E ) . Moreover , at 25°C , the colony size of dna2Δ YEN1on spores was comparable to that of the DNA2 counterparts ( Figure 5E , right ) . However , FACS analysis of DNA content revealed that dna2Δ YEN1on cells accumulated in S- and G2-phases ( Figure 5F ) . DNA2 encodes a large protein with both helicase and endonuclease activities ( Budd and Campbell , 1995 ) . Interestingly , while the helicase activity can be disrupted without affecting yeast viability , the nuclease activity is essential ( Ölmezer et al . , 2016 ) , presumably due to its involvement in processing long 5’-flaps of Okazaki fragments . Our genetic data now reveal that Yen1 is able to partially fulfill the essential roles of Dna2 , provided that its activity is unrestrained in S-phase . Since the spectrum of Yen1 substrates includes 5’-flap structures ( Ip et al . , 2008 ) , Yen1on may be able to substitute the essential function of Dna2 by providing 5’-flap nuclease activity in S-phase . This finding extends previous work showing that Yen1 serves as a backup for the resolution of replication intermediates that arise in helicase-defective mutants of Dna2 ( Ölmezer et al . , 2016 ) . Thus , our method can be used to screen for negative and positive genetic interactions in a rapid , labor-efficient and genome-wide manner , in strains bearing single and multiple mutations . To assess our method’s ability to uncover drug targets , we used the well-characterized immune-suppressive macrolide rapamycin . Rapamycin blocks cell proliferation by inhibiting the target of rapamycin complex I ( TORC1 ) , through binding to the Fk506-sensitive Proline Rotamase Fpr1 ( Heitman et al . , 1991 ) . The TORC1 complex integrates nutrient-sensing pathways and regulates cellular growth accordingly . Rapamycin treatment therefore elicits a starvation response , which stops proliferation . We generated and harvested a wild-type library , then grew it in medium containing rapamycin at low concentration . To compare it to an untreated wild-type library , we counted the number of sequencing reads mapping to each of the 6603 yeast CDSs in both conditions ( Figure 6A ) . Most genes fell on a diagonal , as they do not influence growth on rapamycin . By contrast , a few genes were robustly covered by sequencing reads in the rapamycin-treated library , indicating that their interruption conferred rapamycin resistance . Expectedly , the best hit was FPR1 , encoding the receptor for rapamycin ( Hall , 1996 ) . Other hits included RRD1 ( Rapamycin-Resistant Deletion 1 ) , TIP41 , GLN3 , SAP190 , PSP2 , CCS1 , ESL2 and members of the PP2A phosphatase PPH21 and PPM1 ( Figure 6A ) . These genes are either directly involved in rapamycin signaling or known to confer rapamycin resistance upon deletion ( Xie et al . , 2005 ) . 10 . 7554/eLife . 23570 . 018Figure 6 . Detection of rapamycin resistant strains . ( A ) Comparison of the number of sequencing reads mapping to each of the 6603 yeast CDSs in rapamycin-untreated ( x-axis ) and -treated ( y-axis ) libraries . Note the difference in scale between both axis due to the high representation of rapamycin-resistant clones . ( B ) Distribution of transposons and number of associated sequencing reads on the PIB2 gene . Transposons with high number of sequencing reads in the rapamycin-treated library are clustered at the 5’-end of the CDS . ( C ) Wild-type ( WT ) and pib2Δ strains were transformed with either an empty plasmid ( ∅ ) or plasmids encoding full-length ( FL ) or indicated fragments of Pib2 ( see E , numbers refer to the amino acid position in the full-length Pib2 protein ) . 5-fold serial dilutions of these strains were spotted on YPD or YPD containing 10 ng/ml rapamycin . Centromeric plasmids were used in all strains , except in those denoted with 2 µ , which carried multi-copy plasmids . ( D ) Strains of the indicated genotypes , transformed with either an empty plasmid ( ∅ ) or plasmids encoding full length ( pPIB2 ) or truncated ( pPIB2165-635 ) versions of Pib2 , were grown exponentially in minimal medium with proline as nitrogen source . 3 mM glutamine was added to the culture and the cells were harvested 2 min later . Protein extracts were resolved on an SDS-page and probed with antibodies either specific for Sch9-pThr737 ( P-Sch9 ) , or for total Sch9 to assess TORC1 activity . ( E ) Schematic overview of Pib2 architecture and of the fragments used for genetic studies . ( F ) Summary of yeast-two-hybrid interactions between Pib2 fragments and the TORC1 subunit Kog1 ( Figure 6—figure supplement 1 ) . Fragments indicated by a black box interacted with Kog1 , fragments indicated by a white box did not . ( G ) pib2Δ cells expressing the indicated Pib2 fragments from plasmids ( see E ) were assayed for their sensitivity to rapamycin ( 2 . 5 or 5 ng/ml ) as in C . ( H ) WT or pib2Δ cells expressing the indicated Pib2 fragments from plasmids were assayed as in G , except that cells were spotted on synthetic medium to apply a selective pressure for plasmid maintenance . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 01810 . 7554/eLife . 23570 . 019Figure 6—figure supplement 1 . A ) Transposon coverage of the PIB2 gene . Top row is the rapamycin-treated library and rows below are presented as in Figures 2–5 . The gray scale has been adjusted to account for the large number of sequencing reads mapping in the 5’ region of the gene . ( B ) Yeast-two-hybrid assay assessing the interaction of the indicated Pib2 fragments encoded on the pCAB plasmid , with full-length Kog1 encoded on the pPR3N plasmid . ( C ) gtr1Δ gtr2Δ cells expressing indicated , plasmid-encoded Pib2 fragments ( see Figure 6E ) were assayed for their sensitivity to rapamycin . Note that expressing Pib2Δ533-620 in gtr1Δ gtr2Δ cells appears to inhibit growth even in the absence of rapamycin . ( D ) gtr1Δ gtr2Δ cells , transformed with centromeric plasmids expressing Pib2 fragments ( see Figure 6E ) and alleles of Gtr1 and Gtr2 as indicated , were assayed for their sensitivity to rapamycin . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 019 Finding the TORC1 regulator PIB2 ( Kim and Cunningham , 2015 ) was however unexpected , because PIB2 deletion confers sensitivity , not resistance , to rapamycin ( Kim and Cunningham , 2015; Parsons et al . , 2004 ) . To solve this conundrum , we looked at the distribution of transposons on the PIB2 coding sequence . All the insertions selected by rapamycin treatment mapped to the 5’-end of the gene ( Figure 6B ) . On the contrary , the rest of PIB2 was less covered in the rapamycin-treated than in non-rapamycin-treated libraries ( Figure 6—figure supplement 1A ) . Insertions in the 5’ end of PIB2 therefore conferred rapamycin resistance , while insertions elsewhere , like complete deletion of PIB2 , conferred rapamycin sensitivity . To confirm this result , we engineered N-terminal truncations of Pib2 , guided by the transposon map . Strains expressing Pib2 variants that were truncated from the first until up to amino acid 304 were hyperresistant to rapamycin ( Figure 6C , top ) . We also confirmed that , by contrast , complete PIB2 deletion caused rapamycin hypersensitivity . A larger N-terminal truncation extending beyond the mapped territory of rapamycin resistance-conferring transposons , did not mediate hyperresistance to rapamycin ( Figure 6G , Pib2426-635 ) . To assess whether rapamycin-hyperresistant PIB2 truncations behaved as gain-of-function alleles , we co-expressed Pib2165-635 and full-length Pib2 . In these conditions , the rapamycin hyperresistance was mitigated , indicating that the effect of the truncated Pib2 protein was semi-dominant ( Figure 6C , bottom ) . Expressing the truncation from a high-copy ( 2 µ ) vector did not further increase resistance to rapamycin , indicating that higher expression levels did not change the semi-dominance of the PIB2 truncation allele . Thus , Pib2 truncation leads to a gain-of-function that translates into semi-dominant rapamycin hyperresistance . Because gain-of-function alleles of PIB2 lead to rapamycin resistance , while loss-of-function alleles lead to sensitivity , our data suggest that Pib2 positively regulates TORC1 function . To test this idea further , we investigated the effects of full-length and various truncations of Pib2 on TORC1 signaling . Switching yeast cells from a poor nitrogen source ( proline ) to a rich one ( glutamine ) triggers a fast activation of TORC1 ( Stracka et al . , 2014 ) , leading to a transient surge in the phosphorylation of Sch9 , a key target of TORC1 in yeast ( Urban et al . , 2007 ) . We cultured cells on proline-containing medium and studied TORC1 activation 2 min following addition of 3 mM glutamine by determining the phosphorylation levels of Sch9-Thr737 . Pib2 deficiency severely blunted TORC1 activation in these conditions ( Figure 6D , top ) . In a control experiment , a strain lacking GTR1 – a component of the TORC1-activating EGO ( Exit from rapamycin-induced GrOwth arrest ) complex , which is orthologous to the mammalian Rag GTPase-Ragulator complex ( Chantranupong et al . , 2015; Powis and De Virgilio , 2016 ) – showed a similarly blunted response with respect to TORC1 activation following glutamine addition to proline-grown cells ( Figure 6D , top ) . By contrast , glutamine-mediated TORC1 activation appeared normal in a strain expressing an N-terminally truncated Pib2 variant ( pPIB2165-635 , Figure 6D , bottom ) . Thus , like Gtr1 , Pib2 is necessary to activate TORC1 in response to amino acids . The N-terminus of Pib2 appears to be an inhibitory domain . The ablation of this domain confers rapamycin resistance , yet is not sufficient to constitutively activate TORC1 ( e . g . in proline-grown cells ) . Having identified an inhibitory activity at the N-terminus of Pib2 , we proceeded to address the function of other Pib2 domains . To this aim , we used a split-ubiquitin-based yeast-two-hybrid assay to probe the interaction of Pib2 fragments with the TORC1 component Kog1 , and studied the rapamycin resistance of strains expressing various truncations of Pib2 ( Figure 6E–F ) . We found that Pib2 harbored at least two central Kog1-binding regions , since two mostly non-overlapping fragments ( Pib21-312 and Pib2304-635 ) showed robust interaction with Kog1 ( Figure 6F and Figure 6—figure supplement 1B ) . Kog1 binding is , however , not essential for Pib2-mediated TORC1 activation , since cells expressing a fragment without these Kog1-binding domains ( Pib2426-635 ) were as rapamycin-resistant as cells expressing full length Pib2 ( FL , Figure 6G ) . Pib2 harbors a phosphatidylinositol-3-phosphate ( PI3P ) -binding Fab1-YOTB-Vac1-EEA1 ( FYVE ) domain ( Figure 6E ) . Pib2 truncations lacking the FYVE domain are unable to bind PI3P , and hence to properly localize to the vacuole ( Kim and Cunningham , 2015 ) . When cells expressed FYVE domain-truncated Pib2 ( Pib2Δ426-532 ) , their rapamycin resistance decreased , but not as severely as observed in pib2Δ strains ( Figure 6G ) . This indicates that FYVE-domain-mediated vacuolar recruitment is not absolutely required for Pib2 to activate TORC1 . Strikingly , cells expressing Pib2 variants that were either truncated at the extreme C-terminus ( PIB21-620 ) or carried a deletion within the C-terminus ( PIB2∆533-620 ) were as sensitive to rapamycin as pib2∆ cells ( Figure 6G ) . The C-terminus of Pib2 is therefore important to ensure proper TORC1 activation . We conclude that Pib2 harbors the following functional domains: a large N-terminal Inhibitory Domain ( NID ) , and a C-terminal TORC1-Activating Domain ( CAD ) ; the central portion of the protein harbors FYVE and Kog1-binding domains for proper targeting of Pib2 to the vacuole and TORC1 , respectively . Could the NID act in an auto-inhibitory allosteric fashion by preventing the CAD from activating TORC1 ? We reasoned that if this were the case , plasmid-encoded N-terminally truncated Pib2 should confer similar levels of rapamycin resistance independently of whether a genomic wild-type PIB2 copy was present or not . However , wild-type cells expressing N-terminally truncated Pib2165-635 from a centromeric or a multi copy 2 µ plasmid were less resistant to rapamycin than pib2∆ cells expressing Pib2165-635 from a centromeric plasmid ( Figure 6C ) . Thus , the NID activity of the endogenously-expressed wild-type Pib2 is able to mitigate the rapamycin resistance conferred by ectopic expression ( or overexpression ) of the CAD . Therefore , our data suggest that the NID does not auto-inhibit the CAD within Pib2 , but rather , that the NID and CAD act independently and antagonistically on TORC1 . This latter scenario predicts that , just as expressing NID-truncated Pib2 semi-dominantly activates TORC1 ( Figure 6C ) , expressing CAD-truncated Pib2 should semi-dominantly inhibit it . To test this prediction , we expressed the two CAD truncations Pib2Δ533-620 and Pib21-620 in an otherwise wild-type strain . The resulting strains were significantly more sensitive to rapamycin than their counterparts expressing only full length Pib2 ( Figure 6H ) . Furthermore , when pib2Δ cells expressed the Pib2 CAD truncation alleles , they became even more sensitive to rapamycin . Therefore , Pib2-NID does not act in an auto-inhibitory manner , but rather inhibits TORC1 independently of the presence of a Pib2-CAD . Since the Rag GTPase Gtr1-Gtr2 module of the EGO complex also mediates amino acid signals to TORC1 ( Binda et al . , 2009; Dubouloz et al . , 2005 , see also Figure 6D ) , we tested the possibility that Pib2-NID inhibits TORC1 by antagonizing Gtr1-Gtr2 . This does not appear to be the case; first , the expression of Pib2Δ533-620 or Pib21-620 further enhanced the rapamycin sensitivity of gtr1Δ gtr2Δ cells ( Figure 6—figure supplement 1C ) , and second , the Pib2Δ533-620- or Pib21-620-mediated rapamycin hypersensitivity was not suppressed by expression of the constitutively active , TORC1-activating Gtr1Q65L-Gtr2S23L module ( Binda et al . , 2009; Figure 6—figure supplement 1D ) . These results suggest that Pib2-NID inhibits TORC1 independently of the EGO complex . In summary , Pib2 is targeted to TORC1 by binding to the vacuolar membrane through its FYVE domain and to Kog1 via its middle portion . Pib2 harbors two antagonistic activities , one activating and the other repressing TORC1 . The dose-independent semi-dominant nature of the respective truncations indicates that both repressing and activating activities influence TORC1 independently and do not appear to compete for the same sites on TORC1 . We speculate that high-quality amino acids , such as glutamine , balance the antagonistic TORC1-activating and -repressing activities of Pib2 to tune growth rate according to available resources . In this context , it will be interesting to elucidate how the activities of Pib2 and the EGO complex are coordinated to stimulate TORC1 in response to amino acids .
Here we present a novel method based on random transposon insertion and next-generation sequencing , to functionally screen the genome of Saccharomyces cerevisiae . SAturated Transposon Analysis in Yeast ( SATAY ) can reveal positive and negative genetic interactions , auxotrophies , drug-sensitive or -resistant mutants , and map functional protein domains . SATAY combines advantages from both classical genetics and high-throughput robotic screens . SATAY can in principle be implemented in any strain background , since it does not rely on the existence of available deletion libraries and does not necessitate markers to follow genetic traits . Moreover , SATAY explores a wide genetic space; exotic alleles can be generated as exemplified by PIB2 ( Figure 6 and Figure 6—figure supplement 1 ) , where transposon insertions in different regions of the gene generate opposite phenotypes . Transposon insertion is not limited to CDSs; we observe that promoters of essential genes are often transposon-intolerant ( Figure 2 , see GAL10 , SGV1 , MMS21 , RET2 , HRR25 , NPA3 , SEC9 , SWI1 ) . We also observe that known essential tRNA , snRNA , as well as SRP RNA genes are transposon intolerant ( see Supplementary Dataset ) . Finally , our data reveal transposon-intolerant areas of the genome that do not correspond to any annotated feature , indicating that SATAY could help discover yet-unknown functional elements . SATAY yields unprecedented insight on the domain structure-function relationship of important proteins , and allows the mapping of important functional domains , without prior knowledge . Dispensable domains in essential proteins might not be required for the essential function of these proteins , but may have other roles . The sub-gene resolution enabled by SATAY may thus unveil yet-unsuspected accessory or regulatory functions , even in otherwise well-studied proteins . In addition , structure-function information revealed by SATAY may guide 3D-structure determination efforts by indicating possibly flexible accessory domains . The resolution of a SATAY screen is directly proportional to the number of transposons mapped onto the genome . The current resolution is ~1/40 bp , which is amply sufficient to confidently identify essential genes and protein domains . This resolution is achieved by mapping ~300 , 000 transposons , starting from a 1 . 6E6-colonies library . Not every colony generates a detectable transposon . This is due to several reasons . ( 1 ) Excised transposons only reinsert in 60% of the cases ( our observations and Lazarow et al . , 2012 ) . ( 2 ) 7% of the sequencing reads mapped onto repetitive DNA elements ( such as rDNA , Ty-elements and subtelomeric repeats ) and were discarded because of ambiguous mapping . ( 3 ) Two transposons inserted in the same orientation within two bp of each other will be considered as one by our algorithm . This might be exacerbated at densely covered areas of the genome , such as pericentromeric regions . ( 4 ) Some transposon insertion products may not be readily amplifiable by our PCR approach . We observe that increasing both the size of the original library and the sequencing depth leads to an increase in mapped transposon number ( albeit non-proportional , Table 1 ) . Therefore , the resolution of a screen can be tailored according to the question and the available resource . We could not detect a preferred sequence for transposon insertion ( Figure 1—figure supplement 2A ) , yet two features of the yeast genome biased insertion frequency in select regions . The first is nucleosome position; the Ds transposon has a tendency to insert more frequently within inter-nucleosomal regions , indicating that , like other transposases ( Gangadharan et al . , 2010 ) , Ac might have better access to naked DNA ( Figure 1—figure supplement 2B–C ) . The second pertains to a tendency of the transposon to integrate in the spatial proximity of its original location ( Lazarow et al . , 2012 ) . Indeed , when originated from a centromeric plasmid , the transposon has an increased propensity to reinsert in the pericentromeric regions of other chromosomes . This is not due to a particular feature of pericentromeric chromatin , since this propensity is lost when the transposon original location is on the long arm of ChrXV ( Figure 1B ) . Instead , the increased propensity is likely resulting from the clustering of centromeres in the nuclear space ( Jin et al . , 2000 ) , indicating that centromeric plasmids cluster with the chromosomal centromeres . Thus , SATAY can be utilized to probe , not only the function of the genome , but also its physical and regulatory features , such as nucleosome position and 3D architecture . Finally , SATAY does not require any particular equipment besides access to a sequencing platform , and involves a limited amount of work . It can be easily multiplexed to perform several genome-wide screens simultaneously . Each screen yields two measures , the number of transposons and the number of sequencing reads , both of which , for each gene , reveal the fitness of the cognate mutant . While the number of transposons per gene is appropriate to look for genes that become important for growth , as in genetic interaction screens , the number of sequencing reads is better suited to identify strains that are positively selected , like drug-resistant mutants . Both metrics suffer from intrinsic noise , stemming from the inherently discrete structure of the data , and probably also from unavoidable biases in the amplification and sequencing of the libraries . We show that this noise can be reduced by comparing multiple libraries against each other ( Figure 4—figure supplement 1 ) . Moreover , comparing multiple libraries allows to tailor the composition of each library set to needs . For instance , grouping the VPS13 ( D716H ) with the mmm1Δ VPS13 ( D716H ) libraries allows to selectively detect synthetic interactions with VPS13 ( D716H ) ( e . g . ERMES components ) . By contrast , comparing the mmm1Δ VPS13 ( D716H ) library with all others , selectively finds genes important for the ERMES suppression phenomenon . Thus , while signal-to-noise ratio might be a limiting factor for the detection of genetic interactions , we anticipate that increasing the number of libraries , for instance by generating multiple libraries in each condition , will likely decrease the incidence of false positive and false negative . With the increasing number of screens performed in various conditions will also come the ability to find correlation patterns among genes that are required or dispensable for growth in similar sets of conditions . Such correlations are accurate predictors of common functions and have been extensively used in synthetic genetic screens , such as the E-MAP procedure ( Kornmann et al . , 2009; Schuldiner et al . , 2005 ) . However , while E-MAP screens compute patterns of genetic interaction on a subset of chosen genetic interaction partners , SATAY allows to detect genetic interactions at the genome scale . Because the approach only necessitates a transposon and a transposase , it should not only be feasible in S . cerevisiae and S . pombe ( Guo et al . , 2013 ) , but also amenable to other industrially- , medically- or scientifically-relevant haploid fungi , such as Y . lipolytica , C . glabrata , K . lactis and P . pastoris . The Ds transposon can in principle accommodate extra sequences with no known length limitation ( Lazarow et al . , 2013 ) . An interesting future development will be to incorporate functional units in the transposon DNA , for instance strong promoters , repressors or terminators , IRESs , recognition sites for DNA-binding proteins ( LacI or TetR ) , recombination sites ( LoxP , FRP ) , or coding sequences for in-frame fusion , such as GFP , protein- or membrane-binding domains , signal sequences , etc . Improved designs will not only permit finer mapping of protein domains without reliance on spurious transcription and translation , but might allow the exploration of an even wider genetic space , for instance by generating gain-of-function variants , thus enabling the development of novel approaches to interrogate the yeast genome .
All yeast strains , oligonucleotides and plasmids used herein are listed in Tables 2 , 3 and 4 , respectively . To generate pBK257 , the ADE2 gene interrupted with the MiniDs transposon was PCR amplified from strain CWY1 ( Weil and Kunze , 2000 ) , using PCR primers #4 and #5 . The PCR product and pWL80R_4x ( plasmid encoding the Ac transposase under the control of the GAL1 promoter , Lazarow et al . , 2012 ) were digested with SacI , then ligated together . This plasmid does not confer adenine prototrophy to ade2Δ cells unless the Ac transposase excises the MiniDS transposon , and repairs the ADE2 gene . 10 . 7554/eLife . 23570 . 020Table 2 . Yeast strains used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 020NameParentGenotypeReference CWY1BY4723MATa his3Δ0 ura3Δ0 ade2:Ds-1Weil and Kunze ( 2000 ) ByK157BY4743MATα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0 VPS13 ( D716H ) Lang et al . ( 2015b ) ByK352BY4741MATa his3Δ1 leu2Δ0 met17Δ0 ura3Δ0 ade2Δ::HIS3*This study ByK484By4742MATα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0 ade2Δ::HIS3*This study ByK485ByK352 and ByK484MATa/α his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 LYS2/lys2Δ0 met17Δ0/MET17 ura3Δ0/ura3Δ0 ade2Δ::HIS3*/ade2Δ::HIS3*This study ByK446ByK157MATα his3Δ1 leu2Δ0 ura3Δ0 ade2Δ::HIS3* VPS13 ( D716H ) This study ByK528ByK446MATα his3Δ1 leu2Δ0 ura3Δ0 ade2Δ::HIS3* VPS13 ( D716H ) mmm1Δ::KanMX6This study ByK530ByK352MATa his3Δ1 leu2Δ0 met17Δ0 ura3Δ0 ade2Δ::NAT* dpl1Δ::KanMX6This study ByK533ByK352 MATa his3Δ1 leu2Δ0 met17Δ0 ura3Δ0 ade2Δ::HIS3* psd2Δ::KanMX6 dpl1Δ::NATThis study ByK576ByK485MATa/α his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 LYS2/lys2Δ0 met17Δ0/MET17 ura3Δ0/ura3Δ0 ade2Δ::HIS3*/ade2Δ::HIS3* prp45Δ::KanMX6/PRP45This study ByK579ByK485MATa/α his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 LYS2/lys2Δ0 met17Δ0/MET17 ura3Δ0/ura3Δ0 ade2Δ::HIS3*/ade2Δ::HIS3* PRP451-462-HA ( KanMX6 ) / PRP45This study ByK583ByK485MATa/α his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 LYS2/lys2Δ0 met17Δ0/MET17 ura3Δ0/ura3Δ0 ade2Δ::HIS3*/ade2Δ::HIS3* taf3Δ::KanMX6/TAF3This study ByK588ByK485MATa/α his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 LYS2/lys2Δ0 met17Δ0/MET17 ura3Δ0/ura3Δ0 ade2Δ::HIS3*/ade2Δ::HIS3* TAF31-270-HA ( KanMX6 ) / TAF3This study ByK725ByK533 and ByK484MATa/α his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 LYS2/lys2Δ0 met17Δ0/MET17 ura3Δ0/ura3Δ0 ade2Δ::HIS3*/ade2Δ::HIS3* psd2Δ::KanMX6/PSD2 dpl1Δ::NAT /DPL1This study ByK726ByK533 and ByK484MATa/α his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 LYS2/lys2Δ0 met17Δ0/MET17 ura3Δ0/ura3Δ0 ade2Δ::HIS3*/ade2Δ::HIS3* psd2Δ::KanMX6/PSD2 dpl1Δ::NAT /DPL1This study ByK739ByK725MATa/α his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 LYS2/lys2Δ0 met17Δ0/MET17 ura3Δ0/ura3Δ0 ade2Δ::HIS3*/ade2Δ::HIS3* psd2Δ::KanMX6/PSD2 dpl1Δ::NAT /DPL1 cdc10Δ::URA3/CDC10This study ByK740ByK726MATa/α his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 LYS2/lys2Δ0 met17Δ0/MET17 ura3Δ0/ura3Δ0 ade2Δ::HIS3*/ade2Δ::HIS3* psd2Δ::KanMX6/PSD2 dpl1Δ::NAT /DPL1 cdc10Δ::URA3/CDC10This study ByK741ByK726MATa/α his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 LYS2/lys2Δ0 met17Δ0/MET17 ura3Δ0/ura3Δ0 ade2Δ::HIS3*/ade2Δ::HIS3* psd2Δ::KanMX6/PSD2 dpl1Δ::NAT /DPL1 cdc10Δ::URA3/CDC10This study YJM3916ByK352MATa his3Δ1 leu2Δ0 met17Δ0 ura3Δ0 ade2Δ::HIS3* YEN1onThis study YL516BY4741/BY4742MATa his3Δ1 leu2Δ0 ura3Δ0Binda et al . ( 2009 ) MB32YL516MATa his3Δ1 leu2Δ0 ura3Δ0 gtr1Δ::kanMXBinda et al . ( 2009 ) RKH106YL516MATa his3Δ1 leu2Δ0 ura3Δ0 pib2Δ::kanMXThis study RKH241MB32MATa his3Δ1 leu2Δ0 ura3Δ0 gtr1Δ::kanMX gtr2Δ::hphMX4This study NMY51his3∆200 trp1-901 leu2-3 , 112 ade2 LYS:: ( lexAop ) 4-HIS3 ura3:: ( lexAop ) 8- lacZ ade2:: ( lexAop ) 8-ADE2 GAL4Dualsystems Biotech AG*ADE2 deleted −56 before ATG +62 after STOP with PCR primers #6 and #7 on pFA6a-His3MX6 . 10 . 7554/eLife . 23570 . 021Table 3 . Oligonucleotides used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 021#Original nameSequencePurpose1P5_MiniDsAATGATACGGCGACCACCGAGATCTACtccgtcccgcaagttaaataamplify library2MiniDs_P7CAAGCAGAAGACGGCATACGAGATacgaaaacgaacgggataaaamplify library3688_minidsSEQ1210tttaccgaccgttaccgaccgttttcatccctasequence library4ADE2FwdGGTTCGAGCTCCCTTTTGATGCGGAATTGACclone ADE2 MiniDS5ADE2RevGACCTGAGCTCTTACTGGATATGTATGTATGclone ADE2 MiniDS6Ade2PriFwdGTATAAATTGGTGCGTAAAATCGTTGGATCTCTCTTCTAAcggatccccgggttaattaadelete ADE27Ade2PriRevTATGTATGAAGTCCACATTTGATGTAATCATAACAAAGCCgaattcgagctcgtttaaacdelete ADE28Dpl1_Janke_S1AGCAAGTAGGCTAGCTTCTGTAAAGGGATTTTTCCATCTAATACAcgtacgctgcaggtcgacdelete DPL19Dpl1_Janke_S2GCACTCTCGTTCTTTAAATTATGTATGAGATTTGATTCTATATAGatcgatgaattcgagctcgdelete DPL110Psd2_pringle_FGATGCTGTATCAATTGGTAAAGAATCCTCGATTTTCAGGAGCATCCAACGcgtacgctgcaggtcgacdelete PSD211Psd2_pringle_RCTTGTTTGTACACGCTATAGTCTATAATAAAGTCTGAGGGAGATTGTTCATGatcgatgaattcgagctcgdelete PSD212TAF3_R1TGGATGAGATAATGACGAAAGAAAATGCAGAAATGTCGTTgaattcgagctcgtttaaacTAF3 partial deletion13TAF3_aa90_F2AGGTATTGTTAAGCCTACGAACGTTCTGGATGTCTATGATcggatccccgggttaattaaTAF3 partial deletion14Taf3_FwdGGCAAGATGTGATCAGGACGcheck TAF3 partial deletion15Taf3_RevTCTTGAAGAAGCGAAAGTACACTcheck TAF3 partial deletion16TAF3_R1TGGATGAGATAATGACGAAAGAAAATGCAGAAATGTCGTTgaattcgagctcgtttaaacTAF3 complete deletion17TAF3_aa1_ F1GAAAACAGCGATATCTTTGGGTCAATAGAGTTCCTCTGCTtgaggcgcgccacttctaaaTAF3 complete deletion18PRP45_R1ACTCAAGCACAAGAATGCTTTGTTTTCCTAGTGCTCATCCTGGGCgaattcgagctcgtttaaacPRP45 partial deletion19PRP45_aa154_F2AACGACGAAGTCGTGCCTGTTCTCCATATGGATGGCAGCAATGATcggatccccgggttaattaaPRP45 partial deletion20PRP45_FwdAGGTTGTAGCACCCACAGAAcheck PRP45 partial deletion21PRP45_RevCAATCATCACACCTCAGCGAcheck PRP45 partial deletion22PRP45_R1ACTCAAGCACAAGAATGCTTTGTTTTCCTAGTGCTCATCCTGGGCgaattcgagctcgtttaaacPRP45 complete deletion23PRP45_aa1_F1GCTCTGAGCCGAGAGGACGTATCAGCAACCTCAACCAAATtgaggcgcgccacttctaaaPRP45 complete deletion24CDC10-Ura3_fwdAAGGCCAAGCCCCACGGTTACTACAAGCACTCTATAAATATATTAtgacggtgaaaacctctgacCDC10 complete deletion25URA3-CDC10_revTTCTTAATAACATAAGATATATAATCACCACCATTCTTATGAGATtcctgatgcggtattttctccCDC10 complete deletion26OJM370ATGGGTGTCTCACAAATATGGGAmplify YEN127OJM371TTCAATAGTGCTACTGCTATCACAmplify YEN128OJM372TTCAATAGTGCTACTGCTATCACTGTCACAGGCTCAAACCGGTCGACTG TTCGTACGCTGCAGGTCGACDelitto perfetto on YEN129OJM373ATGGGTGTCTCACAAATATGGGAATTTTTGAAGCCATATCTGCAAGATTCCCGCGCGTTGGCCGATTCATDelitto perfetto on YEN130o3958gacggtatcgataagcttgatatcgGCGCTGGCATCTTTAATCTCPIB2 cloning31o3959actagtggatcccccgggctgcaggTGCTTGGATCCTTCTTGGTCPIB2 cloning32o3224TAATA CGACT CACTA TAGGGvarious PIB2 truncations33o3225ATTAA CCCTC ACTAA AGGGA Avarious PIB2 truncations34o4034atctagttcagggttcgacattctggtctccactacPIB2165-635 truncation35o4010gtagtggagaccagaatgtcgaaccctgaactagatPIB2165-635 truncation36o4012tagtggagaccagaatgttaccgcagcctgctPIB2304-635 truncation37o4035tcaaattagaactagcattcattctggtctccactacaactgtgPIB2221-635 truncation38o4011cacagttgtagtggagaccagaatgaatgctagttctaatttgaPIB2221-635 truncation39o4062atagttggtattaagttgattctcattctggtctccactacaactgPIB2426-635 truncation40o3996cagttgtagtggagaccagaatgagaatcaacttaataccaactatPIB2426-635 truncation41o4063cgtgtttgcgttatggttgtcgctgttcggaatagaPIB2Δ426-532 truncation42o3997tctattccgaacagcgacaaccataacgcaaacacgPIB2Δ426-532 truncation43o4064cacagagccgataacactcgtggttgaaaggttctcPIB2Δ533-620 truncation44o3998gagaacctttcaaccacgagtgttatcggctctgtgPIB2Δ533-620 truncation45o4065gtctcgcaaaaaatgttcatcagcccaaaacatcattaccttctPIB21-620 truncation46o3999agaaggtaatgatgttttgggctgatgaacattttttgcgagacPIB21-620 truncation47o1440GCTAGAGCGGCCATTACGGCCCCGGAGATTTATGGACCTCKOG1 cloning into pPR3N48o1442CGATCTCGGGCCGAGGCGGCCTCAAAAATAATCAATTCTCTCGTCKOG1 cloning into pPR3N49o3787GCTAGAGCGGCCATTACGGCC GAATTGTACAAATCTAGAACTAGTcloning PIB2 fragments into pCabWT*50o3788CGATCTCGGGCCGAGGCGGCCAA GAAACTACTCCAATTCCAGTTTGCcloning PIB2 fragments into pCabWT*51o3872CGATCTCGGGCCGAGGCGGCCAAGCCCAAAACATCATTACCTTCTTCTcloning PIB2 fragments into pCabWT*52o3871CGATCTCGGGCCGAGGCGGCCAAATCTTCGCCCTCCTCAACGTcloning PIB2 fragments into pCabWT*53o3870CGATCTCGGGCCGAGGCGGCCAAGTTGATTCTGTCGCTGTTCGcloning PIB2 fragments into pCabWT*54o3933GCTAGAGCGGCCATTACGGCCAGGAAGAAATTACGCAATTACTACcloning PIB2 fragments into pCabWT*55o3934GCTAGAGCGGCCATTACGGCC AGTGTTATCGGCTCTGTGCCcloning PIB2 fragments into pCabWT*56o3868CGATCTCGGGCCGAGGCGGCCAAATTAGTGCTCGAAGCAGGCTcloning PIB2 fragments into pCabWT*57o3867CGATCTCGGGCCGAGGCGGCCAAGTCATCCGTGAATGGCAACGcloning PIB2 fragments into pCabWT*58o3866CGATCTCGGGCCGAGGCGGCCAAGCCTGCCCCTGTTGAGCTCTcloning PIB2 fragments into pCabWT*59o3865CGATCTCGGGCCGAGGCGGCCAAGTCAGCACCGCTTTCCTCATcloning PIB2 fragments into pCabWT*Oligonucleotides #1 and #2 , ordered as PAGE-purified and lyophilized , are resuspended at 100 μM in water . Oligonucleotide #3 , ordered as HPLC-purified and lyophilized , is resuspended at 100 μM in water and distributed into single-use aliquots . 10 . 7554/eLife . 23570 . 022Table 4 . Plasmids used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 23570 . 022NameParentDescriptionReference pBK257pWL80R_4xCEN/URA3 , carries MiniDs in ADE2 and hyperactive Ac transposase under GAL1 promoterThis study pWL80R_4xCEN/URA3 , carries hyperactive Ac transposase under GAL1 promoterLazarow et al . ( 2012 ) pCORE-UHDelitto pefetto URA3 cassetteStorici and Resnick ( 2003 ) pJM7pENTRY-YEN1ONThis study pRS413CEN/HIS3 , empty vectorSikorski and Hieter , 1989 pRS415CEN/LEU2 , empty vectorSikorski and Hieter , 1989 pRS416CEN/URA3 , empty vectorSikorski and Hieter , 1989 p1822pRS413CEN/HIS3 , GTR1This study p1451pRS415CEN/LEU2 , GTR2This study p1821pRS413CEN/HIS3 , GTR1Q65LThis study p1452pRS415CEN/LEU2 , GTR2S23LThis study p3084pRS416CEN/URA3 , PIB2This study p3099p3084CEN/URA3 , PIB2165-635This study p3097p3084CEN/URA3 , PIB2304-635This study p3101p3084CEN/URA3 , PIB2221-635This study p3253pRS4262 μ/URA3 , PIB2This study p3255pRS4262 μ/URA3 , PIB2165-635This study p3163p3084CEN/URA3 , PIB2426-635This study p3153p3084CEN/URA3 , PIB2Δ426-532This study p3154p3084CEN/URA3 , PIB2Δ533-620This study p3156p3084CEN/URA3 , PIB21-620This study pPR3N2 μ/TRP1 , NubG-HADualsystems Biotech AG pCabWTCEN/LEU2 , Aβ-Cub-LexA-VP16Dualsystems Biotech AG p3081pPR3N2 μ/TRP1 , NubG-HA-KOG1This study p2966pCabWTCEN/LEU2 , Aβ-PIB2-Cub-LexA-VP16This study p3002pCabWTCEN/LEU2 , Aβ-PIB21-620-Cub-LexA-VP16This study p3007pCabWTCEN/LEU2 , Aβ-PIB21-550-Cub-LexA-VP16This study p3001pCabWTCEN/LEU2 , Aβ-PIB21-428-Cub-LexA-VP16This study p3051pCabWTCEN/LEU2 , Aβ-PIB2440-550-Cub-LexA-VP16This study p3054pCabWTCEN/LEU2 , Aβ-PIB2556-620-Cub-LexA-VP16This study p3052pCabWTCEN/LEU2 , Aβ-PIB2621-635-Cub-LexA-VP16This study p3000pCabWTCEN/LEU2 , Aβ-PIB21-312-Cub-LexA-VP16This study p2987pCabWTCEN/LEU2 , Aβ-PIB2304-635-Cub-LexA-VP16This study p2999pCabWTCEN/LEU2 , Aβ-PIB21-162-Cub-LexA-VP16This study p2986pCabWTCEN/LEU2 , Aβ-PIB2165-635-Cub-LexA-VP16This study p2998pCabWTCEN/LEU2 , Aβ-PIB21-101-Cub-LexA-VP16This study p2991pCabWTCEN/LEU2 , Aβ-PIB2102-635-Cub-LexA-VP16This study p2997pCabWTCEN/LEU2 , Aβ-PIB21-49-Cub-LexA-VP16This study p2990pCabWTCEN/LEU2 , Aβ-PIB250-635-Cub-LexA-VP16This study Deletion strains were generated by PCR-mediated gene replacement using the Longtine toolbox for KanMX6 and HIS3 replacement ( Longtine et al . , 1998 ) and the Janke toolbox for NATnt2 ( Janke et al . , 2004 ) , with primers listed in Table 3 . Strain YJM3916 carrying YEN1ON at the endogenous locus was generated using the delitto perfetto method ( Storici and Resnick , 2003 ) . ade2Δ strains were transformed with the pBK257 plasmid . These strains are phenotypically ade- , since the ADE2 gene borne on the plasmid is interrupted by the MiniDs transposon . One liter of freshly prepared SD -Ura +2% Raffinose +0 . 2% Dextrose is inoculated with ade2Δ cells freshly transformed with pBK257 , directly scraped off the transformation plates , at a final OD600 = 0 . 15 . The culture is grown to saturation for 18 to 24 hr at 30°C . Cells are spun for 5 min at 600x g , 20°C , and resuspended in their supernatant at a final OD600 = 39 . 200 μl of this resuspension are plated on ~250–300×8 . 5 cm plates containing 25 ml of SD +2% Galactose -Adenine using glass beads . Plates are incubated in closed but not sealed plastic bags for 3 weeks at 30°C . Clones in which transposon excision has led to the repair of the ADE2 gene on pBK257 start to appear after 10–12 days . The density of clones on the plate reaches 150–200 colonies/cm2 , i . e . 8000-11000 colonies/plates after 3 weeks . All colonies are then scraped off the plates using minimal volume of either water or SD +2% Dextrose -Adenine , pooled , and used to inoculate a 2-liter SD +2% Dextrose -Adenine culture at a density of 2 . 5 106 cells/ml , which is allowed to grow to saturation . This step is used to dilute any remaining ade- cells , which represent about 20% of the total number of cells , and ensures that each transposition event is well represented . For example , reseeding a 2 106 clones library in 2L at a density of 2 . 5 106 cells/ml will ensure that each clone is represented by ( ( 2 . 500 . 000 × 1000×2 ) *0 . 8 ) /2 . 000 . 000 = 2000 cells . The saturated culture is harvested by centrifugation ( 5 min , 1600x g ) , washed with ddH2O , then cell pellets are frozen as ~500 mg aliquots . Cells scraped off the plates were used to inoculate a 1-liter SD +2% Dextrose -Adenine culture at OD 0 . 08 . After growing for 15 hr to OD 0 . 5 , the culture was diluted to OD 0 . 1 in 500 ml SD +2% Dextrose -Adenine , treated with 10 nM ( 9 . 14 ng/ml ) rapamycin ( Sigma ) and grown for 24 hr to OD 0 . 9 . The culture was then diluted again to OD 0 . 1 in 500 ml SD +2% Dextrose -Adenine + 10 nM rapamycin . The treated culture was grown to saturation ( OD 1 . 9 ) , harvested by centrifugation and processed for genomic DNA extraction . A 500 mg cell pellet is resuspended with 500 μl Cell Breaking Buffer ( 2% Triton X-100 , 1% SDS , 100 mM NaCl , 100 mM Tris-HCl pH8 . 0 , 1 mM EDTA ) and distributed in 280 μl aliquots . 200 μl Phenol:Chloroform:Isoamylalcool 25:25:1 and 300 μl 0 . 4–0 . 6 mm unwashed glass beads are added to each aliquot . Samples are vortexed for 10 min at 4°C using a Disruptor Genie from Scientific Industrial ( US Patent 5 , 707 , 861 ) . 200 μl TE are added to each lysate , which are then centrifuged for 5 min at 16100x g , 4°C . The upper layer ( ~400 μl ) is transferred to a fresh tube , 2 . 5vol 100% EtOH are added and the sample mixed by inversion . DNA is pelleted for 5 min at 16100x g , 20°C . The supernatant is removed and the pellets resuspended in 200 μl RNAse A 250 μg/ml for 15 min at 55°C , 1000 rpm on a Thermomixer comfort ( Eppendorf ) . 2 . 5 vol 100% EtOH and 0 . 1 vol NaOAc 3 M pH5 . 2 are added and the samples mixed by inversion . DNA is pelleted by centrifugation for 5 min at 16100x g , 20°C . The pellets are washed with 70% EtOH under the same conditions , the supernatant removed completely , and the pellets dried for 10 min at 37°C . The pellets are resuspended in a total volume of 100 μl water for 10 min at 55°C , 700 rpm on a Thermomixer comfort ( Eppendorf ) . DNA is run on a 0 . 6% 1X TBE agarose gel against a standard 1 kb GeneRuler , and quantified using Fiji . 500 mg cell pellet should yield 20–60 μg DNA . Sequencing involves the following steps: ( 1 ) Digestion of genomic DNA with two four-cutter restriction enzymes , ( 2 ) ligase-mediated circularization of the DNA , ( 3 ) PCR of the transposon-genome junctions using outward-facing primers , ( 4 ) Illumina-sequencing of the combined PCR products . 2 × 2 μg of genomic DNA are digested in parallel in Non-Stick microfuge tubes ( Ambion AM12450 ) with 50 units of DpnII ( NEB #R0543L ) and NlaIII ( NEB #R0125L ) , in 50 μl for 16 hr at 37°C . The reactions are then heat inactivated at 65°C for 20 min and circularized in the same tube by ligation with 25 Weiss units T4 Ligase ( Thermo Scientific #EL0011 ) for 6 hr at 22°C , in a volume of 400 μl . DNA is precipitated overnight or longer at −20°C in 0 . 3 M NaOAc pH5 . 2 , 1 ml 100% EtOH , using 5 μg linear acrylamide ( Ambion AM9520 ) as a carrier , then centrifuged for 20 min at 16100x g , 4°C . Pellets are washed with 1 ml 70% EtOH , for 20 min at 16100 x g , 20°C . After complete removal of the supernatant , pellets are dried for 10 min at 37°C . Each circularized DNA preparation is then resuspended in water and divided into 10 × 100 μl PCR reactions . Each 100 μl PCR reaction contains: 10 μl 10X Taq Buffer ( 500 mM Tris-HCl pH9 . 2 , 22 . 5 mM MgCl2 , 160 mM NH4SO4 , 20% DMSO , 1% Triton X-100 – stored at −20°C ) , 200 μM dNTPs , 1 μM primer #1 , 1 μM primer #2 , 2 . 4 μl homemade Taq polymerase . PCR is performed in an MJ Research Peltier Thermal Cycler PTC-200 using the following conditions: Block: calculated – 95°C 1 min , 35 × [95°C 30 s , 55°C 30 s , 72°C 3 min] , 72°C 10 min . The 2 × 10 PCR reactions are pooled into one NlaIII-digested pool and one DpnII-digested pool . 100 μl from each pool are purified using a PCR clean-up/gel extraction kit ( Macherey-Nagel ) according to the manufacturer protocol , with the following modifications . DNA is bound to the column for 30s at 3000x g; 30 μl of elution buffer ( 10 mM Tris-HCl pH8 . 5 , 0 . 1% Tween ) is applied to the column and incubated for 3 min , then spun for 1 min at 11000x g at 20°C . The eluate is reapplied to the column and a second elution is performed under the same conditions . Purified PCR products are quantified by absorbance at 260 nm . On a 1% agarose gel , the product runs as a smear from 250 bp to 1 . 2 kb , with highest density centered around 500 bp . The 867 bp size band present in the NlaIII-treated sample and the 465 bp size band present in the DpnII-treated sample correspond to untransposed pBK257 . Equal amounts of DpnII- and NlaIII-digested DNA are pooled and sequenced using MiSeq v3 chemistry , according to manufacturer , adding 3 . 4 μl of 100 μM primer #3 into well 12 of the sequencing cartridge . The fastq file generated is uploaded into the CLC genomics workbench , trimmed using adaptor sequences ‘CATG’ and ‘GATC’ ( the recognition sites for NlaIII and DpnII , respectively ) , allowing two ambiguities and a quality limit of 0 . 05 . The trimmed sequence is then aligned to the reference genome , using the following parameters ( mismatch cost , 2; insertion and deletion costs , 3; length fraction , 1; similarity fraction , 0 . 95; non-specific match handling , ignore ) . The alignment is then exported as a BAM file , which is further processed in MatLab , using the Source code 1 , to detect individual transposition events . The outputted bed file is uploaded to the UCSC genome browser . Yeast annotations were downloaded from the Saccharomyces Genome Database ( SGD ) . To generate our list of essential genes , we used YeastMine and searched the SGD for genes for which the null mutant has an ‘inviable’ phenotype ( Balakrishnan et al . , 2012 ) . Volcano plots were computed as follows . Two sets of libraries were defined . For each gene and each library , the number of transposons per gene ( tnpergene variable ) was normalized to the total number of transposons mapped in the library . For each gene , the fold-change is calculated as the mean of the normalized number of transposons per gene in the experimental set , divided by that in the reference set . The p-value is computed using the Student’s t-test by comparing , for each gene , the normalized number of transposons per gene for each library in the experimental and reference sets . Cells were grown to mid-log phase in synthetic minimal medium containing 0 . 5 g/L proline as a sole nitrogen source and stimulated with 3 mM glutamine for 2 min . Cells were treated with 6 . 7% w/v trichloroacetic acid ( final concentration ) , pelleted , washed with 70% ethanol and then lyzed in urea buffer ( 50 mM Tris-HCl [pH 7 . 5] , 5 mM EDTA , 6 M urea , 1% SDS , 0 . 1 mg/ml Pefabloc/phosphatase inhibitor mix ) . After disrupting cells with glass beads and incubating with Laemmli SDS sample buffer , samples were subjected to regular SDS-PAGE and immunoblotting . The phosphorylation level of Sch9-Thr737 and the total amount of Sch9 were assessed using the phosphospecific anti-Sch9-pThr737 and anti-Sch9 antibodies , respectively ( Péli-Gulli et al . , 2015 ) . The split-ubiquitin yeast two-hybrid system from Dualsystems Biotech AG was used following the manufacturer’s instructions . Pib2 fragments ( full-length or truncated ) and full-length Kog1 were cloned into pCabWT and pPR3N plasmids , respectively , and transformed into the strain NMY51 as indicated . Protein-protein interactions were detected as growth of the resultant strains on agar plates lacking adenine . Sequencing data have been deposited at EMBL-EBI ArrayExpress: E-MTAB-4885 . | Genes are stretches of DNA that carry the instructions to build and maintain cells . Many studies in genetics involve inactivating one or more genes and observing the consequences . If the loss of a gene kills the cell , that gene is likely to be vital for life . If it does not , the gene may not be essential , or a similar gene may be able to take over its role . Baker’s yeast is a simple organism that shares many characteristics with human cells . Many yeast genes have a counterpart among human genes , and so studying baker’s yeast can reveal clues about our own genetics . Michel et al . report an adaptation for baker’s yeast of a technique called “Transposon sequencing” , which had been used in other single-celled organisms to study the effects of interrupting genes . Briefly , a virus-like piece of DNA , called a transposon , inserts randomly into the genetic material and switches off individual genes . The DNA is then sequenced to reveal every gene that can be disrupted without killing the cell , and remaining genes are inferred to be essential for life . The approach , named SATAY ( which is short for “saturated transposon analysis in yeast” ) , uses this strategy to create millions of baker’s yeast cells , each with a different gene switched off . Because the number of cells generated this way vastly exceeds the number of genes , every gene will be switched off by several independent transposons . Therefore the technique allows all yeast genes to be inactivated several times in one single experiment . The cells can be grown in varying conditions during the experiment , revealing the genes needed for survival in different situations . Non-essential genes can also be inactivated beforehand to uncover if any genes might be compensating for their absence . In the future , this technique may be used to better understand human diseases , such as cancer , since many disease-causing genes in humans have counterparts in yeast . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"cell",
"biology",
"tools",
"and",
"resources"
] | 2017 | Functional mapping of yeast genomes by saturated transposition |
During apoptosis , Bak and Bax undergo major conformational change and form symmetric dimers that coalesce to perforate the mitochondrial outer membrane via an unknown mechanism . We have employed cysteine labelling and linkage analysis to the full length of Bak in mitochondria . This comprehensive survey showed that in each Bak dimer the N-termini are fully solvent-exposed and mobile , the core is highly structured , and the C-termini are flexible but restrained by their contact with the membrane . Dimer-dimer interactions were more labile than the BH3:groove interaction within dimers , suggesting there is no extensive protein interface between dimers . In addition , linkage in the mobile Bak N-terminus ( V61C ) specifically quantified association between dimers , allowing mathematical simulations of dimer arrangement . Together , our data show that Bak dimers form disordered clusters to generate lipidic pores . These findings provide a molecular explanation for the observed structural heterogeneity of the apoptotic pore .
The Bcl-2 family of proteins are the principal regulators of apoptotic cell death , with either Bak or Bax required to permeabilise the mitochondrial outer membrane ( Lindsten et al . , 2000; Wei et al . , 2001 ) . Bak and Bax are triggered to convert to their activated conformation by the binding of BH3-only relatives , such as Bid and Bim ( Gavathiotis et al . , 2010; Dai et al . , 2011; Du et al . , 2011; Czabotar et al . , 2013; Leshchiner et al . , 2013; Brouwer et al . , 2014 ) , and once activated can be sequestered by pro-survival relatives , such as Mcl-1 and Bcl-xL ( Llambi et al . , 2011 ) . The ability of Bak and Bax to change conformation and oligomerise in the mitochondrial outer membrane appears crucial for their capacity to perforate this membrane ( Westphal et al . , 2014 ) . Both proteins comprise nine α-helices that adopt a globular fold in the non-activated state ( Suzuki et al . , 2000; Moldoveanu et al . , 2006 ) . Non-activated Bak is anchored in the mitochondrial outer membrane via the α9-helix forming a transmembrane domain ( Iyer et al . , 2015 ) . In contrast , non-activated Bax is largely cytosolic due to binding of α9 to its own hydrophobic groove ( Wolter et al . , 1997; Gahl et al . , 2014 ) . Following binding of BH3-only proteins , both Bak and Bax undergo similar conformation changes including α1 dissociation ( Weber et al . , 2013; Alsop et al . , 2015 ) and separation of the ‘latch’ domain ( α6-α8 ) ( Czabotar et al . , 2013; Brouwer et al . , 2014 ) from the ‘core’ domain ( α2-α5 ) . The core and α6 then collapse onto the membrane surface and become shallowly inserted into the membrane to lie in-plane ( Aluvila et al . , 2014; Bleicken et al . , 2014; Westphal et al . , 2014 ) . Symmetric homodimers then form when the exposed BH3 domain of each molecule is re-buried in the hydrophobic groove of another activated molecule ( Dewson et al . , 2008 , 2012; Czabotar et al . , 2013; Brouwer et al . , 2014 ) . Bax oligomers may also form prior to membrane insertion ( Luo et al . , 2014; Sung et al . , 2015 ) . How symmetric homodimers of Bak or Bax then associate to porate the mitochondrial outer membrane is still unknown . Although there is structural information for Bak and Bax homodimers , no structures of higher order oligomers of Bak or Bax have been resolved . Upon an apoptotic stimulus , Bak and Bax have been shown to coalesce into clusters at the mitochondria ( Nechushtan et al . , 2001 ) and small clusters of Bax have been implicated in the rapid release of cytochrome c ( Zhou and Chang , 2008 ) . The availability of full-length recombinant Bax makes biochemical assays more amenable for Bax than for Bak . Thus , recent studies using artificial membranes and recombinant Bax have detected Bax complexes of various shapes and sizes with fluorescence microscopy ( Subburaj et al . , 2015 ) and cryo-electron microscopy has revealed Bax protein exclusively associated with pore edges ( Kuwana et al . , 2016 ) . Furthermore , super-resolution microscopy has shown Bax can be found in ring-like structures , arcs and clusters at the mitochondrial outer membrane of apoptotic cells ( Große et al . , 2016; Salvador-Gallego et al . , 2016 ) , although the orientation of Bax dimers in these structures could not be visualised . Higher order oligomers of both Bak and Bax vary in size when assessed by gel filtration , blue native PAGE or linkage ( George et al . , 2007; Dewson et al . , 2012 ) . Moreover , linkage studies have identified several points at which dimers might associate , including interactions at α-helices 1 , 3 , 5 , 6 , and 9 ( Dewson et al . , 2008 , 2009; Zhang et al . , 2010; Pang et al . , 2012; Ma et al . , 2013; Aluvila et al . , 2014; Bleicken et al . , 2014; Gahl et al . , 2014; Iyer et al . , 2015; Mandal et al . , 2016; Zhang et al . , 2016 ) . Critically , it is not clear whether any of these interaction sites are required for dimer-dimer association and assembly of the apoptotic pore . Here , we compare linkages through the full length of Bak in cells and show that dimers associate in a disordered and lipid-mediated fashion .
To resolve how Bak dimers coalesce to porate the mitochondrial outer membrane , we sought to generate a more detailed biochemical map of the membrane topology of Bak dimers . This work complemented our previous cysteine-accessibility analyses of Bak α5 , α6 and α9 ( Westphal et al . , 2014; Iyer et al . , 2015 ) , by analyzing the Bak N-terminus and additional residues in the α2-α5 core and C-terminus . The two native cysteine residues of human Bak ( C14 and C166 ) were first substituted with serine to generate Bak Cys null ( BakΔCys , i . e . C14S/C166S ) , and then a single cysteine residue substituted at several positions throughout the molecule . Each Bak variant was stably expressed in Bak-/-Bax-/- mouse embryonic fibroblasts ( MEFs ) and tested for function ( Figure 1—figure supplement 1 and Figure 1—figure supplement 2 ) . To convert Bak to the activated oligomeric form , membrane fractions enriched for mitochondria were incubated with tBid , as previously ( Dewson et al . , 2008 ) . To label solvent-exposed cysteine residues , membrane fractions were treated with the thiol-specific labelling reagent IASD ( 4-acetamido-4'- ( ( iodoacetyl ) amino ) stilbene-2 , 2'-disulfonic acid ) . Two negative sulfonate charges prevent IASD from accessing cysteine in hydrophobic environments ( such as the mitochondrial outer membrane or the hydrophobic protein core ) , and also allow isoelectric focusing ( IEF ) to resolve IASD-labelled and IASD-unlabelled Bak ( Tran et al . , 2013; Westphal et al . , 2014 ) . We assessed each cysteine-substituted Bak variant for labelling before , during or after incubation with recombinant tBid or tBidBax , a tBid variant containing the Bax BH3 domain ( Hockings et al . , 2015 ) , that activates Bak analogous to tBid , as well as a control not exposed to IASD and another fully exposed to labelling by denaturation and membrane solubilisation ( Figure 1A , Figure 1—figure supplement 3 and Figure 1—source data 1 ) . The approach thus monitors solvent-exposure of the residue as non-activated Bak converts to oligomeric Bak , and may detect transient exposure during these conformational changes . 10 . 7554/eLife . 19944 . 003Figure 1 . Following oligomerisation , the Bak N-segment , α1 and α1-α2 loop become fully solvent-exposed in contrast to the partially exposed core ( α2-α5 ) and latch ( α6-α9 ) . ( A ) Solvent exposure of hBak cysteine mutants was assessed by IASD labelling before ( lane 2 ) , during ( lane 3 ) and after ( lane 4 ) treatment with tBid . Controls of unlabelled ( untreated , lane 1 ) and fully labelled ( denatured , lane 5 ) Bak were included for comparison . Example IEF western blots are shown . ( B ) Quantitation of IASD labelling before and after treatment with tBid for the panel of previously untested Bak residues . Data are mean ± SD ( n ≥ 3 ) , or range ( n = 2 ) , with n for each residue labelled on the x-axis . IASD labelling data for residue V194C were from ( Westphal et al . , 2014 ) ( denoted # ) . Residues for which there is a significant difference in IASD labelling before versus after tBid are in bold and underlined ( p<0 . 05 ) . ( C ) Heat map overview of Bak IASD labelling with tBid treatment from ( B ) pooled with published analyses from ( Westphal et al . , 2014 ) ( denoted # ) or ( Iyer et al . , 2015 ) ( denoted ^ ) and from treatment with the tBidBax chimera ( denoted *; see also Figure 1—figure supplement 3; Figure 1—source data 1 ) . ( D ) Schematic of Bak structural rearrangement from its non-activated monomeric state to the activated dimer . Helices are numbered for the non-activated Bak . Note the complete solvent-exposure of α1 and the α1-α2 loop in the activated dimer . The following figure supplements are available for Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 00310 . 7554/eLife . 19944 . 004Figure 1—source data 1 . Quantitation of Bak IASD labelling before , during and after Bak activation; table of values . The percentage IASD labelling for each replicate and the summary statistics ( number of replicates , mean , standard deviation ) are shown . Two-tailed unpaired t-tests comparing the mean percentage of IASD labelled Bak ( before versus after , before versus during , and during versus after tBid treatment ) are shown , and significant p-values ( p<0 . 05 ) are highlighted in green . Previously untested Bak residues have been pooled with data from previous studies ( # denotes data from Westphal et al . [2014] , ^ denotes data from Iyer et al . [2015] ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 00410 . 7554/eLife . 19944 . 005Figure 1—figure supplement 1 . Bak cysteine variants retain apoptotic function . Bak-/-Bax-/- MEF expressing Bak cysteine variants were treated with 10 μM etoposide for 24 hr and the percentage cell death quantified by propidium iodide uptake . Data are for unpublished mutants only and are the mean ± SD . The number of replicates ( n ) for each residue is labelled on the x-axis . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 00510 . 7554/eLife . 19944 . 006Figure 1—figure supplement 2 . Bak cysteine variants retain apoptotic function in response to tBid . Mitochondrial fractions from cells expressing the indicated Bak cysteine mutants were incubated with tBid , and samples fractionated to measure Bak-mediated release of cytochrome c from the pellet into the supernatant ( SN ) . This data shows that cell killing in response to etoposide ( Figure 1—figure supplement 1 ) correlates with cytochrome c release initiated by tBid . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 00610 . 7554/eLife . 19944 . 007Figure 1—figure supplement 3 . Quantitation of Bak IASD labelling before , during and after Bak activation; graphical output . ( A , B , C ) . Labelling before , during or after treatment with tBid , segregated by Bak N-extremity ( A ) , Core ( B ) , and C-extremity ( C ) . Previously untested Bak residues have been pooled with data from previous studies ( # denotes data from Westphal et al . [2014] , ^ denotes data from Iyer et al . [2015] ) . Data are mean ± SD , or range ( n = 2 ) , with n for each residue labelled on the x-axis . Residues for which there is a significant difference in IASD labelling before versus after tBid are in bold and underlined ( p<0 . 05 ) . ( D ) IASD labelling for selected Bak variants before , during and after treatment with tBidBax chimera ( Hockings et al . , 2015 ) . Data are mean ± SD , or range ( n = 2 ) , with n for each residue labelled on the x-axis . Residues for which there is a significant difference in IASD labelling before versus after tBid are in bold and underlined ( p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 007 In the N-terminus , most tested residues were accessible to IASD even before tBid treatment ( Figure 1B ) , consistent with the crystal structure ( 2IMT ) ( Moldoveanu et al . , 2006 ) . Those residues remained exposed following incubation with tBid . Four residues in α1 ( R36C , Y41C , Q44C and Q47C ) that were not fully accessible before tBid treatment became exposed following incubation with tBid ( Figure 1B ) . An additional α1 residue , V39C , also showed a tendency for increased labelling . These changes suggest that the α1-α2 loop separates from α1 , consistent with increased labelling of A54C in the α1-α2 loop that opposes α1 . Thus , the increased labelling of cysteine residues placed throughout the N-segment , α1 and the α1-α2 loop , together with exposure of several N-terminal antibody epitopes ( Alsop et al . , 2015 ) , indicate that the Bak N-terminus becomes completely solvent-exposed after activation ( Figure 1D ) . To complete our survey of the core and C-termini of Bak dimers ( Westphal et al . , 2014; Iyer et al . , 2015 ) , we tested the cysteine accessibility of additional selected residues in the α2-α5 core and the α7-α8 region ( Figure 1B ) . A summary of current and previous cysteine labelling data for the full length of the Bak protein is shown in Figure 1C and supports the following picture of Bak dimer topology ( Figure 1D ) . Certain cysteine residues placed in α2 became transiently exposed , consistent with exposure of the BH3 domain followed by its burial in the hydrophobic groove in the α2-α5 core dimer ( Figure 1C ) , as predicted by the published structure of the Bak BH3:groove homodimer ( 4U2V ) ( Brouwer et al . , 2014 ) . Cysteine substitutions on the exposed surface of α3 were successfully labelled with IASD before , during and after tBid treatment . In contrast , two buried residues in α4 were refractory to IASD labelling throughout . In the α4-α5 linker region , although I123C became more exposed following activation , it was still incompletely labelled . This pattern of I123C labelling may reflect initial loss of contact between I123C and two hydrophobic residues in the ‘latch’ ( I167 and W170 ) , followed by shallow membrane insertion of I123C , as it resides on the hydrophobic face of the Bak dimer core . Partial labelling of this residue is consistent with shallow membrane insertion as IASD can label cysteine moieties up to 7 . 5 Å into the hydrocarbon core of a lipid bilayer ( Gründling et al . , 2000; Westphal et al . , 2014 ) . Within the C-termini , the profile of IASD-labelling was consistent with the amphipathic α6 helix lying in-plane on the membrane surface ( Westphal et al . , 2014 ) and the hydrophobic α9 forming the sole transmembrane domain ( Figure 1D ) . In summary , Bak undergoes a series of conformational changes as it transitions from its non-activated monomeric state , to an activated BH3-exposed monomeric intermediate , to the symmetric dimer that is the building block of the Bak oligomer ( Figure 1D ) . In the course of these structural rearrangements , some regions of Bak display changes in solvent exposure ( such as α1 which becomes more exposed , or α2 which is transiently exposed ) whilst others remain buried throughout ( such as the α9 transmembrane anchor ) . Linkage studies have provided significant insight into structural changes in Bak , as well as how the activated proteins associate into higher order oligomers ( Dewson et al . , 2008 , 2009; Ma et al . , 2013; Aluvila et al . , 2014; Brouwer et al . , 2014; Iyer et al . , 2015; Mandal et al . , 2016 ) . To directly compare reported linkages , and to analyse the full length of Bak , we used our expanded library of single-cysteine Bak variants . Each substituted cysteine was tested for the ability to disulphide bond to the same residue in a neighbouring Bak molecule ( or to a cysteine in a nearby protein ) upon addition of the oxidant copper phenanthroline ( CuPhe ) . Linkage between Bak molecules was indicated by the presence of 2x complexes corresponding to twice the molecular weight of the Bak monomer on non-reducing SDS-PAGE . It is important to note that these 2x complexes are the product of CuPhe-mediated linkage between Bak molecules and do not necessarily represent the native Bak BH3:groove dimer that will form even in the absence of linkage , as shown below on BNP . This system of single cysteine substitutions offers an elegant screening approach for assessing the proximity of single cysteine residues in neighbouring molecules in a point-to-point manner , but does not identify interaction surfaces where a single cysteine substitution may not contact its counterpart ( See Discussion ) . Even in the absence of tBid treatment , cysteine residues in flexible regions displayed some linkage to neighbouring non-activated Bak molecules ( Figure 2 , upper panels ) . For example , some 2x complexes were captured by cysteine substitutions in the flexible N-segment ( G4C , C14 , L19C , S23C ) , and α1-α2 loop ( G51C , D57C , P58C , Q66C , S68C , S69C ) , indicating some proximity of Bak monomers in untreated mitochondria . Furthermore , disulphide linkage to proteins other than Bak resulted in Bak complexes larger than the 2x species and was observed for cysteine substitutions in the α2-α3 helices ( D84C , R87C ) , α4-α5 loop ( S121C , N124C ) , α8-α9 linker ( G186C , I188C ) and C-terminal end of α9 ( V205C ) . This is unsurprising as the MOM is a protein-rich environment and there are documented examples of monomeric Bak associating with other membrane proteins such as VDAC2 ( Cheng et al . , 2003; Lazarou et al . , 2010; Ma et al . , 2014 ) . 10 . 7554/eLife . 19944 . 008Figure 2 . The N- and C-extremities of oligomerised Bak are mobile relative to the α2-α5 core dimer . Mitochondrial fractions from cells expressing the indicated Bak cysteine mutants were incubated with tBid to oligomerise Bak , and oxidant ( CuPhe ) added to induce disulphide bonds . Aliquots were analyzed by non-reducing SDS PAGE ( upper panels ) and BNP ( lower panels ) , and immunoblotted for Bak to detect linked species . Data are representative of at least two biological replicates . ( A ) Cysteine linkage occurs throughout the N-extremity . ( B ) Cysteine linkage occurs , but is less complete , in the α2-α5 dimer core . ( C ) Cysteine linkage occurs in the C-extremity . ( See also α6:α6 linkage in Dewson et al . [2008] and α9:α9 linkage in Iyer et al . [2015] ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 00810 . 7554/eLife . 19944 . 009Figure 2—figure supplement 1 . Higher temperature enhances disulphide bonding of cysteine residues in α6 ( H164C ) and the α9 transmembrane domain ( L199C ) . Mitochondrial fractions from cells expressing the indicated Bak cysteine mutants were incubated with tBid to oligomerise Bak , and oxidant ( CuPhe ) added to induce disulphide bonds . Samples were incubated with CuPhe for 15 min either at 0°C or at 30°C . Aliquots were analyzed by non-reducing SDS PAGE ( upper ) and BNP ( lower ) , and immunoblotted for Bak to detect linked species . Note that linkage outside the membrane ( H164C:H164C' ) was efficient even at 0°C . Data are representative of two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 009 After addition of tBid had activated Bak , each cysteine in the N-extremity ( a region comprising the N-terminus , α1 and α1-α2 loop ) showed more efficient linkage to 2x complexes ( Figure 2A , upper panel ) . The freedom with which each cysteine in the N-extremity of Bak can link suggests this region is highly mobile and does not engage in stable protein-protein or protein-lipid interactions following activation . In contrast , cysteine residues positioned in the core ( α2-α5 ) showed relatively modest linkage to 2x complexes ( Figure 2B , upper panel ) , consistent with lack of flexibility in the α2-α5 core dimer ( Brouwer et al . , 2014 ) and its relative immobility due to association with the membrane . We next addressed linkage in the C-extremity encompassing the α6-α8 latch and α9 transmembrane domain ( Figure 2C , upper panel ) . After Bak activation , H164C in α6 showed increased linkage to 2x complexes , as reported previously for several residues in α6 ( Dewson et al . , 2009; Ma et al . , 2013 ) . Cysteine substitutions in the α8-α9 loop and the beginning of the α9 transmembrane domain ( up to residue N190C ) also showed linkage to 2x complexes ( Figure 2C , upper panel ) . However , compared to the N-extremity , linkage was less complete , possibly due to membrane association limiting mobility . Linkage of membrane-buried α9 residues was even less efficient ( Figure 2C ) . However , at a higher temperature α9 residues show much stronger linkage ( Figure 2—figure supplement 1 ) ( Iyer et al . , 2015 ) , attributable to increased Bak mobility or CuPhe penetration into the bilayer . In summary , after Bak activation , the whole N-extremity is highly dynamic due to being fully solvent-exposed; the core is constrained and membrane-associated; and the C-extremity is flexible , but is anchored to the membrane . We next used blue native PAGE ( BNP ) to consider two types of cysteine-mediated Bak linkage: linkage within Bak dimers and linkage between Bak dimers . The Bak BH3:groove interaction within dimers is an extensive protein-protein interface and is stable in 1% digitonin on BNP while the interaction between dimers is not ( Ma et al . , 2013 ) . Thus oligomerised Bak migrates only as BH3:groove dimers on BNP . Addition of CuPhe to mitochondrial extracts prior to BNP can stabilise high order complexes on BNP and does so by cysteine-mediated linkage between dimers ( Ma et al . , 2013 ) . Thus , if the addition of CuPhe at the end of the mitochondrial incubation generated 4x and greater complexes on BNP , we conclude that the linkage was between dimers . If larger complexes were absent and dimers predominated on BNP , any linkage was within the dimer . Thus , parallel examination of CuPhe linked Bak species on BNP and non-reducing SDS PAGE permits differentiation of three conditions: the linkage within the dimer , linkage between dimers , or no cysteine linkage ( Figure 3A ) . 10 . 7554/eLife . 19944 . 010Figure 3 . Linkage constraints for different regions of the active Bak dimer support a flexible extremity model for full-length Bak dimers in the MOM . ( A ) The oxidant CuPhe induces linkage between Bak monomers , and within or between Bak dimers . Correlation of BNP with non-reducing SDS PAGE can differentiate linkage within and between dimers . ( B ) Summary of linkage outcomes within or between dimers for residues in the Bak dimer N-extremity . ( C ) Linkage at the lateral corners of the Bak α2-α5 core dimer is highlighted with residue labels and sticks on the crystal structure ( 4U2V ) [Brouwer et al . , 2014] ) . ( D ) Schematic of a dimer of activated Bak at the membrane in side and top views . The range of movement of the N- and C-extremities is indicated with coloured ovals ( blue and pink respectively ) . ( E ) Our linkage data are consistent with the dimensions of the Bak core dimer . The dimensions of the Bak α2-α5 core dimer ( 4U2V ( Brouwer et al . , 2014 ) are shown in side view and top view . In the crystal structure , the distance from one edge of the core at S68 over the bended structure of the symmetric core dimer to the opposing S68’ is ~75 Å . Thus , for residues in the mobile N-extremity to link within the core dimer , the two homotypic cysteine residues must bridge a distance of ~75 Å . This is not feasible for the residue V61C which is ~30 Å from the symmetric core . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 01010 . 7554/eLife . 19944 . 011Figure 3—figure supplement 1 . Bak V61C linkage is generalisable to a heat stimulus and to the linkage of the equivalent Bax A46C residue . ( A ) Mitochondrial fractions from cells expressing the indicated cysteine variants were incubated with tBid ( or heated to 43°C ) to oligomerise Bak , or Bax , and oxidant ( CuPhe ) added to induce disulphide bonds where indicated . Aliquots were analyzed by BNP , non-reducing SDS PAGE , and immunoblotted for Bak or Bax . Variants lacking cysteines ( Cys null ) were included to show the altered migration of Bak and Bax complexes on BNP in the absence of disulphide bond formation . Data are representative of at least two biological replicates . Note that in the left panels , Bak V61C exhibits the same CuPhe linkage pattern with both tBid ( lane 4 ) and heat treatment ( lane 5 ) . In the right panels , linkage between A46C in a mitochondrial form of Bax efficiently links between dimers . Bax S184L/A46C is a Bax variant that is constitutively membrane localised ( due to the hydrophobic substitution of S184L in the transmembrane domain [Nechushtan et al . , 1999; Fletcher et al . , 2008] ) with a N-terminal cysteine substitution ( A46C ) analogous to Bak V61C . Bax exhibits laddering on BNP in the absence of CuPhe treatment , but addition of CuPhe induces efficient linkage between Bax dimers at residue A46C , resulting in the disappearance of dimers and appearance of high molecular weight complexes on BNP ( lane 9 ) . Thus , we predict Bax dimers exhibit similar flexible extremity characteristics as Bak dimers at the MOM . ( B ) Pellet and supernatant fractions were further analysed for cytochrome c release , and confirmed each variant was responsive to tBid ( and heat treatment in the case of Bak V61C ) . Data are representative of at least two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 011 In the N-extremity ( G4C to S69C ) , each introduced cysteine could link some dimers to larger complexes on BNP ( Figure 2A , lower panel ) . Upon tBid activation , SDS-PAGE ( Figure 2A , upper panel ) showed an increased proportion of 2x complexes , and higher order complexes ( >10x ) were observed on BNP for all residues . Of note was a stretch of residues from approximately E59 to T62 that , following tBid treatment , exhibited very efficient linkage by SDS-PAGE ( obvious 2x , negligible 1x ) . In this stretch , BNP showed no dimer species and a concomitant shift to 4x species and higher order complexes , indicating efficient linkage between dimers but no linkage within dimers . Residues closer to the α2-α5 core ( Q66C to S69C ) , showed progressively more 1x species ( SDS-PAGE ) and an increase in dimers ( BNP ) , indicative of becoming closer to the constrained core of the dimer . Together , these linkage data further argue that dimers possess flexible N-extremities capable of efficient linkage within and between dimers , with linkage becoming exclusively between dimers as residues approach the constrained α2-α5 core ( Figure 3B ) . Interestingly , because the linkage of cysteine residues in the E59-T62 region occurs only between dimers , and is very efficient , this linkage provides a molecular sensor that can specifically quantitate dimer-dimer interactions , and is the first assay to do so . The V61C:V61C' molecular sensor for dimer-dimer interaction is generalisable to other stimuli and to the Bax pore ( Figure 3—figure supplement 1 ) . For example , the same V61C:V61C' crosslinking pattern for Bak is induced by either tBid or heat treatment , arguing that the linkage pattern between dimers is a general characteristic of the Bak apoptotic pore . We also generated the V61C equivalent in membrane localised Bax ( Bax A46C/S184L ) ( Figure 3—figure supplement 1 ) . This Bax variant could be linked between ( but not within ) dimers , suggesting that Bax dimers also have flexible N-terminal extremities . In the Bak C-terminus also , cysteines could link dimers to larger complexes on BNP , although a striking laddering pattern was apparent ( Figure 2C , lower panels ) . Absence of the large ( >10x ) complexes common at the N-extremity is explained by the relatively restricted range of motion of the membrane-associated C-extremity . The very weak laddering pattern for certain residues in the α6-α8 region is explained by those residues facing either the membrane or the cytosol rather than laterally to a nearby Bak dimer . Lastly , linkage in the C-extremity may be only between dimers ( rather than within dimers ) as the proportion of 2x to 1x species on SDS-PAGE is similar to the proportion of higher order complexes to dimers on BNP . Previous studies suggested that the structured α2-α5 core of the Bak ( or Bax ) dimer might form a dominant , symmetric interface capable of driving dimers to form higher order complexes ( Aluvila et al . , 2014; Brouwer et al . , 2014 ) . For example , crystals of the Bak α2-α5 core dimer showed a side-by-side association ( 4U2V ) ( Brouwer et al . , 2014 ) . However , this was attributed to crystal packing effects , as a chemical crosslinking between the α4-α5 loops was not detected ( Brouwer et al . , 2014 ) . Accordingly , in the present studies , disulphide bonding between the α4-α5 loops was not captured with CuPhe ( Figure 2B , lower panel ) . End-to-end associations were suggested by α3:α3’ or α5:α5’ linkages ( Aluvila et al . , 2014; Brouwer et al . , 2014; Mandal et al . , 2016 ) , and disulphide bonding at the equivalent residues was also evident in this study ( H99C:H99C for α3:α3’; H145C:H145C’ for α5:α5’ ) ( Figure 2B , lower panel ) . Notably , this linkage was not as efficient as linkage observed for the N-extremity , and these and other linkage-competent residues locate mostly to the lateral edges of the α2-α5 core dimer when lying in-plane ( Figure 3C ) . Thus , screening individual cysteine substitutions in the Bak dimer core for linkage did not uncover any dominant sites of interaction between Bak dimers , but showed widespread linkage of residues around the lateral edges of the dimer , consistent with the random collision of Bak dimers . These findings support a flexible extremity model for full-length Bak dimers in the MOM , outlined in Figure 3D with side-on ( left ) and top-down views ( right ) . At the centre is the α2-α5 core dimer lying in-plane and partially submerged in the membrane surface . Extending from both sides of the core dimer is a flexible latch ( α6-α8 ) of ~40 residues that also lies in-plane and connects to the transmembrane α9-helices . Extending from both ends of the dimer core is the solvent-exposed , flexible and mobile N-extremity of ~70 residues . Each circle in Figure 3D illustrates the range of possible positions for the flexible N- and C-extremities ( blue and pink respectively ) . Overlap of the largest blue circles ( labelled G4C ) indicates the potential for residues close to the N-terminus to link within dimers . In contrast , the smaller blue circles ( labelled V61C ) do not overlap , indicating a lack of linkage within the dimer at this residue . Notably , these linkage constraints are consistent with the dimensions of the α2-α5 dimer core as determined from the recent crystal structure ( 4U2V ) ( Figure 3E ) . These findings also reveal how dimers associate into high order oligomers . As linkage can be induced between several regions of the Bak dimer , there is no dominant protein-protein interface that mediates dimer assembly into high order oligomers . Rather , the linkage pattern is consistent with the transient collision of dimers in-plane on the membrane . Collisions of the α2-α5 core and α6-α8 latch are limited by their membrane-association , whereas the flexible , entirely solvent-exposed N-terminal region is free to link between dimers and does so very efficiently . We next examined if the inter-dimer interactions identified above were stable in digitonin , as shown for the BH3:groove interface within dimers ( Ma et al . , 2013 ) . Digitonin was added to the mitochondrial incubations after Bak had become oligomerised by incubation with tBid , but prior to disulphide bond formation induced by CuPhe ( Figure 4A ) . As expected , digitonin did not prevent linkage within dimers as shown by 2x complexes of M71C/K113C on SDS-PAGE ( Figure 4A , upper panel , lane 3 ) . However , digitonin prevented all linkage between dimers on BNP of seven single-cysteine variants ( Figure 4A , lower panels ) , indicating that inter-dimer interactions are mediated either by membrane or by weak protein-protein interactions . 10 . 7554/eLife . 19944 . 012Figure 4 . Bak dimer-dimer interactions are disrupted by detergent or Bax . ( A ) Digitonin prevents linkage between Bak dimers . Membrane fractions expressing the indicated Bak single-cysteine variants were first treated with tBid , and then supplemented , as indicated , with detergent ( 1% digitonin ) prior to cysteine linkage . Two additional mutants were included to show linkage within dimers at the BH3:groove interface ( M71C/K113C ) and at extensions to the C-terminus ( GGSGGCK ) . Data are representative of two biological replicates . ( B ) Bax can disrupt Bak dimer-dimer association . Membrane fractions first treated with tBidM97A were then incubated , as indicated , with Bax , Bcl-xL , Mcl-1ΔN151ΔC23 or digitonin prior to cysteine linkage ( upper panels ) . Supernatant and pellet fractions showed partial localisation of each recombinant protein to the membrane fraction ( lower panels ) . Data are representative of three biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 012 These experiments also provided new insight into linkage within dimers . For example , two residues at the far N- and C-termini ( L19C and the C-terminal extension GGSGGCK [Iyer et al . , 2015] ) could link within dimers to generate 2x complexes on SDS-PAGE after digitonin treatment ( Figure 4A , upper panels ) . Thus , these residues were sufficiently distal from the core to reach their counterpart across the length of the structured α2-α5 core dimer and allow linkage within the dimer ( Figure 3D and E ) . On BNP , dimers of these two residues ( L19C , and to a lesser extent GGSGGCK ) also migrated faster than dimers of other variants ( Figure 4A , lower panels ) , suggesting that linkage within the dimer caused a more compact , faster migrating protein complex under these native conditions . To test if interactions between Bak dimers could be disrupted by other proteins , and thus test if Bak dimer aggregation was a dynamic , reversible process , we added recombinant mitochondrial proteins after Bak had formed dimers ( Figure 4B ) . Dimer interactions were monitored by linkage between the N-termini ( V61C:V61C' ) or C-termini ( H164C:H164C' ) , and compared to BH3:groove linkage ( M71C:K113C ) . To oligomerise Bak , mitochondria were first incubated with tBidM97A , a variant that binds poorly to prosurvival proteins such as Bcl-xL and Mcl-1 ( Lee et al . , 2016 ) . Mitochondria were then incubated with the recombinant Bcl-2 proteins Bax , Bcl-xL or Mcl-1ΔNΔC . Finally , CuPhe was applied and the extent of disulphide linkage assessed by non-reducing SDS-PAGE ( Figure 4B ) . Notably , Bax localised to mitochondria and decreased linkage between Bak dimers ( e . g . V61C:V61C’ and of H164C:H164C' ) but not within dimers ( e . g . M71C:K113C ) ( Figure 4B ) . In contrast , recombinant Bcl-xL and Mcl-1ΔNΔC localised to the mitochondrial membrane but did not interfere with the linkage between Bak dimers ( Figure 4B ) . Thus , Bax not only localised to the same membrane microdomain as the pre-formed Bak oligomers , but became partially interspersed with the dimers , indicating that dimer aggregation is dynamic . Our biochemical data support the random collision of Bak dimers at the mitochondrial outer membrane during apoptosis . Indeed , several studies indicate that Bak and Bax can reside in a variety of different structures at the mitochondrial outer membrane of apoptotic cells , including clusters ( Nechushtan et al . , 2001; Zhou and Chang , 2008; Große et al . , 2016; Nasu et al . , 2016; Salvador-Gallego et al . , 2016 ) . To test if random collisions between dimers could feasibly explain the linkage patterns observed , simulations of the random contact between dimers were performed . The efficient and exclusive linkage that occurs between dimers at residue V61C formed the basis of our simulations ( Figure 5 ) . 10 . 7554/eLife . 19944 . 013Figure 5 . V61C:V61C' linkage is a marker of dimer-dimer interaction . ( A ) BNP analysis of V61C:V61C' linkage revealed no dimers , and a high proportion of 4x and >10x species . In contrast , H164C:H164C’ linkage revealed a ladder of linked species . Data are representative of nine biological replicates . ( B ) Densitometry of V61C:V61C' BNP linkage in Bak oligomers shows negligible dimers but reproducible linked species at 4x , 6x , 8x , and 10x , and a large population of species > 10x . Densitometry data ( n = 9 ) were normalised to the area under the curve after alignment as described in Materials and Methods . To correct for small variations in electrophoretic migration , we employed a noise reduction algorithm in which the 4x peak ( 0 ) and a minima at ~720 kDa ( 1 ) were aligned between replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 013 To simulate the aggregation of Bak dimers on the mitochondrial outer membrane , this complex biological system was simplified to limit the number of parameters required to mimic the experimental outcome . We simulated the membrane as a flat surface ( i . e . a 2D plane ) . Each subunit representing a Bak dimer , was evenly distributed on a geometric grid imposed on this surface ( Figure 6A ) , and to avoid edge effects the grid was wrapped onto a torus . A hexagonal grid , rather than square or triangular , was selected to best represent the close packing of objects on a plane with the greatest degrees of freedom . The linkage potential of each subunit in the grid was based on the known linkage constraints of the V61C residue; i . e . linkage occurred exclusively and efficiently between dimers and each dimer could make at most two cysteine linkages , but possibly only 1 or 0 linkages if no other free cysteines were available . Furthermore , the simulation allowed linkages only between immediately adjacent dimers , based on the short 4 Å limit of the CuPhe-mediated disulphide linkage and the known dimensions of the Bak dimer . Thus , each hexagonal subunit had the freedom to link with at most two neighbouring hexagons , in any direction , and this random linkage between neighbouring dimers would result in a mixture of linkage states including double , single or no linkages . With this hexagonal grid , four types of double linkage were possible; opposite , obtuse , acute or reciprocal ( Figure 6B ) . An equal probability for each of the double linkages ( opposite , obtuse , acute or reciprocal ) was imposed , to reflect close but random packing of dimers at the membrane . Furthermore , the simulation was allowed to proceed to complete linkage of available cysteine residues to reflect the near-complete linkage of V61C:V61C’ . With these few parameters , a complex biological system was elegantly reduced to a simplified simulation . 10 . 7554/eLife . 19944 . 014Figure 6 . Bak dimer arrangement examined by two-dimensional stochastic simulations: a random arrangement successfully models the V61C:V61C’ BNP linkage densitometry . ( A ) A grid of 100 × 100 was designated for the 2D simulations . Each unit hexagon within that grid represented a Bak dimer , with the capacity to link to two neighbouring hexagons . The direction of linkage from each hexagon was randomised . An example visual output from a 100 × 100 hexagonal array is shown: neighbouring subsets of linked Bak dimers within the grid are delineated by different colours . ( B ) Linkage possibilities between each dimer unit in a hexagonal 2D array . ( C , D ) Overlay ( C ) , and average ( D ) of 30 predicted densitometry plots . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 01410 . 7554/eLife . 19944 . 015Figure 6—figure supplement 1 . Adjustment to simulation output to allow qualitative comparison to western blot data . Mathematical simulations of Bak dimer linkage generated frequency distributions of the size of Bak complexes ( multiples of Bak dimers ) . These frequency distributions were then transformed to a predicted densitometry by multiplying each count by the number of Bak dimer units present in each linked Bak complex . For instance , the count representing 2x complexes , is multiplied by 1 , the count representing 4x complexes is multiplied by 2 , the count representing 6x complexes is multiplied by 3 , etc . This predicted densitometry approximates the total relative abundance of Bak molecules in each linked Bak complex , which was then transformed with an exponential horizontal axis to mimic the nonlinear spacing of bands , and a kernel estimator to approximate the smearing of bands ( especially at higher molecular weights ) to allow qualitative comparison to western blot densitometry outputs . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 01510 . 7554/eLife . 19944 . 016Figure 6—figure supplement 2 . Comparison of 2D simulation using a square or hexagonal array reveals that either geometry offers a reasonable fit for the empirical data . ( A ) Linkage possibilities are slightly greater in a hexagonal than in a square 2D array . ( B ) A slight reduction in the size of complexes with the square simulation is consistent with the reduced degrees of freedom using the square geometry compared to the hexagonal geometry . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 01610 . 7554/eLife . 19944 . 017Figure 6—figure supplement 3 . 2D simulation with reduced efficiency linkage successfully models reduced linkage in mitochondrial experiments . ( A ) Co-expression of Bak V61C with Bak Cys null resulted in less high order linkage species on BNP . Membrane fractions expressing Bak V61C , FLAG-Cys null , or both , were incubated with or without tBid , and then subjected to CuPhe linkage and analysis by BNP , non-reducing SDS PAGE or reducing SDS PAGE . Data are representative of three biological replicates . Note that the greatly reduced V61C linkage in the presence of Bak Cys null also indicates that most linked species in the V61C cells involve Bak and not some other protein . ( B ) Densitometry of BNP linkage data following tBid treatment for 3 replicates of the V61C:Flag-Cys null co-expression . Data are normalised to the area under the curve . ( C ) Diagram showing the four types of dimers that would form upon co-expression of V61C and Cys null variants of Bak ( i . e . a system in which 50% of Bak molecules are unable to link ) . Note that dimers will be capable of 0 , 1 or 2 linkages . ( D ) The 2D simulation was altered to incorporate 50% linkage-incompetent Bak molecules . This simulation yielded a good concordance with the V61C:Flag Cys null densitometry . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 01710 . 7554/eLife . 19944 . 018Figure 6—figure supplement 4 . 2D simulation with edge blocking probability successfully models mitochondrial experiments in which Bak linkage becomes constrained as cysteines are positioned closer to the dimer core . ( A ) Residues between V61C and the core of the Bak dimer have a progressively more limited range of movement . On the left , the blue circles are shown on a schematic of the Bak dimer to represent the range of movement for G4C , V61C , T62C , Q66C , S68C and S69C . On the right , an illustration of the reduced overlap of Bak dimer extremities ( blue circles ) is shown for residues with progressively smaller ranges of movement ( e . g . V61C>Q66C>S69C ) . ( B ) Densitometry of BNP outputs from V61C , T62C , Q66C , S68C and S69C highlight the reduced linkage efficiency ( relative to V61C ) as cysteine substitutions approach the constrained core of the Bak dimer . Data are shown for three or nine biological replicates . ( C ) The 2D simulation was altered to incorporate an ‘edge blocking probability’ to mimic the increasing linkage constraint experienced by cysteine substitutions as they approached the core . The best approximation of each cysteine substitution mutant is indicated by arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 018 The simulation output was a frequency distribution of the number of Bak dimers present in each linked complex ( Figure 6—figure supplement 1 ) . To approximate the BNP western blot densitometry , the counts were then multiplied by the number of Bak dimer units ( e . g . 1 , 2 , 3 , 4 etc ) in each of the linked Bak complexes , and the predicted densitometry smoothed with a Gaussian kernel to allow qualitative comparison with the Gaussian-like densitometry output from the BNP western blots . Notably , the simulation produced outputs that closely matched the distinctive linkage pattern observed for V61C , i . e . prominent 4x complexes and complexes greater than 10x the Bak molecular weight ( compare Figure 5A and B with Figure 6C and D ) . ( When the simulation was repeated on a square rather than hexagonal grid , the output also closely matched the V61C linkage pattern , as shown in Figure 6—figure supplement 2 ) . Thus , the simulations provided proof of principle that the linkage pattern ( V61C:V61C' ) observed in mitochondria can be explained by random dimer arrangement . These logic arguments based on the linkage constraints of the V61C residue have , for the first time , afforded single molecule resolution of Bak aggregation . To explore the robustness of the simulations , we asked if the simulation could predict the linkage pattern when linkage was limited due either to less V61C residues , or to cysteine residues at positions other than V61C . In the first test , we performed mitochondrial experiments in which ~50% of the Bak molecules lacked cysteine residues . In those experiments , Bak V61C was co-expressed with Bak Cys-null that was FLAG-tagged ( Figure 6—figure supplement 3A , upper panel ) . As expected , only half of the Bak molecules could link to 2x species after activation by tBid and induction of linkage ( Figure 6—figure supplement 3A , middle panel , lane 4 ) . In addition , all molecules formed BH3:groove dimers after incubation with tBid , as shown by BNP , but the capture of higher order oligomers mediated by cysteine-linkage was greatly reduced ( Figure 6—figure supplement 3A , lower panel , lane 4; quantified in Figure 6—figure supplement 3B ) . In simulations of this system , we assumed an equal expression of V61C and the Cys null variants , as indicated by western blot analysis . We also assumed that this population of monomers forms dimers in Mendelian proportions: 25% of dimers are V61C doublets and can link twice , 50% are V61C:Cys null and can link once , and the remaining 25% are Cys null doublets and cannot link ( Figure 6—figure supplement 3C ) . The simulation output ( Figure 6—figure supplement 3D ) reflected that of the BNP densitometry ( Figure 6—figure supplement 3B ) in terms of more dimers and fewer >10x complexes , indicating the simulation was generalisable to a different distribution of linkage-competent Bak molecules . We next examined if our simulations were also robust when approximating the linkage pattern of residues other than V61C ( Figure 6—figure supplement 4A ) . As noted above , because T62C , Q66C , S68C , and S69C substitutions were closer than V61C to the constrained α2-α5 core , they did not link efficiently ( Figure 2A lower , quantified in Figure 6—figure supplement 4B ) . To simulate this steric hindrance caused by a shorter distance from the constrained core , an edge blocking effect was introduced to the model; every edge of every hexagon was prohibited from having a link form across it with this probability for the duration of each simulation . As the edge blocking probability increased from 0 to 0 . 75 , linkage between dimers decreased to generate patterns comparable to BNP analysis of these core-proximal residues ( Figure 6—figure supplement 4C ) . Thus , modification of a single parameter could describe linkage of residues closer than V61C to the Bak dimer core . The adaptability of this simple simulation further illustrated the strength and feasibility of our model of random Bak dimer aggregation at the mitochondrial outer membrane during apoptosis .
Here we investigated the topology of Bak dimers in the mitochondrial outer membrane , and how dimers assemble into the high order oligomers thought necessary to form apoptotic pores in that membrane . We found that Bak dimers are characterised by flexible N- and C-extremities flanking a rigid α2-α5 core , and that these dimers aggregate in disordered , dynamic , clusters . Critically , we found no evidence for a single , dominant protein-protein interface between dimers and predict the lipid environment plays a crucial role in facilitating the aggregation of dimers and subsequent membrane rupture . We had previously proposed the in-plane model for Bak dimers ( Westphal et al . , 2014 ) . Our current data advances this model to show the N- and C-extremities are flexible , based on analysis of the full-length of Bak when activated and dimerised at the mitochondrial outer membrane ( Figure 3D ) . The N-extremity ( ~70 residues N-terminal to α2 ) becomes fully solvent-exposed as shown by IASD labelling , and fully mobile as indicated by linkage , consistent with exposure of N-terminal cleavage sites and antibody epitopes in α1 and the α1-α2 loop ( Griffiths et al . , 1999; Weber et al . , 2013; Alsop et al . , 2015 ) . As discussed previously ( Alsop et al . , 2015 ) , several parts of the α1 region are able to unfold , as they bind antibodies that recognise linear epitopes . Failure of the N-extremity to re-engage in protein-protein or protein-lipid interactions argues that it does not contribute to the assembly of dimers . The remainder of Bak is membrane-associated , with the α2-α5 core and α6 ( and possibly α7-α8 ) in-plane with the outer mitochondrial membrane surface and partially embedded ( Aluvila et al . , 2014; Brouwer et al . , 2014; Westphal et al . , 2014 ) while α9 forms a transmembrane domain ( Iyer et al . , 2015 ) . The topology of the α2-α5 region is consistent with the α2-α5 core dimer crystal structure ( ~35 Å × 45 Å ) ( Brouwer et al . , 2014 ) , as V61C ( ~30 Å from the core dimer ) linked between dimers but not within dimers . Flexibility of the C-terminal α6-α9 region , as indicated by multiple linkages between these regions ( Bleicken et al . , 2014; Iyer et al . , 2015; Zhang et al . , 2016 ) , implies that embedded dimers adopt a range of conformations on the MOM surface ( Figure 7A ) . 10 . 7554/eLife . 19944 . 019Figure 7 . Bak dimers adopt various conformations on the membrane surface and aggregate in compact , disordered clusters to disrupt the mitochondrial outer membrane . ( A ) The top view of Bak dimers lying in-plane on the mitochondrial surface . The α2-α5 core is bounded by the large oval . Extending from the core are the membrane anchored C-terminal helices α6 , α7 and α8 . For simplicity , the α9 transmembrane domains that project into the membrane plane , and the flexible , solvent exposed N-termini of each dimer are not shown . ( B ) We hypothesise that growing clusters of Bak dimers induce membrane tension to rupture mitochondria . ( i ) Upon activation , Bak dimers penetrate the outer leaflet of the membrane and accumulate in a compact irregular cluster . ( ii ) More dimers converge on the cluster thus enlarging the patch of membrane disturbance . ( iii ) Once the patch attains a critical area a non-lamellar lipidic arrangement is generated , relieving membrane tension . ( iv ) Lipids and Bak dimers rearrange to bury exposed hydrophobic surfaces , yielding a variety of proteolipid ( toroidal ) ‘pores’ . Our assays survey a mixture of Bak linkage products derived from stages ( ii ) , ( iii ) and ( iv ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19944 . 019 Linkage experiments have identified several possible interactions between dimers of Bak or Bax ( Dewson et al . , 2008 , 2009; Zhang et al . , 2010; Pang et al . , 2012; Ma et al . , 2013; Aluvila et al . , 2014; Bleicken et al . , 2014; Gahl et al . , 2014; Iyer et al . , 2015; Mandal et al . , 2016; Zhang et al . , 2016 ) , with many also shown in the present study . Notably , however , a dominant protein-protein interface between dimers was not evident . Rather , linkage between the membrane-associated regions could be detected at many positions , arguing for a disordered arrangement . As our single-cysteine scanning approach can only examine the homotypic pairing of cysteine residues , it remains possible that heterotypic pairing of cysteine substitutions may detect a more complex protein interface between dimers , as seen for linkage at the Bak BH3:groove interface ( M71C:K113C , Figure 4B ) ( Dewson et al . , 2008 ) . However , in support of the notion that there is no dominant protein interface between dimers , the interaction between Bak dimers is lost when digitonin is added , in contrast to the BH3:groove interface which persists in the presence of digitonin . Moreover , the lability of dimer-dimer association was highlighted by the capacity of Bax to intermingle with pre-formed Bak dimers . Each of BclxL , Mcl-1 and Bax localised to mitochondria with similar efficiency , however a critical feature of Bax is that , like Bak , following activation by tBid it collapses onto the membrane surface to form homodimers ( Westphal et al . , 2014 ) . The large membrane surface occupied by Bax dimers , together with disturbance of the outer leaflet , may explain why Bax , but not the prosurvival proteins , could disrupt Bak dimer-dimer association in this assay . A variety of biophysical and imaging techniques have been deployed to better understand how Bak ( or Bax ) dimers associate , yet it remains unclear . Well-characterised pore-forming proteins typically reveal an extensive interface between subunits that facilitates pore formation ( Song et al . , 1996; Mueller et al . , 2009 ) . Versions of this regular architecture have recently been considered for Bak and Bax as a closed circuit of dimers lining a circular pore ( Aluvila et al . , 2014; Bleicken et al . , 2014; Brouwer et al . , 2014 ) . In contrast to these hypothetical linear dimer arrangements , our biochemical data revealed that there is no single dominant interface between the α2-α5 core dimers and support the random collision of Bak dimers in two dimensions at the mitochondrial outer membrane during apoptosis . Here a novel approach of mathematical modelling was used to simulate linkage between Bak dimers and showed the feasibility of random dimer association . For example , as the V61C residue linked only between Bak dimers , and did so very efficiently , this provided a powerful tool for testing different arrangements of Bak dimers on the membrane . A minimalist set of assumptions were adopted for stochastic simulations of V61C:V61C’ linkage between Bak dimers: linkage is complete and irreversible but occurs only between dimers; only two linkages are possible per dimer; and neighbouring dimers can only link if both have a free cysteine available for disulphide bonding . Indeed , the simulation showed good concordance with different sets of linkage data obtained from fully oligomerised Bak in mitochondria . From this , we conclude that Bak dimers are closely packed and aggregate with a random orientation . While our data indicate that most Bak ( and probably Bax ) oligomers are disordered clusters , it remains possible that a sub-population of ordered oligomers are responsible for driving pore formation . However , in our model system , preventing ~50% Bak from locating to mitochondria prevented pore formation ( Ferrer et al . , 2012 ) argues that , of the whole population of Bak surveyed in our linkage studies , at least 50% is required for a pore to form somewhere in the mitochondrion . Thus , the patterns observed in our linkage assays are at least 50% representative of Bak engaged in driving pore formation . A disordered cluster of dimers is reminiscent of the ‘carpet’ model of pore formation by peptides ( reviewed by Gilbert [2016] ) . In this model , amphipathic antimicrobial peptides such as melittin lie parallel to the membrane plane , with accumulation causing strain in the outer layer of the membrane until at higher concentrations the lamellar structure of the membrane is destabilised and non-lamellar lipidic pores form ( Lee et al . , 2008 , 2013 ) . Like an amphipathic peptide , the Bak and Bax α2-α5 core dimers and α6-helices penetrate the outer leaflet of the bilayer ( Czabotar et al . , 2013; Aluvila et al . , 2014; Brouwer et al . , 2014; Westphal et al . , 2014 ) . While very high concentrations of antimicrobial peptides can disintegrate membranes in a detergent-like manner ( Lee et al . , 2008 ) , this does not appear to occur with near-physiological levels of Bak and Bax . Antimicrobial peptides also have membrane thinning attributes ( Chen et al . , 2003 ) , as reported for Bax ( Satsoura et al . , 2012 ) . Satsoura and colleagues discuss how membrane deformation by Bax may promote oligomerisation mediated not by protein:protein interactions but by long-range changes in membrane tension ( Satsoura et al . , 2012 ) . For example , the attractive forces between membrane wrapped particles can arise purely as a result of membrane forces ( Reynwar et al . , 2007; van der Wel et al . , 2016; Katira et al . , 2016 ) . Thus , as illustrated in Figure 7B , we hypothesise that Bak dimers penetrate the outer leaflet to attract further dimers and in doing so increase the membrane disturbance . Enlarging clusters then destabilise the membrane sufficiently to form lipidic pores that release apoptogenic factors . The pores may then be stabilised by parts of the Bak and Bax dimers rearranging to line the pore ( Terrones et al . , 2004; García-Sáez , 2012; Westphal et al . , 2014; Kuwana et al . , 2016; Mandal et al . , 2016 ) , consistent with the lipidic pore model first proposed for Bax ( Basañez et al . , 1999 ) . Hence , our model for the disordered clustering of Bak dimers provides a holistic molecular explanation for the detection of Bak and Bax in a variety of formations including complete rings , arcs , lines and clusters ( Große et al . , 2016; Nasu et al . , 2016; Salvador-Gallego et al . , 2016 ) . In conclusion , this study proposes a novel means of oligomerisation by Bak and Bax in apoptotic cells: that dimers aggregate in dense clusters without a dominant interface and this dynamic association requires the involvement of the lipid membrane . We have described molecular tools that can precisely monitor the aggregation of Bak dimers in cells , and could also be used to examine how the pro-survival proteins inhibit Bak oligomerisation , which may impact on the development of cancer therapies that target these proteins . Critically , the combination of linkage data and mathematical simulations described here has offered , for the first time , insight into the assembly of the apoptotic pore with single molecule resolution .
Recombinant proteins were prepared as previously described; caspase 8-cleaved human Bid ( tBid ) ( as per [Kluck et al . , 1999] ) , tBidBax and Bcl-xL full length ( as per [Hockings et al . , 2015] ) , Cys null Bax full length ( as per [Czabotar et al . , 2013] ) , Mcl-1ΔN151ΔC23 ( as per ( Chen et al . , 2005 ) . tBidM97A was kindly provided by E . Lee ( Lee et al . , 2016 ) . A library of human Bak ( or Bax ) mutants was generated by site-directed cysteine substitution ( as per ( Dewson et al . , 2008 ) . Human Bak ( or Bax ) with the endogenous cysteine residues mutated to serine ( Bak Cys null C14S/C166S , or Bax Cys null C62S/C126S S184L ) was used as the template in overlap extension PCR to introduce single cysteine residues throughout the Bak or Bax sequence ( primer sequences listed in Supplementary file 1 ) , and then cloned into the pMX-IRES-GFP retroviral vector ( primer and vector sequences available on request ) . Mutants were introduced into SV40-immortalised Bak−/− Bax−/− mouse embryonic fibroblasts ( MEFs ) by retroviral infection ( as per ( Dewson et al . , 2008 ) . Bak−/− Bax−/− MEFs have been described previously ( Wei et al . , 2001 ) . Cell lines were checked for mycoplasma contamination . Bak−/− Bax−/− MEFs expressing Bak cysteine mutants at levels comparable to endogenous mouse Bak ( Alsop et al . , 2015 ) were permeabilised and membrane fractions containing the mitochondria were isolated as previously described ( Dewson et al . , 2008 ) . Cells were washed in PBS and then re-suspended at a concentration of 1 × 107 cells/ml in wash buffer supplemented with digitonin ( 100 mM sucrose , 20 mM HEPES-NaOH pH 7 . 5 , 100 mM KCl , 2 . 5 mM MgCl2 , 4 μg/ml Pepstatin A ( Sigma-Aldrich , St . Louis , MO , USA ) , Complete protease inhibitors ( Roche , Castle Hill , NSW , Australia ) and 0 . 025% w/v digitonin ( Calbiochem , Merck , Darmstadt , Germany ) . Cells were incubated for 10 min on ice , then membrane fractions were collected by centrifugation at 16 , 000 ×g for 5 min and the supernatant discarded . Pellets were re-suspended in wash buffer ( no digitonin ) , and permeabilisation of the cell membrane confirmed by trypan blue uptake . To assess the apoptotic function of each Bak variant in cells , Bak−/−Bax−/− MEFs expressing Bak cysteine mutants were treated with etoposide ( 10 μM ) for 24 hr , and cell death , as indicated by propidium iodide ( 5 μg/ml ) uptake , determined by flow cytometry ( FACSCalibur , BD Biosciences , San Jose , CA , USA ) . To activate Bak and induce mitochondrial outer membrane permeabilisation in vitro , membrane fractions from Bak−/−Bax−/− MEFs expressing Bak cysteine variants were treated with 100 nM recombinant tBid for 30 min at 30°C ( as described in Dewson et al . , 2008 ) . Selected samples were treated with the functionally equivalent protein tBidBax in which the BH3 domain of Bid was substituted with the Bax BH3 domain ( Hockings et al . , 2015 ) . An alternate stimulus was also performed in which membrane fractions were heated at 43°C for 30 min . Release of cytochrome c was assessed by centrifugation 16 , 000 ×g for 5 min to separate pellet and supernatant fractions and samples analysed by reducing SDS-PAGE , followed by immunoblotting with anti-cytochrome c . The cysteine residue of each Bak variant was assessed for solvent exposure by incubating membrane fractions with the cysteine-labelling reagent IASD ( Molecular Probes Life Technologies , Carlsbad , CA , USA ) for 30 min at 30°C ( as described in [Westphal et al . , 2014] ) . Membrane fractions were supplemented with 100 μM TCEP to prevent oxidisation of cysteines that would inhibit IASD labelling . Samples were unlabelled , or incubated with IASD before , during or after tBid incubation or following denaturation with 1% w/v ASB-16 ( Merck ) . IASD labelling was quenched by the addition of 200 mM DTT and samples solubilised in 1% ASB-16 for 10 min at 22°C . Soluble supernatant fractions were isolated by centrifugation at 16 , 000 ×g for 5 min , and added to an equal volume of IEF sample buffer ( 7 M urea , 2 M thiourea , 2% w/v CHAPS , Complete protease inhibitors , 4 μg/ml pepstatin A , 1% w/v ASB-16 and 0 . 04% w/v bromophenol blue ) . Samples were loaded onto 12-well Novex pH 3–7 IEF gels ( Life Technologies ) and focused with a Consort EV265 power supply with increasing voltage ( 100 V for 1 hr , 200 V for 1 hr and 500 V for 30 min ) . IEF gels were pre-soaked in SDS buffer ( 75 mM Tris/HCl , pH 6 . 8 , 0 . 6% w/v SDS , 15% v/v glycerol ) , and then transferred to PVDF membrane for western blot analysis . IASD labelling was quantified ( ImageLab 4 . 1 , Bio-RAD ) from western blots by measuring signal intensity for the unlabelled and labelled Bak band species in each lane to calculate the percentage of the total signal from each lane attributed to the faster migrating IASD-labelled species . Control samples showed no IASD labelling of Bak Cys null ( Westphal et al . , 2014; Iyer et al . , 2015 ) . A global background subtraction was applied to regions of interest from each western blot . Data were presented as the mean ± SD ( n ≥ 3 ) , or range ( n = 2 ) and the number of replicates indicated on the x-axis . A two-tailed unpaired t-test was employed to determine significant differences ( p<0 . 05 ) in the percentage of IASD labelled Bak before versus after tBid treatment . The cysteine residue of each Bak variant was tested for disulphide linkage with proximal cysteine residues on Bak ( or other proteins ) by incubation with the oxidant copper phenanthroline ( CuPhe ) ( as per [Dewson et al . , 2008] ) . CuPhe was prepared as a stock solution of 30 mM CuSO4 and 100 mM 1 , 10-phenanthroline in 4:1 water/ethanol . Following incubation of membrane fractions with tBid at 30°C , samples were pre-chilled on ice for 5 min , and then incubated with a 100-fold dilution of the CuPhe solution on ice for 15 min . Note that ( Iyer et al . , 2015 ) performed CuPhe linkage without the pre-chill incubation . Disulphide bond formation ( linkage ) was quenched by the addition of 20 mM N-ethyl maleimide ( NEM , to label any remaining free cysteines , 10 min on ice ) and 5 mM EDTA ( to chelate copper , 5 min on ice ) and samples analysed by non-reducing SDS PAGE or blue native PAGE ( BNP ) . CuPhe-linked samples were analysed by SDS PAGE ( 12% TGX gels ( Life Technologies ) ) in the absence of reducing agents to preserve disulphide bonds . The efficiency of linkage between Bak molecules was measured by a shift in Bak migration from 1x to 2x complexes . To discriminate between linkages occurring within or between Bak dimers , CuPhe-linked samples were also analysed in tandem by blue native PAGE ( BNP ) to preserve the native dimer interface ( i . e . the BH3:groove interface ) in addition to any disulphide bonds between Bak molecules . Migration of higher order linked complexes ( i . e . greater than dimer ) on BNP indicated the presence of disulphide bonds between Bak dimers . Following quenching of CuPhe with NEM and EDTA , membrane fractions were isolated by centrifugation at 16 , 000 ×g for 5 min . Supernatants were discarded and membrane pellets were solubilised in 20 mM Bis-Tris pH 7 . 4 , 50 mM NaCl , 10% v/v glycerol , 1% w/v digitonin , and incubated on ice for 1 hr and insoluble material was removed by centrifugation at 16 , 000 ×g for 5 min . The resulting supernatants were prepared for BNP by the addition of Native Sample buffer ( Life Technologies ) and Coomassie Additive ( Life Technologies ) , and then loaded onto Novex 4–16% Native PAGE 1 . 0 mm 10 well gels as per the manufacturer’s instructions ( Life Technologies ) . Western blot transfer to PVDF was performed at 30 V for 150 min in transfer buffer ( 25 mM Tris , 192 mM Glycine , 20% v/v Methanol ) supplemented with 0 . 037% w/v SDS . PVDF membranes were de-stained with 10% v/v acetic acid 30% v/v ethanol , then further de-stained in methanol and rinsed thoroughly with dH2O before immunoblotting . SDS PAGE and IEF membranes were immunoblotted for Bak using the rabbit polyclonal anti-Bak aa23-38 ( B5897 , Sigma-Aldrich , Castle Hill , NSW , Australia , RRID:AB_258581 ) . BNP membranes were immunoblotted for Bak using an in-house anti-Bak monoclonal rat IgG ( clone 7D10 , WEHI , [Dewson et al . , 2009; Alsop et al . , 2015] ) , except in the case of Bak mutants ranging from G51C to P58C that were detected on BNP with the anti-Bak aa23-38 . In-house monoclonal rat antibodies were used to detect recombinant Bax ( clone 49-F9 , WEHI ) , Bcl-xL ( clone 9C9 , WEHI ) and Mcl-1 ( clone 19C 4–15 , WEHI ) . Cytochrome c release was assessed by SDS-PAGE and immunoblotting with anti-cytochrome c ( clone 7 H8 . 2C12 , 556433 , BD Biosciences Pharmingen , San Diego , CA , USA RRID:AB_396417 ) . Horseradish peroxidase conjugated IgG secondary antibodies were used; anti-rabbit ( 4010–05 , AdB Serotec , RRID:AB_609701 ) , anti-rat ( 3010–05 , AdB Serotec , RRID:AB_619911 ) and anti-mouse ( 1010–05 , AdB Serotec , RRID:AB_609673 ) . Immobilised horseradish peroxidase was detected with Luminata Forte western HRP substrate ( WBLUF0500 , Millipore , Billerica , MA , USA ) , images captured with the ChemiDoc XRS+ System ( Bio-RAD , Hercules , CA , USA ) and signal intensity measured with Image Lab 4 . 1 software ( Bio-RAD ) . Images were transformed in Image Lab 4 . 1 ( Bio-RAD ) to correct any rotation of the image and exported as . tif for densitometry in FIJI ( ImageJ 1 . 47n ) ( Schindelin et al . , 2012; Schneider et al . , 2012 ) . The lookup table ( LUT ) was inverted and images rotated 90 degrees . Using the rectangular selection tool , a box was drawn around the lane of interest to encompass the full range of Bak signal ( with fixed dimensions of 370 × 50 pixels ) . This box was then measured with the Dynamic ROI profiler plugin . Plot values were exported for analysis in Prism ( ver 6 . 0f , Graphpad Software Co . , La Jolla , CA , USA ) , and normalised to a fraction of the total intensity for each lane of interest . A clearer representation of multiple replicate plots of lane intensity was achieved by a linear rescaling of each replicate's horizontal axis so that the maximum corresponding to Bak 4x species , and the characteristic minimum ( which is a highly reproducible artefact of our Bak BNP western blots ) at ~720 kDa , coincided across replicates . The formation of a disulphide linked oligomer is modelled by repeatedly ( a ) selecting an available pair of cysteines on neighbouring dimers at random , ( b ) declaring them linked and ( c ) removing those cysteines from the list of those subsequently available . We continue until no neighbouring dimers have cysteines available for linkage , to reflect complete linkage of V61C:V61C’ . We consider a collection of Bak dimers on a two-dimensional surface ( see Figure 6 ) . Dimers are arranged on a hexagonal ( or square ) grid , and linkage is permitted in all directions . Each dimer has two cysteines available for disulphide linkage to a neighbouring dimer . We simulate such that linkage is allowed in all directions with equal probability . Reduced linkage efficiency was modelled by introducing an edge blocking effect . Every edge between neighbouring hexagonal cells was flagged as eligible for a disulphide bond , independently at random with probability p constant across all edges in a given simulation , for p=0 , 0 . 25 , 0 . 5 and 0 . 75 . We modelled the experimental scenario where Bak dimers were formed from a mixture of equal amounts of V61C and cys-null monomers by assigning each dimer either zero , one or two available cysteines with probability 0 . 25 , 0 . 5 and 0 . 25 respectively ( see Figure 6—figure supplement 3C ) . Predicted densitometry for each Bak dimer multiplicity are presented by taking the distribution of oligomer sizes on completion of the simulation , multiplying by the number of Bak dimers in each oligomer to obtain the theoretical density that would give rise to in the western blot , using an exponential horizontal axis to mimic the nonlinear spacing of bands , and applying a gaussian kernel estimator to approximate the smearing of bands and their running into one another at higher molecular weights ( see Figure 6—figure supplement 1 ) . | A healthy organism must carefully remove unwanted , diseased or damaged cells . These unwanted cells can bring about their own death in a controlled process known as apoptosis . Maintaining an appropriate level of apoptosis is crucial to good health: excessive cell death can contribute to neurodegenerative disorders , whereas too little can result in cancer . All cells contain powerhouses called mitochondria , which produce energy . Mitochondria are the scene of a critical ‘point of no return’ in apoptosis . When a cell receives a death signal , a ‘killer’ protein known as Bak punches holes ( or pores ) in the membrane of the mitochondria . These pores allow toxic molecules to leak out from the mitochondria into the interior of the cell , where they trigger a series of events that dismantles the cell from the inside out . To create the pores , Bak undergoes extensive shape changes that allow the proteins to form dimers that then cluster and perforate the membrane . To investigate how Bak clusters assemble on the mitochondrial membrane , Uren et al . used cultured cells and biochemical techniques to show where the Bak dimers contacted each other before and after the pore formed; these findings were complemented with mathematical modelling . The results show that during apoptosis , Bak dimers contact each other at several different places ( rather than at one or two places ) to assemble into disorderly , ever-changing clusters . Based on these observations , Uren et al . suggest that the enlarging clusters stress the membrane and cause pores to form . The next step is to investigate whether physical forces that act within the mitochondrial membrane could drive the clustering of Bak proteins . This knowledge could ultimately enable us to learn how to manipulate apoptosis in cells , potentially as part of treatments for the diseases in which this cell death process occurs inappropriately . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2017 | Disordered clusters of Bak dimers rupture mitochondria during apoptosis |
Rhythmic neuronal activity provides a frame for information coding by co-active cell assemblies . Abnormal brain rhythms are considered as potential pathophysiological mechanisms causing mental disease , but the underlying network defects are largely unknown . We find that mice expressing truncated Disrupted-in-Schizophrenia 1 ( Disc1 ) , which mirror a high-prevalence genotype for human psychiatric illness , show depression-related behavior . Theta and low-gamma synchrony in the prelimbic cortex ( PrlC ) is impaired in Disc1 mice and inversely correlated with the extent of behavioural despair . While weak theta activity is driven by the hippocampus , disturbance of low-gamma oscillations is caused by local defects of parvalbumin ( PV ) -expressing fast-spiking interneurons ( FS-INs ) . The number of FS-INs is reduced , they receive fewer excitatory inputs , and form fewer release sites on targets . Computational analysis indicates that weak excitatory input and inhibitory output of FS-INs may lead to impaired gamma oscillations . Our data link network defects with a gene mutation underlying depression in humans .
Psychiatric disorders not only diminish life quality of affected individuals , but also pose a substantial issue in public health because of their high prevalence in modern society . Mutations in ‘risk genes’ enhance the probability to develop these disorders , pointing to a strong genetic component in the etiology of mental illnesses ( Ross et al . , 2006 ) . More than two decades ago , DISC1 has been identified as a major genetic risk factor involved in psychiatric disorders ( St Clair et al . , 1990; Blackwood et al . , 2001 ) . The original discovery came from a Scottish family carrying a large c-terminal 1:11 translocation in the DISC1 gene downstream of exon eight , which results in a c-terminal truncation of DISC1 ( St Clair et al . , 1990 ) . Family members who are affected by the mutation suffer from mental illness including major depression ( 10 cases ) , schizophrenia ( 7 cases ) or bipolar disorder ( 1 case , St Clair et al . , 1990 ) . By comparison ( 48 cases ) , no major psychiatric disease was diagnosed in any of the relatives lacking DISC1 truncation . Thus , truncation of DISC1 constitutes one of the largest known risk factors for mental illness . Recently , a mouse model has been developed which reproduces the human form of DISC1 truncation ( Shen et al . , 2008 ) . Those Disc1 mice allow to directly examine the impact of a depression- and schizophrenia-related risk gene mutation on behaviour and the activity of neuronal networks involved in the control of cognitive functions . We find that Disc1 mice show increased immobility during the tail-suspension ( TST ) and forced swim test broadly accepted as depression-related behavioural changes in rodents ( Porsolt et al . , 1977; Steru et al . , 1985 ) . This behavioural phenotype correlates with abnormalities in the synchrony of low-gamma oscillations ( 30–50 Hz ) in the prelimbic cortex ( PrlC ) , which have been implicated in supporting encoding of information ( Fries et al . , 2007 ) . Moreover , we provide evidence that network malfunction is related by a profound defect of FS-INs , including reduced numbers of FS-INs as well as alterations in their synaptic connections . Thus , our study provides a correlative link between behavioural alterations in Disc1 mice and the possible underlying cellular and synaptic mechanisms .
To investigate the effect of truncated Disc1 on potential depression- or schizophrenia-like phenotypes of Disc1 mice we conducted a comprehensive analysis covering a wide spectrum of behavioural deficits characteristic for psychiatric syndromes ( Figure 1 , refer to Table 1 for a summary of values ) . We probed depression-related traits using highly validated tests for anhedonia ( LeGates et al . , 2012 ) and behavioural despair ( Porsolt et al . , 1977; Steru et al . , 1985 ) in rodents . Disc1 mice showed no deficit in sucrose preference , which is used to quantify anhedonia ( Figure 1A ) . However , in the TST and forced swim test , in which animals exhibit epochs of immobility that are thought to reflect states of behavioural despair intersected by periods of active escape , Disc1 mice showed longer periods of immobility ( Figure 1B , C ) . Behavioural variability among individuals was high and therefore resulted in a moderately but significantly enhanced mean immobility by 35% in the TST and by 18% in the forced swim test of Disc1 mice ( TST: p = 0 . 015 , 22 Disc1 and 14 control mice; forced swimming: p = 0 . 049 , 22 and 20 mice; Cohen's d of 0 . 82 and 0 . 65 corresponding to a strong and moderate effect size , respectively [Table 1] ) . Both genotypes reached similar movement speeds in the open field arena , indicating that high immobility in the behavioural despair tests cannot be caused by motor impairment ( Figure 1D ) . 10 . 7554/eLife . 04979 . 003Figure 1 . Disc1 mice show depression-related behavioural despair . ( A ) Disc1 mice show similar sucrose preference as controls ( 146 ± 7 vs 144 ± 15% sucrose intake , n = 10 each group ) . ( B and C ) Enhanced behavioural despair of Disc1 mice in TST ( 37 . 8 ± 3 . 3 vs 25 . 8 ± 2 . 9% , n = 22 Disc1 , 14 control mice ) and forced swim test ( 44 . 1 ± 2 . 6 vs 37 . 3 ± 2 . 0% , n = 22 , 20 ) . ( D ) Unaltered locomotion of Disc1 mice . Left , examples of the path of a Disc1 and a control mouse during a 10 min exploration period ( n = 19 Disc1 , 18 control mice ) . Right , Disc1 mice move slower during the initial phase of the task but reach similar movement speeds as controls during the later phase . ( E ) Radial arm water maze to probe spatial reference and working memory . The animals were released from a random start arm and had to find the hidden platform in the southern arm . Green line shows the path of one representative animal during one trial . Arm entries were detected by a threshold-crossing algorithm ( bottom , N = 6 , 9 ) . ( F ) Path length and reference memory errors plotted against the five subsequent test days . Identical shortening of the swim path length ( left ) and identical number of reference memory errors ( entries in wrong arms; middle ) of Disc1 and control mice indicates intact spatial learning . Time spent in the target arm is identical between genotypes ( n = 6 , 9 ) . ( G ) The number of working memory errors ( re-entries in previously explored arms within a trial , 14 ± 2 vs 17 ± 3 ) did not depend on the genotype . *p < 0 . 05 , **p < 0 . 01 . Data are mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 00310 . 7554/eLife . 04979 . 004Figure 1—figure supplement 1 . Additional assessment of working memory performance . ( A ) Working memory assessment in a delayed match-to-place task in the water y-maze . In each trial , animals had to find the hidden platform in one of the target arms ( sample phase ) . After a 30 s delay , the mice were released from the same start arm ( match phase ) and could either choose the correct arm or the opposite arm ( working memory error ) . Both Disc1 and control mice performed significantly better in the match phase , indicating intact spatial working memory ( n = 6 , 10 ) . ( B ) Both genotypes showed similar spontaneous alternation when they explored a y-maze for 10 min , confirming intact working memory ( n = 8 each group ) . #p < 0 . 001 . Data are mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 00410 . 7554/eLife . 04979 . 005Figure 1—figure supplement 2 . Extra-dimensional rule-shifting task in the y-maze . ( A ) Mice were food-restricted and learned to forage in a y-maze . Then , they learned the first reward rule ( right arm correct ) . In each trial , one randomly chosen arm of the two possible target arms was illuminated . Once the animals had learned the spatial rule , the reward regime was changed to ‘light on’ ( extradimensional shift , n = 6 , 5 ) . ( B ) Normalized correct trials are plotted against trial number . Arrow at t = 0 indicates the reward rule change . Both groups of mice learned the initial spatial rule as well as the rule change . Data are binned over 10 runs . ( C ) Examples of individual learning curves of a Disc1 and a control mouse with 95% confidence intervals . The trial in which the lower confidence interval exceeded chance level was considered the first trial in which the animal had learned the task ( learning trial ) . Right , identical learning trials of both rules in Disc1 and control mice . Data are mean ± SEM unless stated . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 00510 . 7554/eLife . 04979 . 006Figure 1—figure supplement 3 . Unaltered anxiety of Disc1 mice . Unaltered anxiety in Disc1 mice quantified from the time spent in the center of the open field ( n = 19 , 18 ) . Data are mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 00610 . 7554/eLife . 04979 . 007Figure 1—figure supplement 4 . Unaltered sociability of Disc1 mice . ( A ) 3-chamber social interaction task . The mice explored the 3-chamber arena containing two empty ( E ) wire cups in a habituation run . In a second run , a stranger mouse ( S ) was placed in one of the cups ( n = 9 each group ) . Right , examples of exploration paths . ( B ) Disc1 and control mice spent identical time in both compartments during habituation and more time with the stranger mouse in run two , indicating absent left-right preference and intact sociability , respectively . #p < 0 . 001 . Data are mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 00710 . 7554/eLife . 04979 . 008Table 1 . Quantitative summary of cellular and synaptic properties of Disc1 and control PrlC neuronsDOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 008ParameterDisc1ControlP/NTestCohen's dSucrose preference1 . 46 ± 0 . 071 . 44 ± 0 . 150 . 901 , N = 10/10Student's t-testnaFreezing in TST ( norm . ) 1 . 39 ± 0 . 121 . 00 ± 0 . 100 . 015 , N = 22/14Student's t-test0 . 82Immobility in forced swim test ( norm . ) 1 . 18 ± 0 . 071 . 00 ± 0 . 050 . 049 , N = 22/20Student's t-test0 . 65Open field total path length ( m ) 14 . 5 ± 0 . 919 . 3 ± 0 . 70 . 0012 , N = 19/18Student's t-test1 . 47Open field path length ( first half/second half ) 4 . 8 ± 0 . 4/8 . 9 ± 0 . 59 . 8 ± 0 . 5/9 . 1 ± 0 . 31 . 1*10−12 , 0 . 682 for s halfOne-way ANOVA followed by t-testna ( second half ) Radial arm maze reference/working memory errors22 ± 5/14 ± 225 ± 2/17 ± 30 . 239/0 . 361 , N = 6/9Mann–Whitney U testnaProportion spontaneous alternation0 . 60 ± 0 . 030 . 65 ± 0 . 040 . 361 , N = 8/8Student's t-testnaExtradim . spatial/shifted rule learning trial ( y-maze ) 10 ± 1/39 ± 715 ± 3/34 ± 90 . 154/0 . 632 , N = 8/7 , 6/5Student's t-testnaTime in center of open field ( % ) 43 . 5 ± 0 . 937 . 4 ± 3 . 50 . 377 , N = 19/18Student's t-testna3-chamber social interaction: stranger preference49 . 4 ± 4 . 653 . 2 ± 3 . 30 . 513 , N = 9/9Student's t-testnaTST-dependent cFos increase PrLC ( norm . ) 3 . 92 ± 1 . 544 . 78 ± 1 . 010 . 03/0 . 04 , N = 4/3Mann–Whitney U test1 . 52/3 . 49Freezing in TST ( % , electrode-implanted sample ) 52 . 2 ± 0 . 129 . 4 ± 0 . 10 . 004 , N = 8/6Mann–Whitney U test1 . 82TST: low-gamma power/amplitude ( *10−3 ) 0 . 11 ± 0 . 02/0 . 87 ± 0 . 050 . 29 ± 0 . 04/1 . 34 ± 0 . 090 . 003/0 . 002 , N = 8/6Mann–Whitney U test2 . 33/2 . 77TST: theta power/amplitude ( *10−3 ) 0 . 11 ± 0 . 03/0 . 04 ± 0 . 0040 . 29 ± 0 . 04/0 . 07 ± 0 . 0040 . 012/0 . 001 , N = 8/6Mann–Whitney U test1 . 97/3 . 12Home cage: low-gamma power ( *10−3 ) 0 . 12 ± 0 . 020 . 22 ± 0 . 030 . 009 , N = 8/6Mann–Whitney U test1 . 62Home cage: theta power ( *10−3 ) 0 . 1 ± 0 . 020 . 18 ± 0 . 040 . 031 , N = 8/6Mann–Whitney U test1 . 29Urethane anesthesia: low-gamma power ( *10−3 ) 0 . 6 ± 0 . 21 . 1 ± 0 . 20 . 024 , N = 11/7Student's t-test1 . 27TST: hippocampal theta power ( *10−3 ) 0 . 8 ± 0 . 24 . 1 ± 1 . 50 . 009 , N = 5/4Mann–Whitney U test1 . 71CA1-PrLC theta coherence TST/home cage0 . 52 ± 0 . 06/0 . 52 ± 0 . 070 . 52 ± 0 . 04/0 . 49 ± 0 . 020 . 972 , N = 4/3One-way ANOVAnaPV-IN count PrLC all layers ( normalized ) 0 . 69 ± 0 . 081 . 00 ± 0 . 040 . 004 , N = 9/8Student's t-test1 . 81PV-IN count layer 2–3/layer 53 . 0 ± 0 . 7/31 . 9 ± 5 . 36 . 2 ± 0 . 54/48 . 8 ± 2 . 520 . 007/0 . 011 , N = 6/5Mann–Whitney U test2 . 35/1 . 86Somatostatin-IN count ( normalized ) 1 . 04 ± 0 . 101 . 00 ± 0 . 080 . 392 , N = 6/5Mann–Whitney U testnaDAPI density PrLC ( normalized ) 0 . 93 ± 0 . 041 . 00 ± 0 . 040 . 158 , N = 6/5Mann–Whitney U testnaCalbindin-IN count ( normalized ) 0 . 75 ± 0 . 121 . 00 ± 0 . 050 . 027 , N = 6/5Mann–Whitney U test1 . 2PV-VGAT-positive boutons PrLC ( normalized ) 0 . 62 ± 0 . 091 . 00 ± 0 . 100 . 021 , N = 6/4Mann–Whitney U test2 . 02FS-IN bouton density in vitro ( µm−1 ) 0 . 086 ± 0 . 0110 . 073 ± 0 . 0090 . 388 , N = 12/10*Student's t-testnaFS-IN bouton density in vivo ( µm−1 ) 0 . 091 ± 0 . 0100 . 093 ± 0 . 0100 . 868 , N = 16/20*Student's t-testnaFS-IN axon length in vitro ( mm ) 1 . 54 ± 0 . 142 . 02 ± 0 . 270 . 200 , N = 4/4*Mann–Whitney U testnaAmplitude ( pA ) 50 . 8 ± 13 . 3121 . 1 ± 32 . 10 . 011 , N = 20/13*Mann–Whitney U test0 . 79rise time ( ms ) 0 . 32 ± 0 . 020 . 32 ± 0 . 030 . 985 , N = 19/13*Mann–Whitney U testnadecay time constant ( ms ) 5 . 92 ± 0 . 405 . 28 ± 0 . 280 . 193 , N = 10/11*Mann–Whitney U testnaonset latency ( ms ) 1 . 01 ± 0 . 031 . 02 ± 0 . 050 . 378 , N = 19/13*Mann–Whitney U testnaFS-IN-to-PC uIPSCfailure rate0 . 32 ± 0 . 070 . 08 ± 0 . 040 . 015 , N = 20/13*Mann–Whitney U test1 . 06coefficient of variation0 . 924 ± 0 . 1650 . 426 ± 0 . 0720 . 008 , N = 17/13*Mann–Whitney U test0 . 92skewness−0 . 262 ± 0 . 135−0 . 124 ± 0 . 1740 . 503 , N = 17/13*Mann–Whitney U testnapaired-pulse ratio 20/50 ms0 . 84 ± 0 . 08/0 . 85 ± 0 . 080 . 89 ± 0 . 09/0 . 82 ± 0 . 030 . 836/0 . 937 , N = 7/6 , 6/6Mann–Whitney U testnamultiple-pulse 50 Hz 10th0 . 58 ± 0 . 060 . 54 ± 0 . 050 . 671 , N = 8/7*Student's t-testnaConnection probability ( % ) 34 . 612 . 85 . 3*10–29 , N = 78/148*Chi2 testnaNr4 ± 210 ± 20 . 036 , N = 14/11*Mann–Whitney U test0 . 87Binomial fittingQr27 . 6 ± 2 . 628 . 5 ± 3 . 50 . 831 , N = 14/11*Student's t-testnaPr0 . 45 ± 0 . 070 . 58 ± 0 . 060 . 178 , N = 14/11*Student's t-testnaamplitude ( pA ) 30 . 7 ± 2 . 328 . 7 ± 3 . 40 . 34 , N = 15/9*Mann–Whitney U testnaspEPSCs on FS-INsfrequency ( Hz ) 6 . 0 ± 1 . 09 . 5 ± 1 . 10 . 022 , N = 15/9*Student's t-test1 . 1coefficient of variation0 . 75 ± 0 . 040 . 75 ± 0 . 050 . 596 , N = 15/9*Student's t-testnaamplitude ( pA ) 17 . 9 ± 1 . 118 . 4 ± 3 . 00 . 202 , N = 11/13*Mann–Whitney U testnaspEPSC on PCsfrequency ( Hz ) 3 . 1 ± 0 . 42 . 6 ± 0 . 40 . 246 , N = 11/13*Mann–Whitney U testnaamplitude ( pA ) 20 . 1 ± 1 . 121 . 9 ± 1 . 90 . 388 , N = 24/15*Student's t-testnamIPSC on PCsfrequency ( Hz ) 0 . 74 ± 0 . 141 . 18 ± 0 . 230 . 025 , N = 24/15*Mann–Whitney U test0 . 57N indicates number of animals except for: *N indicates number of cells , #N indicates number of axons . To test for schizophrenia-associated symptoms we assessed context representation and learning ( Waters et al . , 2004 ) and examined spatial reference and working memory in the radial arm water maze ( Murray et al . , 2011 ) . Both groups showed identical spatial learning and numbers of working memory errors ( Figure 1E–G ) . We confirmed intact working memory of Disc1 mice in a delayed match-to-sample and a spontaneous alternation task ( Figure 1—figure supplement 1 ) . Furthermore , Disc1 mice could normally learn reward rules in a spatial extra-dimensional paradigm-shifting task ( Figure 1—figure supplement 2 ) . This test resembles features of the Wisonsin card sorting test , in which schizophrenia patients typically show deficits ( Okubo et al . , 1997 ) . Finally , Disc1 mice had no abnormalities in anxiety or sociability ( Figure 1—figure supplements 3 , 4 ) . Thus , Disc1 mice showed a specific phenotype broadly interpreted as depression-related behavioural despair ( Porsolt et al . , 1977; Steru et al . , 1985 ) . PrlC integrates information from cortical and subcortical regions to exert higher-level control of behaviour including the decision to execute actions ( Yee , 2000 ) and the regulation of mood ( Covington et al . , 2010 ) , both of which are impaired in depression ( Elliott et al . , 1997 ) . We therefore hypothesized that PrlC dysfunction may be involved in the emergence of behavioural despair of Disc1 mice . By using antibody labelling against the immediate early gene cFos as a marker of neurons which underwent enhanced activity , we found that TST increased the number of cFos-positive cells in the PrlC compared to baseline in the home cage in both Disc1 and control mice , indicating that the PrlC is involved in controlling behavioural despair ( Figure 2A ) . The elevation of cFos-positive cells by TST occurred in both , Disc1 as well as control mice , suggesting that no major differences in PrlC activation upon exposure to TST may exist between genotypes ( Figure 2A ) . 10 . 7554/eLife . 04979 . 009Figure 2 . Behavioural despair of Disc1 mice correlates with impairment in theta and low-gamma oscillations in the PrlC . ( A ) TST activates cFos in PrlC independent from genotype ( fold increase Disc1: 3 . 92 ± 1 . 54 , n = 4; control: 4 . 78 ± 1 . 07 , n = 3 ) . ( B ) LFP recording during TST . Enhanced freezing of Disc1 mice is preserved in the electrode-implanted cohort ( 52 . 2 ± 5 . 8 vs 29 . 4 ± 4 . 0 , n = 8 , 6 ) . M1 , 2: motor cortex , Cg: cingulate cortex . ( C and D ) Reduced power of Disc1 mice in the theta ( 0 . 11 ± 0 . 03 vs 0 . 29 ± 0 . 04 mV2*10−3 ) and low-gamma band ( 0 . 11 ± 0 . 02 vs 0 . 29 ± 0 . 04 mV2*10−3 , n = 8 , 6 ) . Insets: filtered traces . ( E ) Oscillation amplitudes over frequency . ( F ) Oscillatory defects are observed in the home cage ( theta: 0 . 10 ± 0 . 02 vs 0 . 18 ± 0 . 04 mV2*10−3 , gamma: 0 . 12 ± 0 . 02 vs 0 . 22 ± 0 . 03 mV2*10−3 , n = 8 , 6 ) . ( G ) Theta and low-gamma power correlate with TST freezing duration ( theta: r = −0 . 6923 , p = 0 . 0061; low-gamma: r = −0 . 79 , p = 0 . 0008 ) but not with home cage immobility ( r = −0 . 029 , r = −0 . 222 ) . Black lines: linear fits . ( H ) Home cage low-gamma but not theta can predict TST freezing ( gamma: r = −0 . 569 , theta: r = −0 . 440 ) . ( I ) Low-gamma activity in Disc1 PrlC is impaired during UP-states in anesthesia ( 0 . 6 ± 0 . 1 vs 1 . 1 ± 0 . 2 mV2*10−3 , n = 11 , 7 ) . ( J ) Top , cross-correlation of LFP simultaneously recorded in hippocampus and PrlC suggests that theta oscillations are driven by hippocampus ( peak lag: 36 . 5 ± 20 . 9 vs 35 . 3 ± 14 . 3 ms , n = 5 , 4 ) . Bottom , hippocampal theta power is impaired in Disc1 mice ( 0 . 87 ± 0 . 23 vs 4 . 14 ± 1 . 54 mV2*10−3 , n = 4 , 3 , p = 0 . 01 ) . *p < 0 . 05 , **p < 0 . 01 . Data are mean ± SEM , circles are individual mice . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 00910 . 7554/eLife . 04979 . 010Figure 2—figure supplement 1 . Unchanged hippocampal-prefrontal theta coherence . ( A ) Average coherence during TST and baseline revealed a sharp peak in the delta frequency range in both genotypes . Solid lines are mean , shaded areas SEM . ( B ) There is no difference in the average coherence ( 6–12 Hz ) between Disc1 and control mice ( n = 4 Disc1 , 3 control mice ) . Data are mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 01010 . 7554/eLife . 04979 . 011Figure 2—figure supplement 2 . Low-gamma defect in the PrlC of Disc1 mice does not depend on the behavioral state during TST . ( A ) Average power spectral density in the PrlC of Disc1 ( green ) and control mice ( black ) during freezing ( left ) and movement ( right ) . ( B ) Summary plots of mean low-gamma power revelas significantly impaired power during both behavioral states . ( n = 8 Disc1 , 6 control mice ) . Data are mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 011 In humans and rodents activity in the PrlC becomes synchronized during various behavioural states ( Uhlhaas and Singer , 2010; Roux et al . , 2012 ) . We therefore speculated that Disc1 truncation might cause changes in rhythmic activity patterns rather than gross activity levels in the PrlC . To test for this possibility , we recorded local field potentials ( LFPs ) in the PrlC of behaving mice ( Figure 2B ) . Longer freeze times of Disc1 mice were apparent in the electrode-implanted sample ( 8 Disc1 and 6 control mice , p = 0 . 0041; Figure 2B ) similar to non-implanted Disc1 mice . During TST , Disc1 mice showed reduced normalized power and amplitude in the theta ( 6–12 Hz , p = 0 . 012 and p = 0 . 001 , respectively ) and low-gamma ( 30–50 Hz , p = 0 . 003 and p = 0 . 002 , respectively ) but not high gamma band ( 80–100 Hz , p = 0 . 11 and p = 0 . 239 , respectively; Figure 2C–E ) . Oscillations were similarly impaired during home cage exploration ( Figure 2F ) . These observations were independent of the behavioral state and present during movement and passive coping ( Figure 2—figure supplement 1 ) . When data from Disc1 and control mice were pooled , theta and low-gamma power linearly correlated with freeze duration in TST ( thetaTST p = 0 . 0061 and low-gammaTST p = 0 . 0007 , respectively; Figure 2G , left ) but not with immobility in the home cage ( thetabaseline p = 0 . 829 and low-gammabaseline p = 0 . 782 , respectively; Figure 2G , right ) , showing that reduced synchrony of theta and low-gamma oscillations impairs TST-specific cortical processing rather than alterations in general locomotion . Moreover , low-gamma but not theta power in the home cage could significantly predict TST freeze duration when Disc1 and control data were pooled ( p = 0 . 033 , p = 0 . 114 , respectively; Figure 2H ) , indicating that local low-gamma power correlates with defects of the intrinsic function of the prefrontal network irrespective of the animal's behaviour . Recordings of spontaneous low-gamma oscillations during UP states under anaesthesia ( Hasenstaub et al . , 2005 ) confirmed the independence of low-gamma power from behaviour-dependent brain state ( Figure 2I ) . Jointly these data suggest that defective thetaTST and low-gammaTST+baseline are correlated with behavioural despair of Disc1 mice . We next examined the mechanisms underlying oscillatory impairments in Disc1 PrlC . Prefrontal theta oscillations are driven by the hippocampus ( Siapas et al . , 2005; Sigurdsson et al . , 2010 ) whereas gamma activity patterns are generated by synaptic interactions between GABAergic FS-INs and glutamatergic PCs in local neuronal networks ( Atallah and Scanziani , 2009; Tiesinga and Sejnowski , 2009 ) . Consistent with the hippocampal drive of theta oscillations to the PrlC , cross-correlation analysis of theta-filtered signals in simultaneous LFP recordings from dorsal CA1 and PrlC revealed a ∼30 ms peak time lag in both Disc1 and control mice ( Figure 2J ) . In agreement with the intact working memory of Disc1 mice , for which high synchrony of theta oscillations between hippocampus and prefrontal cortex is required ( Jones and Wilson , 2005; Siapas et al . , 2005; Sigurdsson et al . , 2010 ) , coherence in the theta band was comparable between genotypes ( Figure 2—figure supplement 2 ) . However , theta power was markedly reduced in CA1 of Disc1 mice ( Figure 2J ) , suggesting that the prefrontal theta power deficit may be caused by a theta dysfunction in the hippocampus . To obtain deeper insight into the pathophysiology of Disc1-associated behavioural despair , we next focussed on the mechanisms underlying impaired low-gamma oscillations in the PrlC because human depression patients show reduced low-gamma activity in frontal regions ( Liu et al . , 2012 ) . Our cFos labelling suggested that TST directly activated the PrlC network ( Figure 2A ) and that local PrlC mechanisms may contribute to the TST phenotype of Disc1 mice . The PrlC of Disc1 mice contained significantly fewer PV-positive INs ( ∼40% reduction , p = 0 . 0037 , 9 Disc1 and 8 control mice , Figure 3A , Figure 3—figure supplement 1 ) . A similar reduction in PV-positive cells was observed in CA1 ( ∼40%; p = 0 . 022 ) but not in the ventro-orbital cortex ( p = 0 . 375; Figure 3—figure supplement 1 ) . PV-positive cells of both genotypes expressed Disc1 ( Figure 3—figure supplement 2 ) . In contrast , the number of somatostatin-expressing INs ( p = 0 . 392 ) and total cell density ( DAPI area , p = 0 . 158 ) were unchanged ( Figure 3A , B ) . Studies on schizophrenia patients suggested that PV-expression might be down-regulated in FS-INs ( Hashimoto et al . , 2003 ) . However , detection of PV immunoreactivity in electrophysiologically identified FS-INs in PrlC slices did not depend on the genotype ( Disc1: 9/16 cells; control: 10/19 cells; Figure 3C , Figure 3—figure supplement 3 ) . Moreover , the number of INs expressing calbindin , a marker for FS-INs partially coexpressed with PV ( Markram et al . , 2004 ) , was reduced in the Disc1 PrlC in vivo ( ∼25% reduction , p = 0 . 027 , 6 Disc1 and 5 control mice , Figure 3D ) , supporting our conclusion of reduced PV-cell quantity rather than PV content of FS-INs . Finally , the frequency of miniature IPSCs ( mIPSCs ) recorded in PCs was significantly reduced in the PrlC of Disc1 mice , consistent with a loss of PV-positive cells ( p = 0 . 025 , 24 Disc1 and 15 control cells; Figure 2E ) . These data further suggested a lack of mechanisms compensating for the reduced PV cell population . 10 . 7554/eLife . 04979 . 012Figure 3 . Loss of FS-INs and their output synapses in the Disc1 PrlC . ( A ) Reduction of PV but not somatostatin-positive INs in the PrlC of Disc1 mice ( normalized count PV: 0 . 69 ± 0 . 08 vs 1 . 00 ± 0 . 04 , n = 9 Disc1 , 8 control mice , p = 0 . 0037; somatostatin: 1 . 04 ± 0 . 10 vs 1 . 00 ± 0 . 08 , n = 6 , 5 , p = 392 ) . ( B ) Total cell density quantified from DAPI area is unchanged ( normalized density: 0 . 93 ± 0 . 04 vs 1 . 00 ± 0 . 04 , n = 6 , 5 mice , p = 0 . 158 ) . ( C ) FS-INs of Disc1 and control mice express PV ( n = 25 , 29 cells ) . ( D ) Expression of the FS-IN marker calbindin is reduced in Disc1 PrlC ( 0 . 75 ± 0 . 12 vs 1 . 00 ± 0 . 05 , n = 6 , 5 mice , p = 0 . 027 ) . ( E ) Frequency of mIPSCs recorded in PCs was significantly reduced in the Disc1 PrlC ( 0 . 73 ± 0 . 14 vs 1 . 18 ± 0 . 23 Hz , n = 24 , 5 cells , p = 0 . 025 ) . ( F ) Fewer PV-VGAT-coexpressing boutons in Disc1 mice ( normalized count 0 . 62 ± 0 . 09 vs 1 . 00 ± 0 . 10 , n = 6 , 4 mice , p = 0 . 021 ) . ( G ) Identical bouton density of intracellularly labelled FS-INs in vitro ( 0 . 09 ± 0 . 01 vs 0 . 07 ± 0 . 01 µm−1 , n = 12 , 10 , p = 0 . 388 ) and PV-positive cells in vivo ( 0 . 09 ± 0 . 01 vs 0 . 09 ± 0 . 01 µm−1 , n = 16 , 20 , p = 0 . 868 ) . *p < 0 . 05 , **p < 0 . 01 . Scale bars: A , B , C ( left ) , D: 100 µm , E: 25 µm , C ( right ) : 10 µm . Data are mean ± SEM , circles individual mice or cells ( C and F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 01210 . 7554/eLife . 04979 . 013Figure 3—figure supplement 1 . PV-positive INs are lost throughout cortical layers in the Disc1 PrlC . ( A ) Immunohistochemical staining of PV in the PrlC of a Disc1 and a control mouse . Scale bar: 100 µm . Right , quantification revealed reduced PV cell counts in layers 2/3 and 5 ( n = 6 Disc1 , 5 control mice ) . ( B and C ) Reduction of PV cells in the hippocampus but not ventro-orbital cortex of Disc1 mice . Pcl: pyramidal cell layer . *p < 0 . 05 , **p < 0 . 02 . Data are mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 01310 . 7554/eLife . 04979 . 014Figure 3—figure supplement 2 . PV-positive INs express Disc 1 . Confocal image stack of DAPI , PV and Disc1 stained-sections demonstrated expression of Disc1 in PV-positive neurons in the PrlC of Disc1 and control mice . Images are representative for 35 PV/Disc1-positive cells ( of 39 tested PV-expressing neurons ) in Disc1 PrlC and 15 PV/Disc1-positive cells ( of 15 tested ) in ctrl PrlC . p = 0 . 832 . Image size: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 01410 . 7554/eLife . 04979 . 015Figure 3—figure supplement 3 . Electrophysiological characteristics of FS-INs . ( A ) In vitro whole-cell patch clamp recording from a FS-IN . Infrared-differential interference contrast microscopical image reveals small and round soma shape of a FS-IN surrounded by larger PC somata . ( B ) FS-INs can be identified by the kinetic properties of single action potentials . Shown are superimposed examples of a single action potential of a FS-IN ( blue ) , a non-FS-IN ( red ) , and a morphologically identified PC ( grey ) . FS-INs fire briefer spikes than the other cell types . ( C ) FS-INs respond to long-lasting somatic suprathreshold current injection ( 1 s ) with a characteristic non-adapting , high-frequency train of action potentials . Non-FS-INs and PCs show varying degrees of accomodation and/or adaptation . ( D ) Adaptation coefficient defined as the first interspike interval of a train divided by the last one was used to separate FS-INs ( coefficient >0 . 6 , n = 59 Disc1 , 72 control cells ) from non-FS-INs ( <0 . 6 ) . Note , FS-INs in Disc1 and control PrlC show similar adaptation coefficients . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 01510 . 7554/eLife . 04979 . 016Figure 3—figure supplement 4 . Morphological characteristics of FS-INs . ( A ) Axon reconstructions of FS-INs revealed dense axonal plexus . Asterisks mark som position ( not shown ) . ( B ) FS-INs form perisomatic boutons on their target cells . Asterisks indicate soma positions of putative PCs . This example is depicted from a control mouse . ( C ) Disc1 and control FS-INs had similar total axon lengths . Data are mean ± SEM . n = 4 cells each group . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 016 The loss of FS-INs was paired with an equal reduction in the number of their PV/VGAT-coexpressing terminals ( ∼40% reduction , p = 0 . 021 , 6 Disc1 and 4 control mice , Figure 3F ) . Three-dimensional reconstructions of FS-INs revealed no difference in bouton density on FS-IN axons ( Figure 3G ) . Similarly , we detected comparable bouton densities on PV-positive axons in the PrlC of both genotypes in vivo ( Figure 3G ) and indistinguishable axon lengths from reconstructed cells in vitro ( Figure 3—figure supplement 4 ) . Thus , the amount of PV-positive FS-INs and their synapses is reduced by ∼40% in the Disc1 PrlC . To test whether altered synaptic transmission of local FS-INs might contribute to the low-gamma defect , we recorded from layer 5 FS-INs and PCs in acute prefrontal slices ( Figure 4 ) . Paired recordings from synaptically connected FS-INs and postsynaptic PCs revealed a strong reduction in the amplitude of unitary inhibitory postsynaptic currents ( uIPSCs ) in Disc1 PrlC ( ∼60% reduction , p = 0 . 012 , 20 and 13 pairs , respectively; Figure 4A ) . This decline in synaptic inhibition was not caused by a reduced connection probability with distance among communicating partners because inter-somatic distances between pre- and postsynaptic cells were identical ( <60 µm; see ‘Material and methods’; Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 04979 . 017Figure 4 . Output and input signalling of FS-INs are impaired in the Disc1 PrlC . ( A ) Paired recordings of FS-INs and PCs revealed a reduction in uIPSC amplitude in the PrlC of Disc1 mice ( 50 . 8 ± 13 . 3 vs 121 . 1 ± 32 . 1 , n = 20 , 13 pairs , p = 0 . 012 ) but enhanced connection probability ( 35 . 4 vs 11 . 5% , n = 65 , 122 simultaneous recordings , p < 0 . 001 ) . Scale , 100 µm . Middle , confocal images of the pairs . ( B ) Amplitude distributions of uIPSCs of a Disc1 and control pair . Lines represent best fit results obtained by multiple probability compound binomial analysis . ( C ) Nr but not Qr or Pr are reduced at Disc1 FS-IN synapses ( Nr: 4 ± 2 vs 10 ± 2 , p = 0 . 036; Qr: 27 . 6 ± 2 . 6 vs 28 . 5 ± 3 . 5 , p = 0 . 831; Pr: 0 . 45 ± 0 . 07 vs 0 . 58 ± 0 . 06 , p = 0 . 340 , n = 14 , 11 pairs ) . ( D ) Left , superimposed single traces of a Disc1 and control pair . Failure rate , coefficient of variation , and skewness of uIPSCs support reduced Nr ( 0 . 32 ± 0 . 07 vs 0 . 08 ± 0 . 04 , p = 0 . 015 , 0 . 92 ± 0 . 17 vs 0 . 43 ± 0 . 07 , p = 0 . 008 , −0 . 262 ± 0 . 135 vs −0 . 124 ± 0 . 174 , p = 0 . 503 , respectively , n = 17–20 , 13 pairs ) . ( E ) Reduced frequency of spEPSCs in Disc1 FS-INs ( 6 . 0 ± 1 . 0 vs 9 . 5 ± 1 . 1 Hz , p = 0 . 022 , n = 15 , 9 ) but not PCs ( 3 . 1 ± 0 . 4 vs 2 . 6 ± 0 . 4 , p = 0 . 246 , n = 11 , 13 ) . ( F ) Unchanged spEPSC amplitudes ( FS-INs: 30 . 7 ± 2 . 3 vs 28 . 7 ± 3 . 4 pA , p = 0 . 340 , n = 15 , 9 cells; PCs: 17 . 9 ± 1 . 1 vs 18 . 4 ± 3 . 0 pA , p = 0 . 246 , n = 11 , 13 paris ) . *p < 0 . 05 , #p < 0 . 001 . Data are mean ± SEM , circles individual cells . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 01710 . 7554/eLife . 04979 . 018Figure 4—figure supplement 1 . Recording depths and axonal distance between pairs of synaptically connected neurons are similar in Disc1 and control FS-IN to PC recordings . ( A ) Amplitude of uIPSCs is plotted against distance of the recorded soma from the pial surface . Broken line indicates border between layer 2/3 and 5 . There is no correlation between uIPSC amplitude and the anatomical depth of the presynaptic ( left ) or postsynaptic ( right ) neuron in FS-IN to PC paired recordings in acute prefrontal slices . ( B ) Synaptic latency was defined as the time between the maximal point of rise of the presynaptic action potential and the onset of the postsynaptic uIPSC . Paired recordings in Disc1 and control slices show similar mean synaptic latencies ( Disc1: 1 . 0 ± 0 . 03 ms vs control 1 . 0 ± 0 . 05 ms , p = 0 . 379 , n = 19 Disc1 , 13 control pairs ) , suggesting similar axonal distance between pre- and postsynaptic partners under the assumption of similar conduction velocities of action potentials and similar transmitter release periods . ( C ) Failure rate did not depend on the distance ( expressed as synaptic latency ) between pre- and postsynaptic cells ( r = −0 . 04 , p = 0 . 825 ) . Bars represent mean ± SEM , circles represent individual data points . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 01810 . 7554/eLife . 04979 . 019Figure 4—figure supplement 2 . Bootstrapping analysis reveals small errors in the estimation of synaptic parameters with binomial fitting . ( A ) Examples of bootstrapping of the original data from the Disc1 paired recording shown in Figure 3B . Bootstrap replications ( n = 100 ) of the original data set were fitted in identical manner as the original data shown in Figure 4B , C . ( B ) Summary graphs of the synaptic parameters Nr , Pr and Qr for bootstrapping results of all pairs revealed a difference in Nr but nor Qr or Pr . Note the small errors in the estimate of all three parameters obtained . Bars represent mean ± SEM . *p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 01910 . 7554/eLife . 04979 . 020Figure 4—figure supplement 3 . Identical dynamic and kinetic properties of uIPSCs at FS-IN output synapses . ( A ) Paired-pulse modulation in FS-IN-to-PC paired recordings was was determined from dual pulse stimulation ( inter-pulse interval 20 or 50 ms ) in paired recordings . Two presynaptic action potentials were evoked in the FS-IN at 20 ms and 50 ms inter-pulse interval . Amplitudes of uIPSCs were measured from the preceding baseline and the paired-pulse ratio ( uIPSC2/uIPSC1 ) was quantified . ( B ) Summary of paired-pulse experiments show no significant difference in short-term dynamics of uIPSCs between both genotypes ( 20 ms , p = 0 . 836 , n = 7 Disc1 , 6 control pairs , 50 ms , p = 0 . 037 , n = 6 , 6 ) . ( C and D ) Analysis of multiple-pulse modulation of uIPSCs elicited by trains of action potentials ( 10 pulses , 50 Hz ) in paired recordings revealed identical modulation across genotypes ( 10th uIPSC/1st uIPSC , p = 0 . 671 , n = 8 , 7 ) . ( E ) Determination of kinetic properties of uIPSCs . The solid lines represent biexponential fits to the decay phase of the average uIPSC . ( F ) Neither the 20–80% rise time ( p = 0 . 985 , n = 19 , 13 ) nor the decay time constant ( p = 0 . 193 , n = 10 , 11 ) differed between genotypes . Inner graphs shows the amplitude-scaled fits to the decay phase shown in E . Data are mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 02010 . 7554/eLife . 04979 . 021Figure 4—figure supplement 4 . Unaltered Q at Disc1 FS-IN-to-PC synapses . ( A ) Recordings of mIPSCs in PCs revealed identical amplitudes in Disc1 and control cells ( N = 24 , 15 , p = 0 . 388 ) . ( B ) Direct recording of quantal uIPSCs at FS-IN-to-PC synapses revealed identical Q . Quantal release was evoked in pairs in an extracellular medium containing SrCl2 . Quantal uIPSCs are visible as asynchronous events following a presynaptic action potential train . ( C ) Superimposed average quantal uIPSCs . ( D ) Simulation of apparent N as a function of p . As a result of the quantification of Q shown in B and C , Q was set to 11 pA in these calculations . Independent of the value of p , N at Disc1 FS-IN output synapses is always smaller than at control pairs . *p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 02110 . 7554/eLife . 04979 . 022Figure 4—figure supplement 5 . Similar fluctuation of spEPSCs at Disc1 and control FS-IN inputs . Superimposed individual spEPSCs are shown for a representative FS-IN recording in Disc1 and control PrlC slices . Note the similar variability in the amplitude of the input spEPSCs , quantified as the coefficient of variation ( p = 0 . 596 , n = 15 , 9 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 022 To determine which of the synaptic parameters , number of release sites ( Nr ) , quantal size ( Qr ) and release probability ( Pr ) , may contribute to the reduction in uIPSC size , we used multiple probability-compound binomial analysis ( Kraushaar and Jonas , 2000 ) ( Figure 4B , C ) . Nr but not Qr or Pr was reduced in Disc1 pairs by ∼60% ( p = 0 . 036 , p = 0 . 831 , p = 0 . 178 , 14 and 11 pairs , respectively; Figure 4C ) . Bootstrapping demonstrated that errors in the parameter estimation were similar to previous reports ( Kraushaar and Jonas , 2000 ) ( Figure 4—figure supplement 2 ) . Failure rate and coefficient of variation of uIPSCs were higher in Disc1 pairs ( p = 0 . 015 and p = 0 . 008 , respectively; Figure 4—figure supplement 1 ) , whereas the skewness was unchanged ( p = 0 . 503 , Figure 4D ) , confirming a change in Nr rather than Pr ( Kerr et al . , 2008 ) . Paired-pulse behaviour and kinetic properties of uIPSCs did not depend on the genotype , further excluding altered Pr or somatodendritic synapse location , respectively ( Figure 4—figure supplement 3 ) . Amplitudes of quantal IPSCs recorded in the presence of extracellular 5 . 5 mM strontium were not significantly different between genotypes ( 4 and 5 pairs; p = 0 . 195 ) . Moreover , mIPSCs had similar mean size in PCs located in the PrlC of Disc1 and control mice , further confirming similar Qr ( 24 and 15 cells , p = 0 . 388; Figure 4—figure supplement 4 ) . Thus , Disc1 FS-INs form ∼60% fewer release sites per target PC , resulting in an according reduction of unitary inhibitory strength . How can the contradiction between similar numbers of axonal release sites per FS-IN but fewer synaptic contacts per FS-IN-to-PC connection in Disc1 PrlC be reconciled ? Interestingly , connection probability defined as the probability to record from connected FS-IN-to-PC pairs was ∼threefold higher in Disc1 mice ( Disc1: 35 . 4% , control: 11 . 5% , p < 0 . 001; Figure 4A ) , suggesting that redistribution of release sites at the expense of individual connection strength might contribute to low-gamma defects . Recruitment of FS-INs by local excitatory collaterals is an important requirement for the generation of gamma oscillations ( Tiesinga and Sejnowski , 2009 ) . We therefore examined FS-IN excitation by glutamatergic synapses ( Figure 4E , F ) . The frequency of spontaneous excitatory postsynaptic current ( spEPSC ) was strongly reduced ( p = 0 . 022 , 15 Disc1 and 9 control cells ) . In contrast the mean amplitude and coefficient of variation of spEPSCs were unchanged ( p = 0 . 34 and p = 0 . 596 , respectively; Figure 4E , F , Figure 4—figure supplement 5 ) , suggesting reduced PC-to-FS-IN connectivity rather than changes in Nr , Qr or Pr . SpEPSC frequency in PCs was unaffected ( Figure 4E , F ) . Thus , synaptic excitation particularly of FS-INs is impaired and may contribute to low-gamma defects in Disc1 PrlC . In the Disc1 PrlC fewer FS-INs redistribute their weaker outputs to a higher number of PCs and receive fewer glutamatergic inputs . To address whether these alterations influence the synchrony of low-gamma oscillations , we designed computational neuronal network models with synaptically connected FS-INs and PCs and compared scenarios with experimentally-driven synaptic properties and connectivities from Disc1 and control prefrontal cortices ( Wang and Buzsáki , 1996 ) ( Figure 5 , Table 1 , Table 2 ) . Both network models generated synchronous low-gamma activity patterns ( Figure 5B ) . These oscillations were generated by a recurrent PC → FS-IN → PC network over a broad range of excitatory drives provided to both FS-INs and PCs ( Figure 5C , Figure 5—figure supplement 1 ) , in agreement with current theories on the generation of gamma rhythms in cortical networks ( Tiesinga and Sejnowski , 2009 ) . Consistent with gamma oscillations in prefrontal areas of rodents ( Massi et al . , 2012 ) and monkeys ( Wilson et al . , 1994 ) , INs discharged at higher rates than PCs ( mean fAP; Figure 5B ) . Spike histograms as well as LFP analogs demonstrated high synchrony of low-gamma activity in the control network model ( Figure 5B , C; black ) . In contrast , reduced synchrony of low-gamma emerged in the Disc1 circuit ( Figure 5B , C; green; Figure 5—figure supplement 1 ) . Precise timing of PC activity was proposed an important requirement for information processing ( Uhlhaas and Singer , 2010 ) . Cross-correlation analysis of FS-IN and PC discharges and quantification of PC spike times in relation to FS-IN activity revealed that spike timing fidelity of PCs was high in the control but strongly reduced in the Disc 1 network model ( Figure 5C; Figure 5—figure supplement 1 ) . These findings were robust over a wide range of excitatory regimes ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 04979 . 023Figure 5 . Disc1-mediated circuit changes impair low-gamma power in a network model . ( A ) Schematic of network structure , cellular and synaptic properties of the control ( black ) and Disc1 circuit ( green ) . ( B ) After synapses are enabled ( arrowhead ) , networks synchronize in the low-gamma range . From top to bottom: Raster plots representing action potentials; binned spike frequencies; LFPs; power of firing rates . ( C ) Cross-correlation of PC and FS-IN activity and power under different regimes of Poisson-distributed excitatory drives . Arrows point to simulation in ( B ) . Inset , PC-FS-IN spike cross-correlogram . ( D ) Effects of separate Disc1-induced circuit changes on low-gamma power . Left , effect of fewer FS-INs at different PC population sizes . Middle , reduced feedback excitation expressed as connection probability ( Pcon ) for different E-I quantal conductances ( qg ) . Right , redistribution of inhibitory synapses ( reduction of release sites/connection with proportional increase in Pcon I-E for networks with different mean connection probabilities ) . Shaded bars indicate control and Disc1 parameters . Filled squares represent the default network . ( E ) Low-gamma power depends on the strength of the Disc1 phenotype . Dashed line: reduced NFS-IN and Pcon EI . Continuous line: additional redistribution of inhibitory synapses . *p < 0 . 05 , #p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 02310 . 7554/eLife . 04979 . 024Figure 5—figure supplement 1 . Characteristics of Disc1 and control low-gamma oscillations at different excitatory drives . ( A ) Control ( black ) and Disc1 ( green ) network activity was quantified by means of the LFP ( left ) and the average discharge rate ( right ) at different excitation strengths . Excitatory drive was modelled as Poisson-distributed excitatory postsynaptic conductances . A variable fraction of the PC population ( 0–50% ) received excitatory inputs at 0 . 6 kHz . In contrast , 80% of the FS-INs were excited at varying mean frequencies ( 0 . 4–1 kHz ) . Top left , during conditions of strong excitation of FS-INs and weak excitation of PCs , oscillation frequency was high ( >60 Hz ) because FS-INs were highly active and PCs were silent . This corresponds to a gamma rhythm mediated by a network of mutually connected FS-INs ( interneuron gamma ( ING ) -mechanism ) . Shifting the excitation away from the FS-INs and recruiting more PCs reduced the oscillation frequency and established a gamma rhythm mediated by recurrent PC-IN connections ( pyramidal-interneuron gamma ( PING ) -mechanism ) . Bottom left , highest power of gamma activity was observed in the PING regime , where most of the analysis was performed ( see Figure 5C–E ) . ( B ) Top , schematic illustrating the PC sync measure as a quantification of PC spike timing relative to FS-IN activity ( ‘Materials and methods’ ) . Bottom , in the PING regime ( high PC drive , low FS-IN drive ) , PCs in the control network display much higher spike time fidelity than Disc1 PCs . DOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 02410 . 7554/eLife . 04979 . 025Table 2 . Summary of default paramters defining intrinsic and synaptic properties in network modelsDOI: http://dx . doi . org/10 . 7554/eLife . 04979 . 025SynapsePcon ( % ) Pcon Disc1 ( % ) Sdcon ( IN–IN spacings ) gq ( mS cm−2 ) NrNr Disc1τrise ( ms ) τdecay ( ms ) Esyn ( mV ) ggap ( nS ) IN-IN2050300 . 0051040 . 152−600 . 01IN-PC1025300 . 021040 . 25 . 5−60–PC-IN106 . 5200 . 01550 . 110–exc . Drive0 . 05110 . 120–Abbreviations: Pcon , probability of synaptic connections; SDcon , standard deviation in connection probability expressed as cell-to-cell distances; qg , quantal conductance; Nr , number of release sites; τrise and τdecay , rise and decay time constant; Esyn , synaptic equilibrium potential; ggap , electrical coupling . Finaly , we isolated the identified changes in Disc1 PrlC ( reduced FS-IN number , diminished local feedback excitation , and redistribution of inhibitory contacts ) and examined their individual influence on low-gamma power ( Figure 5D , E ) . Reducing FS-IN number and their feedback excitation to experimentally defined Disc1 values resulted in a cumulative decline of low-gamma power ( Figure 5D , E ) . In contrast , redistribution of FS-IN output synapses , which reproduced reduced unitary inhibitory strength and enhanced connectivity , improved network synchrony but failed to lift it to control levels ( Figure 5E; p < 0 . 001 ) . Taken together , these data propose that loss of FS-INs , their reduced recruitment and diminished inhibitory output strength jointly result in an impairment of low-gamma synchrony in the PrlC .
Our results demonstrate that a mutation in a high risk gene for depression , the DISC1 truncation correlates with reduced synchrony of theta and low-gamma oscillations in the PrlC . Notably and in line with the intact working memory of Disc1 mice ( Sigurdsson et al . , 2010 ) , phase-locking between hippocampal and prefrontal regions was unchanged . In contrast , our results suggest that reduced low-gamma synchrony in the PrlC may contribute to enhanced immobility of Disc1 mice , interpreted as depression-related behaviour ( Porsolt et al . , 1977; Steru et al . , 1985 ) , and the extent of synchrony reduction predicted the magnitude of the phenotype . The mean Disc1 phenotype differed mildly but significantly from controls which can be largely explained by the high inter-individual variability in behaviour . We identified impaired synaptic excitatory input and inhibitory output of PV-FS-INs in the Disc1 PrlC as a strong candidate mechanism underlying low-gamma defects . Thus , truncation of DISC1 in human patients may contribute to the development of depression by affecting anatomical and physiological properties of prefrontal PV-FS-INs . Our conclusions fit to the key role of FS-INs in the generation of fast network oscillations in cortical networks ( Cardin et al . , 2009 ) . Single cell recordings during spontaneous gamma oscillations in vivo ( Massi et al . , 2012; Pernia-Andrade and Jonas , 2014 ) and during pharmacologically induced gamma activity patterns in vitro ( Hajos et al . , 2004 ) revealed that strength of perisomatic inhibition and timed synaptic feedback excitation of FS-INs are key parameters setting gamma synchrony in healthy cortical networks , including the PrlC ( Goldman-Rakic , 1995 ) and the hippocampus ( Hajos et al . , 2004 ) . Furthermore , optophysiological activation or silencing of PV-IN populations result in the enhancement or suppression of gamma power in the prefrontal cortex in vivo , respectively ( Sohal et al . , 2009 ) . Similarly , recruitment of FS-INs by activating PC assemblies increases gamma power in the somatosensory cortex ( Cardin et al . , 2009 ) . Thus , the observed reduced excitatory input to FS-INs , which may cause diminished recruitment of these cells , as well as the impaired synaptic output of FS-INs will ultimately result in a loss of gamma synchrony . Gamma synchrony is also sensitive against changes in strength of gap coupling among FS-INs ( Bartos et al . , 2002 ) . Whether this synaptic property is altered in Disc1 mice would need further investigations . Our conclusions fit also to recent investigations in cortical slice preparations from mouse models for psychiatric disorders caused by mutations of genes encoding the Lysophosphatidic acid 1 receptor ( Cunningham et al . , 2006 ) , neuregulin-1 or ErbB4 receptor ( Fisahn et al . , 2009 ) . Impairment of pharmacologically induced gamma activity in these studies correlated with a marked loss of cortical PV-expressing cells . Although the mechanisms underlying the loss of synchronous gamma oscillations were not examined in these investigations , they indicate that excitation-inhibition imbalance may be the pathophysiological mechanism . PV-cells are expressed throughout the brain raising the question of specificity of the observed Disc1 effects for the prefrontal cortex . The number of PV-positive cells was also reduced in CA1 but unaltered in the ventro-orbital cortex , suggesting that PV cell defects similar to the ones identified in the PrlC occur in some but not necessarily all brain regions containing PV cells . Thus , in depth investigations are required to understand brain area-specific differences in Disc1 effects . What cellular mechanisms might explain the structural and functional reorganization of the Disc1 PrlC ? Studies of Disc1 protein interactions indicate a central role in developmental processes such as morphological differentiation and neuronal migration ( Ozeki et al . , 2003; Duan et al . , 2007 ) . Recent examinations established a direct link between Disc1 and axon growth . Disc1 knockout resulted in reduced axon elongation in cultured hippocampal cells ( Shinoda et al . , 2007 ) . Neurotrophin-3 signaling via phosphorylation of extracellular signal related kinase 1/2 ( ERK-1/2 ) is a crucial regulator of axon development , and Disc1 knockdown has been shown to abolish this phosphorylation ( Shinoda et al . , 2007 ) . Recent findings in the hippocampus further suggest a role for Disc1 in axonal path-finding . The axons of dentate gyrus granule cells , the mossy fibers , were redistributed from their normal location in strata lucidum and oriens towards the PC layer of CA3 in a mouse model expressing truncated Disc1 ( Faulkner et al . , 2008 ) . Disc1 is highly expressed in the developing brain ( Ozeki et al . , 2003; Schurov et al . , 2004 ) , highlighting its likely role in neuronal maturation and network wiring . Although no reports specifically for GABAergic axons exist , these results suggest that the effects of DISC1 truncation are not PC-specific but may also affect FS-IN axons during development . Finally , Disc1 is required for the migration of IN precursors from the ganglionic eminences to the cortex ( Steinecke et al . , 2012 ) . Therefore , truncated DISC1 might cause a migratory block of FS-INs , which could lead to the observed reduced PV cell numbers in the adult prefrontal cortex of this study . Given the limited efficacy of current anti-depressive medication , deciphering the mechanisms underlying network defects caused by mutations of DISC1 or other ‘risk genes’ will be a crucial step towards new treatment options .
All animal experimentation was in agreement with national legislation ( approved by the Regierungspräsidium Freiburg ) . During behavioural tests , adult Disc1 mice ( Shen et al . , 2008 ) ( >5 weeks ) were housed with 2–5 animals per cage . Animals were accustomed to handling in daily sessions for at least two days prior to experimentation . For TST , the tail of the mice was fixed with tape to a horizontal bar at ∼25 cm height . Movement was recorded with an IP camera . All mice were tested once . The forced swim test was performed in a 2 l glass beaker filled with 1 l tap water . Both tests lasted 6 min . For open field analysis , the mice were placed in a 30 × 30 cm arena and videotaped for 10 min . Two regions of interest of identical area ( center , periphery ) were defined . The percentage time spent in both regions and the total distance travelled were quantified . To test for anhedonia , mice were housed individually with two drinking bottles per cage , one containing 1% sucrose solution , the other tap water . Liquid consumption from both bottles was measured after 48 hr . Prior to this test , mice were kept on a two-bottle paradigm for 2–5 days . The position of the sucrose containing flask was chosen randomly for each cage . Spatial reference and spatial working memory were tested in a radial 6-arm water maze in a bathing pool ( 120 cm diameter ) . The water ( 19–21°C ) was whitened with non-toxic tempera colour . Mice were released from a random start arm and allowed to find the hidden platform within 1 . 5 min . After this time window mice were guided to the platform by the experimenter . Animals were allowed to rest on the platform for 15 s . Four runs per day were performed for 5 days with fixed target and random starting arm locations . After the last run a probe trial was conducted with the platform removed from the maze . Arm entry detection and tracking of movement was performed automatically with custom-made ImageJ routines based on MTrack2 and Python routines . Spatial reference memory errors were defined as entries in non-target arms and spatial working memory errors as re-entries in previously explored non-target arms within a trial . A match-to-place task was performed in a 3-arm water maze with a start arm , a target arm with hidden platform , and a non-target arm . Each mouse first underwent a 30 s extinction trial with only start arm and non-target arm open . After a 30 s inter-trial interval , the mouse was put back in the start arm , this time with both arms accessible ( sample trial ) . Animals were allowed to find the platform within 1 . 5 min and were guided to the platform in case they failed to perform the task . Mice rested on the platform for 15 s upon arrival . Finally , after another 30 s inter-trial interval , the match trial task was conducted similarly to the sample trial . A trial was defined as correct if the mouse first entered the target arm in the match trial ( 3 runs per day for 4 days ) . The extradimensional paradigm-shifting test was carried out in a Y-maze . Mice were food-restricted for 5 days . During this time , they were trained to search for food reward available at the end of both target arms upon release from the start arm . From day 6 onward , mice received food reward in the right arm ( ‘right correct’ task ) . During all tasks , both target arms were randomly illuminated with LEDs . When mice reached the learning criterion ( 10 subsequent correct runs or 1 error in 12 runs ) , the reward rule was switched to ‘light on—correct arm’ . Learning was measured in 10 subsequent runs and from learning curves computed with the Learning Analysis toolbox ( Smith et al . , 2004 ) . Social behaviour was measured in a 3-chamber social interaction arena composed of two side chambers and one central chamber ( 30 × 19 cm each ) . Both side chambers contained a wire pencil cup . In the first habituation run , mice explore the arena . In the second run , a stranger mouse was placed in one randomly chosen cup ( individual run duration 10 min ) . The total travel distance and time spent in each compartment were quantified . For all tests , animals were randomly chosen and test apparatuses , except the water maze , were cleaned with 70% ethanol between animals . Adult animals were anesthetized with isoflurane ( induction 3% , maintenance 1–2% ) for chronic implantations or urethane ( injected intraperitoneally; 2 g/kg urethane in saline ) for recordings in fully anesthetized conditions . Mice were fixed in a stereotaxic frame ( Kopf Instruments ) and received O2 through a mouthpiece throughout the procedure . Body temperature was kept stable with a heating pad set to 38°C . Stereotaxic coordinates for the PrlC were ( from bregma ) : anterior-posterior: +1 . 9 to +2 mm , medio-lateral: 2–2 . 25 mm , and 1 . 6–1 . 7 mm forward at 45° from brain surface for anesthetized recordings or +1 . 9 , 0 . 7–0 . 8 and 1 . 9 from brain surface at 10–15° for chronic implantations . CA1 coordinates were from bregma: −2 mm , 1 . 5 mm and 1 . 5 mm ( PC layer of PC layer/stratum oriens border ) . Coordinates were determined with a mouse brain atlas . LFPs were recorded in anaesthetized mice with either glass pipettes filled with physiological saline ( resistance 0 . 5–3 MΩ ) or tungsten microelectrodes ( 80 µm tip diameter , HEKA; 5 kHz sampling frequency ) . LFPs in freely moving mice were recorded with an implanted teflon-insulated platinum or stainless-steel wire ( 125 µm diameter ) fixed with superglue/dental cement mixture . A reference electrode was placed over the parietal cortex or cerebellum . Buprenorphine ( 0 . 02–0 . 03 ml ) was injected subcutaneously at the end of the surgery . The recording site of all animals was identified by perfusing them intracardially ( see below ) with 4% paraformaldehyd ( PFA ) after recordings . Brains were sectioned and stained with cresyl violet . In a subset of experiments , animals were sacrificed by decapitation under deep urethane anaesthesia and brains were fixed in 4% PFA for 2–10 days . Horizontal slices were cut ( 300 µm thickness ) , washed in phosphate-buffered saline ( PBS ) and embedded in Mowiol . Recording sites were identified using a light microscope ( Zeiss 2FS Plus ) . Only experiments with identified recording sites in the PrlC or the CA1 pyramidal cell layer to stratum oriens border were used for data analysis . LFPs were measured under anaesthesia with an EPC10 USB amplifier ( HEKA ) in current clamp mode . A wireless amplifier system ( W4 , Multichannel Systems ) was used for LFPs recordings during behaviour >2 days after surgery ( 1 kHz sampling frequency ) . LFPs were analyzed with open-source MATLAB routines ( MathWorks; www . chronux . com ) . Power spectral density and coherence were computed with the ‘multi-taper’ functions of the Chronux toolbox using nine data tapers . Power spectra and envelopes were corrected for the 1/f decline in power over frequency by multiplying power at each frequency by frequency ( ‘normalized power’ ) . This normalization did not affect differences in power between genotypes ( data not shown ) . For cross-correlation analysis MATLAB's xcorr function was used . Gamma envelopes were extracted from the absolute of the Hilbert transform calculated with custom made Python routines . Frontal PrlC slices were cut from 3-to-4 week-old animals as described before ( Sauer and Bartos , 2010 ) . In paired recordings action potentials were induced in the presynaptic FS-IN with short duration current injection ( 2 ms ) while the postsynaptic PC was monitored in voltage-clamp with a holding potential ( Vhold ) set to −70 mV . Only neurons with intersomatic distances <60 µm were targeted for paired recordings . Recording temperature was 32–34°C . Neurons were judged not connected when repeated ( >20 ) trials of action potentials in the presynaptic cell did not elicit a postsynaptic signal in the target neuron as monitored at high-resolution settings with an oscilloscope . Only recordings with an access resistance <25 MΩ were accepted for analysis . Spontaneous EPSCs ( spEPSCs ) were pharmacologically isolated by bath application of SR95531 ( 5 µM ) . Amplitude and time course of mean spEPSCs were analyzed with a threshold-crossing algorithm using Python . Tetrodotoxin ( 0 . 5 µM ) and kynurenic acid ( 4 mM ) were bath-applied to isolate mIPSCs . Unbinned uIPSC amplitude distributions were fit with a multiple probability-compound binomial analysis model of release consisting of the sum of Nr Gaussian functions representing 1 to Nr individual independent release sites , thereby following the procedure of ( Kraushaar and Jonas , 2000 ) . Free parameters for the procedure were Qr , standard deviation of Qr ( SDQr ) , Pr and Nr with initial boundaries set to 10–50 , 3–30 , 0 . 1–0 . 9 and 1–31 , respectively . The step sizes for iteration were 4 , 5 . 4 , 0 . 1 and 3 , respectively . For each parameter combination a probability density function was created as a sum of Gaussians with means at Qr*i and width SDQr*i for i in range 1 to Nr . Each Gaussian was multiplied with the binomial probability of release at i release sites with a given Pr ( Kraushaar and Jonas , 2000 ) . Failures were included as the binomial probability of release at zero sites and set as point zero of the compound probability density function . The best fit of the data was determined with the maximum-likelihood-estimation procedure . To receive confidence intervals of the estimated parameters we performed bootstrap analysis ( Kraushaar and Jonas , 2000 ) . For bootstrap analysis the original amplitudes of an experiment were loaded and bootstrap datasets were created from the n data points as n random picks with replacement . Binomial fitting was then performed on the bootstrap data with maximum-likelihood-estimation . The whole procedure was repeated 100- times for each experiment . Quantal uIPSCs were recorded in pairs with extracellular CaCl2 replaced with SrCl2 ( 5 . 5 mM ) . Asynchronous release was triggered under these conditions by evoking action potential trains ( 10 pulses , 50 Hz ) in the presynaptic neuron . Asychronously released quanta were detected up to 400 ms after the train . Mice were deeply anesthetized by brief isoflurane exposure followed by intraperitoneal injection of pentobarbital ( 0 . 2 ml of 15 mg/ml solution in water ) or urethane ( 2 g/kg in physiological saline ) . Surgery was only commenced after pain reflexes had been abolished . Mice were transcardially perfused with PBS for ∼1–2 min , then with 4% PFA for ∼13–30 min ( ∼3–8 ml/min ) . Brain was removed and stored in PBS overnight and in a subset of experiments in 4% PFA . Horizontal slices ( 50 µm ) were permeabilized in PBS and 0 . 4% Triton X-100 ( 30 min at room temperature ) , blocked in PBS and 0 . 2% Triton X-100 and 4% normal goat serum ( NGS ) for 30 min ( room temperature ) . Primary antibodies were applied overnight at 4°C in PBS and 0 . 1% Triton X-100 and 2% NGS . Slices were washed three times in PBS with 1% NGS ( 10 min ) , incubated in secondary antibody solution ( PBS , 1 . 5% NGS; 2–3 hr at room temperature ) and subsequently washed in PBS ( 2 × 10 min ) . DAPI was applied for 5 min ( 1:1000 in PBS ) . After final washing steps in PBS ( 3 × 10 min ) slices were embedded in Mowiol . Incubation with secondary antibody alone gave no unspecific staining ( data not shown ) . Antibody-labelling was visualized with a confocal microscope ( Zeiss LSM510 or 710 ) . The primary antibodies used were: mouse-anti-PV ( Swant; 1:1000 ) , mouse-anti-calbindin ( Swant; 1:1000 ) , rabbit-anti-SOM ( Peninsula Laboratories; 1:1000 ) , rabbit-anti-VGAT ( Synaptic Systems; 1:1000 ) , rabbit-anti-Disc1 ( Sigma–Aldrich; 1:1000 ) , and rabbit-anti-cFos ( Calbiochem , 1:2000 ) . The secondary antibodies used were: Cy3-goat-anti-rabbit ( Jackson Immunoresearch or Dianova; 1:1000 ) , AlexaFluor647-goat-anti-mouse ( Invitrogen; 1:1000 ) and AlexaFluor488-goat-anti-mouse ( Invitrogen; 1:1000 ) . Cell bodies were counted from maximum intensity projections of z-stacks taken with a 10× objective or from epifluorescence images taken at 5× magnification . To assess PV/VGAT double-positive boutons , colocalized structures were visually identified in 100 × 100 µm regions in 40× single-z plane images within layer 5 of the PrlC . Cells and boutons were counted manually without knowing the genotype . Data were compared with an automated approach in which colocalization was quantified from the same images with custom-made routines written in ImageJ's macro language . Detection thresholds were set to both channels independently with the ImageJ ‘triangle’ method until 5 , 10 , 25 , 50 or 70 brightest percent of pixels remained . Both channels were multiplied with each other to reveal the fraction of colocalized areas . As s control , the same analysis was performed with VGAT images rotated by 90° to determine random colocalizations which were subtracted from colocalizations . Automated analysis gave results that were comparable to manual counting ( data not shown ) . To define bouton density in FS-INs , boutons of intracellularly labelled cells were visually identified as brighter and thicker spots in the biocytin labelled axon . Axon segments ( 60–130 µm length ) were chosen pseudorandomly from the labelled neuron . In vivo , boutons of individual PV-expressing axons ( 27–124 µm length ) were identified in confocal z-stacks as structures colocalizing VGAT . For in vitro and in vivo analyses the length of the traced axon segment was determined in 3D with ImageJ's ‘simple neurite tracer’ . During whole-cell recordings , 1 s long hyperpolarizing and depolarizing current injections were applied ( step size 100 pA; range −100–600 pA ) . INs were identified as FS when the adaptation coefficient was >0 . 6 , determined as the ratio between the first and the last inter-spike-interval . During recordings , cells were filled with 0 . 2% biocytin . To visualize recorded neurons brain slices were fixed overnight in 4% PFA ( 4°C ) , then washed in PBS ( 1 , 10 , 15 , 15 min ) . Blocking was done for 60 min in PBS and 10% NGS . Primary antibody ( mouse-anti-PV or rabbit-anti-PV , Swant , 1:1000 ) was incubated in PBS , 0 . 3% Triton X-100 and 5% NGS ( 24 hr at room temperature ) . The slices were washed again in PBS ( 1 , 10 , 15 , 15 min ) and transferred to secondary antibody solution composed of PBS , 0 . 3% Triton X-100 , 3% NGS and streptavidin conjugated to AlexaFluor647 or 488 ( 1:500 , 24 hr at 4°C ) . Slices were washed in PBS ( 1 , 10 , 15 , 15 min ) and embedded in Mowiol . FS-INs were selected for reconstruction if the signal-to-noise ratio allowed a clear visualization of the axon . A high-resolution confocal image stack was taken with a 40× oil immersion objective ( NA 1 . 4 ) at optimal resolution settings . Semi-automated 3-D reconstruction was performed with the Simple Neurite Tracer plugin of ImageJ . Skeletons of the axon were extracted and analyzed with Lmeasure ( http://cng . gmu . edu:8080/Lm/ ) . Total axon length and branch point number were normalized to the total volume of the stack . Networks of FS-INs and PCs were implemented as conductance-based single compartment models in NEURON 7 . 2 . Three types of chemical connections were modelled: I-I , I-E and E-I , with a distance-dependent connection probability ( Pcon ) following a Gaussian function and a mean Pcon based on our experimental observations ( Table 2 ) . Synaptic events were modelled as Nr quantal conductance changes with an exponential rise and decay ( τrise and τdecay ) and a quantal peak conductance qg obtained from our measured data ( Table 2 ) . Excitatory drive to the network was modelled as irregular trains of Poisson-distributed excitatory postsynaptic conductances ( Table 2 ) with varying frequencies ( range: 0 . 4–1 kHz ) . All presented data are averages of 20 individual simulation runs . The neuronal network model represents a local circuit of the cortex . Networks of FS-INs ( N = 100 for control and N = 65 for Disc1 networks ) and PCs ( default N = 900; 400–1900 in Figure 5D ) were arranged on two concentric circles to represent the columnar organization ( Compte et al . , 2000 ) with minimal distances between two neighbouring cells of 50 µm . FS-INs and PCs were equipped with Hodgkin-Huxley-type conductances to reproduce the FS phenotype in INs ( Wang and Buzsáki , 1996 ) and regular-spiking in PCs ( Hemond et al . , 2008 ) . Synaptic connections were formed randomly following a distance-dependent Gaussian profile ( see Table 2 ) . FS-INs were electrically coupled to four of their nearest eight neighbour INs using a coupling conductance of 0 . 01 nS ( Bartos et al . , 2002 ) . Strength and distribution of electrical coupling among FS-INs have not been experimentally examined in Disc1 and control mice and kept the same in both models ( see Table 2 ) . Mutual E–E synapses among PCs were excluded for simplicity of network design and interpretation . Events were triggered after the presynaptic action potential following a latency which consisted of a constant part ( the release phase; 0 . 5 ms ) plus a distance-dependent part ( the action potential conduction phase; distance in IN–IN spacing × 0 . 05 ms; action potential conduction velocity 0 . 25 ms−1; [Bartos et al . , 2002] ) . For a single simulation run , connections were formed randomly with a random number of quantal contacts picked from the range Nr ± 50% . All cells had random initial membrane potentials ( range: −70 to −60 mV ) and began to receive their excitatory drive at random onset times 0 < t ≤ 50 ms . At t = 100 ms , all synapses were switched on . When changing the excitatory drive to the FS-IN population , the mean input frequency onto every FS-IN was altered . When changing the excitatory drive to the PC population , the mean excitatory drive on single PCs remained constant but the percentage of cells receiving that drive was varied ( Figure 5B–E; Figure 5—figure supplement 1 ) . During the simulation , spike times and the sum of all unitary inhibitory conductances ( LFP analog ) were recorded and the mean firing rate histograms with 1 ms time bin were calculated ( Figure 5B ) . As a measure for the strength of oscillatory activity in the network , LFP analogs and mean firing rate histograms were recorded between 400 < t ≤ 700 ms and were subjected to power spectral density analysis using MATLAB's periodogram algorithm with 1 Hz frequency resolution . The maximum of the resulting power spectrum indicated the prominent oscillation frequency and the power at this maximum plus the two adjacent frequencies ( range: ± 1 Hz ) were used to quantify the mean power of the oscillation ( Figure 5B–E ) . To determine time lag histograms of PC spikes , we first identified the peaks of FS-IN activity in every gamma cycle and then calculated for every PC spike the time lag to its closest FS-IN peak ( Figure 5—figure supplement 1 ) . We described the resulting time lag distribution with a single measure: PC sync = probability of a single PC spike per gamma cycle/variance ( σ2 ) of time lags . Using this definition , PC sync describes the precision of PC spikes in relation to IN activity and PC time locking to the ongoing gamma oscillation . To directly quantify the correlation between FS-IN and PC activity , mean firing rate histograms were obtained for both cell types and a cross-correlation ( X-corr ) analysis was performed using MATLAB's xcorr function ( Figure 5C; Figure 5—figure supplement 1 ) . Maximal cross-correlation was obtained from the peak of the resulting cross-correlogram ( Figure 5—figure supplement 1 ) . For default parameter settings see Table 2 . Statistical significance was tested with a two-tailed Student's t-test , Mann–Whitney U-test or signed rank test . One-way ANOVA was used for multiple comparisons . For nominally scaled data , a χ2 test or Fisher's exact test was used . Data are expressed as mean ± SEM . To measure of effect size of behavioural differences between Disc1 and control mice we used Cohen's d . Custom-made analysis tools are contained in the Source Code file analysiscodes . txt . | Our thoughts and emotions are produced and processed by complex networks of neurons inside our brains . Signals are sent from one neuron to another via chemical messengers , and pass through the neuron as an electrical signal . The electrical signals produced by a brain region often show steady rhythms , or oscillations . In the brains of many people diagnosed with certain mental disorders , such as schizophrenia and major depression , these oscillations are disrupted , but how these changes in rhythm are linked to defects in the networks of neurons behind the electrical activity is not well understood . Studies of a family in Scotland over several decades revealed that a gene called DISC1 was shortened in family members who had been diagnosed with mental illnesses . Recently , scientists have been able to create mice that have mutations that are equivalent to this DISC1 mutation . It is hoped that studying the behavior and neural activity of these mutant mice could lead to a better understanding of human mental disorders . Sauer et al . confirmed that the mutant mice showed depression-related behavior; in experiments that involved trying to escape from hopeless situations , the mutant mice gave up on their escape attempts much sooner than the normal mice . Recording the brain activity of these ‘depressed’ mice showed that the activity of a brain region called the prelimbic cortex was weak and disordered—very much like the brain activity seen in human depression . In particular , two types of brain activity , called theta and low-gamma oscillations , were not synchronized . To determine precisely what causes these abnormal oscillations , Sauer et al . took brain slices from depressed mice , and then stained them with dyes that showed the circuits in the prelimbic cortex more clearly . This revealed that depressed mice had developmental defects in a specific type of inhibitory neuron called fast-spiking interneurons—there were fewer of these cells , and the neurons that were there did not have the correct number of connections to other neurons . Further investigation showed that these neurons had difficulties receiving and releasing the chemical messengers that allow neurons to communicate , and Sauer et al . thought that this might cause the low-gamma oscillation problems . To confirm this theory , Sauer et al . created a computer model that simulated the defective interneurons . The simulations support the theory that the defects in the fast-spiking interneurons cause the abnormal low-gamma rhythms seen in depressed mice . In the future , a better understanding of the defects of inhibitory cells in DISC1 mutants and other mouse models of mental illness might open up new avenues for targeted drug design . As the prelimbic cortex combines inputs from various other brain areas , a further challenge will be to examine whether these inputs influence the activity of the prelimbic cortex and thus contribute to depression-related behavior . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Impaired fast-spiking interneuron function in a genetic mouse model of depression |
Open research data provide considerable scientific , societal , and economic benefits . However , disclosure risks can sometimes limit the sharing of open data , especially in datasets that include sensitive details or information from individuals with rare disorders . This article introduces the concept of synthetic datasets , which is an emerging method originally developed to permit the sharing of confidential census data . Synthetic datasets mimic real datasets by preserving their statistical properties and the relationships between variables . Importantly , this method also reduces disclosure risk to essentially nil as no record in the synthetic dataset represents a real individual . This practical guide with accompanying R script enables biobehavioural researchers to create synthetic datasets and assess their utility via the synthpop R package . By sharing synthetic datasets that mimic original datasets that could not otherwise be made open , researchers can ensure the reproducibility of their results and facilitate data exploration while maintaining participant privacy .
Openly accessible biomedical research data provide enormous utility for science , society , and the economy ( Arzberger et al . , 2004; Munafò et al . , 2017; Murdoch and Detsky , 2013; Piwowar et al . , 2011 ) . With open data , scholars can verify results , generate new knowledge , form new hypotheses , and reduce the unnecessary duplication of data collection ( Asendorpf et al . , 2013; Nosek et al . , 2012 ) . However , the benefits of data sharing need to be considered in light of disclosure risk . Researchers who wish to share data while reducing the risk of disclosure have traditionally used data anonymization procedures , in which explicit identifiers such as names , addresses , and national identity numbers are removed ( Hrynaszkiewicz et al . , 2010 ) . To add additional disclosure protection , particularly sensitive variables ( e . g . , age ) are sometimes aggregated and random noise may be added to the dataset . Despite these anonymization efforts , specific individuals can still be identified in anonymized datasets with high accuracy ( Ohm , 2009; Rocher et al . , 2019 ) . Data aggregation and random noise can also distort the relationships between variables in the dataset ( Purdam and Elliot , 2007 ) , which can interfere with reproducibility and exploratory data analysis . The creation of synthetic datasets can substantially overcome replicability issues , as this method creates a new dataset that mimics an original dataset by preserving its statistical properties and relationships between variables ( Little , 1993; Reiter , 2005b; Reiter , 2005a; Reiter and Raghunathan , 2007; Rubin , 1993 ) . Synthetic datasets also reduce disclosure risk to essentially zero , as no complete casewise record in the new dataset represents a real individual ( Duncan and Elliot , 2011 ) . Synthetic datasets also allow researchers to fit exploratory models in the synthetic datasets , which the data custodians can verify in the original data . Finally , synthetic datasets enable readers and reviewers to better understand the data , as they can recreate the reported analyses and explore data distributions , variance , outliers , and means . Synthetic datasets were originally developed for sharing sensitive population-level data ( for a summary , see Bonnéry et al . , 2019 ) . The use of synthetic data for sharing sensitive information is beginning to emerge in the biobehavioural sciences ( e . g . , Arslan et al . , 2018; Newbury et al . , 2018 ) ; however , this approach is not widely known in the field . Given the benefits of synthetic data , the purpose of this article is to introduce this concept using examples and an accompanying R script . The R script and datasets to reproduce the analyses described in this paper are available online at https://github . com/dsquintana/synthpop-primer ( Quintana , 2019; copy archived at https://github . com/elifesciences-publications/synthpop-primer ) . This website also includes a link to a RStudio Server instance of the primary analysis and results , which recreates the complete computational environment used for this manuscript ( i . e . , the R version and R package versions used ) and facilitates straightforward reproducibility of the analysis described in this article via any modern web browser .
Van Cappellen et al . ( 2016 ) investigated the impact of oxytocin administration on self-reported spirituality and deposited the raw study data online ( https://osf . io/rk2x7/ ) . In a between-participants design , volunteers were randomly assigned to self-administer an intranasal oxytocin ( N = 41 ) or intranasal placebo spray ( N = 42 ) . Approximately forty minutes after receiving the nasal spray , participants were asked “Right now , would you say that spirituality is important in your life ? ” . The reported outcome from an ANCOVA suggested that when accounting for religious affiliation , participants who self-administered the oxytocin nasal spray reported that spirituality was more important in their lives compared to those who self-administered the placebo spray . A synthetic version of the original dataset was created using the syn ( ) function from the synthpop package . A comparison of the four main variables of interest revealed similar distributions between the synthetic and the original datasets and no individual extreme values ( Figure 1A ) . There were also no replicated unique sets of values . A bivariate comparison of self-reported spiritualty ( Figure 1—figure supplement 1 ) and religious affiliation ( Figure 1—figure supplement 2 ) between the nasal spray groups suggested that the counts between the synthesized and original datasets were similar . The relationship between age and self-reported spirituality was also similar between datasets ( Figure 1—figure supplement 3 ) . Altogether , these visualisations were indicative that the synthetic dataset has good general utility . Nasal spray group differences in self-reported spiritualty will be examined using both an independent samples Welch’s t-test and a linear regression model equivalent , with the former required to assess specific utility . This analysis in the original dataset suggested no significant difference in spirituality ratings between the nasal spray groups [t = 1 . 14 , 95% CI ( −0 . 45 , 1 . 63 ) , p=0 . 26] . An equivalent linear regression model yielded the same outcome , as expected . Estimating this linear model in the synthesized dataset revealed the same p-value outcome ( t = −1 . 12 , p=0 . 26 ) . The lack-of-fit test comparing the models generated in the original and synthetic datasets was not statistically significant , X2 ( 2 ) =0 . 01 , p=0 . 995 . This suggests that the method used for synthesis retained all the relationships between variables that influenced the model fitting parameters . The standardized difference of the nasal spray condition coefficient between the synthesized and the observed data for this t-test was 0 . 003 , which was not statistically significant ( p=0 . 998 ) . A comparison of confidence intervals revealed 99 . 94% CI overlap between the synthetic and original datasets ( Figure 1B ) . Overall , these results indicate that the model from the synthesized data demonstrates high specific utility for this t-test . This particular outcome was not reported in the original article; however , this provides a demonstration of how synthetic data can be used for data exploration . The analysis script underlying exploratory analysis can be shared with the owners of the original dataset , for the easy verification of the analysis . In this case , applying this model to the synthetic data produces almost precisely the same results as the original dataset . Next , let’s explore the correlation between age and self-reported spirituality . A Pearson correlation test revealed no statistically significant correlation between age and self-reported spirituality [r = 0 . 04 , 95% CI ( −0 . 19 , 0 . 26 ) , p=0 . 75] , which is the same result as the linear model equivalent . Estimating this model using the synthetic dataset revealed a similar Pearson’s r value and a non-significant result ( r = 0 . 08 , p=0 . 5 ) . The lack-of-fit test comparing the models generated in the original and synthetic datasets was not statistically significant , X2 ( 2 ) =0 . 13 , p=0 . 94 . Moreover , the test of the standardized differences between the synthetic and original data for the relationship coefficient was not statistically significant ( p=0 . 72 ) and there was 90 . 8% CI overlap between the synthetic and observed data ( Figure 1C ) , suggesting high specific utility . Finally , let’s explore the main reported outcomes from this study , that oxytocin increased self-reported spirituality controlling for religious affiliation . First , the original outcome was verified [F ( 1 , 75 ) =4 . 87 , p=0 . 03] and then the analysis was structured as a linear regression model . This analysis yielded the same p-value of . 03 , which was associated with a t-statistic of −2 . 2 . Estimating this model using the synthetic dataset revealed similar t-statistic and a p-value that was on the border of statistical significance ( t = −1 . 8 , p=0 . 07 ) . The lack-of-fit test for the full model was not statistically significant , X2 ( 3 ) =0 . 91 , p=0 . 82 . The test of the standardized differences between the synthetic and original data for the nasal spray coefficient was not statistically significant ( p=0 . 63 ) and there was 87 . 8% overlap between the synthetic and observed data ( Figure 1D ) . Although the synthetic model nasal spray coefficient was not statistically significant , like the original model , this matters little given the considerable overlap between the confidence intervals of the synthetic and original models . It is worth a reminder at this point that just because one coefficient is significant and the other is not , this does not necessarily mean that these coefficients are significantly different from each other ( Gelman and Stern , 2006 ) . The primary interest in the comparison of synthetic and original models is effect size estimation and confidence interval overlap . The test of the standardized differences between the synthetic and original data for the religious affiliation coefficient was not statistically significant ( p=0 . 49 ) and there was 82 . 3% overlap between the synthetic and observed data ( Figure 1D ) . Overall , these results are indicative of high specific utility for the overall synthesized model and its coefficients . Sociosexual orientation is described as an individual’s propensity to participate in uncommitted sexual relationships . Given the personal nature of sociosexual orientation , this a good example of the type of sensitive data that individuals may be hesitant to share in some cases , thus demonstrating the benefit of releasing of synthetic data . Jones and DeBruine , 2019 collected sociosexual orientation data from 9627 individuals using a revised version of the sociosexual orientation inventory ( Penke and Asendorpf , 2008 ) both in the laboratory and online , along with data on self-rated attractiveness and basic demographic information . Fourteen variables were synthesized from the original dataset , which has been archived online ( https://osf . io/6bk3w/ ) . The synthetic data demonstrated good general utility , as the distributions of variables were comparable between the original and synthetic datasets ( Figure 2A ) . A model was fitted to examine if self-rated attractiveness , data collection location ( laboratory or online ) , and age predicted the number of times someone has had sexual intercourse on only a single occasion with another individual . Both self-rated attractiveness ( t = 13 . 64 , p<0 . 001 ) and age ( t = 27 . 69 , p<0 . 001 ) were statistically significant predictors , whereas location was not a significant predictor ( t = 0 . 05 , p=0 . 96 ) . Estimating this linear model in the synthesized dataset revealed relatively similar t-statistic outcomes ( self-rated attractiveness: t = 14 , p>0 . 001; age: t = 27 . 51 , p<0 . 001; location: t = 1 . 12 , p=0 . 26 ) . The lack-of-fit test was not statistically significant , [X2 ( 4 ) =1 . 5 , p=0 . 83] , suggesting that the method used for synthesis retained all the relationships between variables that influenced the parameters of the fit . The standardized difference of the self-rated attractiveness coefficient between the synthesized and the observed data was 0 . 32 , which was not statistically significant ( p=0 . 75 ) . A comparison of confidence intervals revealed 91 . 8% CI overlap between the synthetic and original datasets ( Figure 2B ) . The standardized difference of the age coefficient between the synthesized and the observed data for this t-test was −0 . 28 , which was not statistically significant ( p=0 . 78 ) . A comparison of confidence intervals revealed 92 . 9% CI overlap between the synthetic and original datasets ( Figure 2B ) . The standardized difference of the location coefficient between the synthesized and the observed data for this t-test was 1 . 08 , which was not statistically significant ( p=0 . 28 ) . A comparison of confidence intervals revealed 72 . 4% CI overlap between the synthetic and original datasets ( Figure 2B ) . Overall , these results indicate that this model from the synthesized data demonstrated high specific utility . A comparison of the original and synthetic datasets revealed 213 replicated individuals ( 2 . 3% of the total sample ) . To reduce disclosure risk , these replicated values were removed and models were refitted to examine the effects of removal on outcomes . The lack-of-fit test between the original model and synthetic model with replicated individuals was not statistically significant , X2 ( 4 ) =2 . 8 , p=0 . 6 . The refitted model coefficients yielded similar results to the model derived from the original data ( self-rated attractiveness: standardized difference = 0 . 3 , p=0 . 77 , CI overlap = 92 . 4%; age: standardized difference = −0 . 74 , p=0 . 46 , CI overlap = 81 . 2%; location: standardized difference = 0 . 95 , p=0 . 34 , CI overlap = 75 . 8%; Figure 2C ) . Therefore , reducing the risk of disclosure by removing replicated individuals in the synthetic dataset maintains specific utility , in this case . Heart Rate Variability ( HRV ) is a non-invasive measure of autonomic cardiac control ( Akselrod et al . , 1981 ) and thought to be positively correlated with fitness level ( Dixon et al . , 1992 ) . The Root Mean Square of Successive Differences ( RMSSD ) is a commonly used HRV measure , however , its distribution tends to be positively skewed ( e . g . , Kobayashi et al . , 2012 ) . Moreover , missing data are common in HRV investigations due to equipment malfunction . To demonstrate the effects of the distribution pattern of HRV ( normal , low skew , high skew ) , and missing data ( none , 5% , 20% ) for a range of sample sizes ( 40 , 100 , 10000 ) , twenty-seven data sets with four variables ( heart rate , weight , fitness level , and HRV ) were simulated for the creation of synthetic datasets , which included outliers ( Supplementary file 1 ) . For all datasets , HRV and fitness level were modelled to have a relationship that is typically associated with a medium effect size ( r = 0 . 3 ) . The specific utility of synthetic datasets was examined by comparing the relationship between HRV and fitness in each synthetic dataset to its respective original dataset . None of the lack-of-fit tests were statistically significant ( Supplementary file 1 ) , suggesting that the method used for synthesis retained all the relationships between variables that influenced the parameters of the fit . A comparison of confidence intervals revealed strong overlap between the synthetic and original datasets for most ( but not all ) of the 27 analyses and none of the standardized coefficient differences between the synthesized and the observed datasets were statistically significant ( all p’s > 0 . 05; Figure 3 , Figure 3—figure supplements 1–2 ) . However , the overlap between the synthetic and original models were on the border of statistical significance when synthesizing data with a low skew in the simulated samples with 10 , 000 cases ( Supplementary file 1; Figure 3—figure supplement 2 ) . All 27 synthetic datasets also demonstrated good general utility , regardless of the parameters ( Figure 3—figure supplements 3–5 ) . Thus , synthetic dataset generation in synthpop seems to be relatively robust against differences in sample size , missingness , and skew in these simulated samples , however , there were indications of poorer performance in some of larger datasets with 10 , 000 ( Supplementary file 1 ) . Altogether , it is crucial that general and specific utility is assessed for each synthesised dataset , as it is difficult to predict before synthesis how well the procedure will perform .
Researchers need to consider the trade-off between the risk of identification and the utility of open datasets when deciding whether to make their research data openly available . Open data can provide substantial utility , but this may expose research participants to the risk of identification . Conversely , closed datasets decrease the risk of disclosure to essentially zero , but have almost no public utility . The generation of synthetic datasets provides an appealing compromise , as synthetic data can offer comparable levels of utility as the original datasets while substantially reducing disclosure risk . The adoption of synthetic datasets in the biobehavioural sciences will improve reproducibility and secondary data exploration , as it will facilitate the sharing of data that would otherwise not be made available . Study participants are generally in favour of researchers sharing their deidentified data ( Ludman et al . , 2010; Mello et al . , 2018 ) . Thus , when planning future studies researchers should include data sharing provisions when receiving participant consent ( Taichman et al . , 2016 ) . Obtaining updated consent to share data from participants who have not provided this when originally participating in a study can be resource intensive . Some have suggested that sharing deidentified datafiles should not require new re-consent from participants ( Taichman et al . , 2017 ) , but as mentioned above , many commonly used data deidentification approaches may not sufficiently reduce disclosure risk ( Ohm , 2009; Rocher et al . , 2019 ) . In some circumstances , datasets may include extremely sensitive information that is difficult to anonymise . For instance , Arslan et al . ( 2018 ) collected highly sensitive data examining the role of ovulatory changes on sexual desire and behavior in women but did not request consent from participants to share data considering valid privacy concerns . Instead , a synthetic version of the dataset was created using synthpop and made available on a publicly accessible repository . Releasing this synthetic dataset provides considerable utility , as other researchers can verify the analysis and fit novel models using this dataset , which can be confirmed by the data custodians with the original data . An additional step of removing individuals that have been fully replicated in the synthesized data set can further reduce disclosure risk , without necessarily reducing general or specific utility . Therefore , synthetic data can offer a valuable solution for sharing data collected under conditions where participants did not specifically provide consent for sharing data ( and where re-consent is impractical ) , as well as for situations in which a dataset contains especially sensitive information . In addition to the verification of results and hypothesis generation , synthetic datasets can also benefit the training of machine learning algorithms in research areas with a dearth of data , such as rare condition research ( Ekbatani et al . , 2017; Sabay et al . , 2018 ) , via the creation of additional synthetic datasets that closely match real datasets . One criticism of sharing raw data is that research groups would not have the first opportunity to analyse the data and report outcomes to their satisfaction ( Lo , 2015; Ross et al . , 2012 ) . It has been recommended that secondary data analysts should seek collaborations with teams that collected the original data in recognition of their investment in collecting the data ( Taichman et al . , 2016 ) , but this is difficult to enforce in practice . To make meaningful inferences with synthetic data , secondary data analysts need to verify their synthetic models against models from the original data , which is in the possession of the original authors who can verify these inferences ( Reiter et al . , 2009 ) . This would increase the likelihood of co-authored collaborations , at least compared to the status-quo in which secondary analysts could publish their results without necessarily collaborating with the original authors . Thus , open synthetic data provide an opportunity for secondary analysists scholars to fit models that the original authors may not have considered , while also encouraging them to collaborate with the original authors to verify their models in the real dataset . Of course , secondary analysts could still report results from synthetic datasets without verification from the primary authors , but it would need to be made explicit that analyses were conducted on a synthetic dataset , and generated models may not necessarily mirror the models generated from the original dataset . Journals have adopted a spectrum of public data archiving ( PDA ) policies , ranging from the policies that data should be made “available upon request” all the way to mandated data deposition in peer-reviewed journals dedicated to open data ( Sholler et al . , 2019 ) . While an “available upon request” PDA policy is better than no policy at all ( Stodden et al . , 2018 ) , such datasets are often difficult to retrieve in practice as corresponding authors can become unreachable or original datasets are lost ( Couture et al . , 2018; Stodden et al . , 2018; Wicherts et al . , 2006 ) . Sharing data with published papers would remove these impediments for accessing data , with synthetic data offering a solution for when it is not possible to share the original dataset due to disclosure concerns . Despite the benefits of synthetic datasets , this approach is not without limitations . First , it is possible for synthetic data to have poor general and specific utility , which would diminish the benefits of sharing in terms of reproducibility and secondary data exploration . While a synthetic dataset with poor utility would still provide a valuable guide for reproducing reported analyses , these are likely to provide substantially different estimates and exploratory analyses may not produce accurate models . Second , current synthetic data methods limit the types of statistical inference that can be performed on synthetic data to linear models . In practice , this means that only linear models can be for comparison in order to demonstrate specific utility . Of course , scholars are free to perform any type of analysis on the synthetic data , which should provide approximately the same outcome as the original data as long as the synthetic data offer good general utility . Third , as mentioned above , the risk of identity disclosure from synthetic datasets is negligible but this only holds under two conditions: that none of the complete synthetic data records match with the original data and that there are no extreme single individual values in the dataset that can be linked to an individual ( Drechsler , 2011; Duncan and Elliot , 2011 ) . Therefore , to reduce disclosure risk and the possibility that participants will recognise themselves in the dataset , complete matching records ( i . e . , when all variables in the original dataset for a case matches a case in the synthetic dataset ) should be identified and removed . Additionally , in the case of categorical variables with only a few observations , scholars should consider collapsing these into another category ( e . g . , if there are only a few observations in an age band of 70–79 years old , this can be collapsed into the previous age band of 60–69 years old ) . If there are uncommon continuous values above or below a certain threshold , it may be prudent to collapse these into another category or creating a new category ( e . g . , top-coding a new ‘70+’ age variable for any age equal to or above 70 ) . While recoding may lead to synthetic datasets with less utility , this approach might be required to reduce disclosure risk , something that data synthesizers will have to carefully consider in light of the sensitivity of the dataset along with national laws and guidelines . When the creation of synthetic datasets for disclosure control was first proposed in the early 1990s , it was considered “rather radical” at the time ( pg . 461; Rubin , 1993 ) . Researchers have continued improving this method since these initial proposals ( Reiter , 2005b; Reiter , 2005a; Reiter and Raghunathan , 2007 ) , but only more recently has an easy-to-implement tool for creating synthetic data become available . The synthpop R package enables researchers to generate and share synthetic datasets that mimic original datasets with sensitive information . Importantly , the use synthetic data will improve the reproducibility of biobehavioral research and help generate novel hypotheses for future research ( Bonnéry et al . , 2019 ) . | It is becoming increasingly common for scientists to share their data with other researchers . This makes it possible to independently verify reported results , which increases trust in research . Sometimes it is not possible to share certain datasets because they include sensitive information about individuals . In psychology and medicine , scientists have tried to remove identifying information from datasets before sharing them by , for example , adding minor artificial errors . But , even when researchers take these steps , it may still be possible to identify individuals , and the introduction of artificial errors can make it harder to verify the original results . One potential alternative to sharing sensitive data is to create ‘synthetic datasets’ . Synthetic datasets mimic original datasets by maintaining the statistical properties of the data but without matching the original recorded values . Synthetic datasets are already being used , for example , to share confidential census data . However , this approach is rarely used in other areas of research . Now , Daniel S . Quintana demonstrates how synthetic datasets can be used in psychology and medicine . Three different datasets were studied to ensure that synthetic datasets performed well regardless of the type or size of the data . Quintana evaluated freely available software that could generate synthetic versions of these different datasets , which essentially removed any identifying information . The results obtained by analysing the synthetic datasets closely mimicked the original results . These tools could allow researchers to verify each other’s results more easily without jeopardizing the privacy of participants . This could encourage more collaboration , stimulate ideas for future research , and increase data sharing between research groups . | [
"Abstract",
"Introduction",
"Methods,",
"materials,",
"and",
"results",
"Discussion"
] | [
"medicine",
"tools",
"and",
"resources"
] | 2020 | A synthetic dataset primer for the biobehavioural sciences to promote reproducibility and hypothesis generation |
Self-organization of discrete fates in human gastruloids is mediated by a hierarchy of signaling pathways . How these pathways are integrated in time , and whether cells maintain a memory of their signaling history remains obscure . Here , we dissect the temporal integration of two key pathways , WNT and ACTIVIN , which along with BMP control gastrulation . CRISPR/Cas9-engineered live reporters of SMAD1 , 2 and 4 demonstrate that in contrast to the stable signaling by SMAD1 , signaling and transcriptional response by SMAD2 is transient , and while necessary for pluripotency , it is insufficient for differentiation . Pre-exposure to WNT , however , endows cells with the competence to respond to graded levels of ACTIVIN , which induces differentiation without changing SMAD2 dynamics . This cellular memory of WNT signaling is necessary for ACTIVIN morphogen activity . A re-evaluation of the evidence gathered over decades in model systems , re-enforces our conclusions and points to an evolutionarily conserved mechanism .
In the early embryo , secreted morphogens regulating a limited number of signaling pathways carry the task of instructing dynamic and coordinated cell differentiation across the developing tissue . In the day 6 . 5 mouse embryo , a hierarchy of signaling mediated by BMP , WNT , and NODAL carries the positional information in the epiblast leading to axis formation and germ layer specification ( Arnold and Robertson , 2009 ) . Each of these morphogens induces the expression of the next ligand ( BMP signaling induces WNT expression , and WNT signaling induces NODAL expression ) , and all induce the expression of their own inhibitors . To what extent this signaling cascade is involved in human gastrulation can now be investigated using in vitro models of early human embryos derived from human embryonic stem cells ( hESCs ) . We have previously shown that the first step of this signaling hierarchy is conserved in humans and that in response to BMP4 , hESCs grown in geometrically confined colonies , self-organize to induce and pattern embryonic and extra-embryonic germ layers . Ectoderm was specified at the center of circular colonies , extra-embryonic tissue at the edge and mesendoderm in between . Molecular signatures of gastrulation , such as the induction of SNAIL and activation of pERK could also be observed ( Warmflash et al . , 2014 ) . Consistent with the conservation of the second step of the signaling hierarchy , BMP4 , which signals through SMAD1/5/8 , was shown to induce the expression of WNT ligands , which can also induce the emergence of a primitive streak and the self-organization of embryonic germ layers in this system ( Etoc et al . , 2016; Martyn et al . , 2018 ) . We , therefore , termed this system a human gastruloid . The third pathway , SMAD2/3 , which signals on behalf of ACTIVIN and NODAL ligands , has an intriguing role . On the one hand , SMAD2/3 signaling is required for hESC pluripotency maintenance ( James et al . , 2005; Vallier et al . , 2005 ) . On the other hand , the ACTIVIN/NODAL pathway is necessary for fate specification , acting as a morphogen to pattern the blastula of vertebrate embryos from the amphibian to the mouse ( McDowell and Gurdon , 1999; Robertson , 2014 ) . The ACTIVIN/NODAL pathway is also necessary for the self-organization of human gastruloids as inhibition of SMAD2/3 signaling blocks primitive streak formation and mesendoderm induction by BMP4 and eliminates anterior mesendoderm fates induced by WNT3A ( Warmflash et al . , 2014; Martyn et al . , 2018 ) . Furthermore , co-presentation of ACTIVIN with WNT3A to micropatterned hESC colonies leads to the induction of the organizer specific marker GOOSECOID , and a functional human primitive streak , which when grafted into chick embryos induces the formation of a secondary axis ( Martyn et al . , 2018 ) . Despite these findings , the mechanism by which SMAD2/3 signaling can specify both hESC pluripotency and differentiation remains perplexing , and a number of key questions remain unanswered . Among them are: how can a single signaling pathway carry these two opposite functions before and after the onset of gastrulation ? To what extent do the dynamics of SMAD signaling affect these readouts ? Finally , to what extent do cells have a memory of past signaling ? In this study we aimed to provide highly quantitative answers to these questions by following the dynamics of TGFβ signaling during the self-organization of human gastruloids . Imaging of hESC reporter lines engineered by CRISPR/Cas9-mediated tagging of SMAD1 , SMAD2 , or SMAD4 demonstrated that each branch of the pathway has distinct signaling dynamics . In response to ACTIVIN , SMAD2 displayed a dramatic transient nuclear translocation , which stood in sharp contrast to the stable BMP4-induced SMAD1 response . ACTIVIN stimulation did induce transient mesendodermal gene transcription , which correlated with SMAD2 dynamics . This induction , however , was not sustained and cells reverted back to pluripotency at later times . Interestingly , pre-presentation of WNT3A to the cells , while not changing SMAD2 dynamics or expression of the pluripotency markers , stabilized the subsequent transcriptional response to ACTIVIN to maintain mesendodermal fates . This implies an unexpected ability of human embryonic stem cells to record their signaling history without overt changes in fate .
A functional SMAD2/3 pathway is necessary for both maintenance of pluripotency , as well as differentiation and self-organization of human gastruloids downstream of BMP4 and WNT3A ( Warmflash et al . , 2014; Martyn et al . , 2018 ) . In the context of BMP4 induced differentiation , treatment with a small molecule inhibitor of SMAD2/3 signaling , SB431542 ( SB ) , eliminated all mesendodermal fates , as indicated by the loss of BRA and SOX17 positive cells , leaving only the putative extra-embryonic and ectoderm fates , marked by CDX2 and SOX2 expression , respectively ( Figure 1A ) ( Warmflash et al . , 2014 ) . Presentation of the downstream morphogen WNT3A induced primitive streak formation at the colony border , and addition of SB in this context removed the SOX17 positive mesendodermal population ( Figure 1A ) ( Martyn et al . , 2018 ) . In order to ask whether gastruloid self-organization can be initiated at the ACTIVIN/NODAL point in the signaling hierarchy , we stimulated micropatterned colonies grown in conditioned media with high concentrations of ACTIVIN A ( referred to as ACTIVIN throughout this work ) . In contrast to stimulation with BMP4 and WNT3A , no differentiation was observed after 48 hr with ACTIVIN alone ( Figure 1B ) . Surprisingly , the lack of differentiation in response to ACTIVIN was not due to a lack of signal sensing , as an increase in nuclear SMAD2/3 was detected by immunofluorescence at the colony border after one hour of stimulation ( Figure 1C–D ) . We conclude that while SMAD2/3 signaling is necessary for mesendoderm induction downstream of BMP4 and WNT3A in human gastruloids , it is not sufficient to induce it . The inability of ACTIVIN/SMAD2 to induce differentiation stood in stark contrast with the ability of BMP/SMAD1 to induce gastruloid self-organization , including patterning of the mesoderm and endoderm germ layers . This contrast in activity of the two branches of the TGFβ pathway prompted us to assess possible differences in their signaling dynamics . We used CRISPR/Cas9 genome engineering on RUES2 to fluorescently tag the N-terminus of the endogenous receptor-associated , R-SMAD1 , with tagRFP ( RUES2-RFP-SMAD1 ) and R-SMAD2 with mCitrine ( RUES2-mCit-SMAD2 ) ( Figure 2A and Figure 2—figure supplement 1 ) . As activation of the pathway leads to the binding of SMAD1 and 2 to the co-SMAD , SMAD4 , before nuclear translocation and regulation of gene expression , a GFP-tagged SMAD4 line ( RUES2-GFP-SMAD4 ) was also included ( Figure 2A ) ( Nemashkalo et al . , 2017 ) . Each line was also transfected with ePiggyBac transposable elements carrying a nuclear marker ( H2B-mCitrine or H2B-mCherry ) in order to analyze the response of individual cells ( Figure 2—figure supplement 2A ) . N-terminal SMAD fusion proteins were shown to function similarly to endogenous proteins in biochemical and cell-based assays ( Schmierer and Hill , 2005 ) . Additionally , the SMAD response dynamics measured with our reporter lines , matched the behavior by of the endogenous proteins measured by immunofluorescence and western blotting as discussed below . We began our dynamic studies of the two branches of TGFβ signaling in RUES2 cells grown on micropatterned colonies in chemically-defined medium ( TeSR-E7 ) , which is a version of serum-free E8 medium that can maintain hESCs and that lacks any TGFβ ligands ( Chen et al . , 2011 ) . Therefore , in E7 the exogenous TGFβ levels could be precisely controlled . In response to BMP4 , we detected an increase in the SMAD1 nuclear signal that was stable over 12 hr ( Figure 2B–D , Figure 2—figure supplement 2B , Figure 2—videos 1 and 2 ) . The SMAD1 response was observed at the colony edge , consistent with our previous immunofluorescence results and our discovery that the TGFβ receptors are localized to the apical surface only at the border of the colony ( Warmflash et al . , 2014; Etoc et al . , 2016 ) . In our previous work we quantified the SMAD1 response detected by immunofluorescence as the number of positive nuclei within the colony ( Etoc et al . , 2016 ) . Although our reporter line allows us to track the level of nuclear SMAD1 in individual cells , as opposed to a binary on/off quantification , we obtain similar results to those shown in Figure 2D with a binary analysis of the RFP-SMAD1 reporter line ( Figure 2—figure supplement 2C–D ) . In response to ACTIVIN , SMAD2 on the other hand , responded only transiently: a pulse of nuclear translocation over the first 1–2 hr was followed by a gradual decrease over the next 6 hr ( Figure 2E–G , Figure 2—figure supplement 2E , and Figure 2—video 3 ) . Interestingly , the long-term SMAD2 nuclear level did not return completely to the pre-stimulus level . As in the case of SMAD1 , the SMAD2 response was highest at the colony edge , again consistent with the apical localization of the receptors ( Etoc et al . , 2016 ) . In order to study signaling dynamics at the single-cell level and to eliminate modifier influences on both SMAD branches coming from neighboring cells within the micropatterned colony , we performed the same experiment on dissociated cells grown under regular culture conditions ( Figure 3A ) . The SMAD1 response to BMP was stable and the level of nuclear RFP-SMAD1 was dependent on the BMP4 ligand concentration in agreement with previous single-cell immunofluorescence measurements ( Figure 3B–C ) ( Etoc et al . , 2016 ) . The increase in the average nuclear signal as a function of time resulted from nearly all cells responding to the added BMP4 , which is evident in the shift in the histogram of RFP-SMAD1 nuclear intensity ( Figure 3—figure supplement 1A ) . The SMAD2 response to ACTIVIN was again transient with a nuclear signal that remained elevated at longer times as demonstrated by our reporter line and by immunofluorescence and western blot analysis of the parental RUES2 line ( Figure 3D–E , Figure 3—figure supplement 1B–C ) . This is consistent with other data based on immunofluorescence ( Heemskerk et al . , 2017 ) . Although , the adaptive response is somewhat more pronounced in our experimental setup . As with the SMAD1 response , the average SMAD2 response resulted from nearly all cells responding with similar dynamics , which is evident in the shift up and down in the histogram of SMAD2 nuclear intensity ( Figure 3—figure supplement 1D ) . The SMAD2 peak response displayed a strong sigmoidal dependence on ACTIVIN concentration ( Figure 3—figure supplement 1E ) . However , the post-stimulation baseline , defined as the average SMAD2 nuclear-to-cytoplasmic ratio at T > 8 hr after ACTIVIN addition , as well as the time scale of the transient response , was not dependent on the ACTIVIN dose above 0 . 5 ng/mL ( Figure 3—figure supplement 1E–F ) . This transient SMAD2 signaling behavior did not result from depletion of ACTIVIN from the medium , as culture medium recovered from cells that were incubated with ACTIVIN for 12 hr , still induced a SMAD2 response when presented to unstimulated cells ( Figure 3—figure supplement 1G ) . SMAD4 followed the dynamics of the relevant R-SMADs following presentation of BMP4 or ACTIVIN ( Figure 3F–G , Figure 3—figure supplement 1H ) . The stable and adaptive responses of SMAD4 to BMP and ACTIVIN , respectively , is consistent with recent experiments that also utilized this reporter line ( Nemashkalo et al . , 2017; Heemskerk et al . , 2017 ) . Thus , in response to their activating ligands , the two branches of the TGFβ pathway display distinct dynamics of signal transduction for both the R-SMAD and the co-SMAD . This also demonstrates that modifying signals do not influence the dynamics of the response at the edge of the micropatterned colonies . We have previously shown that BMP4 signaling induces a sustained transcriptional response leading to gastruloid differentiation ( Warmflash et al . , 2014; Etoc et al . , 2016 ) . This is consistent with the stable nature of SMAD1 signaling presented above . The adaptive behavior of SMAD2 signaling prompted us to ask whether the short SMAD2 signaling peak was sufficient to elicit a transcriptional response and fate changes in RUES2 cells exposed to ACTIVIN . RNA-seq analysis was performed on dissociated cells cultured in E7 and E7 +ACTIVIN at 1 , 2 . 5 , 4 , 8 and 12 hr following stimulation . 3529 genes showed a change in expression level of at least two-fold during the experimental time course . They fell into three distinct groups . The first , which consisted of the majority of transcripts ( 2 , 956 ) , peaked at 2 . 5 hr and declined at later time points ( Figure 4A , magenta box ) . This group matched the timing of the transient SMAD2 response and it included key regulators of mesendodermal differentiation , such as EOMES , HHEX , GATA2 , and GATA3 ( Figure 4—source data 1 ) ( Teo et al . , 2011; Loh et al . , 2014 ) . The second group , which consisted of 452 transcripts , showed stable induction ( Figure 4A , orange box ) . This group included genes expressed during pluripotency , such as NANOG , NODAL , LEFTY1 , LEFTY2 and SMAD7 ( Figure 4—source data 2 ) ( Sato et al . , 2003 ) . Finally , the third group , which consisted of 121 transcripts , represented genes that were stably or transiently down regulated upon ACTIVIN presentation and included genes that are involved in signaling pathways not previously associated with pluripotency or differentiation , such as insulin signaling and cAMP response ( Figure 4A , gray box and Figure 4—source data 3 ) . These results suggest that cells transiently activate differentiation in response to ACTIVIN . Examination of the signaling hierarchy involved in gastruloid self-organization revealed the presence of feedback loops at all three levels . ACTIVIN induced the expression of its own ligands and inhibitors , as well as those of the BMP and WNT pathway ( Figure 4—source data 1–2 ) . However , despite the induction of the expression of the ligands and inhibitors , the overall threshold of signaling was not sufficient to induce and maintain mesendodermal fates from either the BMP or the WNT pathway . We took two additional approaches to evaluate the transient and stable transcriptional responses following ACTIVIN treatment . First , we performed motif enrichment analysis on our gene groups . We selected 10 transcription factors that regulate primitive streak and mesendodermal differentiation: MIXL1 , LEF1 , BRACHYURY ( BRA ) , GATA6 , FOXH1 , FOXA1 , FOXA2 , GOOSECOID ( GSC ) , SOX17 , and EOMES , and asked if their binding sites are enriched in the promoter region of the genes belonging to each of the dynamic groups . In support of our hypothesis , the motifs were significantly enriched only in the transiently expressed genes of group one and not within group 2 or 3 ( Figure 4—figure supplement 1A and Figure 4—source data 4 ) . This suggests that the gene regulatory network activated during the peak of SMAD2 signaling is associated with mesendodermal differentiation . We additionally compared our gene groups with tissue specific genes identified in isolated endoderm , mesoderm , and ectoderm/epiblast tissue from E7 . 5 mouse embryos ( Lu et al . , 2018 ) . Although all groups contained some significant enrichment of genes from one or more of the mouse germ layers , group one displayed the most significant enrichment of endodermal genes ( Figure 4—figure supplement 1B and Figure 4—source data 5 ) . Overall our data demonstrate that during the peak of SMAD2 nuclear accumulation , hESCs are en route for differentiation . However , the mesendodermal differentiation program is not maintained and cells return to pluripotency . In the mouse embryo , SMAD3 is dispensable for early development as demonstrated by the fact that SMAD3 knockout mice make it to adulthood ( Zhu et al . , 1998; Yang et al . , 1999; Datto et al . , 1999 ) . In the absence of SMAD2 in the epiblast , SMAD3 can mediate some mesoderm induction during gastrulation . However , more anterior mesendodermal lineages are completely eliminated and the embryos fail at gastrulation suggesting a critical role for SMAD2 in this process ( Vincent et al . , 2003; Dunn et al . , 2004 ) . In order to decipher whether the transient SMAD2 response is sufficient to drive the transcriptional program downstream of ACTIVIN presentation in hESCs , we generated two independent RUES2 SMAD3 knockout lines ( RUES2-SMAD3-/- ) using CRISPR-Cas9 mediated genome editing ( Figure 4—figure supplement 2A–B ) . These lines maintained expression of pluripotency markers NANOG , OCT4 , and SOX2 , which is consistent with the previous finding that SMAD2 , but not SMAD3 , regulates NANOG expression to promote pluripotency in hESCs and mouse epiblast stem cells ( Figure 4—figure supplement 1C ) ( Sakaki-Yumoto et al . , 2013 ) . In response to ACTIVIN , RUES2-SMAD3-/- cells displayed a transcriptional response identical to the parental RUES2 line for both pluripotency- and mesendoderm-associated ACTIVIN target genes ( Figure 4B–C ) . Although our results cannot rule out possible redundancy between SMAD2 and SMAD3 , we conclude that SMAD3 is not required for maintenance of pluripotency or the transcriptional response dynamics following ACTIVIN presentation . We have shown that SMAD2 nuclear levels are transient following a step input of ACTIVIN , with a long-term baseline that remains elevated relative to the pre-stimulus level . We next asked if the increase in the SMAD2 baseline after the peak response is required for the maintenance of pluripotency . In order to address this question , RUES2-mCit-SMAD2 cells were treated with SB in order to inhibit ACTIVIN signaling , 8 hr after stimulation , and analyzed for their ability to maintain pluripotency . SB treatment led to a decrease in SMAD2 nuclear-to-cytoplasmic levels back to the unstimulated baseline ( Figure 5A ) . As observed previously , presentation of SB led to a loss of pluripotency , as indicated by the loss of NANOG expression in RUES2-mCit-SMAD2 ( Figure 5B ) ( James et al . , 2005; Vallier et al . , 2005; Xu et al . , 2008 ) . Analysis of the parental RUES2 line confirmed the loss of pluripotency under the same experimental conditions ( Figure 5—figure supplement 1A ) . We conclude that the elevated baseline at the tail of the SMAD2 response is ligand dependent and responsible for maintaining the pluripotency program long-term . The fact that the SMAD2 post-stimulation baseline is the same regardless of ACTIVIN concentration above 0 . 5 ng/mL , suggests that pluripotency is insensitive to graded ligand levels above this threshold ( Figure 3E ) . To test this hypothesis , we treated single cells with different levels of ACTIVIN and compared the expression of NANOG , OCT4 , and SOX2 after 2 days of stimulation . Expression of all three markers was similar at three different concentrations of added ACTIVIN ( 1 , 10 , and 100 ng/mL ) and expression of NANOG and OCT4 was elevated relative to the –ACTIVIN condition ( Figure 5C–D and Figure 5—figure supplement 1B; results from additional hESC line RUES1 shown in Figure 5—figure supplement 1C ) . To test for zero ACTIVIN/SMAD2 input we presented SB for the same amount of time . Pluripotency was not maintained under these conditions as demonstrated by down-regulation of NANOG and OCT4 below the –ACTIVIN condition ( Figure 5C–D and Figure 5—figure supplement 1B–C ) . Toxicity as a result of SB treatment or increased ACTIVIN levels was not observed ( Figure 5—figure supplement 1D ) . This confirms the insensitivity of the pluripotent state to graded ACTIVIN . Even at the highest ligand concentrations pluripotency was maintained and no differentiation was observed . In order to compare immunofluorescence data across biological replicates , which may differ in absolute intensity values , we computed the Kolmogorov-Smirnov ( KS ) distance between each marker’s cumulative distribution function ( CDF ) measured under different conditions to a reference CDF , in this case the CDF measured under the 1 ng/mL ACTIVIN condition ( Figure 5—figure supplement 1E ) . In biological replicates using RUES2 or RUES1 the distributions of NANOG and OCT levels under the 100 or 10 ng/mL ACTIVIN conditions are reproducibly close to the distributions measured under the 1 ng/mL ACTIVIN condition ( small KS distance , Figure 5E ) . Whereas the distributions measured under the –ACTIVIN and SB conditions are reproducibly far from the distributions measured under the 1 ng/mL ACTIVIN condition ( large KS distance , Figure 5E ) . This analysis demonstrates the reproducibility of our findings that the pluripotent state is insensitive to graded ACTIVIN levels . We have demonstrated that ACTIVIN alone cannot drive stable differentiation of human gastruloids . However , WNT can lead to differentiation and self-organization of gastruloids in a SMAD2/3 dependent manner ( Martyn et al . , 2018 ) . Since WNT is operating up-stream of ACTIVIN/NODAL in the proposed signaling hierarchy , we asked whether cells with a history of WNT signaling might respond differently to ACTIVIN treatment . To address this , cells were treated with WNT for 24 hr , washed to remove WNT , and then cultured with or without ACTIVIN for an additional 24 hr ( Figure 6A ) . Surprisingly , when cells were treated with WNT alone in the absence of ACTIVIN no mesendodermal differentiation was observed and cells remained pluripotent ( Figure 6B–C ) . However , if the cells were exposed to WNT before ACTIVIN stimulation mesendodermal fates were robustly induced ( Figure 6D–E ) . KS distance analysis using the pluripotency condition ( -/ACT ) as the reference condition demonstrated the reproducibility of these observations in RUES2 and RUES1 ( Figure 6F–G ) . Following WNT and ACTIVIN treatment , mesendoderm marker expression was also observed in the RUES2-SMAD3-/- line , suggesting that SMAD3 is not necessary for ACTIVIN-dependent mesendoderm induction ( Figure 6—figure supplement 1A ) . Differentiation was ACTIVIN concentration dependent , as demonstrated by the induction of BRA at low ACTIVIN and the induction of BRA , EOMES and GSC at high ACTIVIN concentrations ( Figure 6—figure supplement 1B–D ) . Therefore , following WNT priming ACTIVIN functions as a morphogen to pattern mesendodermal fates and SMAD3 is again dispensable for this process . In order to address the mechanism of WNT priming , we first asked whether the transcriptional effector of canonical WNT signaling , β-catenin , is involved in this process . We therefore treated cells with the small molecule inhibitor endo-IWR-1 , which blocks β-catenin function through stabilization of its destruction complex . endo-IWR-1 was added with WNT on day one and again with ACTIVIN on day 2 . In the presence of endo-IWR-1 mesendoderm maker expression was eliminated and pluripotency marker expression was maintained , indicating a requirement for β-catenin in the differentiation process ( Figure 7A and Figure 7—figure supplement 1A ) . We then asked if endogenous WNT , induced by a possible positive feedback loop , is required for stable induction of mesendoderm , particularly during the ACTIVIN treatment phase when exogenous WNT has been removed . To address this we blocked endogenous WNT secretion using the small molecule inhibitor IWP-2 , which was added with WNT on day one and again with ACTIVIN on day 2 . We find that addition of IWP-2 does not affect ACTIVIN-dependent mesendoderm differentiation and loss of pluripotency ( Figure 7A and Figure 7—figure supplement 1A ) . However , IWP-2 at the same concentration does block WNT-dependent mesendoderm differentiation downstream of BMP4 ( Figure 7—figure supplement 1B ) ( Martyn et al . , 2018 ) . Toxicity as a result of either endo-IWR-1 or IWP-2 treatment was not observed ( Figure 7B ) . Together , these results demonstrate the presence of an unexpected WNT signaling memory in cells that is mediated via β-catenin and is established prior to and is required for the morphogen activity of ACTIVIN . Since we have shown that the SMAD response dynamics can be stable or transient depending on the branch being activated , we next asked if WNT memory affects the dynamics of SMAD2 signal transduction . The transient response of SMAD2 and SMAD4 and the elevation in the baseline post-stimulation was the same whether or not cells were previously exposed to WNT ( Figure 7C ) . However , transcription of mesendodermal genes was stabilized in response to ACTIVIN following WNT priming ( Figure 7D ) . As in the case of pluripotency , stable mesendodermal fate acquisition required on-going SMAD2 signaling in the elevated baseline as treatment with SB , 8 hr after ACTIVIN stimulation , eliminated mesendodermal differentiation at 24 hr after ACTIVIN ( Figure 7—figure supplement 1D ) . In contrast , and further arguing for a mechanism of WNT priming that is temporally up-stream of ACTIVIN , addition of SB during WNT priming did not block mesendoderm differentiation ( Figure 7—figure supplement 1E ) . Overall , we conclude that cells maintain a memory of WNT exposure , which makes them competent to differentiate in response to ACTIVIN ( Figure 7E ) .
Following the tradition of experimental embryology , during the late 1960 s , Nieuwkoop began a series of explant and transplant experiments in the amphibian blastula that ultimately led to the discovery of ACTIVIN as a mesoderm inducer in the 1980 s ( Nieuwkoop , 1969; McDowell and Gurdon , 1999 ) . When presented to isolated animal cap explants that normally give rise only to ectoderm , ACTIVIN was sufficient to induce different types of mesendodermal cells based on its concentration , and was thus qualified in principle as a morphogen ( Green and Smith , 1990; Wilson and Melton , 1994 ) . When the ACTIVIN receptors and the SMAD pathways were characterized in 1990 s , it was shown that micro-injection of different amounts of synthetic mRNAs encoding SMAD2 into Xenopus animal caps , also recapitulates the mesendodermal-inducing effects of A ACTIVIN morphogen presentation , independently confirming that the activation of different thresholds of the pathway was sufficient for mesoderm induction ( Shimizu and Gurdon , 1999 ) . Inhibition of the pathway by microinjection of a dominant negative type II ACTIVIN receptor into Xenopus blastula led to complete loss of mesendodermal derivatives , demonstrating that ACTIVIN/SMAD2 signaling was necessary for mesoderm induction in the amphibian embryo ( Hemmati-Brivanlou and Melton , 1992 ) . Loss-of-function analysis in the mouse later confirmed the frog conclusions about the pathway as SMAD2/3 double knockout mice failed to properly induce mesendoderm and gastrulate ( Dunn et al . , 2004 ) . However another TGFβ ligand , NODAL , rather than ACTIVIN was shown to be the inducer in the mammalian embryo ( Conlon et al . , 1994 ) . Elegant genetic experiments also performed in the mouse placed NODAL signaling in a positive feedback loop that drives gastrulation . The loop is initiated when NODAL signaling induces BMP4 expression in the extra-embryonic ectoderm . BMP signaling subsequently induces WNT3 expression , which in turn induces high levels of NODAL in the proximal-posterior part of the embryo , which marks the site of primitive streak initiation ( Arnold and Robertson , 2009 ) . Using live reporters of SMAD1 , SMAD2 and SMAD4 , we find that BMP/SMAD1 signaling is stable , ACTIVIN/SMAD2 signaling is transient ( or adaptive ) , and that co-SMAD4 follows the dynamics of the receptor-associated SMADs . In response to ACTIVIN , hESCs transiently induce mesendodermal gene expression , without generating mesendodermal fates , and the cells return to the state of pluripotency . However , the expression of ACTIVIN/NODAL inhibitors remained elevated and plausibly buffer the additional ACTIVIN , as expected from prior theory ( François and Siggia , 2008 ) . Pre-presentation of WNT , however , stabilizes transcription and fate acquisition without modifying SMAD2 dynamics . Thus , a transient signaling response is compatible with a stable change in cell fate , as shown previously in a murine myoblast cell line ( Sorre et al . , 2014 ) . Both pluripotency maintenance and mesendodermal fate acquisition are not affected by the loss of SMAD3 . Our data provide evidence for a previously undetected level of signal integration that implies the presence of a cellular memory . This memory operates not at the level of the SMAD2 signaling dynamics but by some other means to change the response of the cells to ACTIVIN . It is tempting to speculate that it might be mediated through epigenetic mechanisms , in a manner similar to what has been described in Xenopus experiments where β-catenin recruits PrMT2 to induce dorsal fates ( Blythe et al . , 2010 ) . The ACTIVIN/SMAD2 pathway has been shown to be necessary for maintenance of pluripotency , but not sufficient to induce mesendoderm in hESCs , except when influenced by modifiers of other signaling pathways ( D'Amour et al . , 2005; McLean et al . , 2007; Singh et al . , 2012; Funa et al . , 2015 ) . Our study shows that mere pre-treatment of hESCs with WNT , endows them with a memory that enables a graded response to subsequently applied ACTIVIN; simultaneous exposure to WNT and ACTIVIN is not required . Furthermore , the WNT treatment alone is not sufficient to elicit a stable fate change since in minimal media , without ACTIVIN , the cells revert to pluripotency . The ability of embryonic cells to record WNT signals may be a broadly conserved and fundamental aspect of animal development , as shown by elegant experiments in Drosophilia where the authors demonstrate a morphogen effect for Dpp ( a BMP homologue ) after cells have lost contact with the source of WNT ( Alexandre et al . , 2014 ) . In the context of vertebrate development , our findings force the reevaluation of the traditional literature regarding the sufficiency of the ACTIVIN/SMAD2 pathway for mesendodermal induction and patterning . A review of the evidence in model systems for the ACTIVIN/NODAL mediated morphogen effect reveals that in all the experimental settings , at least some of the cells were either still under the influence , or had been previously exposed to WNT signaling before or during ACTIVIN/NODAL signaling . For example , cells of the Xenopus animal cap derived from the blastula stage embryo have been under maternal WNT influence as early as the two-cell stage , hours before ACTIVIN is presented to the explants . This influence is the consequence of cortical rotation that occurs after sperm entry , and activates the WNT pathway on the dorsal side of the embryo , as evidenced by the dorsal-specific nuclear localization of β-catenin ( Schneider et al . , 1996; Larabell et al . , 1997; Rowning et al . , 1997 ) . It is clear from the asymmetric elongation of the animal cap explants in response to ACTIVIN that the prior WNT exposure sets up a dorsal-ventral pre-pattern that affects the response . It is also clear that the dorsal and ventral regions of the animal cap when separated respond differently to ACTIVIN ( Sokol and Melton , 1991; Bolce et al . , 1992 ) . Although some mesoderm is induced in the ventral caps , only cells of the dorsal cap that have seen WNT signals undergo ACTIVIN-mediated induction of GSC + mesendoderm . These experiments demonstrate that both pathways are required for complete mesendodermal patterning without explicitly distinguishing their temporal relationship . Our study revises this point of view by bringing a temporal order . We suggest that only cells endowed with a WNT memory can respond to ACTIVIN to specify the full range of mesendodermal fates . Even in the extensively studied Xenopus embryo , it is still debated whether in the late blastula stage marginal zone , prior to the onset of gastrulation , an ACTIVIN/NODAL gradient already defines the medial to lateral mesendodermal fates or whether there is simply a bipartite division into dorsal organizer and ventral mesoderm ( Harland and Gerhart , 1997; Smith , 2009 ) . This debate also translates into whether ACTIVIN/NODAL is sufficient to pre-pattern the mesendoderm or whether complete patterning requires prior WNT exposure present only in the dorsal part of the embryo . Controversies persist , since the embryo is rapidly developing , there are no live reporters of signaling , and fates are often assayed in early gastrulation rather than in the late blastula . In the mouse the earliest manifestation of the streak is the proximal-posterior expression of WNT , which extends distally . By mid-streak stage , NODAL signaling is highest in the Node ( Tam and Loebel , 2007 ) . Thus , cells that leave the streak at various proximal-distal positions and NODAL levels , manifestly have been exposed to WNT first ( Tam and Behringer , 1997 ) . This interpretation , however , does not eliminate the co-requirement for ACTIVIN and WNT , but rather suggests that instructive WNT , required for mesendodermal differentiation , occurs temporally up-stream of ACTIVIN/SMAD2 signaling . In conclusion , a cellular memory of WNT signaling , endows embryonic cells , including hESCs , with the competence to maintain ACTIVIN/SMAD2-mediated mesendodermal fate specification and patterning .
Experiments were performed with the RUES2 hESC line ( XX female; US National Institutes of Health , human ESC registry no . 0013 ) or CRISPR/Cas9 edited cell lines based on RUES2 . Key experiments were repeated with the RUES1 hESC line ( XY male; US National Institutes of Health , human ESC registry no . 0012 ) . RUES1 and RUES2 have been authenticated by STR profiling and both tested negative for mycoplasma contamination . hESCs were grown in HUESM medium that was conditioned by mouse embryonic fibroblasts and supplemented with 20 ng/mL bFGF ( MEF-CM ) . Cells were grown on tissue culture dishes ( BD Biosciences , San Jose , CA ) for maintenance and expansion at 37°C and 5% CO2 . Dishes were coated overnight at 4°C with Geltrex ( Thermo Fisher Scientific , Waltham , MA ) diluted 1:40 in DMEM/F12 and then incubated at 37°C for at least 20 min before passaging . Cells were passaged as aggregates using Gentle Cell Dissociation Reagent ( STEMCELL Technologies , Vancouver , Canada ) . Individual micropatterned coverslips ( CYTOO , Grenoble , France ) were washed one time with water for 5 min at room temperature ( RT ) to activate the surface according to the manufacturers recommendation . Coverslips were then coated at 37°C for 2 hr with 20 µg/mL Laminin-521 ( BioLamina , Sundbyberg , Sweden ) in 0 . 5 mL PBS + Mg/+Ca . The laminin was then removed by serial dilutions without allowing the coverslip to dry ( 1:4 dilution in PBS –Mg/–Ca , six times ) . Chips were seeded immediately or stored overnight at 4°C in 2 mL PBS –/– and seeded on the following day . Cell seeding was performed as follows . Cells growing in MEF-CM were washed once with PBS +/+ and dissociated to single cells with Accutase ( STEMCELL Technologies ) . Cells were centrifuged and 600 , 000 cells were resuspended in 2 mL of MEF-CM supplemented with 10 µM Rock-inhibitor ( Y-27632 , Abcam , Cambridge , MA ) . The cell suspension was then placed over the coverslip in a 35 mm tissue culture dish . The sample was left unperturbed for 10 min at RT in order to achieve homogeneous seeding of the cells throughout the chip before being moved to the incubator . After 2 hr , the medium was replaced with MEF-CM without Rock-inhibitor and the cells were incubated overnight . On the following day , the medium was changed to MEF-CM with additional ligands , BMP4 , WNT3A , or ACTIVIN A ( R and D Systems , Minneapolis , MN ) with or without 10 µM SB431542 ( Stemgent , Lexington , MA ) . We stimulated cells with ACTIVIN A rather than NODAL , because although both ligands differentiate hESCs towards the mesendoderm lineage , NODAL requires much higher concentrations to be effective ( McLean et al . , 2007 ) . For experiments in TeSR-E7 ( STEMCELL Technologies ) , cells were moved to E7 medium the day after seeding and incubated for an additional 24 hr before adding ligands . Live imaging was carried out in E7 imaging medium , which was prepared with FluoroBrite DMEM ( Thermo Fisher Scientific ) according the published protocol for E8 ( Beers et al . , 2012 ) . Optical-quality plastic tissue culture dishes ( ibidi , Martinsried , Germany ) were coated with 10 µg/mL Laminin-521 in PBS +/+ for 2 hr at 37°C or overnight at 4°C . Single cells were collected as described for micropatterned cell culture and dishes were seeded with 50 , 000 single cells resuspended in 2 mL E7 supplemented with 10 µM Rock-inhibitor . The samples were incubated overnight . On the following day , the medium was changed to E7 supplemented with 10 µM Rock-inhibitor and ligands or small molecules . For the 2 day protocol in which cells were switched from E7 ±100 ng/mL WNT3A to E7 ±10 ng/mL ACTIVIN A , the samples were washed with PBS +/+ before adding fresh medium . Live imaging was carried out in E7 imaging medium . Small molecules were used at the following concentrations and were replaced every 24 hr: 10 µM SB431542 ( Stemgent ) , 1 µM IWP-2 ( Stemgent ) , and 1 µM endo-IWR-1 ( Tocris ) . Cells on dishes or coverslips were rinsed once with PBS –/– , fixed with 4% paraformaldehyde ( Alfa Aesar , Thermo Fisher Scientific , Tewksbury , MA ) for 20 min at RT , and then rinsed twice and stored in PBS –/– . Cells were blocked and permeabilized with blocking buffer ( 2% bovine serum albumin and 0 . 1% Triton X-100 in PBS –/– ) for 30 min at RT . Cells were incubated with primary antibodies in blocking buffer overnight at 4°C and then washed three times with 0 . 1% Tween-20 in PBS –/– ( PBST ) . Cells were incubated with secondary antibodies ( diluted 1:1000 ) : Alexa Fluor 488 , 555 , or 647-conjugated from ( Invitrogen Molecular Probes , Thermo Fisher Scientific ) and DAPI nuclear stain ( 1:10 , 000 dilution ) in blocking buffer for 30 min at RT , and then washed twice with PBST and once with PBS –/– . Dishes were stored and imaged in PBS –/– . Coverslips were mounted on slides using Fluoromount-G mounting medium ( SouthernBiotech , Birmingham , AL ) . Western blotting was performed using standard techniques . The nuclear fractions presented in Figure 3—figure supplement 1C were obtained using the NE-PER Nuclear and Cytoplasmic Extraction Reagents ( Thermo Fisher Scientific ) and following the manufacturers protocol . Following SDS-PAGE and Western transfer , membranes were stained with anti-Smad2 Rabbit mAb ( Cell Signaling Technology , Danvers , MA ) at 1:1000 dilution . Membranes were stripped for 15 min at RT using Restore PLUS Western Blot Stripping Buffer ( ThermoFisher Scientific ) and re-stained with anti-Lamin B1 Rabbit polyAb ( Proteintech , Rosemont , IL ) at 1:2000 dilution . Wide-field images of fixed samples were acquired on an Olympus IX-70 inverted microscope with a 10x/0 . 4 numerical aperture objective lens . Tiled image acquisition was used to acquire images of large areas of dishes or coverslips in four channels corresponding to DAPI and Alexa Fluor 488 , 555 , and 647 . Live imaging was performed on a spinning disk confocal microscope equipped with 405- , 488- , and 561 nm lasers and an environmental chamber ( CellVoyager CV1000 , Yokogawa ) . Images were acquired every 10 min with a 20x/0 . 75 numerical aperture objective lens . The cells were maintained at 37°C and 5% CO2 during live imaging . For the tagRFP-SMAD1 reporter cells , CRISPR/Cas9 mediated genome engineering was used to fuse a cassette containing a blasticidin resistance gene ( BsdR ) , a T2A self-cleaving peptide , and a tagRFP fluorescent protein onto the N-terminus of SMAD1 , so that the locus produces both a tagRFP-SMAD1 fusion protein together with BsdR . Similarly , for the mCitrine-SMAD2 reporter line , CRISPR/Cas9 was used to fuse a cassette containing a puromycin resistance gene ( PuroR ) , a T2A self-cleaving peptide , and an mCitrine fluorescent protein onto the N-terminus of SMAD2 , so that the locus produces both an mCitrine-SMAD2 fusion protein together with PuroR . RUES2 hESCs were nucleofected with a pX335 plasmid ( Cong et al . , 2013 ) that co-expresses the nickase version of Cas9 and the specific sgRNA of interest ( protospacer sequences: 5’-GCAGCACTAGTTATACTCCT-3’ for SMAD1 and 5’-GGACGACATGTTCTTACCAA-3’ for SMAD2 ) , as well as the appropriate homology donor plasmid . Nucleofection was carried out using the Cell Line Nucelofector Kit L ( Lonza , Walkersville , MD ) and the B-016 setting of a Nucleofector II instrument . Nucleofected cells were plated into MEF-CM supplemented with 10 µM Rock-inhibitor . After 4 days , blasticidin or puromycin was added for 7 days to select for cells that had been targeted . Cells that survived selection were passaged as single cells using Accutase , plated in MEF-CM supplemented with 10 µM Rock-inhibitor , and allowed to grow into colonies . Colonies arising from a single cell were handpicked , expanded , and screened for correct targeting by PCR amplification of the genomic region and Sanger sequencing . Correctly targeted clones were subsequently transfected with ePiggyBac plasmids containing either H2B-mCitrine or H2B-mCherry cassettes to enable nuclear labeling for cell tracking ( Lacoste et al . , 2009 ) . Individual clones were again isolated and controlled for normal karyotype ( G-banding ) and pluripotency maintenance . RUES2 hESCs were nucleofected with a pX330 plasmid ( Cong et al . , 2013 ) that co-expressed the wild-type version of Cas9 and one of two sgRNAs targeting SMAD3 ( protospacer sequences: 5’-CCACCAGATGAACCACAGCA-3’ for sgRNA #1 and 5’-TTATTATGTGCTGGGGACAT-3’ for sgRNA #2 ) . The sgRNAs were designed to target the first or second coding exon shared by all SMAD3 isoforms , a strategy that was successfully used to knockout SMAD3 in human primary cell lines ( Voets et al . , 2017 ) . Two different sgRNAs were used to control for off-target effects . We modified the pX330 plasmid to also express a puromycin-2A-EGFP cassette to enrich for cells that had been successfully nucleofected . Nucleofection was carried out as described above and cells were plated in MEF-CM supplemented with 10 µM Rock-inhibitor . On the following day puromycin was added for 24 hr . Cells that survived selection were allowed to recover for several days . Cells were then passaged as single cells using Accutase , plated in MEF-CM supplemented with 10 µM Rock-inhibitor , and allowed to grow into colonies . Colonies arising from a single cell were handpicked , expanded , and screened for correct targeting by PCR amplification of the genomic region and Sanger sequencing . The resulting chromatograms were decomposed using the Tide web-based tool ( Brinkman et al . , 2014 ) . RUES2 cells were seeded in 6-well plates ( 1 , 00 , 000 single cells per well ) in E7 medium supplemented with 10 µM Rock-inhibitor and incubated overnight . On the following day the medium was changed to E7 supplemented with 10 µM Rock-inhibitor with or without 10 ng/mL ACTIVIN A . Samples ( three pooled-wells ) were collected in 1 mL Trizol at 0 , 2 . 5 , 4 , 6 and 12 hr . The 2 day protocol was carried out similarly: the medium was changed on the day after seeding to E7 supplemented with 10 µM Rock-inhibitor with or without 100 ng/mL WNT3A . On the second day following seeding the wells were washed once with PBS and the media was changed to E7 supplemented with 10 µM Rock-inhibitor with or without 10 ng/mL ACTIVIN A . Samples were collected in 1 mL Trizol before the addition of WNT3A , after WNT3A treatment ( referred to as 0 hr ) , and at 2 . 5 , 4 , 8 , and 12 hr after the addition of ACTIVIN A . Total cellular RNA for each sample was extracted using the RNeasy Mini Kit ( QIAGEN , Germantown , MD ) and cDNA was synthesized using the Transcriptor First Strand cDNA Synthesis Kit ( Roche ) . RT-PCR for selected genes was performed on three technical replicates using the LightCycler 480 SYBR Green I Master mix in a LightCycler 480 instrument ( Roche , Basel , Switzerland ) . Primers were designed using Primer-BLAST ( Ye et al . , 2012 ) or obtained from qPrimerDepot ( Cui et al . , 2007 ) or from previously published sequences ( Mendjan et al . , 2014 ) . Primer sequences and source are listed in Table 1 . RUES2 cells were seeded in 6-well plates ( 200 , 000 single cells per well ) in E7 medium supplemented with 10 µM Rock-inhibitor and incubated overnight . On the following day the media was changed to E7 supplemented with 10 µM Rock-inhibitor with or without 10 ng/mL ACTIVIN A . Samples ( three pooled-wells ) were collected in 1 mL Trizol at 1 , 2 . 5 , 4 , 8 and 12 hr for the ACTIVIN-treated conditions and after 0 , 6 and 12 hr for the no-ACTIVIN conditions ( to be used as negative controls ) . Total cellular RNA for each sample was extracted using RNeasy Mini Kit ( QIAGEN ) and 2 ug of total RNA was used to prepare each individual RNA-seq library . RNA-seq library construction was conducted with the TruSeq RNA Library Preparation Kit ( Illumina , San Diego , CA ) as per the manufacturer’s instructions and sequenced in an Illumina HiSeq 2500 apparatus . Raw reads were mapped to hg19 using STAR aligner , and the gene read counts were normalized using the DESeq2 Bioconductor package ( Love et al . , 2014 ) . Library preparation , sequencing , and mapping were performed by the New York Genome Center ( New York , NY , USA ) . All raw data files are available from the GEO database ( accession number GSE111717 ) . For images acquired from micropatterned cell culture experiments , stitching and colony detection were carried out as described previously using custom software written in MATLAB ( Etoc et al . , 2016 ) . For analysis of the SMAD reporter lines on micropatterned colonies a single z-plane through the middle of the colony was analyzed at each time point . Background in each channel was removed by subtracting a minimum intensity image that was generated by taking the minimum value at each pixel over all images in that channel in the same z-plane . Vignetting was corrected by dividing by a flat-field image that was generated by normalizing the intensity of the minimum image to 1 . This procedure corrects the intensity drop-off at the border of the images without further altering the average image intensity . Nuclei segmentation and signal quantification were performed on the corrected images as follows . The H2B image was thresholded to generate a binary image separating the foreground ( nuclei ) from the background . The original , corrected H2B image was then filtered with a median and sphere filter with parameters matching the expected size of individual nuclei . Local maxima corresponding to individual nuclei were detected using the MATLAB extended-maxima transform function . Maxima were dilated to increase the likelihood of obtaining a single maximum per nuclei . Maxima falling within the foreground were used as seeds for watershed segmentation , which was also restricted to the foreground and was used to obtain a labeled object corresponding to each nucleus within the image . The segmented objects were further processed to eliminate objects much larger or smaller than the expected size of individual nuclei . The results of the segmentation were used as a mask to obtain the median per cell nuclear intensity in each channel . Images acquired from fixed samples that were stained by immunofluorescence were similarly segmented and analyzed . The DAPI channel was used to perform nuclei segmentation , and the segmented objects were subsequently used to obtain the median per cell signal intensity in each channel . For the mCitrine-SMAD2 reporter line an enrichment of the mCitrine signal in the cytoplasm relative to the nucleus prior to ligand presentation could be detected , which decreased concomitantly with an increase in the nuclear signal following ACTIVIN presentation . Therefore , the mCitrine-SMAD2 response was quantified as the nuclear-to-cytoplasmic ratio , which was used previously as a readout for TGFβ pathway activity ( Warmflash et al . , 2012 ) . In order to estimate the cytoplasmic signal for each cell , a narrow donut surrounding the nuclear mask was formed by dilating the nuclear mask once by an inner radius and a second time by an outer radius and subtracting the first dilated object from the second . The donut , which formed the cytoplasmic mask , was not restricted to the foreground pixels ( H2B signal ) , but it was prevented from overlapping with the masks of neighboring nuclei . The median mCitrine intensity within the nuclear mask was divided by the median intensity within the cytoplasmic mask on a per cell basis . The qualitative behavior of the SMAD2 nuclear-to-cytoplasmic response was not sensitive to the exact size of the inner and outer radius of the donut . Therefore , the values were chosen manually and kept fixed throughout all analyses . For the RFP-SMAD1 reporter line a faint nuclear signal could be detected prior to ligand presentation . However , no cytoplasmic signal could be detected above the background . Therefore , the RFP-SMAD1 response was quantified as the median nuclear signal normalized to the median H2B signal to normalize for cells moving in and out of the z-plane . In order to analyze the SMAD response as a function of radial position within the micropatterned colonies , the cells within a single colony were binned based on the radial position of their center and the average response per cell within each bin was calculated . The radial profile for individual colonies was then averaged over several colonies . Analysis of the SMAD reporter lines in single-cell culture was carried out in a similar manner . The maximum intensity projection image , rather than a single z-plane was analyzed . Since cells remained flat under these conditions it was not necessary to normalize the SMAD1 signal by H2B . For each gene , a baseline expression profile , which was calculated using a linear interpolation between the 0 , 6 and 12 hr control samples , was subtracted from the expression values of the ACTIVIN-treated samples . The gene list was filtered to contain only those genes that: ( 1 ) had at least one time point with an absolute fold-change larger than 2 ( up- or down-regulated ) compared to 0 hr and ( 2 ) had at least one time point with a normalized read count higher than 100 . That generated a list of 3529 genes of interest , which was then hierarchically clustered by their z-scored expression values , using Cluster 3 . 0 ( de Hoon et al . , 2004 ) with the following options: centered correlation as the similarity metric and average linkage as clustering method . The resulting hierarchical tree was visualized using Java TreeView ( Saldanha , 2004 ) to identify the minimal clusters of interest . To identify motifs enriched within gene clusters , the 2000 bp upstream sequences for all genes were extracted using the PWMEnrich Bioconductor package . Motif enrichment of each cluster's sequence set was performed using AME ( Bailey et al . , 2015 ) with the HOCOMOCOv10 database ( Kulakovskiy et al . , 2016 ) against the background ( upstream sequences of all genes ) . We assessed the enrichment of the genes in each of the groups identified in Figure 4A for previously defined marker gene sets from isolated endoderm , mesoderm , and ectoderm/epiblast tissue from E7 . 5 mouse embryos ( Lu et al . , 2018 ) using the GOseq Bioconductor package ( Young et al . , 2010 ) . Mouse genes were mapped to human orthologues using data downloaded from the Mouse Genome Informatics database ( http://www . informatics . jax . org/downloads/reports/HOM_MouseHumanSequence . rpt ) . The RNA-sequencing data related to Figure 4 and discussed in this publication have been deposited in NCBI's Gene Expression Omnibus ( Edgar et al . , 2002 ) and are accessible through GEO Series accession number GSE111717 ( https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE111717 ) . Image analysis software StemCellTracker ( commit no . 1c31fdd ) is available through GitHub ( Kirst , 2014; copy archived at https://github . com/elifesciences-publications/StemCellTracker ) . | Embryonic stem cells can renew themselves to generate more stem cells , or specialize to become any type of cell found in an adult . They therefore hold great potential for studying how we develop from a single cell into a complex organism made of many different cell types . In a key stage of development , individual cells form into organized tissues . The earliest phase of tissue organization involves the formation of three ‘germ layers’ . Human embryonic stem cells allow us to recreate this early stage of embryo development in the lab . When grown in confined spaces , the cells organize into clusters that can then develop germ layers . Previous work using these clusters showed that a network of signaling proteins – including one called WNT – trigger human embryonic stem cells to form the initial clusters . Then , another signaling protein called ACTIVIN tells the cells to specialize to form the two inner germ layers . But in experiments that apply only the ACTIVIN signal , the cells instead keep dividing to make more stem cells . ACTIVIN can trigger the activity of a protein called SMAD . To visualize how cells respond to ACTIVIN in real time , Yoney et al . used a gene editing technique called CRISPR to add fluorescent tags to SMAD in human embryonic stem cells . The results show that the ACTIVIN response triggers a peak in the amount of SMAD in the cell’s nucleus that then decreases over several hours . This briefly activated several genes that are known to help to form germ layers . However , this gene activity was not maintained for long enough to cause the stem cells to specialize and organize into layers . Yoney et al . then repeated the experiments on cells that had previously been exposed to WNT signaling proteins . The germ layer gene activity was maintained in this case , leading to the cells specializing and forming the inner two germ layers . This suggests that the cells somehow remembered the WNT signal , and this memory changed how they responded to ACTIVIN . The next step is to understand how cells store the memory of the WNT signal . As well as aiding our understanding of development , it could also help us to understand situations where signaling goes wrong , such as cancer . The technique used here to follow signals in real time could also be used to study other biological signaling processes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"developmental",
"biology"
] | 2018 | WNT signaling memory is required for ACTIVIN to function as a morphogen in human gastruloids |
Cells adapt to familiar changes in their environment by activating predefined regulatory programs that establish adaptive gene expression states . These hard-wired pathways , however , may be inadequate for adaptation to environments never encountered before . Here , we reveal evidence for an alternative mode of gene regulation that enables adaptation to adverse conditions without relying on external sensory information or genetically predetermined cis-regulation . Instead , individual genes achieve optimal expression levels through a stochastic search for improved fitness . By focusing on improving the overall health of the cell , the proposed stochastic tuning mechanism discovers global gene expression states that are fundamentally new and yet optimized for novel environments . We provide experimental evidence for stochastic tuning in the adaptation of Saccharomyces cerevisiae to laboratory-engineered environments that are foreign to its native gene-regulatory network . Stochastic tuning operates locally at individual gene promoters , and its efficacy is modulated by perturbations to chromatin modification machinery .
The capacity to adapt to changes in the external environment is a defining feature of living systems . Cells can rapidly adapt to familiar changes that are commonly encountered in their native habitat by sensing the parameters of the environment and engaging dedicated regulatory networks that have evolved to establish adaptive gene expression states ( Jacob and Monod , 1961; Thieffry et al . , 1998 ) . However , dedicated sensory , signaling , and regulatory networks become inadequate , or even detrimental , when cells are exposed to unfamiliar environments that are foreign to their evolutionary history ( Tagkopoulos et al . , 2008 ) . In principle , at least one gene expression state that maximizes the health/fitness of the cell always exists , despite the inability of the native regulatory network to establish such a state . This is true because under any conceivable environment , the activities of some genes are beneficial , whereas those of others are futile or even actively detrimental ( Jacob and Monod , 1961; Tagkopoulos et al . , 2008; Hottes et al . , 2013 ) . In fact , if the initial fitness defect is not lethal , a population of cells may slowly adapt to an unfamiliar environment through the accumulation of genetic mutations that rewire regulatory networks , thereby achieving more optimal gene expression states ( Tagkopoulos et al . , 2008; Applebee et al . , 2008; Philippe et al . , 2007; Goodarzi et al . , 2010; Tenaillon et al . , 2012; Rodríguez-Verdugo et al . , 2016; Blount et al . , 2012; Van Hofwegen et al . , 2016; Damkiær et al . , 2013 ) .
In this work we speculate whether cells have evolved alternative strategies for finding adaptive gene expression states , on more physiological timescales , without relying on their hard-coded sensory and regulatory systems . Since the perception of the external world may be of limited value under unfamiliar conditions , perhaps a more effective strategy would be to focus on maximizing the internal health of the cell—without regard to the specific parameters of the outside world . This would be a challenging strategy , as every gene in the genome would need to independently reach the expression level that maximizes the overall health of the cell , and these expression levels could vary significantly from condition to condition . In particular , we asked whether individual genes could , in principle , carry out a search process equivalent to gradient descent ( Cauchy , 1847 ) , where the health consequence of stochastic alterations in gene expression could gradually tune the expression of individual genes towards a level that is optimal for internal health . We reasoned that such an optimization process would require the existence of: ( 1 ) a source of stochastic transitions in gene expression; ( 2 ) the ability of local chromatin to maintain a record of recent changes in transcription; and ( 3 ) a central metabolic hub that integrates diverse parameters of intracellular health and continuously broadcasts whether the overall health of the cell is improving or deteriorating . In fact , we find that the foundations for meeting these requirements are already present in eukaryotic cells: ( 1 ) The expression of many genes is dominated by noisy bursts of transcription—a widespread phenomenon of largely unknown functional significance ( Sanchez and Golding , 2013; Raj and van Oudenaarden , 2008; Blake et al . , 2006; Raser and O'Shea , 2005; Elowitz et al . , 2002 ) ; ( 2 ) Co-transcriptional histone modification can modify eukaryotic chromatin in promoters and gene bodies , establishing a short-term memory of recent transcriptional events ( Li et al . , 2007; Rando and Winston , 2012 ) ; and ( 3 ) Global integrators of cell health have evolved in eukaryotes . A classic example is the mTOR pathway , which integrates a vast array of intracellular parameters reflecting nutrient availability , energy , and the presence of diverse stresses ( Conrad et al . , 2014; González and Hall , 2017; Albert and Hall , 2015; Saxton and Sabatini , 2017 ) . With the necessary components for gradient-based optimization of gene expression in place ( Figure 1A ) , the promoter of each gene would be able to conduct a simple search process that culminates in finding the expression level that maximizes the overall health of the cell: if global fitness/health is increasing and there was a previous increase in transcriptional output ( representing larger or more frequent transcriptional bursts ) , the promoter further increases its transcriptional activity ( Figure 1B ) . If fitness is decreasing and there was a previous increase in transcriptional activity , the promoter decreases its transcriptional output . Transcriptional output is altered in the opposite direction in the event that there was a previous decrease in transcriptional output . For each gene , this tuning process can be expressed as: ΔEt = k∙sgn ( ΔFt∙ΔEt-1 ) +η ( see Figures 1 and 2A ) ; here , E denotes the vector of gene-level transcription rates , F the current fitness/health of the cell , k is a proportionality constant , η a noise term , and sgn is a function yielding −1 if its argument is negative , 0 if its argument is zero , and +1 if its argument is positive . One can easily see how the process described here can tune the optimal expression of a single gene . What is remarkable , however , is the ability of this hypothetical stochastic tuning process to find near-optimal gene-expression states for a system with thousands of genes . As can be seen in the simulations presented in Figure 2 , this is achieved through a fitness-directed stochastic search culminating in individual genes reaching specific gene expression levels that maximize the health/fitness of the cell . Such a stochastic tuning mechanism would be highly valuable to free-living organisms , enabling them to optimize their global gene expression patterns to match the specific requirements of any environment in which their dedicated sensory and regulatory networks are inadequate or sub-optimal . Informed by the simulations above , we sought to test for evidence of stochastic tuning in the eukaryotic model organism Saccharomyces cerevisiae . We engineered conditions in which the expression of a single gene was required for growth , but for which no regulatory input existed to drive appropriate expression levels . This was achieved by using a yeast strain ( BY4743 ) that lacks the URA3 gene , which is essential when cells are grown in the absence of uracil . We placed a chromosomally integrated copy of URA3 at a different locus under the control of a weak synthetic promoter , consisting primarily of a pseudorandom sequence . All recognizable binding sites for native transcription factors were removed from the generated promoter sequence ( see Materials and Methods and Supplementary file 1 for details ) , in an attempt to decouple it from any existing sensory and regulatory input . We henceforth refer to this synthetic promoter sequence as synprom ( see Supplementary file 1 for sequence ) . In the experiments described below , URA3 is typically tagged with a fluorescent fusion , either mRuby ( Kredel et al . , 2009 ) or a superfolder GFP ( Pédelacq et al . , 2006 ) , and a copy of a mouse DHFR gene coupled to a different fluorescent protein is inserted at the same location on the sister chromosome to act as an internal control . A schematic of the insertion constructs is shown in Figure 3A . We also added the URA3 competitive antagonist 6-azauracil ( 6AU ) to the media to control the threshold level of URA3 production required for growth . The growth condition , SC+glu-ura media , containing x µg/ml of 6AU , will henceforth be referred to as ura-/6AUx . Even with the challenging and specific experimental layout described here , with growth highly dependent on URA3 expression , we expect that stochastic tuning might contribute to fitness through mechanisms acting in cis at the promoter driving URA3 , those acting in trans through modulation of factors that ( despite our best efforts ) weakly affect the promoter driving URA3 , and through tuning of unrelated pathways that benefit survival and growth in the –URA condition . Nevertheless , URA3 expression itself will clearly be the key driver of growth since it is the critical bottleneck for nucleotide biosynthesis in the absence of uracil supplementation . To look for evidence of fitness-directed stochastic tuning , we tracked the colony formation of cells containing synprom-driven URA3 after plating on ura-/6AU15 plates . Lacking sufficient URA3 expression to overcome high 6AU levels , these non-growing cells would be expected to succumb to starvation and die . Remarkably , however , after prolonged incubation we observed apparently stochastic transitions to rapid growth , leading to the formation of macroscopic colonies over time ( Figure 3B ) . We eventually observed colony formation by roughly one cell in 103 , a rate too high to be driven by mutation-driven adaptation in the absence of growth . The synthetic promoter referred to as ‘synprom’ throughout the text is the combination of a pseudorandom sequence with a small natural promoter-proximal region taken from the SAM3 gene , with both stripped of all recognizable matches to known transcription factor binding sites ( see Materials and Methods for details ) . We also tested all combinations of five other synthetic promoter sequences and one other promoter proximal region , enumerated in Supplementary file 2 . As shown in Figure 3—figure supplement 1 , four of the six synthetic promoters support stochastic tuning , and the ability of synprom5 ( the purely artificial component of the synprom referred to in the remainder of the text; see Supplementary file 2 for all synthetic promoter sequences ) to undergo tuning remains even with a different promoter proximal region . These findings highlight the universality of the observed tuning phenomenon and minimize the possibility that our observations actually arise due to the presence of some residual sequence-specific transcription factor binding site present in synprom . As shown in Figure 3C , we also observed similar tuning behavior for two high-noise natural promoters , PHSP12 and PRGI1 ( Tirosh et al . , 2009; Tirosh et al . , 2006 ) , indicating that stochastic tuning can function even when superimposed on naturally evolved regulatory sites . Across all promoters ( natural and synthetic ) tested here , the observed tuning rates , relative to the number of viable plated cells , varied from 1 in 101 ( PHSP12 ) to 1 in 105 ( synprom5-arf1 ) . The apparently stochastic nature of colony formation in our experiments is reflected both in the steady emergence of colonies over the course of days or weeks ( Figure 3B–C and Figure 3—figure supplement 1 ) , and in the wide variance of colony sizes observed on ura-/6AU15 plates ( Figure 3D ) . Microscopy revealed that cells remain quiescent for days before transitioning to URA3 expression and rapid growth , with a transition rate dependent on the choice of promoter ( Figure 3E ) . Furthermore , the change that enables growth under the ura-/6AU15 condition must be passed from mother to daughter cells , as colonies expand from a few points of initiation instead of showing random division of cells throughout the microscopic field over time . While the presence of some deterministic process , yielding colony formation over the observed timescales ( dependent on the initial state of each cell ) , cannot be ruled out , a far simpler explanation for the observed phenomenon of a long lag followed by appearance of colonies over a wide range of times is that each cell independently undergoes a random process that can eventually lead to growth . We confirmed that the appearance of colonies is not simply due to aging of the plates; 6AU-containing plates which were pre-incubated for a week or longer prior to plating of cells showed no change in colony formation rates ( data not shown ) . To provide further insights into the regulatory changes occurring during the onset of cell growth , we performed flow cytometry time courses on cells challenged by , and subsequently growing in , liquid ura-/6AU5 media , using cells with synprom-driven URA3-mRuby , and with a DHFR-GFP fusion driven by either the constitutive ADH1 promoter ( Figure 4A–B ) or synprom ( Figure 4C–D ) itself . The use of PADH1 to drive the second reporter allows us to control for extrinsic noise and global changes in gene expression , whereas coupling synprom to the non-beneficial DHFR-GFP fusion allows us to test whether the observed stochastic tuning is driven by any trans-acting input from some existing regulatory network or whether it is truly specific to the allele needed for growth , as required by our proposed tuning model . Several patterns in the growth curves and flow cytometry data are immediately apparent . First , as with the agar-based growth discussed above , cells show a lag of at least 72 hr with undetectable growth , followed by the onset of steady growth until saturation . In the case of URA3-mRuby driven by synprom and DHFR-GFP by the constitutive promoter PADH1 , URA3-mRuby fluorescence increases substantially in tandem with the onset of cell growth , and expression subsequently remains high until saturation; in contrast , DHFR-GFP signals do not even recover to their initial levels ( Figure 4B; compare dashed and solid line distributions ) . This demonstrates that the URA3 induction resulting in growth is promoter-specific and does not simply reflect a general increase in protein expression . We observed qualitatively equivalent behavior when URA3 was driven by PRGI1 or PHSP12 ( Figure 4—figure supplement 1 ) . Even more strikingly , for cells with synprom driving both fluorescent fusions , we observed a specific enhancement of URA3-mRuby expression over that of DHFR-GFP ( Figure 4D ) , showing that the transition to high URA3 expression is not only promoter-specific but allele-specific , and thus must be driven at least partly by changes occurring in cis at the specific locus whose expression is required for growth . As an additional test , we performed quantitative RT-PCR experiments to measure the ratio of URA3 and DHFR mRNA expression in tuned cells either in liquid ura-/6AU5 media or on ura-/6AU15 plates ( see Figure 4—figure supplement 2 ) . In both cases , we observed a substantial increase in the URA3:DHFR ratio in the tuned cells , indicating that the observed tuning occurs at least partly through a local cis-acting process at the locus required for growth ( although we cannot rule out additional changes in other promoters that also contribute to survival and growth , which may account for the observed heterogeneity in expression levels between replicates ) . Consistent with our proposed tuning model , the allele-specific nature of the transcriptional induction supports a key role for a local tuning process that is independent of dedicated sensory and regulatory input . The presence of the competitive URA3 inhibitor 6-azauracil allows us to vary the threshold level of URA3 required for growth . Thus , it is instructive to consider how the concentration of 6AU may alter stochastic tuning behavior , both in the context of the computational model described above and in the actual behavior of the system . We made two crucial modifications to the numerical model employed in Figure 2 to mimic our experimental setup . First , rather than having the entire gene expression profile begin far from the optimal point , we begin with all genes but one ( representing URA3 ) at their optimal values , reflecting the fact that aside from the artificial stress of lacking appropriate URA3 regulation , the cells’ native regulatory network can provide an appropriate response to ura-/6AU media . Second , we note that due to the presence of the competitive inhibitor 6AU , the URA3 in the cell will not even be able to contribute meaningfully to nucleotide biosynthesis ( and thus impact the cell’s health/fitness ) until it passes a threshold level . Thus , the tuning term ( Figure 2A ) is not applied to the gene representing URA3 until after the concentration of URA3 passes a threshold . Aside from the modifications noted above , we model tuning in the ura-/6AU environment as we did for the general case in Figure 2 , and in particular , the fitness effects of changing URA3 expression must compete with noisy gene expression from the other 999 genes in the model gene expression profile to impact the direction of tuning . The resulting URA3 expression profiles during simulated tuning in the presence of low or high concentrations of 6AU are shown in Figure 5A . In the low 6AU case , the tuning mechanism pushes URA3 expression almost deterministically to its optimal ( high ) value , whereas in the presence of high 6AU , the URA3 expression level undergoes a random walk until expression becomes high enough to allow the tuning mechanism to ‘sense’ the gradient and drive the cells into a URA3+ state . The effects on tuning rates of varying the 6AU concentration are plotted in Figure 5B , where we observe that increasing 6AU concentrations both slow tuning and dramatically increase the variance in the amount of time required for each individual cell to reach a URA+ state . This is precisely the behavior observed experimentally with high 6AU concentrations ( Figure 3 ) . On the other hand , tuning in our experimental system switched from slow and stochastic to rapid and deterministic in the presence of low 6AU concentrations , with observable tuning occurring over the course of a few hours ( Figure 5C ) . Importantly , the tuning process is confined to the URA3-mRuby allele , despite the fact that DHFR-GFP is also being driven by the same synthetic promoter . This again demonstrates that the tuning process occurs independently of conventional gene regulation by dedicated sensory and regulatory input . We utilized time-lapse fluorescence microscopy to monitor the correspondence between expression of URA3-mRuby and cell division in PHSP12-URA3-mRuby/PADH1-DHFR-GFP cells that initiated the tuning process . Consistent with our proposed tuning model , gene expression fluctuations that surpassed a threshold for alleviating the URA3 deficit were reinforced over long timescales and were sustained ( inherited ) across multiple generations as the tuned colony expanded ( Figure 6 ) . As expected , there is no accompanying increase in DHFR-GFP . Similar trajectories were observed for other tuning micro-colonies ( Figure 7A ) . The apparently long autocorrelation time of URA3-mRuby fluctuations through the duration of a tuning trajectory is consistent with our proposed fitness feedback reinforcement mechanism . In order to quantitatively determine the timescale of gene expression fluctuations , also known as mixing times ( Sigal et al . , 2006 ) , we utilized fluorescence-activated cell sorting ( FACS ) to sort a population of cells for the bottom 20% , the top 20% , and complete ( mock-sorted ) distribution of URA3-mRuby expression and measured the timescales over which the sorted fluorescence distributions converged to each other ( Figure 7—figure supplement 1 ) . For cells growing under uracil-replete conditions ( SC+glu ) , we observed a relatively fast mixing time on the order of ~100 min ( Figure 7—figure supplement 1; Supplementary file 3 ) . On the other hand , cells starving in ura-/6AU10 media had mixing times that ranged from 400 to 1200 min ( Figure 7—figure supplement 1; Supplementary file 3 ) . To determine the association of URA3-mRuby levels across generations with growth , we primed cells with 12 hr of exposure to ura-/6AU5 media and then tracked the division of tuned vs . untuned microcolonies of PHSP12-URA3-mRuby/PADH1-DHFR-GFP cells over 24 hr time courses in ura-/6AU5 media . By comparing the fluorescence of cells that are about to divide with those that are not , we found that dividing cells have significantly higher levels of mRuby than non-dividing cells , whereas the separation was much smaller for GFP ( Figure 7C ) . Furthermore , the URA3-mRuby levels within the tuning colony were highly heritable; as seen in Figure 7D , as the indicated colony tunes and grows , cells within that colony maintain a high-mRuby state through subsequent divisions , and even their internal rankings are mostly preserved . mRuby levels in other , non-tuned microcolonies are almost uniformly lower than cells in the tuned colony . The fitness-driven optimization component of our model ( Figure 1 ) further predicts that fluorescence levels should not only be heritable , but also that cells will continue to increase URA3 expression ( possibly noisily ) until they reach either a local optimum fitness or some biological constraint on maximum gene expression . Consistent with our expectation , we observed that the ratio of mRuby to GFP levels ( the latter of which is fused to a gene whose product is not needed for growth ) became steadily higher in cell lineages that had been dividing for longer ( Figure 7E ) . These observations demonstrate that the level of URA3 expression is correlated with fitness , is transmitted across several generations , and shows an ongoing upward trend in tuned cells over the course of time . That last finding is particularly important because a directed increase in URA3 once a lineage begins growing is predicted by our model for fitness-directed tuning , but cannot be explained by other competing hypotheses . The images and data shown in Figure 7 were taken for colonies within a single field of view of a 40x objective to ensure internal consistency in illumination and normalization , but their behavior is representative of our observations across multiple such windows . ( e . g . , Figure 7—figure supplement 2 , panel A ) . Similar quantitative analysis from another experiment beginning directly from growth in SC+glu ( instead of short-term pregrowth in ura-6AU media ) is shown in Figure 7—figure supplement 2 , panels B-D . It is crucial to exclude the possibility that genetic mutations underlie the observed tuning transition on –ura/6AU plates . The ongoing emergence of the tuned state in non-growing cells , over the course of many days , makes mutational mechanisms unlikely . In addition , as seen by microscopy ( Figure 3E ) , no more than 1–3 cell divisions occur prior to the onset of sustained growth in a small fraction of cells . Nevertheless , given the phenomenon of stress-induced mutagenesis in non-growing bacterial cells ( Al Mamun et al . , 2012 ) , we wished to conclusively exclude any possibility of mutational mechanisms . To this end , we note that changes in URA3 expression occurring due to mutations should be stably heritable in the progeny of the tuned cells , which would not be expected to revert to a URA3 low state even after restoration of uracil in the media . To test the reversibility of the URA3 high state , we designed an experimental setup in which tuned colonies isolated from ura-/6AU plates were grown for varying numbers of passages in uracil-replete media ( SC+glu including uracil ) and then re-exposed to uracil starvation ( Figure 8—figure supplement 1 ) . If any genetic mutation were responsible for increasing URA3 expression in the tuned cells , the phenotype should be stable for many generations . On the other hand , stochastic tuning would predict that cells revert to a naïve state following sufficient growth in uracil-containing conditions , as they no longer benefit from URA3 expression . As seen in Figure 8A , cells with synprom-driven URA3 show reversion toward the naïve colony formation rates upon growth in ( uracil containing ) SC+glu media , with recovery apparent even after a single round of growth on an SC+glu plate , and subsequently becoming stronger with additional SC+glu passages . To conclusively exclude mutational mechanisms , we performed untargeted whole-genome re-sequencing of a total of eight isolates with synprom-driven URA3-mRuby ( four colonies from 6AU15 plates and four separate biological replicates taken after the onset of growth in 6AU5 liquid media; see Materials and Methods for details ) . For each case , we scanned the region within 25 kb of the LEU2 locus ( where the URA3 cassettes were integrated ) for mutations , since control of URA3 expression was shown in these cells to operate locally in cis ( Figure 4 and Figure 4—figure supplement 2 ) . The results are summarized in Supplementary file 4: Of the eight isolates , five show no mutations within 25 kb of the URA3-mRuby insertion , two show SNPs of unknown fitness contribution in a minority of the population , and one shows a duplication of the URA3-mRuby cassette ( based on the presence of a read density that is twice the level observed elsewhere for the same chromosome ) . These data clearly indicate that the origin of growth-supporting URA3 expression levels in these cells cannot be reliant on a mutational mechanism , as only one of the eight cases – that with the URA3 duplication -- shows a mutation at high enough levels in the population to explain the onset of growth ( mutations present in less than half of the population must have arisen after one or more cells in the population had already tuned and began growing , and thus by definition could not be responsible for the initial onset of the growing state ) . The phenotypes caused by the sequence variants observed in populations C2 and L4 are not immediately obvious , but even if they are beneficial , their presence in a minority of cells excludes the possibility that they were responsible for the onset of tuning . Note that it should not be surprising ( and , indeed , would be expected ) that beneficial mutations might arise in a population once it had begun expanding in a new environment due to stochastic tuning . Our findings are consistent with a non-genetically heritable basis for the observed tuning in seven out of eight of the cases examined , as in all other growing lines , mutations near the URA3 gene were either non-existent or present only in a minority of the population . A formal possibility for colony formation in a subset of the population is that growth occurs solely on the basis of pre-existing URA3 levels in cells prior to being exposed to uracil deprivation . Microscopic observations of starving cells ( Figure 3E ) argue against this possibility , as a substantial lag passes before any cells begin sustained growth . Also , colony formation continues over the course of many days ( Figure 3B–D ) , demonstrating that even cells that were non-growing for a substantial time period after exposure to URA- stress can eventually grow under this condition . Nevertheless , to conclusively discount the possibility of pre-existing URA3 levels determining tuning , we sorted populations of cells on the basis of initial URA3 expression , isolated those with the highest mRuby levels ( the top 0 . 5–1% , well outside of the main distribution of the population ) and plated them . These experiments clearly showed that the ability to form colonies on ura-/6AU plates is not restricted to cells with initially high URA3-mRuby expression ( Supplementary file 5 ) , as the highly fluorescent cells do not form colonies on ura-/6AU plates at rates substantially higher than unsorted cells , and certainly not at a sufficiently higher rate to fully explain the observed colony formation rates . These data argue against the possibility that growth occurs only in cells that , by chance , already have high levels of URA3 expression at the time of plating ( although such cells may have some slight advantage , given the nature of their initial state ) . The proposed fitness-directed tuning mechanism relies on the capacity of local chromatin to maintain a memory of recent changes in transcription , and to modulate the transcription rate based on the fitness consequences of those changes , as conveyed by the proposed central metabolic integrator of health/fitness . We hypothesized that chromatin modification machinery may be intimately involved in these processes . To probe the mechanistic basis of stochastic tuning , we focused on perturbations to histone acetylation/deacetylation ( deletions of GCN5 , SIN3 , HST3 , HST4 ) , and chromatin remodeling ( deletions of ASF1 , ISW2 , SWR1 , UBP8 ) , all of which provide potential pathways for coupling feedback from the cell’s physiological state to allele-specific modulation of chromatin and transcription ( See Table 1 for details ) . We selected these targets because of their association with genes showing particularly high levels of noise ( and thus , more likely to be driven by tuning ) in single-cell proteomic analysis ( Newman et al . , 2006 ) . In our screening , homozygous replacements of HST3 , HST4 , SWR1 , ISW2 , and UBP8 with a kanMX cassette showed little effect on colony formation rates on ura-/6AU plates , and SIN3::kanMX/SIN3::kanMX strains showed severely compromised cell survival under growth-arrested conditions; all were excluded from further analysis . On the other hand , we found that genetic perturbations to the histone acetylation machinery through deletion of the key histone acetyltransferase GCN5 essentially abolished tuning , whereas deletion of the histone chaperone ASF1 , in contrast , increased tuning rate by more than an order of magnitude ( Figure 8B ) . At the same time , we show that the observed tuning process does not rely on transcriptional memory mechanisms grounded in chromatin localization , given the lack of effect of a NUP42 deletion ( Figure 8B; cf . ( Guan et al . , 2012 ) ) . In interpreting our data on the effects of genetic perturbations on tuning ( Figure 8B ) , it was crucial to consider the possibility that cells may lose viability at variable rates under different conditions , which could contribute to the observed differences in colony formation rates . We thus performed experiments to measure the rate of cell death in the presence of uracil starvation and compared the results with the different colony formation rates observed . As shown in Figure 8—figure supplement 2 , the effects of a mutation on survival and tuning rates are not significantly correlated . For example , deletion of GCN5 resulted in the nearly complete loss of stochastic tuning , deletion of NUP42 had no effect , and deletion of ASF1 substantially enhanced tuning , yet none of these mutations shows a change in survival rates during incubation in uracil-free media compared with wild type cells sufficient to explain the observed change in colony formation rate ( Figure 8—figure supplement 2 ) . Even for the poorest surviving strain , GCN5::kanMX/GCN5::kanMX , colony formation rates after ten days are 100−1000 times lower than wild type cells even though survival rates are lower only by a factor of ten . Given the apparent importance of chromatin modifications in fitness-directed tuning , we also tested the effects of nicotinamide treatment ( which inhibits the sirtuin class of histone deacetylases , or HDACs ( Bitterman et al . , 2002 ) ) on reversion of the tuned cells back to a naïve state . As shown in Figure 8A , we found that chemical inhibition of sirtuin HDACs by nicotinamide treatment substantially accelerated the decay of a tuned population to the naïve state , further highlighting the importance of histone modification in stochastic tuning . Combined with the data on knockout strains described above , our results suggest a central role for chromatin modifications in the establishment and maintenance of the tuning process , although the molecular details cannot be discerned from these data alone . The abstract model introduced in Figures 1–2 demonstrates the potential utility of fitness-directed stochastic tuning to establish adaptive gene expression states without directly sensing the external environment . In order to substantiate the biological feasibility of stochastic tuning , we implemented its critical components in a plausible simulation incorporating generic features of chromatin modification and the information flow of the Central Dogma of Molecular Biology . We therefore designed and simulated a dynamical model tracking transcription rates , transcript levels , protein levels , and histone modifications in a single cell , with parameter distributions sampled from experimental data ( Figure 9A; see Methods for details ) . We incorporated the possibility of adding or removing chromatin marks that can alter the transcription rates of the associated genes . Our model incorporates two classes of marks: tuning marks ( T ) , which link cellular fitness to transcriptional output by having mark addition rates that are a function of the recent direction of change in global fitness and current number of such marks at each promoter; and stabilizing marks ( S ) , which are added at a rate dependent on the number of tuning marks at each promoter ( Figure 9B ) . At any time , the transcriptional output of the promoter is a function of the density of both tuning marks and stabilizing marks . As such , the tuning marks provide a critical connection between changes in global fitness and transcription rates , whereas the more slowly changing stabilizing marks capture the average transcriptional output over longer timescales , enabling a more stable optimization trajectory . Both T and S chromatin marks come in two varieties: positive ( activating ) and negative ( repressive ) . Our aim was to develop a generic simulation consistent with our general knowledge of coupling between chromatin modification and transcription ( Li et al . , 2007; Rando and Winston , 2012; Zhou and Zhou , 2011; Mitra et al . , 2006 ) . As such , the tuning and stabilizing marks described here need not correspond to any specific chemical moiety or be attributed to any particular histone modification enzyme . Modulation of enzyme activity by global fitness could be due to some as yet unknown signaling pathway or , alternatively , be dependent on known metabolic substrates or cofactors , such as acetyl-CoA and NAD+ ( Lin et al . , 2000; Thaminy et al . , 2007; Tanner et al . , 1999 ) . As shown in Figure 9C , the detailed model is capable of stochastic tuning of a single gene which strongly impacts the fitness of the cell ( as would be the case for URA3 in our experimental setup ) . For most randomly generated gene-level parameters , stochastic tuning results in substantially higher fitness compared to when cells undergo random fluctuations in transcription levels or when transcription is fixed at a rate appropriate for a different environment , and in most cases , tuning is able to consistently achieve near-optimal expression levels . The model is robust to variations in both the sampled biological parameters ( Figure 9C ) and the parameters of the model itself ( Figure 9D ) and can locate an optimal expression level regardless of the ratio between the initial and target protein levels ( Figure 9E ) . These results demonstrate that a generic , biologically feasible implementation of fitness-directed stochastic tuning can in fact function even in the presence of the multiple layers of noise and temporal delays acting between transcription rates ( at which tuning occurs ) and protein levels ( which dictate fitness ) . Note that we do not expect to find conditions where stochastic tuning is the primary mechanism of gene expression modulation for every gene in the genome , even for novel or extreme environments . Rather , we expect that the cells’ hard-wired transcriptional regulatory logic exerts the primary role in the transcriptional reprogramming of the majority of genes in the genome . For its part , we expect that stochastic tuning plays the dominant role in modulating the expression of few genes/pathways that represent critical bottlenecks for fitness ( for example , induction of a drug efflux pump , or repression of an enzyme that activates a pro-drug chemotherapeutic agent ) .
We have described a mechanism of adaptation through fitness-directed optimization of gene expression . In numerical simulations , the proposed framework has the remarkable capacity to simultaneously tune the expression of thousands of genes , enabling optimization of fitness without directly sensing environmental parameters . The demonstration that a phenomenon consistent with fitness-directed stochastic tuning operates in S . cerevisiae has important implications for the adaptation of eukaryotic microbes to novel or extreme environments where their genetically encoded regulatory networks become inadequate . However , we speculate that stochastic tuning operates in parallel with conventional regulation even in frequently encountered environments . Indeed , hard-coded sensory and regulatory networks are unlikely to have the encoding capacity to optimally respond to every conceivable subtle change in the environment—even within the native habitat . We therefore favor a model in which dedicated regulatory networks quickly move the system to a state reasonably well matched to a given condition , and stochastic tuning subsequently optimizes expression to achieve a more precisely adapted state for every individual encounter . The ability to discover optimal gene expression states through a stochastic fitness-directed search may have provided significant advantage to early eukaryotic microbes . Microorganisms have evolved stochastic search strategies in other contexts . Indeed , the proposed stochastic tuning mechanism is reminiscent of the biased random walk phenomenon in bacterial chemotaxis , where stochastic transitions in the rotation of the flagellar motor are biased towards the direction that increases chemoattractant signaling over time ( Macnab and Koshland , 1972 ) . Detailed molecular mechanisms of chemotaxis have been revealed over the course of the last few decades , demonstrating the versatility of molecular processes in implementing rather complex computations ( reviewed in ( Sourjik and Wingreen , 2012 ) ) . Although our main focus here has been on establishing the phenomenology of fitness-directed stochastic tuning , we have already identified some critical components . In particular , histone acetylation/deacetylation ( via GCN5 and sirtuins ) seem to play a critical role , as deletion of GCN5 almost entirely abolished tuning . This is consistent with the high degree of intrinsic noise exhibited by the genes that are regulated by the SAGA complex , in which GCN5 is the catalytic subunit ( Newman et al . , 2006 ) . Previous work has shown that increased transcriptional noise is beneficial for adaptation to acute environmental stress ( Blake et al . , 2006 ) . Interestingly , however , early work demonstrated that deletion of GCN5 further increases expression noise in the context of the PHO5 promoter ( Raser and O'Shea , 2004 ) . Taken together , these data suggest that stochastic tuning is not driven by noise alone; rather we support a model in which the proper integration of noise , transcriptional memory , chromatin modification , and cellular-health feedback work together to implement a directed search mechanism to drive the expression level of individual genes to levels that maximize the overall health of the cell . Indeed , histone modification is tightly coupled with gene expression . Co-transcriptional histone modification can store recent memory of transcriptional activity ( Li et al . , 2007; Rando and Winston , 2012 ) and histone modification can , in turn , affect transcription rate ( Stasevich et al . , 2014 ) . There has been a longstanding debate on the functional significance of this reciprocal coupling . Our model and results help to unify these phenomena and support their functional relevance as requisite components of a stochastic tuning-based cellular adaptation framework . We note that our experimental setup for demonstrating stochastic tuning has superficial similarities to a series of experiments performed in S . cerevisiae by the Braun lab , in which they sought to determine whether glucose-driven repression of the GAL1 promoter could be overcome to allow expression of a HIS3 construct in glucose-containing media ( Stern et al . , 2007; Stolovicki et al . , 2006 ) . While the authors observed consistent emergence of growth in a large fraction of cells that they initially noted could be attributed to either genetic or epigenetic mechanisms ( Stolovicki et al . , 2006 ) , subsequent analysis has shown that in that experimental system , genetic mutations are the primary mechanism of adaptation , possibly driven by hypermutability of the genes involved in the response of interest ( David et al . , 2010; Moore et al . , 2014; David et al . , 2013 ) . These mutational mechanisms stand in clear contrast to the rapidly reverting epigenetic stochastic tuning observed in our experiments . In addition to perception of environmental parameters , cells also possess a variety of hard-wired homeostatic mechanisms sensing and responding to internal parameters , optimizing resource allocation in response to parameters such as growth rate ( Klumpp et al . , 2009; Klumpp and Hwa , 2014; Brauer et al . , 2008; Barenholz et al . , 2016; Keren et al . , 2013 ) and metabolite/nutrient pools ( Potrykus et al . , 2011; Broach , 2012 ) . However , while these mechanisms allow cells to sense their internal state , they still reflect specific evolved responses to alter resource allocation and gene expression in a predefined way in response to stress , standing in contrast with the ability of stochastic tuning to conduct a search and discover arbitrary gene expression states that are adaptive under extreme and unfamiliar environments . The widely varying tuning rates for different promoters ( Figure 3B–C and Figure 3—figure supplement 1 ) clearly indicate that sequence features can influence tuning efficacy . By design , all but one promoter driving URA3 in our experiments contained a TATA box , which has been linked to high intrinsic noise ( Newman et al . , 2006 ) , condition-specific expression variability ( Tirosh et al . , 2006 ) and reliance on chromatin-mediated regulation ( Tirosh et al . , 2008; Basehoar et al . , 2004 ) . Indeed , replacement of the ( TATA-containing ) PSAM3 derived sequence in synprom with a similarly generated sequence from the TATA-free PARF1 promoter substantially reduced tuning rates under the conditions tested ( Figure 3—figure supplement 1 ) . We also note that when we performed experiments similar to those described above with the repressed natural promoter PGAL1 , we observed dramatically lower rates of colony formation ( less than 1 in 107 ) , and those colonies that did form appeared to be non-reverting genetic mutants ( data not shown ) . Exploring the full importance of transcriptional noise for tuning efficiency , as well as that of other features such as propensity for nucleosome positioning , will be important in future work . Fitness-directed stochastic tuning requires feedback of the global state of health to every promoter in the genome . The dependence of many histone modification enzymes on metabolic intermediates and cofactors ( e . g . , NAD+ for the sirtuin family of histone deacetylases ( Lin et al . , 2000; Thaminy et al . , 2007 ) ; SAM for histone methyltransferases ( Luka et al . , 2009 ) , and acetyl-CoA for histone acetyltransferases ( Tanner et al . , 1999 ) ) provides support for potential direct feedback of global fitness-related parameters to the epigenome ( Katada et al . , 2012; Kurdistani , 2014 ) , and indeed we showed that chemical manipulation of sirtuin activity had substantial effects on retention of epigenetic memory . These enzymes may very well serve as distinct channels of health-related information utilized by stochastic tuning . In this regard , chromatin itself may function as a global health integrator , with histone modifications and their effect on gene expression being highly contingent on the current trajectory of cellular fitness . Alternatively , cells may utilize a single global health integrator ( such as the mTOR system ) as hypothesized in our idealized model . The mTOR pathway integrates diverse parameters of internal health including energy , nutrient availability , and cellular stresses ( González and Hall , 2017 ) . Intriguingly , the mTOR pathway has recently been shown to regulate histone acetylation states through a variety of mechanisms ( Chen et al . , 2012; Workman et al . , 2016 ) Fitness-directed stochastic tuning has important implications for gene regulation . Beyond a potentially widespread mechanism of cellular adaptation , stochastic tuning brings together seemingly unrelated phenomena under a unifying conceptual framework . These are areas of study at the frontier of genetics and biochemistry , including stochastic gene expression , transcriptional memory , and metabolic modulation of epigenetic states . Stochastic tuning may have initially evolved as a mechanism for adaptation of single-cell eukaryotes to extreme environments . However , once available , it may have found additional utility as a versatile mechanism for controlling and fine-tuning gene expression in the context of physiological and developmental processes in metazoans . This is consistent with the evolutionary arc of an ancient set of molecular mechanisms that now serve as key mediators of differentiation ( Álvarez-Errico et al . , 2015; Ziller et al . , 2015; Meissner , 2010 ) . Exploring this possibility represents an important area for future research . Optimization of cellular health through the fitness-directed stochastic tuning mechanism may also play an important role in allowing cancer cells to survive and thrive in a variety of microenvironments unfamiliar to their evolved regulatory networks , and in the face of extreme challenges imposed by chemotherapy and radiation . Indeed , stochastic tuning may underlie the epigenetically mediated metastatic potential and chemotherapy resistance observed in a variety of cancer types ( Wu and Roberts , 2013; Perez-Plasencia and Duenas-Gonzalez , 2006; Lv et al . , 2016; Li et al . , 2015; Borley and Brown , 2015; Bonito et al . , 2016; Shaffer et al . , 2017 ) . Our observations support the existence of a fitness-directed tuning process that operates at the level of transcription . However , in principle , tuning could also occur at any point along the hierarchy of gene expression where noise , memory , and feedback of global fitness can drive the activity of gene products towards levels that optimize the overall health of the cell . In particular , searching for evidence of tuning at the level of translation would be an important focus for future research .
For routine growth of strains , we used YPD broth ( 10 g/L yeast extract , 20 g/L peptone , 20 g/L dextrose ) or YPD agar plates ( YPD broth +20 g/L Bacto agar ) . We used standard recipes based on SC+glucose ( SC+glu ) ( Kaiser et al . , 1994 ) for all physiological experiments . SC/loflo refers to SC made with low fluorescence yeast nitrogen base ( US Biologicals ) . In the case of SC+glu , we used dropout supplement powders interchangeably from ForMedium ( DSCK012 ) and US Biologicals ( D9515 ) , although they differ slightly in the concentrations of adenine and para-amino benzoic acid supplied . SC+glu derivatives lacking particular nutrients are specified as SC+glu-NUTRIENT; e . g . , SC+glu-ura for SC+glu lacking uracil . We also refer to the commonly used mixture of SC+glu-ura with 6-azauracil added as ura-/6AUi , where i is the final concentration of 6AU in microgram/mL . The agar for all plates used in physiological experiments was either Noble agar ( Difco ) or quadruple-washed Bacto agar . For the removal of the GAL-GIN11 cassette in counter-selections ( see below ) , cells were plated on YPGA agar plates ( 10 g/L yeast extract , 20 g/L peptone , 20 g/L galactose , 20 g/L agar , 100 microgram/mL ampicillin ) . All growth was at 30°C; liquid phase growth included shaking at 200–220 rpm in an Innova 42 incubator ( New Brunswick ) . As diagrammed in Figure 3A , we constructed two classes of insertion cassettes . Each follows the pattern of having a promoter , a functional reporter protein fused to a fluorescent protein , and then ends with a CYC1 terminator . For URA3 , the native sequence from S . cerevisiae was used , with the exception of one silent SNP and an A160S mutation that does not appear to alter enzyme function . The red fluorescent protein mRuby is described in ( Kredel et al . , 2009 ) . For DHFR , we used murine DHFR from pSV2-dhfr ( Subramani et al . , 1981 ) with an L22R mutation making it methotrexate-resistant ( Simonsen and Levinson , 1983 ) . GFP refers in all cases to superfolder GFP ( Pédelacq et al . , 2006 ) codon-optimized for S . cerevisiae using web-based tools from IDT ( Integrated DNA Technologies ) ; see Supplementary file 3 for the corresponding nucleotide sequence . In each case , the reporter and fluorescent protein were separated by a short A/G/S containing linker . All constructs were cloned in bacterial hosts using pBAD-derived plasmids; separate plasmids were constructed with each promoter of interest downstream of a region homologous to the upstream target site in the S . cerevisiae genome , and URA3-mRuby-cyc or DHFR-GFP-cyc upstream of a region homologous to the downstream target site in the S . cerevisiae genome . All constructs were chromosomally integrated at the leu2∆0 locus of our yeast strains . Double-stranded DNA for transformation in yeast was then generated by first amplifying the promoter and reporter constructs separately , using primers yielding 20–40 bp overlaps; we then used crossover PCR to generate the complete construct of interest and subsequent amplification to generate a sufficient quantity for transformation . All PCR used for strain construction was performed using Q5 high fidelity polymerase ( NEB ) ; routine PCRs for strain validation were instead performed using OneTaq or Taq polymerase ( NEB ) . Promoters for ADH1 , HSP12 , and RGI1 were cloned from our wild type strain ( BY4743 or its haploid progenitors BY4741/BY4742 ) and included the entire region from 1700 to 1800 bp upstream of the start codon to the base immediately prior to the start codon . The ADH1 promoter was selected as a classic constitutive promoter ( DeMarini et al . , 2001 ) ; HSP12 and RGI1 were chosen as they show high variance in expression between conditions ( Tirosh et al . , 2009; Tirosh et al . , 2006 ) , a characteristic expected to be favorable for stochastic tuning . Synprom was designed in two stages: the bulk of the DNA is a 600 bp random sequence generated using a Markov model to match the trinucleotide frequencies present across all natural S . cerevisiae promoters . To this sequence we appended the 200 bp immediately prior to the start codon of SAM3 , to provide native transcription and translation start sites . The resulting sequence was then modified to remove all recognizable binding sites for yeast transcription factors ( TFs ) as follows: we used the set of position weight matrices and match thresholds in ScerTF ( Spivak and Stormo , 2012 ) to identify all recognizable TF binding sites in the promoter , and randomized the sequences of only those regions and their immediate surroundings until no recognizable TF binding sites remained . The resulting perturbed sequence is given as Supplementary file 1 . The required sequences were synthesized as gBlocks from Integrated DNA Technologies and combined via Gibson assembly ( Lartigue et al . , 2009 ) . All yeast strains were derived from BY4741 or BY4742 ( Brachmann et al . , 1998 ) , which includes a complete deletion of the URA3 ORF ( BY4741: Mat a his3Δ1 leu2Δ0 met15Δ0 ura3Δ0; BY4742: Mat α his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0 ) . Insertions of URA3 or DHFR fusion proteins were always at the leu2∆0 locus unless otherwise noted . To facilitate consistent insertion , we replaced the leu2∆0 allele of BY4741/BY4742 with a LEU2-GAL-GIN11 cassette ( Akada et al . , 2002 ) , which allows growth in leucine-free media but inhibits growth in the presence of galactose . We note that at least in our copy of the BY474x strains , the leu2∆0 deletion runs only from ChrIII:84799—ChrIII:93305 , rather than extending to position 93576 as annotated . Nevertheless , the deletion is sufficient to remove the entire leu2 open reading frame . Strains containing the fusion proteins were constructed by transforming the LEU2-GAL-GIN11 containing cells with appropriate double-stranded oligos ( see above ) and selection on YPGA plates , allowing replacement of the LEU2-GAL-GIN11 cassette with the desired insert . Insertions were confirmed by PCR product sizing . Diploid strains were derived by mating one BY4741-derived ( mat a ) strain with one BY4742-derived ( mat α ) , and subsequently plating on SC+glu-lys-met or SC+glu-lys-met-cys . All transformations were carried out using the LiAc-PEG-ssDNA method ( Gietz and Woods , 2002 ) . Knockout strains were generated by beginning from appropriate haploids containing either a leu2::promoter-URA3 or leu2::promoter-DHFR construct or simply leu2∆0 , amplifying an appropriate kanMX knockout cassette from the corresponding strain in the S . cerevisiae gene deletion collection ( Giaever et al . , 2002 ) , and selecting on YPD+G418 plates . We confirmed the presence of kanMX at the appropriate site and absence of the native gene by PCR . Diploid knockout strains containing appropriate deletions and a URA3-mRuby insertion at leu2∆0 were generated by mating these haploids as noted above . Experiments showing colony formation rates over time all follow a common formula . Cells were grown overnight in SC+glu media , and then in the morning back-diluted 1:200 into fresh , prewarmed SC+glu . The cells were grown for four to five hours at 30°C with shaking and then pelleted , washed once with 25 mL deionized ( DI ) water , pelleted , washed with 1 mL water , pelleted , and resuspended in 1 mL water . Specified dilutions were made in DI water from this final cell suspension . Cells were then either plated on full plates at pre-chosen dilutions ( 100 microliters of an appropriate cell suspension ) , or a dilution series was spotted onto appropriate agar plates ( 10 microliters per spot ) . Plates were imaged and counted every 1–2 days for the duration of the experiment ( lasting between a few days and weeks , depending on the experiment in question ) . Plates were wrapped in parafilm after ~3 days to minimize drying . Plating was performed identically on SC+glu plates ( to establish the number of cells being plated ) and plates containing one or more test conditions ( e . g . , ura-/6AU ) . Cells were counted either directly from the plates or from stored digital images . Direct plate counts were done manually for all visible colonies; for those counted from saved images , we imposed a minimum size threshold of 0 . 2 mm in diameter ( rounding up to the nearest pixel ) . Times for counts were rounded to the number of days since plating . To determine the survival rates of cells undergoing uracil starvation in the presence of various other perturbations , we measured the death rates of cells lacking any copy of URA3 in SC-ura+glu media . Cells were pregrown and washed as described above for plating assays , but then resuspended in liquid SC-ura+glu media and incubated at 30°C . Aliquots were regularly removed and spotted on SC+glu plates to determine the number of viable colonies . Survival rates are for leu2∆0 homozygotes ( the original BY4743 diploid , possibly with a homozygous deletion of a specified gene ) with no available copy of URA3 . Cells were analyzed by flow cytometry on an LSR Fortessa ( Becton Dickinson ) at the Columbia University Microbiology and Immunology Flow Cytometry Core Facility or University of Michigan Flow Cytometry Core . Cells to be used in these experiments were initially prepared and washed following the same pregrowth procedure as given above for colony formation assays , except that growth was in low fluorescence SC/loflo media instead of SC . A flask containing 25 mL of prewarmed media ( generally ura-/6AU5 made from an SC-ura/loflo base ) was then inoculated with 200 microliters of the cell suspension , and cells were grown with shaking at 30°C . Subsequent data acquisition varied depending on the experiment to be performed . For the long time courses shown in Figure 4 and its supplement , for an initial timepoint , 200 microliters of the washed cell suspension were combined with 500 microliters of 2x PBS/E ( 1x PBS with 10 mM EDTA added ) , 290 microliters DI water , and 10 microliters of flow cytometry counting beads ( Invitrogen CountBright beads ) . At subsequent timepoints , snapshots were taken by combining 490 microliters of the growing cells , 10 microliters counting beads , and 500 microliters 2x PBS/E . In either case , cells were run on the Fortessa , with signals recorded for forward and side scatter , mRuby ( using the Texas Red laser/filter set ) , and GFP ( using the FITC laser/filter set ) . Data were analyzed using the flowCore and flowViz modules of R ( Ellis et al . , 2006; Ellis et al . , 2009 ) . Beads and cells were first identified based on their forward scatter and side scatter ( FSC/SSC ) values ( using permissive gates that capture the vast majority of each population ) and fluorescence ( beads were required to show very high fluorescence ) . For each growth phase ( exponential in SC+glu , starving in ura-/6AU , growing in ura-/6AU ) , we obtained empirical autofluorescence corrections by analyzing populations in a similar growth state lacking the fluorescent tag on URA3 . Guided by exploratory analysis , we fit a linear model for starving cells predicting mRuby and GFP autofluorescence as a function of the observed forward and side scatter , and used constant autofluorescence values characteristic of each of the two growing phases ( obtained from cells with no fluorescent protein in a similar physiological state , either uracil-starved or undergoing stochastic tuning-driven growth ) . During analysis of liquid phase fluorescent populations ( shown in Figure 4 and its supplement ) , the predicted autofluorescence values were subtracted from the observed value; in these cases , an additional gate was applied to remove events with very low forward scatter values , which had a very high variance in fluorescence and were well below the size of the main population . For the use of FACS followed by plating to test the colony formation rates of highly fluorescent cells , cells were prepared as described above , sorted using a BD FACSAria , and then subsequently plated in equal quantities on SC+glu and ura-/6AU15 plates . For the short timescale tuning data shown in Figure 5C , the cells were grown for 3–4 hr side by side in SC/loflo + glu and –ura/loflo/6AU1 media , and then placed on ice and run directly on the flow cytometer . For each biological replicate ( performed on different days ) , we grew leu2::synprom-URA3-mRuby/leu2::synprom-DHFR-GFP and nonfluorescent leu2::URA3/leu2∆0 cells in parallel to allow direct comparison of the observed fluorescence levels . Analysis was performed separately for each biological replicate . We first normalized all fluorescence signals by the FSC-A signal raised to the power of 1 . 5 , which we found empirically to be an effective correction removing most of the dependence of the fluorescence on cell size . Next , a mapping of FSC signals to expected autofluorescence on each channel was fitted using the R loess function ( with default parameters ) , and the expected autofluorescence subtracted from the observed value for each cell to yield what we refer to as the blanked fluorescence . We then calculated and compared the changes in the median blanked fluorescence of the populations for the same cells grown in SC+glu vs . ura-/6AU1 media . Confidence intervals were calculated by bootstrapping with 200 bootstrap replicates . Cells for whole genome sequencing were taken directly from the growth condition of interest ( ura-/6AU15 plate or ura-/6AU5 liquid media ) and flash frozen in 15% glycerol or 1x TES ( 10 mM Tris , pH 7 . 5; 10 mM EDTA , 0 . 5% SDS ) . One reference sample grown under unselective conditions was taken for each starting strain to use as a baseline . Genomic DNA was isolated using a YeaStar Genomic DNA kit ( Zymo Research ) according to the manufacturer’s instructions . Samples were then barcoded and prepared for sequencing using a Nextera XT kit ( Illumina , Inc . ) and sequenced as part of a pooled library on a NextSeq ( Illumina , Inc . ) . Sequencing reads were clipped to remove adapters and commonly observed artifactual end sequences with cutadapt ( Martin , 2014 ) , and then further trimmed using Trimmomatic 0 . 30 ( Bolger et al . , 2014 ) to remove very low quality ( <3 ) end bases , retain only the portion of the read with a quality score above 15 in a four base sliding average window , and remove reads less than 10 bp long . Surviving trimmed reads were then aligned to the reference genome using Bowtie 2 . 1 ( Langmead et al . , 2009 ) ; the reference genome was constructed from the S . cerevisiae S288c genome ( GenBank BK006934 – BK006949 ) , deleting the URA3 ORF and inserting the sequence for the appropriate URA3 and DHFR constructs in separate copies of chromosome III at the LEU2 locus . Read data used in this analysis are available from the Short Read Archive under accession SRP117724 . After alignment , mutational calls and read depths were obtained using the mpileup and depth modules of samtools 0 . 1 . 18 ( Li et al . , 2009 ) , respectively . Reads for called variants within 25 kb of the insertion site were examined manually and compared to the sequenced parental strain; validated variants are listed in Supplementary file 4 . RNA was isolated using an adaptation of the hot acid phenol method ( Collart and Oliviero , 2001 ) . Cells for RNA isolation were grown under appropriate conditions ( either in liquid phase or on agar plates ) , and then snap-frozen in 1x TES ( 10 mM Tris , pH 7 . 5; 10 mM EDTA; 0 . 5% SDS ) and stored below −70°C . Snapshots of 200 to 600 microliters were taken from growing liquid phase cultures , whereas from agar plates we harvested 1–20 colonies of <0 . 5 mm diameter taken from the same plate as each biological replicate . RNA was isolated by rapidly thawing the cell suspension and mixing 1:1 with a 5:1 acid phenol:chloroform solution , then incubating 60 min at 65°C with occasional vigorous vortexing . The solution was then chilled on ice for 5 min , and centrifuged 5 min at 16 , 000 x g at 4°C . The aqueous phase was mixed 1:1 with additional acid phenol:chloroform , chilled , and centrifuged as before . The aqueous phase was then mixed 1:1 with a 24:1 chloroform:isoamyl alcohol solution , and centrifuged 5 min at 4°C . The resulting aqueous phase was transferred to a fresh tube and combined with 1/10 vol 3 M sodium acetate , 2 volumes of 1:1 ethanol:isopropanol , and 1/800—1/200 vol Glycoblue ( Ambion ) , and then precipitated for at least 1 hr at −20°C and then at least 1 hr at −80°C . RNA was recovered by centrifuging 15 min at 16 , 000 x g at 4°C , washed with ice cold 75% ethanol , spun an additional 5 min , and then air-dried and resuspended in RNAse-free water . The samples were then further purified using a Zymo RNA clean and concentrator five according to the manufacturer’s instructions , including an on-column DNase digestion . Total RNA was purified from cells in the desired growth condition using the hot acid-phenol procedure described above . cDNA pools were generated for each sample using random hexamer-primed reverse transcription with Protoscript II ( New England Biolabs ) following the manufacturer’s instructions . cDNA pools were used directly in qPCR reactions without further purifications , assembling reactions using iTaq Universal SYBR Green Supermix ( BioRad ) following the manufacturer’s instructions , in GeneMate PCR plates . Plates were sealed with Microseal ‘B’ adhesive film ( BioRad ) and run on a BioRad CFX96 detection system . Ct values calculated by the instrument software were then exported for subsequent analysis . All isolated RNA was quantified on a Bioanalyzer ( Agilent ) and found to have an RIN >= 6 . 8 . For comparison of URA3 and DHFR expression , we calculated separate ∆Ct values for each qPCR run replicate by taking the median of all technical replicates from that run . Values plotted in Figure 4—figure supplement 2 reflect ∆Ct data from 1 to 2 technical replicate wells on each of two to four separate , independently assembled runs; we plot the median of day-wise data points for each separate biological sample . Primer locations and sequences are given in Supplementary file 7 . We performed a no-reverse transcriptase control reaction for each sample to ensure that DNA contamination did not contribute to the observed signal ( data not shown ) . qRT-PCR data were analyzed using a Bayesian hierarchical model treating the ∆Ct value between the URA3 and DHFR primers as follows:∆Ct ( sample , day ) ~T ( µs ( sample ) , σrep , νrep ) µs ( sample ) ~T ( µc ( class ) , σc ( class ) , νbio ) Parameters not otherwise specified were assigned appropriate uninformative priors . Here ‘sample’ refers to a single biological sample and ‘class’ to a single growth condition . The key parameter of interest is µc for each class of cells under study , the overall average URA3:DHFR difference for cells grown under that condition . We fitted the model using JAGS ( Plummer , 2003 ) , and then report credible intervals and other inferences from the posterior distribution on µc . Each of the ∆Ct ( sample , day ) values used the median across 1–2 technical replicates for each primer pair . Data were analyzed using custom-written python and R scripts . Source code for the nontrivial analysis of flow cytometry data is provided as Source code 1 . Uncertainties for cell counts ( shown in plating and flow cytometry data ) were calculated by treating each observed count as a Poisson random variable; using Bayesian inference with the Jeffreys prior ( Jeffreys , 1961 ) , the posterior distribution for the rate parameter I ( the concentration of cells ) is given in closed form by I ~ Gamma ( 0 . 5+∑i=0nin , n ) Where n is the number of observations and the in are the observed counts . Error bars then indicate a central 95% credible interval for I given the observed data . Experiments to examine the reversion of tuned colonies toward a naïve state were performed as shown in Figure 8—figure supplement 1 . Single colonies from a ura-/6AU15 plate were streaked out onto SC +glu and allowed to grow . From that plate , single colonies were again picked and underwent repeated passages in liquid media; each ‘passage’ refers to a 200-fold dilution , which is then allowed to grow for 48 hr ( 96 hr for the very first transfer ) . Cells were also taken for plating from the original ura-/6AU15 plate , the first SC +glu plate stage , and several subsequent time points during liquid culture . Cells taken from plates were immediately diluted in water and spotted on SC +glu and ura-/6AU15 to track colony formation rates; cells taken from liquid passages were streaked out on SC +glu plates prior to use in spottings , in order to obtain a consistent physiological state . Plots for ‘naïve’ cells refer to cells treated identically , except that they had initially been grown on SC +glu plates instead of ura-/6AU15 plates . Recovery was assessed based on the amount of time required for 1 in 10 , 000 cells spotted on the new ura-/6AU15 plate to form countable colonies ( using linear interpolation of colony counts between observed data points ) ; in the event that one dilution yielded no colonies passing our size threshold , but the next ( 10-fold more concentrated ) spot gave an uncountable haze , we assigned a count of 1 to the more concentrated spot . The numerical simulations shown in Figures 2 and 5 were performed by implementing the model described in the text using the Matlab programming language and simulated using Matlab ( Mathworks , Inc . ) or GNU Octave version 3 . 8 . 1 ( Eaton et al . , 2009 ) , with qualitatively equivalent results obtained in either case . All simulations were performed using the same initial conditions ( but different random seeds , for the sampling shown in Figure 5 ) . Octave code implementing this model is provided as Source code 2 . The physiological tuning model employed for Figure 9 and the accompanying text was implemented in python , and simulated using python 2 . 7 . 6 , making heavy use of the numpy ( Svd et al . , 2011 ) and scipy ( Jones et al . , 2001 ) libraries , with data analysis and plotting using matplotlib ( Hunter , 2007 ) and pandas ( McKinney , 2010 ) . The details of the physiological model itself are given below . To provide a suitable mechanistic model for stochastic tuning , we developed a discrete-time model tracking the temporal evolution of transcription rates ri , t ( continuous , changed in response to random fluctuations and potentially tuning input ) , copy number of each transcript per cell xi , and copy number of each protein per cell pi , considered separately for each gene i . Transcriptional regulation lies at the center of our consideration for fitness-directed tuning . In the physiological model , there is a time-dependent probability ri , t for a single transcript to be generated from gene i at each timestep; the probabilities ri , t are updated in response to changing fitness as described below . In addition , each copy of the transcript present in the cell has a fixed probability di of being degraded at each timestep . The net change at each timestep t in the transcript level xi for each gene i is thus given by xi , t ~xi , t-1 - binom ( xi , t-1 , di ) +bern ( ri , t-1 ) Here binom/bern are binomial and Bernoulli random variables , respectively . Terms using binomial distributions allow a uniform probability for each present copy of a protein or transcript to be degraded or translated , whereas the Bernoulli term captures the probability of a transcript arising from each gene in a single timestep . We used a timestep of 1 s for all simulations described here . Protein production in our physiological model arises from similar principles . At each timestep , each copy of a transcript from gene i has a fixed gene-dependent probability li of being translated to produce a single copy of the corresponding protein . In addition , each copy of that protein already present in the cell has a gene-dependent probability ei of being degraded . Thus , the net rate of change in the protein copy number pi at each time t is governed by the equation pi , t ~pi , t-1 - binom ( pi , t-1 , ei ) +binom ( xi , t-1 , li ) The fixed , gene-specific parameters di , ei , and li were drawn from distributions that are themselves fits to appropriate experimental data; we then modified the fitted parameters to yield distributions that are contained within the physiological distributions , while excluding the extreme ends of the available range . The parameters used for the physiological rate distributions are summarized below: Transcription rates ( used to initialize the transcription rate distribution , and separately to set the target transcription rate distribution ) : Transcripts per hour are gamma distributed with shape = 5 and rate = 2 ( obtained by fitting data from ( Holstege et al . , 1998 ) and excluding extreme values ) Transcript degradation rates di: Half lives in minutes have a gamma distribution with shape = 12 . 0 and rate = 0 . 75 ( obtained by fitting data from ( Holstege et al . , 1998 ) and then modifying to exclude extreme values ) . Protein degradation rates ei: Half lives in hours have a scaled t distribution with mean = 1 , sigma = 0 . 382 , and 80 degrees of freedom ( fit based on data from ( Christiano et al . , 2014 ) , but modified to exclude long half-lives , consistent with the induction of autophagy in stressed cells ( Cebollero and Reggiori , 2009 ) ) . Protein synthesis rates li: log2 synthesis rates per transcript have a scaled t distribution with mean=-5 , sigma = 0 . 5 , and 80 degrees of freedom ( in units of s−1 ) ; based on protein abundance data from ( Kulak et al . , 2014 ) combined with the other parameters defined above , and modified to exclude extreme values ) . As described in the main text , our model permits two classes of ‘marks’ ( representing histone modifications ) that alter transcription rates: tuning marks ( T ) , which change in level on the basis of recent changes in fitness and the current tuning mark state at each gene , and stabilizing marks ( S ) , which change in abundance based on the tuning mark levels at each promoter . The number of each mark type at each promoter may be positive or negative , reflecting the possibility of distinct activating ( + ) or repressing ( - ) chromatin modifications . The rate of change in the tuning marks proceeds according to the following principles . At each timestep , marks may be added or removed on the basis of recent changes in fitness; each mark may decay with a fixed probability; and marks may be added or removed in an undirected manner due to random drift . Referring to the number of tuning marks at a particular gene i as µi , the change in tuning marks at each timestep due to the tuning contribution alone is given by Δµi , tuning ~sgn ( ΔFt ) * sgn ( µi ) * randint ( 1 , 5 ) * bern ( ptunestep ) Here sgn ( x ) is one if x is positive , −1 if x is negative , and 0 if x is zero . Δ F indicates the difference in mean fitness between the previous nwindow steps and the nonoverlapping block of nwindow steps before that; thus , sgn ( ΔFt ) will be positive if the cells are becoming healthier , and negative if the cells are becoming less healthy . The fitness itself , Ft , is calculated as the Euclidean distance between the observed vector of protein levels pt at a particular timestep , and the median observed in the last quarter of a long ( 10 times the normal simulation length ) trajectory where all transcription rates are fixed at their target values ( note that oscillation still occurs , even in this case of known-correct transcription rates , due to the inherent randomness in transcript and protein production and degradation ) . In the context of our model , ΔF represents the direction of change in global cellular health , and ptunestep indicates the probability that tuning marks will be added/removed at a particular timestep . The combination of signs of the change in fitness ( ΔF ) and marks ( µi ) ensures that if the fitness is increasing and a given promoter has a positive number of tuning marks , the number of tuning marks at that promoter will increase further , whereas if the fitness was decreasing , the number of tuning marks will be decreased . The inverse directions apply for promoters with negative levels of T marks . Note that for control simulations where the effects of tuning are removed , the sign of the fitness-dependent term above is instead taken to be random . The removal and random drift of tuning marks are governed by the equations Δµi , removal ~ −1 * sgn ( µi ) * binom ( µi , pdecay ) and Δµi , random ~ ( 1–2*bern ( 0 . 5 ) ) * bern ( prandom ) respectively . The first equation here indicates that each individual mark may be removed with probability pdecay at each timestep , and in addition , the second equation dictates that each promoter may have a single mark of random sign added at each timestep , with probability prandom . The overall equation for the change in tuning marks at promoter i at each timestep is thus given by the sum of the terms above: Δµi , t ~ Δ Ft * sgn ( µi , t-1 ) * randint ( 1 , 5 ) * bern ( ptunestep ) - sgn ( µi ) * binom ( µi , pdecay ) + ( 1–2*bern ( 0 . 5 ) ) * bern ( prandom ) The stabilizing marks ( S ) , in contrast , do not vary directly in response to fitness , but rather , at each timestep may be added or removed from each promoter depending on its current state of T marks ( see Figure 9B ) : if the promoter has a high transcription rate due to high T levels , the net S count will be increased ( with a probability at each timestep proportional to the current magnitude of the T level ) , and if the promoter has low T levels , the net S count is decreased . The effect of the stabilizing marks is to slowly shift the baseline transcription rate of genes over time . The change in number of S marks νi at gene i at each timestep is given by: Δνi , t ~ sgn ( µi , t-1 ) * bern ( abs ( µi , t-1 ) * ps_mark / µmax ) Here ps_mark is a probability of changing S marks at each time step , and µmax the maximum number of T marks allowed at a given promoter , whether positive ( activating ) or negative ( repressive ) . Every gene in the model is taken to have a baseline transcription rate , ri , 0 , drawn from the physiological distributions defined above . The time-dependent instantaneous transcription rate of a given gene , ri , t , is then calculated from the number of tuning marks ( µi ) and stabilizing marks ( νi ) . The effects of tuning and stabilizing marks in the model are multiplicative , such that the transcription rate ri at gene i with µi tuning marks and νi stabilizing marks is given by ri , t = ri , 0 * α * exp ( β ) ; where α = 2* ( ( µi , t / µmax ) +1 ) and β = mS * νi , t Here mS represents the magnitude of the effects of a single S mark , and the number of T marks is constrained to the interval [-µmax , µmax] . The various fixed model parameters ( e . g . , mS , pdecay , etc . ) were chosen to be physiologically plausible while supporting tuning . The values of these parameters used in Figure 9C and E are taken as a baseline and shown in Supplementary file 8; note , however , that as shown in Figure 9D , the performance of the model is robust to changes in those parameters . A python implementation of the model , along with sample inputs corresponding to the simulations described here , are included as Source code 3 . In order to measure the mixing times under different stress conditions , synprom-URA3-mRuby/synprom-DHFR-GFP cells were grown overnight in SC +glu media . The next morning , the cells were back-diluted 1:100 into fresh , prewarmed low fluorescence SC +glu or ura-/6AU10 . The cells in ura-/6AU10 media were kept in a 30°C incubated shaker for 24 hr before sorting , whereas the cells in the complete media were sorted after four hours of growth at 30°C . The cells were sorted based on their mRuby fluorescence level into three populations of the top 20% , bottom 20% , and the complete distribution ( mock-sorted ) of cells . In order to minimize the effects of both autofluorescence and size-fluorescence correlations , the cells ( including those in the mock-sorted population ) were tightly gated on FSC-A levels . The sorted cells were kept on ice until they were spun down and transferred to pre-warmed media identical to that in which they had previously been incubated ( that is , cells from complete media to complete media and cells from ura-/6AU to fresh ura-/6AU ) . The cells were incubated at 30°C thereafter . A sample of each population was analyzed using flow cytometry at different time intervals , with T = 0 being the time that the fresh media was added to the samples . The last time point for the cells in SC +glu media was 630 min , and for the ura-/6AU cells was 6660 min . We calculated the distribution of mRuby fluorescence values for each sample at each time point by smoothing the observed values using a kernel density estimator . We then measured the pairwise mRuby fluorescence distribution overlap of the top 20% , bottom 20% and the complete distribution at each time point for each growth condition . The distribution overlap was calculated by numerically integrating the area under the ( normalized ) kernel density distribution estimates of both populations being compared . An increasing overlap relative to t = 0 signifies the amount that the two populations have moved towards each other , and therefore the higher the overlap , the more mixed the two populations have become . Therefore , we calculated ft= ( maxx-xt ) xt=0 , where x is the overlap between the two distributions and max ( x ) is the maximum observed overlap . f ( t ) can be modeled as an exponential decay process according to:f ( t ) =ae-tτwhere τ provides a timescale for the mixing time ( in particular , τ ln ( 2 ) is the half-life of the decay process ) . We used nonlinear curve fitting in Matlab to estimate the values of the parameters in the above equation for cells grown under each of the physiological conditions described above and report the estimated half-lives to give insight into the mixing times active in the populations studied here . The images shown and analyzed in Figure 6 , Figure 7 , and panel A of Figure 7—figure supplement 2 were obtained on a Zeiss Axio Observer Z1 , using a 40x objective lens . PHSP12-URA3-mRuby/PADH1-DHFR-GFP cells were grown overnight in SC +glu liquid media , and then back-diluted 100x into SC/loflo + glu media and grown four additional hours with shaking at 30°C . Cells were spun down , and then incubated in ura-/loflo/6AU5 liquid for 12–13 hr . The cells were then pipetted onto the prepared slides . In order to prepare slides , we added 200 µL of ura-/loflo/6AU5 media containing 1% agar to each well of a two-well slide . Using a 22 µm coverslip , the surface of the media in the wells containing the solid media was flattened . After adding the cells on to the wells , we allowed extra media to be absorbed and then added a cover slip on top . The cells were imaged on DIC , GFP , brightfield , and mRuby channels; snapshots were taken once every 30 min for approximately 24 hr . The additional imaging time series analyzed in panels B-D of Figure 7—figure supplement 2 were obtained for PHSP12-URA3-mRuby/PADH1-DHFR-GFP cells immobilized to thin-bottomed growth chambers and grown in ura-/6AU5 media . To prepare the slides , cells were grown overnight in SC +glu liquid media , and then back-diluted 100x into SC/loflo + glu media and grown four additional hours with shaking at 30°C . During that incubation , a coverslip/incubation chamber ( Nunc ) was treated for five minutes with poly-D-lysine solution ( MPI Biomedical ) , washed three times with sterile deionized water , and then allowed to dry . After the pregrowth period , cells were diluted 10x into additional prewarmed SC/loflo + glu , and then pipetted onto the poly-D-lysine treated cover slip and allowed to settle for 30 min at room temperature . The media was removed , and non-adherent cells were washed away with two 1 mL rinses of sterile deionized water . The cells were then covered with 2 mL of ura-/loflo/6AU5 media , and then placed in a preheated microscopy incubation chamber ( OKO ) at 30°C and 90% relative humidity . Cells were imaged on DIC , GFP , and mRuby channels; snapshots taken once every 30 min for 24 hr on a Nikon Eclipse Ti microscope using a 20x objective . For comparative visualization purposes ( Figure 7A–B ) , the DIC or brightfield channel of each image was rescaled using the ImageMagick ‘normalize’ operator , and the fluorescence channels were normalized by subtracting the minimum pixel value within a given field of view , and then subjecting the remaining data to a median filter over a 5 × 5 pixel window . The fluorescence channels were then stacked on the DIC or brightfield to generate the images shown . Un-normalized data were used for all quantitative analysis . For the quantitative analysis in Figure 7C–E and Figure 7—figure supplement 2 , segmentation and lineage tracking were performed manually to identify cell division events and define cell interiors at the plotted timepoints . The fluorescence of each cell for each channel was then taken to be the average value of all pixels within the defined cell interior , with the mode value of all pixels in a defined window around the cell subtracted as background . For the purpose of classifying cells based on their division state , a cell was classified as ‘dividing’ if it gave rise to a daughter cell before the next analyzed snapshot . Timepoints prior to three hours were excluded from quantitative analysis of dividing vs . nondividing cells for the populations pregrown in SC/loflo + glu , as a large fraction of cells in all of our microscopy experiments did undergo a single division before arresting , likely using residual nutrients from their previous growth in complete media . | To survive , cells have to adapt to changes in their environment . Organisms can do so by constantly modifying the expression of their genes . For example , bacteria exposed to high temperatures turn on heat-shock genes to help them cope . Responses to familiar environmental changes take place thanks to specific , hard-wired molecular pathways . These transmit external signals to transcription factors , proteins that can bind DNA near a gene to regulate its expression . Yet , such established responses may not exist for stressful conditions that cells have never encountered during their evolutionary history . In this case , how can organisms adjust which genes to express , and at what levels ? Here , Freddolino et al . theorize that , in a new environment , individual genes can randomly increase or decrease their level of expression . If a change ends up being good for the survival of the cell , it is further reinforced . This ‘stochastic tuning’ would allow organisms to find the optimal levels of gene expression without using genetically predetermined pathways that involve transcription factors . Mathematical simulations suggest that this mechanism can improve the growth and survival of a cell in a new environment . Diverse experiments demonstrate that a phenomenon consistent with stochastic tuning occurs in yeasts . The organisms are genetically modified so that their transcription factors can no longer activate URA3 , a gene required to grow in conditions lacking a chemical called uracil . Yet , these altered yeast cells still manage to boost their URA3 expression in a uracil-free environment . Stochastic tuning could thus work alongside other types of conventional gene regulation to help cells adapt to new and challenging living conditions . For instance , this may be how cancerous cells survive and thrive when facing chemotherapy drugs . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"computational",
"and",
"systems",
"biology"
] | 2018 | Stochastic tuning of gene expression enables cellular adaptation in the absence of pre-existing regulatory circuitry |
Neonatal inflammation is common and has lasting consequences for adult health . We investigated the lasting effects of a single bout of neonatal inflammation on adult respiratory control in the form of respiratory motor plasticity induced by acute intermittent hypoxia , which likely compensates and stabilizes breathing during injury or disease and has significant therapeutic potential . Lipopolysaccharide-induced inflammation at postnatal day four induced lasting impairments in two distinct pathways to adult respiratory plasticity in male and female rats . Despite a lack of adult pro-inflammatory gene expression or alterations in glial morphology , one mechanistic pathway to plasticity was restored by acute , adult anti-inflammatory treatment , suggesting ongoing inflammatory signaling after neonatal inflammation . An alternative pathway to plasticity was not restored by anti-inflammatory treatment , but was evoked by exogenous adenosine receptor agonism , suggesting upstream impairment , likely astrocytic-dependent . Thus , the respiratory control network is vulnerable to early-life inflammation , limiting respiratory compensation to adult disease or injury .
At birth , neonates transition from a sterile maternal environment into an environment filled with pathogens , microbes , and toxins and must simultaneously begin robust , rhythmic breathing . Respiratory problems represent a significant clinical problem for neonatologists ( Martin et al . , 2012 ) , especially in preterm infants where breathing is unstable ( Poets and Southall , 1994; Poets et al . , 1994 ) and infections are common ( Stoll et al . , 2002; Stoll et al . , 2004 ) . Further , inflammation appears to augment respiratory dysfunction in neonates , whereby inflammation depresses hypoxic responses ( Olsson et al . , 2003; Rourke et al . , 2016 ) and induces recurrent apneas ( Hofstetter et al . , 2007 ) . Despite the prevalence of early life inflammation , little is known about the long-lasting consequences of neonatal inflammation on adult neurorespiratory control . We are beginning to understand the potential for long-term consequences of early life inflammation in other physiological systems . Neonatal inflammation blunts adult immune function ( Bilbo et al . , 2010; Mouihate et al . , 2010; Spencer et al . , 2011 ) , increases adult stress reactivity ( Shanks et al . , 2000; Wang et al . , 2013; Grace et al . , 2014 ) , impairs adult learning and hippocampal plasticity ( Bilbo , 2005a; Bilbo et al . , 2006 ) , increases the risk of neuropsychiatric disorders ( Rantakallio et al . , 1997; Hornig et al . , 1999 ) , and worsens age-related cognitive decline ( Bilbo , 2010 ) . Yet , we know very little about the long-term effects of neonatal inflammation on adult neurorespiratory control . Respiratory plasticity is an important feature of the neural control of breathing , providing adaptability and maintenance of breathing when the respiratory system is challenged ( Fuller and Mitchell , 2017 ) . Phrenic long-term facilitation ( pLTF ) is a frequently studied adult model of respiratory motor plasticity ( Mitchell and Johnson , 2003 ) and is elicited by at least two distinct cellular signaling pathways: the Q-pathway and the S-pathway ( reviewed in Dale-Nagle et al . , 2010 ) . The Q-pathway is evoked by moderate acute intermittent hypoxia ( mAIH; 3 × 5 min hypoxic episodes , PaO235–45 mmHg ) and is serotonin dependent , while the S-pathway is evoked by severe AIH ( sAIH , PaO225–35 mmHg ) and is adenosine dependent ( Nichols et al . , 2012 ) . Interestingly , Q-pathway-evoked plasticity is undermined by even low levels of acute , adult , systemic inflammation and restored by the non-steroidal anti-inflammatory , ketoprofen ( Vinit et al . , 2011; Huxtable et al . , 2013; Hocker and Huxtable , 2018 ) , while S-pathway-evoked adult plasticity is inflammation resistant ( Agosto-Marlin et al . , 2017 ) . Though we are beginning to understand more about the mechanisms of acute , adult inflammation on respiratory motor plasticity ( Hocker et al . , 2017; Hocker and Huxtable , 2018 ) , we do not know how inflammation in early postnatal life impacts respiratory motor plasticity in the adult . Furthermore , few studies have investigated sex-differences in pLTF ( Behan et al . , 2002; Dougherty et al . , 2017 ) and we know even less about sex-differences in respiratory control in response to inflammation . Additionally , males are more sensitive acutely to neonatal inflammation leading to higher male mortality in neonates ( Bouman et al . , 2005; Kentner et al . , 2010; Rathod et al . , 2017 ) , but our understanding of other sex-differences after neonatal inflammation are unknown . Given the profound effects of neonatal inflammation on other physiological systems , we tested the hypothesis that neonatal inflammation undermines Q-pathway , but not S-pathway , respiratory motor plasticity in adult male and female rats . Our results indicate that one neonatal inflammatory challenge completely abolishes adult , AIH-induced Q-pathway and S-pathway respiratory motor plasticity . Despite no lasting increases in adult , inflammatory gene expression , Q-pathway impairment is inflammation-dependent and is restored by acute adult anti-inflammatory treatment . Conversely , S-pathway impairment is inflammation-independent , but can be evoked by intermittent adenosine receptor agonism , suggesting phrenic motor neurons are not impaired . Since astrocytes are a primary source of adenosine during hypoxia ( Takahashi et al . , 2010; Angelova et al . , 2015 ) , they are likely impaired by neonatal inflammation and contributing to impairment of respiratory plasticity . These studies are the first steps toward understanding the lasting effects of neonatal inflammation on adult respiratory plasticity and suggest neonatal inflammation induces lasting-changes , increasing susceptibility to adult ventilatory control disorders .
Male and female postnatal day 4 ( P4 ) rats were injected with either LPS ( Lipopolysaccharide; 1 mg/kg , i . p . ) or saline . The dose of LPS was based on previous studies demonstrating CNS inflammatory gene expression in neonates ( Rourke et al . , 2016 ) , as well as our unpublished data ( N . Morrison , S . Johnson , J . Watters , A . Huxtable , unpublished observations ) . Within 24 hr of neonatal LPS injections , there was significantly greater mortality of male pups ( 8 of 67 ) than female pups ( 1 of 55 , Fisher’s exact test , p = 0 . 04 , Figure 1A ) . No mortality was evident in the saline treated males ( n = 63 ) or females ( n = 63 ) . For the surviving pups , neonatal LPS males weighed significantly less at week 7 ( no pairwise weight differences seen in females ) , but importantly , weights were not different in adults ( Figure 1B ) . Q-pathway-evoked pLTF is evident as the increase in integrated phrenic activity 60 min after mAIH ( PaO235–45 mmHg ) in adult , anesthetized rats ( Bach and Mitchell , 1996 ) . As expected , in adult males treated with neonatal saline , Q-pathway-evoked pLTF was evident after mAIH ( 55 ± 33 . 2% change from baseline , n = 7 , p = 0 . 0006 , Figure 2A and C ) . However , Q-pathway-evoked pLTF was absent in adult males treated with neonatal LPS ( 14 ± 49% , n = 12 , p = 0 . 2247 Figure 2A and C ) . To control for the known effects of estrus cycle hormones on pLTF in females ( Zabka et al . , 2001; Behan et al . , 2002; Dougherty et al . , 2017 ) , adult females were ovariectomized 7–8 days before electrophysiology studies . Similar to males , adult females treated with neonatal saline displayed Q-pathway-evoked pLTF ( 97 ± 63% change from baseline , n = 7 , p < 0 . 0001 , Figure 2B and C ) , while adult females challenged with neonatal LPS did not express pLTF ( −15 ± 43% , n = 6 , p = 0 . 4689 , Figure 2B and C ) . Phrenic amplitude did not change from baseline in the time control group ( 8 ± 6% change , n = 5 , p = 0 . 6482 ) , regardless of sex or neonatal LPS exposure , and was significantly reduced compared to males or females treated with neonatal saline . Between groups , Q-pathway-evoked pLTF was significantly abolished in adults after neonatal LPS compared to adults after neonatal saline for both males ( p = 0 . 0200 ) males and females ( p < 0 . 0001 ) . Thus , neonatal inflammation induces lasting impairment of adult , Q-pathway-evoked respiratory motor plasticity in both males and females . To test whether this lasting impairment of adult , respiratory motor plasticity was due to ongoing adult inflammation as a result of the neonatal inflammatory LPS challenge , we acutely treated adults with the non-steroidal anti-inflammatory , ketoprofen ( 12 . 5 mg/kg , i . p . , 3 hr ) , a high dose previously shown to restore plasticity after acute , adult inflammation ( Huxtable et al . , 2013 ) . Ketoprofen treatment restored Q-pathway-evoked pLTF in adult males treated with neonatal LPS ( 58 ± 18% change from baseline , n = 4 , p = 0 . 0004 , Figure 3A and C ) . Ketoprofen also restored Q-pathway-evoked pLTF in adult females treated with neonatal LPS ( 111 ± 44% from baseline , n = 5 , p < 0 . 0001 , Figure 3B and C ) . Adults treated with neonatal saline ( male: 54 ± 17% from baseline , n = 4 , 0 . 0008; female: 89 ± 40% , n = 5 , p < 0 . 0001 ) were unaffected by adult ketoprofen treatment . Additionally , phrenic motor amplitude did not change in adult time controls treated with ketoprofen ( 13 ± 14% change from baseline , n = 4 , p = 0 . 3436 ) and was significantly reduced compared to all other groups . Between groups , pLTF was not different between adult males ( p = 0 . 7605 ) or females ( p = 0 . 2932 ) after neonatal saline or neonatal LPS , suggesting the impairment in Q-pathway-evoked pLTF is inflammation-dependent in both males and females . Because the lasting impairment of Q-pathway-evoked pLTF was inflammation-dependent , we examined whether neonatal inflammation had lasting effects on adult neuroinflammation in regions involved in respiratory neural control and motor plasticity . Since plasticity was abolished in both males and females , data from both sexes were combined for analysis of inflammatory genes . In medullary and cervical spinal homogenates , neonatal LPS did not significantly alter mRNA for adult inflammatory genes ( IL-6 , IL-1β , TNF-α , or iNOS; Figure 4A and B ) . However , COX-2 gene expression was reduced in adult spinal cords after neonatal LPS ( Figure 4B , p = 0 . 001 ) , suggesting a decrease in COX-dependent inflammatory signaling . Thus , there was no evidence for lasting increases in neuroinflammatory gene expression in adults after a single exposure of neonatal inflammation in respiratory control regions . S-pathway-evoked pLTF is evident as the increase in integrated phrenic activity 60 min after sAIH ( PaO225–35 mmHg ) in adult rats . As expected , in adult males after neonatal saline , S-pathway-evoked pLTF was evident after sAIH ( 61 ± 69% change from baseline , n = 5 , p = 0 . 0001 , Figure 5A and C ) . Contrary to our hypothesis , S-pathway-evoked pLTF was abolished in adult males after neonatal LPS ( 7 ± 18% change from baseline , n = 4 , p = 0 . 6770 , Figure 5A and C ) . In adult females after neonatal saline , S-pathway-evoked pLTF was evident ( 102 ± 47% change from baseline , n = 4 , p < 0 . 0001 , Figure 5B and C ) . Similar to adult males treated with neonatal LPS , S-pathway-evoked pLTF was abolished in adult females after neonatal LPS ( 0 ± 33% , n = 4 , p = 0 . 9796 , Figure 5B and C ) . Phrenic amplitude in the time control group was significantly less than males ( p = 0 . 0147 ) or females ( p < 0 . 0001 ) treated with neonatal saline . Between groups , S-pathway-evoked pLTF was significantly reduced after neonatal LPS in both adult males ( p = 0 . 0180 ) and females ( p < 0 . 0001 ) compared to adults after neonatal saline . Thus , neonatal inflammation induces lasting impairment of adult , S-pathway-evoked respiratory motor plasticity in both males and females . To test whether this lasting impairment of adult , S-pathway-evoked plasticity is due to ongoing inflammation in adults after neonatal LPS , we examined sAIH-induced plasticity after an acute , adult treatment with ketoprofen ( 12 . 5 mg/kg , i . p . , 3 hr ) . Ketoprofen did not alter normal expression of S-pathway-evoked pLTF in adult males after neonatal saline ( 63 ± 22% change from baseline , n = 5 , p = 0 . 0014 , Figure 6A and C ) . However , contrary to the Q-pathway results , adult ketoprofen did not restore S-pathway-evoked pLTF in adult males after neonatal LPS ( 0 ± 65% change from baseline , n = 5 , p = 0 . 9804 , Figure 6A and C ) . Similarly , adult females treated with neonatal saline also exhibited normal S-pathway-evoked pLTF after adult ketoprofen ( 130 ± 22% change from baseline , n = 5 , 0 . 0803 , Figure 6B and C ) , and S-pathway-evoked pLTF was not restored by ketoprofen in adult females after neonatal LPS ( 25 ± 30% change from baseline , n = 6 , p = 0 . 0803 , Figure 6B and C ) . Adult males and females treated with neonatal LPS and adult ketoprofen were not different from time controls ( males , p = 0 . 4964; females p = 0 . 5227 ) . Between groups , S-pathway-evoked pLTF after acute ketoprofen was significantly reduced in adults after neonatal LPS comapred to adults after neonatal saline in both males ( p = 0 . 0019 ) and females ( p < 0 . 0001 ) . Thus , neonatal inflammation induces a lasting impairment of adult , S-pathway-evoked respiratory motor plasticity , which is not due to ongoing adult , inflammatory signaling . S-pathway-evoked plasticity elicted by sAIH is adenosine dependent ( Golder et al . , 2008; Nichols et al . , 2012 ) and can be evoked by intermittent CGS-21680 , an adenosine 2A receptor agonist . To test if neonatal inflammation is impairing phrenic motor neurons and preventing pLTF , we examined phrenic output after intermittent CGS-21680 on the cervical spinal cord , around the phrenic motor pool . Intrathecal CGS-21680 ( 100 µM , 3 × 10 µL ) evoked phrenic motor plasticity in adult males after neonatal saline ( 110 ± 17% change from baseline , n = 4 , p < 0 . 001 , Figure 7A and C ) and females after neonatal saline ( 127 ± 47% , n = 4 , p < 0 . 001 , Figure 7B and C ) . After neonatal LPS , intrathecal CGS-21680 also elicited plasticity in adult males ( 85 ± 64% , n = 6 , p < 0 . 001 , Figure 7A and C ) and adult females ( 147 ± 74% , n = 6 , p < 0 . 001 , Figure 7B and C ) , demonstrating phrenic motor neurons are not impaired after neonatal inflammation and are capable of S-pathway-evoked plasticity . The vehicle control group was not different from baseline ( −3 ± 5% change , n = 4 , p = 0 . 8891 ) and significantly reduced compared to all other groups . Between groups , pLTF was not different between adult males ( p = 0 . 2841 ) or females ( p = 0 . 4032 ) after neonatal saline or neonatal LPS . Thus , adult phrenic motor neurons are not impaired after neonatal inflammation and are capable of plasticity after neonatal inflammation . Therefore , the source of intermittent adenosine release is impaired during sAIH-induced pLTF after neonatal inflammation . While there was no evidence for elevated neuroinflammation based on the inflammatory genes evaluated here , the anti-inflammatory drug ketoprofen successfully restored Q-pathway-evoked plasticity . Additionally , our results indicate the impairment in S-pathway-evoked plasticity was likely due to a lasting change in adenosine signaling , possibly as a result of altered astrocytes . Thus , we hypothesized a lasting change in astrocytes and microglia in respiratory control regions , influencing neuronal function and impairing adult plasticity . We evaluated GFAP ( astrocytes ) and IBA1 ( microglia ) immunoreactivity in the adult preBötC , the site of respiratory rhythmogenesis ( Smith et al . , 1991 ) , and in cervical spinal cords in the region of the phrenic motor nucleus , the presumptive site of pLTF ( Baker-Herman et al . , 2004; Devinney et al . , 2015; Dale et al . , 2017 ) . Neonatal inflammation did not alter GFAP ( p = 0 . 5969 ) or IBA1 ( p = 0 . 6487 ) immunoreactivity in adult preBötC in either sex ( Figure 8A , B and E ) , suggesting astrocyte and microglial density were not changed in adults after neonatal inflammation . Furthermore , there were no changes in GFAP ( p = 0 . 7195 ) or IBA1 ( p = 0 . 9254 ) immunoreactivity in adult cervical spinal cords ( Figure 8C , D and F ) , suggesting no lasting changes in astrocyte and microglia density in the region of the phrenic motor nucleus . Additionally , no obvious differences in astrocyte or microglial morphology in adult phrenic motor nuclei or the preBötC were seen following neonatal LPS inflammation , suggesting other signaling mechanisms are responsible for impairing adult pLTF . Neonatal inflammation did not significantly alter moderate acute hypoxic phrenic amplitude responses within adult males ( neonatal saline = 114 ± 41% change from baseline; neonatal LPS = 93 ± 36% ) or females ( neonatal saline = 185 ± 53%; neonatal LPS = 148 ± 63% , Table 1 ) . Hypoxic phrenic amplitude responses were also unaffected by the anti-inflammatory ketoprofen in adult males ( neonatal saline +Keto = 118 ± 36%; neonatal LPS +Keto , 118 ± 44% ) or females ( neonatal saline +Keto , 165 ± 52%; neonatal LPS +Keto , 189 ± 82% , Table 1 ) . However , adult females exhibited significantly greater acute phrenic amplitude responses to moderate hypoxia ( main effect , p = 0 . 0004 ) . Phrenic amplitude in response to severe hypoxia was similarly unaltered by neonatal inflammation within adult males ( neonatal saline = 139 ± 37% change from baseline; neonatal LPS = 106 ± 10% ) or females ( neonatal saline = 172 ± 125%; neonatal LPS = 172 ± 26% , Table 1 ) . Acute ketoprofen pretreatment did not alter acute hypoxic phrenic amplitude responses in adult males ( neonatal saline = 151 ± 25% change from baseline; neonatal LPS , 174 ± 96% ) or females ( neonatal saline = 194 ± 45%; neonatal LPS = 235 ± 63% , Table 1 ) . Similarly , adult females exhibited a significantly greater acute amplitude response to severe hypoxia than males ( main effect , p = 0 . 021 ) . All physiological parameters remained within experimental limits ( Table 2 ) . Neonatal saline or LPS caused no significant changes in adult temperature , PaCO2 , PaO2 , or pH at baseline . There were no between group differences in baseline MAP , suggesting no long-lasting cardiovascular changes after neonatal inflammation . No significant changes occurred over time in temperature , pH , or PaCO2 for any group . As expected , MAP and PaO2 were significantly decreased during hypoxic episodes in experimental groups ( Huxtable et al . , 2015; Hocker and Huxtable , 2018 ) , but these changes were not evident in time control groups and were not different from baseline values at 60 min post-AIH . Baseline phrenic burst frequency was not significantly different between groups and frequency plasticity , an increase in burst frequency 60 min after AIH ( Baker-Herman and Mitchell , 2008 ) , was not evident in any group . Phrenic burst frequency did not change after intrathecal CGS-21680 .
Although neonatal inflammation is common ( Stoll et al . , 2002; Stoll et al . , 2004 ) , little is known concerning how neonatal inflammation alters ventilatory control . Here , we investigated the long-term consequences of neonatal systemic inflammation on adult respiratory motor plasticity , a key feature of the neural control of breathing providing adaptability to respiratory system challenges ( Mitchell and Johnson , 2003 ) . We show for the first time that a single inflammatory challenge to neonates completely abolishes AIH-induced Q-pathway and S-pathway-evoked respiratory motor plasticity in adult males and females . Our results indicate a persistent change in adult inflammatory signaling contributes to this impairment since adult anti-inflammatory treatment restores Q-pathway-evoked , but not S-pathway-evoked , pLTF . Further , this is the first evidence of impairment of S-pathway-evoked motor plasticity , suggesting neonatal inflammation likely leads to a further vulnerable adult as this pathway was thought of as a ‘backup pathway’ after acute , adult inflammation ( Agosto-Marlin et al . , 2017 ) . However , we demonstrate S-pathway plasticity can be revealed by intermittent , spinal adenosine receptor agonism , suggesting astrocyte dysfunction after neonatal inflammation since they are the likely source of adenosine during hypoxia ( Takahashi et al . , 2010; Angelova et al . , 2015 ) . These studies are the first steps toward understanding the lasting effects of neonatal inflammation on adult neurorespiratory control and suggest neonatal inflammation may increase susceptibility to adult ventilatory control disorders . LPS-induced neonatal inflammation transiently upregulates cytokines in regions involved in respiratory control and plasticity ( N . Morrison , S . Johnson , J . Watters , A . Huxtable , unpublished observations ) , consistent with inflammatory profiles in other CNS regions ( Wang et al . , 2006; Schwarz and Bilbo , 2011; Bilbo and Schwarz , 2012; Jafri et al . , 2013 ) . Neonatal inflammation also increases male mortality , consistent with clinical male mortality after neonatal inflammation ( Person et al . , 2014 ) and is relevant to the increased risk of sudden infant death syndrome for males ( Kinney and Thach , 2009 ) . However , similar to other studies ( Bilbo et al . , 2005b; Mouihate et al . , 2010; Smith et al . , 2014 ) , we found no measurable adult changes in cytokines , or glial number or morphology despite a lasting inflammation-dependent impairment in Q-pathway-evoked pLTF . While we have previously demonstrated adult Q-pathway pLTF is sensitive to low levels of acute , systemic inflammation ( Huxtable et al . , 2011; Huxtable et al . , 2013; Huxtable et al . , 2015; Huxtable et al . , 2018a; Vinit et al . , 2011; Hocker and Huxtable , 2018 ) , this is the first study to demonstrate neonatal inflammation induces lasting changes in inflammatory signaling to undermine adult Q-pathway-evoked pLTF . Additionally , S-pathway-evoked pLTF is not restored by even the high dose of ketoprofen used here , but is revealed by intermittent adenosine 2A receptor agonism on phrenic motor neurons ( Seven et al . , 2018 ) . Thus , these results indicate phrenic motor neurons are not impaired after neonatal inflammation and the loss of S-pathway-evoked plasticity is likely due to impaired adenosine signaling during hypoxia . Furthermore , a likely source of adenosine during hypoxia is astrocytes ( Takahashi et al . , 2010; Angelova et al . , 2015 ) , suggesting neonatal inflammation induces lasting astrocyte-specific changes in the adult spinal cord to impair adult S-pathway-evoked pLTF . Further understanding the mechanisms impairing distinct forms of respiratory motor plasticity is required to develop plasticity as a therapeutic tool , such as for spinal cord injury and amyotrophic lateral sclerosis ( Mitchell , 2008; Gonzalez-Rothi et al . , 2015 ) . Additionally , considering the cross-talk between Q- and S-pathways ( Dale-Nagle et al . , 2010; Fields and Mitchell , 2015; Perim et al . , 2018 ) , the response of the respiratory control system likely depends on the functional status of both Q- and S-pathways . Thus , future studies are needed to understand how inflammation modifies cross-talk between Q- and S-pathways and how respiratory motor plasticity can be exploited therapeutically ( Gonzalez-Rothi et al . , 2015 ) . The timing of neonatal inflammation is likely a significant factor in how neonatal inflammation impacts adult physiology . Low-levels of cytokines are important for neurodevelopment ( Bilbo and Schwarz , 2009 ) , and perturbing the balance of neonatal cytokines during development leads to lasting aberrant effects on neural circuits and developing cells ( Reemst et al . , 2016 ) . Furthermore , while many components of the respiratory system begin developing in utero ( Prakash et al . , 2000; Pagliardini et al . , 2003; Mantilla and Sieck , 2008; Johnson et al . , 2018 ) , the respiratory control system undergoes significant postnatal maturation . In these studies , we induced systemic inflammation with LPS at P4 , similar to other studies showing long-term consequences of neonatal inflammation in other physiological systems ( Shanks et al . , 2000; Walker et al . , 2006; Fan et al . , 2008; Kohman et al . , 2008; Bilbo , 2010 ) , supporting the idea that important neural changes occur within the first week of life . Yet , it remains to be determined whether there is a precise critical period where neonatal inflammation impacts respiratory control circuits . However , our data on male mortality after neonatal LPS are consistent with other critical developmental windows , including a male-specific sensitive period to LPS ( Rourke et al . , 2016 ) , disproportionate male mortality from neonatal inflammation ( Person et al . , 2014 ) , the increased risk of sudden infant death syndrome for males ( Kinney and Thach , 2009 ) , and increased incidence of obstructive sleep apnea in adults after neonatal inflammation ( McNamara and Sullivan , 2000 ) . Thus , these data have important implications for understanding the sex-specific impairment early in life and into adulthood . Additionally , we and others ( Spencer et al . , 2006 ) observed a short delay in weight gain after neonatal inflammation , which normalized by weaning , suggesting no lasting effects on growth . Future studies are needed to refine our understanding of the critical periods during development when early-life inflammation induces long-lasting physiological changes to improve our understanding of adult disease and better understand important developmental processes . While other reports have shown sex differences in neonatal programming of adult neuro-inflammatory responses ( LaPrairie and Murphy , 2007; Rana et al . , 2012 ) , we observed no sex-differences in the effects of neonatal inflammation on adult plasticity . Importantly , this is the first evidence of inflammation abolishing pLTF in females and the first to report sAIH-induced respiratory motor plasticity in females . Females exhibited greater acute hypoxic phrenic amplitude responses relative to males , consistent with previous findings ( Mortola and Saiki , 1996; Bavis et al . , 2004 ) , despite variability in reports of sex differences in hypoxic ventilatory responses ( Behan and Kinkead , 2011 ) . In contrast to our results following neonatal inflammation , neonatal stress alters adult hypoxic responses in a sex-dependent manner , whereby male responses are enhanced and female responses are blunted ( Rousseau et al . , 2017 ) . Thus , the long-term effects on respiratory control may be dependent on the type of stressors in early life . Importantly , our experiments were performed in adult , ovariectomized females with exogenously restored estradiol levels to permit respiratory motor plasticity ( Behan et al . , 2002; Zabka et al . , 2003; Dougherty et al . , 2017 ) . Therefore , as sex hormones are known to modulate respiratory control and hypoxic responses ( Nelson et al . , 2011; Behan and Kinkead , 2011 ) , we cannot rule out a confounding role for exogenous estradiol supplementation after ovariectomy . Finally , after neonatal inflammation , we found no differences in adult hypoxic responses , suggesting no lasting change in carotid body responses due to neonatal inflammation . Accordingly , the deficit in adult respiratory motor plasticity after neonatal inflammation is likely a consequence of long-term changes in the spinal cord where pLTF occurs ( Baker-Herman et al . , 2004; Devinney et al . , 2015; Dale et al . , 2017 ) . While adult anti-inflammatory treatment restored Q-pathway-evoked pLTF , we did not observe increases in inflammatory gene expression in adult medullary or cervical spinal cord homogenates . Thus , while inflammatory signaling contributes to the impairment of adult plasticity , the source of this signaling change remains unclear and will be the topic of future studies . Similarly , others demonstrated no changes in baseline CNS inflammatory markers after neonatal inflammation , but observed priming of glial responses to adult stimuli ( Bilbo et al . , 2005b; Mouihate et al . , 2010; Smith et al . , 2014 ) , suggesting lasting changes in glia have the potential to underlie impairments in adult respiratory plasticity . Contrary to other reports ( Boissé et al . , 2005; Kentner et al . , 2010 ) , we found spinal COX-2 gene expression was decreased in adulthood , suggesting a decrease in inflammatory signaling , which is unlikely to contribute to the lasting inflammation-dependent impairment in plasticity . Further , the acute inflammatory impairment of adult respiratory plasticity is COX-independent ( Huxtable et al . , 2018a ) , emphasizing a role for other inflammatory molecules mediating the lasting impairment in respiratory motor plasticity . Unmeasured inflammatory genes or post-transcriptional changes in inflammatory proteins may be responsible for undermining adult pLTF after neonatal inflammation . Conversely , other perinatal stimuli involving inflammatory signaling , such as maternal care and diet , do have lasting programming effects on adult inflammatory cytokine expression ( Bilbo and Schwarz , 2009 ) , but are more complex stimuli than the acute neonatal inflammation in our study . We also observed no change in microglial or astrocyte density and no obvious qualitative changes in morphology in adult medullas or spinal cords after neonatal inflammation . Thus , there are no obvious signs of inflammation in regions contributing to pLTF despite the restoration of Q-pathway-evoked pLTF with ketoprofen . Furthermore , the abolition of S-pathway-evoked pLTF is likely due to lasting changes in adenosine signaling from astrocytes , suggesting an astrocyte-specific change underlies this impairment . Thus , future studies are needed to identify inflammatory mechanisms undermining the Q-pathway and further details of the inflammation-independent mechanism responsible for undermining S-pathway-evoked motor plasticity . The adult respiratory control network is vulnerable to early life stressors ( Bavis et al . , 2004; Genest et al . , 2004; Fournier et al . , 2011 ) , which may undermine the ability to compensate during adult ventilatory control disorders . Our study is the first to demonstrate lasting consequences of neonatal inflammation on adult respiratory control . These deficits in respiratory control are independent of later life events , in contrast to other studies in which the physiological effects of early life inflammation are not revealed until after an adult stimulus ( Bilbo et al . , 2005b; Bilbo , 2010 ) . We found a single episode of neonatal systemic inflammation induced lasting impairment of both Q- and S-pathway-evoked respiratory motor plasticity in adults . Our results suggest the adult impairment of Q-pathway plasticity is dependent on acute inflammatory signaling; however , we observed no lasting increase in adult inflammatory gene expression or the density of astrocytes and microglia . The pharmacological induction of S-pathway-evoked pLTF demonstrates phrenic motor neurons are capable of plasticity and suggest upstream impairment , such as the source of adenosine . While strong evidence supports astrocytes as the primary source of adenosine during hypoxia ( Takahashi et al . , 2010; Angelova et al . , 2015 ) , we cannot rule out other sources of adenosine . Identifying cell-type specific changes underlying lasting physiological impairments will be explored in future studies . Future studies will investigate the lasting effects of neonatal inflammation on isolated microglia and astrocytes to uncover potential mechanisms of adult impairments after neonatal inflammation . Together , these results indicate two mechanistic pathways to spinal motor plasticity induced by AIH are undermined by neonatal inflammation in rats . Our experimental approach assessed phrenic nerve output in anesthetized rats and may not be generalizable to respiratory control in freely behaving animals or to other forms of motor plasticity . However , AIH induces long-term facilitation of ventilation in humans ( Mateika and Komnenov , 2017 ) and strengthens corticospinal pathways to non-respiratory motor-neurons ( Christiansen et al . , 2018 ) , suggesting our results likely have relevance to mechanisms of human spinal motor plasticity after AIH . While AIH-induced respiratory motor plasticity does not necessarily alter normal homeostatic control of ventilation , the general facilitation of spinal motor output has significant therapeutic potential for treating patients with respiratory and non-respiratory motor limitations ( Trumbower et al . , 2012; Trumbower et al . , 2017; Nichols et al . , 2013; Hayes et al . , 2014 ) In conclusion , this basic science study has major implications for the understanding the neonatal origins of adult ventilatory control disorders . These studies are the first evidence that one neonatal inflammatory exposure induces long-term impairments in adult respiratory control with potential relevance to many respiratory disorders . These findings are particularly relevant since inflammation is common in neonates ( Person et al . , 2014 ) , especially those born prematurely who are at higher risk for adult disease ( Luu et al . , 2016 ) . Improving our appreciation of how early life inflammation can influence adult respiratory control will have important consequences for understanding adult disease and susceptibility to respiratory disorders . Additionally , AIH-induced spinal motor plasticity is also a promising therapy to enhance motor recovery after spinal injury ( Trumbower et al . , 2012 ) . However , not all patients respond to AIH ( Hayes et al . , 2014; Trumbower et al . , 2017 ) and our findings suggest neonatal inflammatory exposure could contribute to these therapeutic limitations and understanding the mechanisms undermining plasticity will increase the therapeutic potential of AIH-induced spinal motor plasticity .
LPS ( 0111:B4 , Sigma Chemical ) was dissolved and sonicated in sterile saline for neonatal intraperitoneal ( i . p . ) injections ( 1 mg/kg ) . S- ( + ) Ketoprofen ( Keto , Sigma Chemical ) was dissolved in ethanol ( 50% ) and sterile saline for acute , adult injections ( 12 . 5 mg/ml/kg , i . p . , 3 hr ) . 17-β estradiol was dissolved in sesame oil ( Tex Lab Supply , Texas , USA ) for acute injections ( 40 μg/mL/kg , i . p . , 3 hr ) in adult females after ovariectomy . The adenosine 2A receptor agonist CGS-21680 was dissolved in fresh artificial cerebrospinal fluid ( aCSF: 120 mM NaCl , 3 mM KCl , 2 mM CaCl2 , 2 mM MgCl2 , 23 mM NaHCO3 , and 10 mM glucose ) with DMSO ( 10% ) for intrathecal injections . Timed pregnant rats ( E14-17 upon arrival ) were purchased in pairs from a commercial vendor ( Envigo ) and monitored daily . To control for between litter effects , litters were stratified such that each dam fostered similar numbers of male and female pups . On postnatal day 4 ( P4 ) , all of the stratified pups with one dam were injected with LPS ( 1 mg/kg , i . p . ) , while pups with the control dam were injected with sterile saline ( i . p . ) . The dose of LPS was based on previous studies demonstrating CNS inflammatory gene expression in neonates ( Rourke et al . , 2016 ) , as well as our unpublished data ( N Morrison , S Johnson , J Watters , A Huxtable , unpublished observations ) indicating CNS inflammation following LPS ( 1 mg/kg ) . Pups were weighed weekly and weaned at P21 . Electrophysiology experiments were conducted once males reached 300 g . Females were ovariectomized at approximately 250 g , 7–8 days prior to electrophysiology experiments . Ovariectomies were performed as previously described ( Dougherty et al . , 2017 ) to control for the known effects of estrus cycle hormones on pLTF ( Zabka et al . , 2001; Behan et al . , 2002; Dougherty et al . , 2017 ) . Adult rats were anesthetized with isoflurane and maintained on a nose cone ( 2 . 5% in O2 ) during surgery . Depth of anesthesia was confirmed by the absence of toe-pinch responses . Bilateral dorsolateral incisions exposed ovarian fat pads . Ovaries were ligated and removed , muscle layers were approximated , and skin incisions were closed with a single dissolvable suture . A single dose of buprenorphine ( 0 . 05 g/kg , s . c . ) was administered at the end of surgery for pain control and rats recovered in individual cages for 7-8 days before electrophysiology studies . Since pLTF exists in females only when estradiol is high ( Dougherty et al . , 2017 ) , estradiol levels were restored by injection of 17-β estradiol ( 40 μg/mL/kg , i . p . ) three hours before electrophysiology experiments . All experimental groups consisted of adult male and female rats after a single injection of either neonatal LPS or neonatal saline . To investigate the impact of neonatal systemic inflammation on adult Q-pathway-evoked respiratory motor plasticity , the following experimental groups were used: male neonatal saline + mAIH ( n = 7 ) , male neonatal LPS + mAIH ( n = 12 ) , female neonatal saline + mAIH ( n = 7 ) , female neonatal LPS + mAIH ( n = 6 ) . To investigate if acute anti-inflammatory treatment restores Q-pathway-evoked respiratory motor plasticity after neonatal inflammation , adults were treated with ketoprofen ( 12 . 5 mg/kg , i . p . ) three hours before electrophysiology experiments: male neonatal saline + Keto + mAIH ( n = 4 ) , male neonatal LPS + Keto + mAIH ( n = 4 ) , female neonatal saline + Keto + mAIH ( n = 5 ) , female neonatal LPS + Keto + mAIH ( n = 5 ) . To investigate the impact of neonatal systemic inflammation on adult S-pathway-evoked respiratory motor plasticity , we used the following experimental groups: male neonatal saline + sAIH ( n = 5 ) , male neonatal LPS + sAIH ( n = 4 ) , female neonatal saline + sAIH ( n = 4 ) , female neonatal LPS + sAIH ( n = 4 ) . To investigate if acute anti-inflammatory treatment restores S-pathway-evoked respiratory motor plasticity after neonatal inflammation , adults were treated with ketoprofen ( 12 . 5 mg/kg , i . p . ) three hours before electrophysiology experiments: female neonatal saline + Keto + sAIH ( n = 5 ) , male neonatal LPS + Keto + sAIH ( n = 5 ) , female neonatal saline + Keto + sAIH ( n = 5 ) , female neonatal LPS + Keto + sAIH ( n = 6 ) . To investigate if intermittent , intrathecal CGS-21680 reveals S-pathway-evoked respiratory motor plasticity , we used the following experimental groups: male neonatal saline + CGS-21680 ( n = 4 ) , male neonatal LPS + CGS-21680 ( n = 6 ) , female neonatal saline + CGS-21680 ( n = 4 ) , female neonatal LPS + CGS-21680 ( n = 6 ) . To reduce use of additional animals , and because time control experiments were not statistically different between males or females , time control groups consisted of animals from each experimental condition . The time control group for studies investigating the Q-pathway ( Figure 2 ) and S-pathway ( Figure 5 ) consisted of adults after neonatal saline ( male: n = 1; female n = 2 ) , neonatal LPS ( n = 1 male , 1 female ) . The time control + Keto group ( Figures 3 and 6 ) consisted of adults after neonatal saline + Keto ( n = 1 male , 1 female ) and neonatal LPS + Keto ( n = 1 male , 1 female ) . Vehicle controls for intrathecal CGS-21680 experiments ( Figure 7 ) consisted of adults after neonatal saline ( n = 1 male , 1 female ) and neonatal LPS ( n = 1 male , 1 female ) . Electrophysiological studies have been described in detail previously ( Bach and Mitchell , 1996; Baker-Herman and Mitchell , 2002; Huxtable et al . , 2013 ) . Rats were anesthetized with isoflurane , tracheotomized , ventilated ( Rat Ventilator , VetEquip ) , and vagotomized bilaterally . A venous catheter was placed for drug delivery and fluid replacement , and a femoral arterial catheter was used to monitor blood pressure and for arterial blood sampling . Arterial blood samples were analyzed ( PaO2 , PaCO2 , pH , base excess; Siemens RAPIDLAB 248 ) during baseline , during the first hypoxic response , and 15 , 30 and 60 min post-AIH . Temperature was measured with a rectal temperature probe ( Kent Scientific Corporation ) and maintained between 37°C and 38°C with a custom heated table . Using a dorsal approach , hypoglossal and phrenic nerves were cut distally , and de-sheathed . Rats were converted to urethane anesthesia ( 1 . 8 g/kg i . v . ; Sigma-Aldrich ) , allowed to stabilize for one hour , and paralyzed with pancuronium dibromide ( 1 mg; Selleck Chemicals ) . In rats receiving intrathecal injections , a laminectomy was performed at cervical vertebrae 2 ( C2 ) and a primed , silicone catheter was inserted two millimeters through a small incision in the dura . The catheter tip extended toward the rostral margin of C4 ( Baker-Herman and Mitchell , 2002 ) . CGS-21680 ( 100 μM ) or vehicle ( 10% DMSO in aCSF ) was injected around the phrenic motor pool in three boluses ( 10 µL ) separated by 5 min . Nerves were bathed in mineral oil and placed on bipolar silver wire electrodes . Raw nerve recordings were amplified ( 10 k ) , filtered ( 0 . 1–5 kHz ) , integrated ( 50 ms time constant ) , and recorded ( 10 kHz sampling rate ) for offline analysis ( PowerLab and LabChart 8 . 0 , AD Instruments ) . Apneic and recruitment CO2 thresholds were determined by changing inspired CO2 with continuous end-tidal CO2 monitoring ( Kent Scientific Corporation ) . End tidal CO2 was set 2 mmHg above the recruitment threshold , whereby arterial blood samples were used to establish baseline PaCO2 , which was maintained within 1 . 5 mmHg of the baseline value throughout . Blood volume and base excess were maintained ( ±3 MEq/L ) by continuous infusion ( 1–3 mL/h , i . v . ) of hetastarch ( 0 . 3% ) and sodium bicarbonate ( 0 . 99% ) in lactated ringers . Experiments were excluded if mean arterial pressure deviated more than 20 mmHg from baseline . All rats ( excluding time control rats ) received three , 5 min bouts of either mAIH ( ~10 . 5% O2 , PaO235–45 mmHg ) or sAIH ( ~7% O2 , PaO225–35 mmHg ) . The average amplitude and frequency of 30 consecutive integrated phrenic bursts were taken during baseline , the first acute hypoxic response , and 15 , 30 , and 60 min after AIH and made relative to baseline amplitude . Phrenic nerve activity data for each experimental group were compared using two-way , repeated measures ANOVA with Fisher LSD post hoc tests . Sample sizes were selected based on similar , previous studies and the variance of pLTF in our experience ( Huxtable et al . , 2018a; Huxtable et al . , 2018b; Hocker and Huxtable , 2018 ) . Physiological variables were compared using two-way , repeated measures ANOVA with Tukey’s post hoc test . Mean arterial pressure is reported for baseline , the end of the third hypoxic exposure , and 60 min after AIH . Acute hypoxic responses were compared using an ANOVA with Fisher LSD post hoc test . Values are means ± SD . Neonatal rats ( P4 ) were injected with either vehicle ( saline ) or LPS ( 1 mg/kg , i . p . ) and allowed to mature to ~12 weeks . Adult male and female rats were anesthetized with isoflurane and perfused with PBS ( transcardiac ) . Medulla and cervical spinal cords ( C3-C7 ) were dissected and flash frozen until they were homogenized in Tri-Reagent ( Sigma , St . Louis , MO , USA ) . Glycoblue reagent ( Invitrogen , Carlsbad , CA , USA ) was used to isolate total RNA , according to the manufacturer’s protocol . cDNA was reverse transcribed from 1 µg of total RNA using MMLV reverse transcriptase together with a cocktail of oligo dT and random primers ( Promega , Madison , WI , USA ) , as previously described ( Crain and Watters , 2015 ) , and analyzed using qPCR with PowerSYBR green PCR master mix on an ABI 7500 Fast system . Inflammatory gene expression was analyzed in medulla and spinal cord homogenates using the following primers: Wherever possible , primers were designed to span introns ( Primer three software ) and were purchased from Integrated DNA Technologies ( Coralville , IA , USA ) . Primer efficiency was assessed by use of standard curves , as previously reported ( Crain and Watters , 2015 ) . Expression of inflammatory genes was made relative to 18 s ribosomal RNA calculated using the 2-ΔΔCT method ( Livak and Schmittgen , 2001 ) . Gene transcripts were considered undetectable , and not included in statistical analyses if their CT values fell outside of the linear range of the standard curve for that primer set , which in most cases was ≥34 cycles . Upon completion of electrophysiology experiments , rats were perfused ( transcardiac ) with cold phosphate buffered saline ( PBS , pH 7 . 4 ) , followed by 4% paraformaldehyde ( pH 7 . 4 ) . All brains were removed and immersed in paraformaldehyde until sectioning ( Leica VT 1200S vibratome ) . For immunohistochemistry , transverse medullary and coronal cervical spinal cord sections ( 40 µm ) were washed ( PBS ) and blocked ( PBS , 0 . 3% Triton , 1% BSA , 2 hr , room temperature ) to prevent non-specific antibody binding . For medullary sections , two combinations of primary antibodies were used ( PBS , 0 . 3% Triton , 0 . 01% BSA , room temperature , 24 hr ) : ( 1 ) rabbit anti-GFAP ( 1:1000 , Millipore AB5804 ) to label astrocytes and guinea pig anti-NK1R ( 1:500 , Millipore AB15810 ) to label preBötzinger Complex ( preBötC ) neurons ( Gray et al . , 1999 ) , and ( 2 ) rabbit anti-IBA1 ( 1:1000 , Wako 019–19741 ) to label microglia and guinea pig anti-NK1R ( 1:500 , Millipore AB15810 ) to label preBötC neurons . For the spinal cord , two different combinations of primary antibodies were used ( PBS , 0 . 3% Triton , 0 . 01% BSA , room temperature , 24 hr ) : ( 1 ) rabbit anti-GFAP ( 1:1000 , Millipore AB5804 ) to label astrocytes and goat anti-ChAT ( 1:300 , Millipore AB144p ) to label motor neurons , ( 2 ) rabbit anti-IBA1 ( 1:1000 , Wako 019–19741 ) to label microglia and goat anti-ChAT ( 1:300 , Millipore AB144p ) to label motor neurons . After primary antibody incubation , sections were rinsed ( PBS ) and incubated with secondary antibodies ( PBS , 0 . 3% triton , 0 . 01% BSA , room temperature , 3 hr ) : donkey-anti-rabbit 647 IgG ( 1:1000 , Life Technologies A31573 ) to label GFAP and IBA1 primary antibodies , donkey-anti-goat 555 IgG ( 1:1000 , Life Technologies A21432 ) to label ChAT primary antibody and donkey-anti-guinea pig 488 IgG ( 1:1000 , Alexa Fluor 706-545-148 ) to label NK1R primary antibody . Sections were washed and mounted onto charged microscope slides , air dried and covered with prolong gold ( Life technologies , P36930 ) to preserve the fluorescence . A glass cover slip was placed over the samples and sealed with clear nail polish . Slides were stored in the dark at 4°C until imaged . All immunohistochemistry experiments contained adult male and female tissues after neonatal saline ( medulla: n = 5 males , seven females; spinal cord: n = 5 males , six females ) or neonatal LPS ( medulla: n = 6 males , four females; spinal cord: n = 6 males , three females ) . All immunofluorescent images ( 1024 × 1024 pixels , 40x magnification ) were acquired using a Leica Microsystems CMS GmbH confocal microscope using the LAS X acquisition and viewing software ( 0 . 5 µm z-stack step increments ) . All images were taken using identical laser and gain settings and identically adjusted for contrast/brightness using ImageJ open source software to allow for comparisons across all groups . To quantify the density of microglia and astrocytes , maximum intensity projections for 20 µm of z-stacks from the medulla and cervical spinal cords were analyzed . Mean fluorescent intensity for each image within a single batch was made relative to the average fluorescent intensities of adults after neonatal saline samples within each sex ( Paizs et al . , 2009 ) . Data are presented as percent change from adults after neonatal saline within each sex . GraphPad Prism 7 . 0 software was used for statistical analyses . Differences in mortality between treatments and between sexes was evaluated with Fisher’s exact test . Phrenic nerve activity data for each experimental group were compared using two-way , repeated measures ANOVA with Fisher LSD post hoc tests . Physiological variables were compared using two-way , repeated measures ANOVA with Tukey’s post hoc test . Mean arterial pressure is reported from baseline , the end of the third hypoxic exposure , and 60 min after AIH . Acute hypoxic phrenic responses were compared using an ANOVA with Fisher LSD post hoc test . Microglial and astrocytic density comparisons were made between groups using a one-way ANOVA with multiple-comparisons post hoc tests . For all tests , p < 0 . 05 was considered significant and all data are expressed as mean ± SD . | Breathing is essential to life . At birth , the brain quickly adapts and learns to control breathing in different situations . This adaptability is called neuroplasticity . Most breathing-related adjustments in the brain are short-term , like breathing faster during exercise . The brain can also learn from prior experience to prepare for future situations . For example , intermittent exposure to low oxygen causes long-term changes in signals from the brain to muscles controlling breathing , which may help them prepare for future low oxygen situations . This is called long-term facilitation ( LTF ) . This neuroplasticity may also help the brain to compensate or stabilize breathing during an illness or injury . Illnesses shortly after birth can affect how the brain controls breathing and may contribute to respiratory diseases later in life . They may also have lasting effects on the ability to of the brain to learn and respond to stress , and may even contribute to psychiatric disorders or age-related cognitive decline . Now , Hocker et al . show that inflammation shortly after birth has effects on breathing control that extend into adulthood . In the experiments , rats were injected four days after birth with either saline solution or a drug causing inflammation . When the rats grew into adults , their ability to make long-term breathing adjustments , or LTF , was assessed . In the rats exposed to early life inflammation two important pathways that enable LTF were eliminated . One pathway was restored when the rats received an anti-inflammatory treatment . Activating nerve cells reinstated the other pathway , suggesting these cells are not impaired . The experiments suggest inflammation during early life impairs breathing control later on and may contribute to adult respiratory disease . Inflammation is common among infants in their first year , particularly among those born prematurely . This early-life inflammation may put them at risk of diseases associated with breathing control , like sleep apnoea , later in life . More studies are needed to understand the relationship between early life inflammation , respiratory control , and respiratory disease later in life . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2019 | One bout of neonatal inflammation impairs adult respiratory motor plasticity in male and female rats |
Influenza A viruses ( IAV ) are subject to species barriers that prevent frequent zoonotic transmission and pandemics . One of these barriers is the poor activity of avian IAV polymerases in human cells . Differences between avian and mammalian ANP32 proteins underlie this host range barrier . Human ANP32A and ANP32B homologues both support function of human-adapted influenza polymerase but do not support efficient activity of avian IAV polymerase which requires avian ANP32A . We show here that the gene currently designated as avian ANP32B is evolutionarily distinct from mammalian ANP32B , and that chicken ANP32B does not support IAV polymerase activity even of human-adapted viruses . Consequently , IAV relies solely on chicken ANP32A to support its replication in chicken cells . Amino acids 129I and 130N , accounted for the inactivity of chicken ANP32B . Transfer of these residues to chicken ANP32A abolished support of IAV polymerase . Understanding ANP32 function will help develop antiviral strategies and aid the design of influenza virus resilient genome edited chickens .
Influenza A viruses ( IAV ) infect a wide range of host species but originate from wild birds . Zoonotic transmission from the avian reservoir is initially restricted by host specific species barriers . Infection of new host species requires the virus to bind to cell surface receptors , utilise foreign host cellular proteins whilst evading host restriction factors in order to replicate its genome , and finally transmit between individuals of the new host . The negative sense RNA genome of influenza A virus ( IAV ) is replicated in the cell nucleus using a virally encoded RNA-dependent RNA polymerase , a heterotrimer composed of the polymerase basic 1 ( PB1 ) , polymerase basic 2 ( PB2 ) and polymerase acidic ( PA ) proteins together with nucleoprotein ( NP ) that surrounds the viral RNA , forming the viral ribonucleoprotein complex ( vRNP ) ( Te Velthuis and Fodor , 2016 ) . Crucially , the viral polymerase must co-opt host factors to carry out transcription and replication ( Te Velthuis and Fodor , 2016 ) . The PB2 subunit is a major determinant of the host restriction of the viral polymerase ( Almond , 1977 ) . Avian IAV polymerases typically contain a glutamic acid at position 627 of PB2 , and mutation to a lysine , the typical residue at this position in mammalian-adapted PB2 ( Subbarao et al . , 1993 ) , can adapt the avian polymerase to function efficiently in mammalian cells . We have suggested that the restriction of avian IAV polymerase is due to a species specific difference in host protein ANP32A ( Long et al . , 2016 ) . Avian ANP32A proteins have a 33 amino acid insertion , lacking in mammals , and overexpression of chicken ANP32A ( chANP32A ) in human cells rescues efficient function of avian origin IAV polymerases ( Long et al . , 2016 ) . Removal of the 33 amino acids from chANP32A prevents polymerase rescue , whilst conversely artificial insertion of the 33 amino acids into either huANP32A or B overcomes host restriction ( Long et al . , 2016 ) . A naturally occurring splice variant of avian ANP32A lacks the first four amino acids of the 33 amino acid insertion , reducing the rescue efficiency of avian IAV polymerase in human cells ( Baker et al . , 2018 ) . This may be due to the disruption of a SUMOylation interaction motif , shown to enhance chANP32A’s interaction with IAV polymerase ( Domingues and Hale , 2017 ) . In human cells , both family members ANP32A and ANP32B ( huANP32A/B ) are utilised by human adapted IAV polymerases , and are thought to stimulate genome replication from the viral cRNA template , although the exact mechanism remains unclear ( Sugiyama et al . , 2015 ) . Here we demonstrate that the avian ANP32B clade is evolutionarily distinct from mammalian and other ANP32Bs . We demonstrate that two amino acids differences , N129 and D130 , in the LRR5 domain of chANP32B render it unable to interact with and support IAV polymerase function . We used CRISPR/Cas9 to remove the exon encoding the 33 amino acid insertion from chANP32A or to knockout the entire protein in chicken cells . Edited cells that expressed the short chANP32A isoform lacking the additional 33 amino acids supported mammalian-adapted but not avian IAV polymerase activity . Cells completely lacking chANP32A did not support either mammalian or avian IAV polymerase activity and were resilient to IAV infection . These results suggest a strategy to engineer IAV resilience in poultry through genetic deletion or amino acid changes of the LRR domain of ANP32A protein .
To examine the relatedness of ANP32 proteins from different species , we constructed a phylogenetic tree using vertebrate ANP32 protein sequences using Drosophila mapmodulin protein as an outgroup . ANP32A and E homologues formed well-supported monophyletic clades which included multiple avian and mammalian species ( Figure 1 , Figure 1—figure supplement 1 ) . Most vertebrate ANP32B proteins formed a monophyletic clade but this clade did not include avian ANP32B proteins . Rather , avian ANP32B proteins were strongly supported as members of a distinct clade with ANP32C from Xenopus and unnamed predicted proteins from non-placental mammals . This suggests that avian ANP32B and mammalian ANP32B are paralogues: birds have lost the protein orthologous to human ANP32B and eutherian mammals have lost the protein orthologous to avian ANP32B . Synteny provides further evidence to support the evolutionary relationship between avian ANP32B , Xenopus ANP32C , and the unnamed marsupial gene as they are all found adjacent to ZNF414 and MYO1F on their respective chromosomes ( Figure 1—figure supplement 2 ) . In humans , we found a short stretch of sequence between ZNF414 and MY01F which appears homologous to avian ANP32B ( Figure 1—figure supplement 2 ) . This provides further evidence that a functional gene orthologous to avian ANP32B has been lost in placental mammals . We and others have previously shown that both human ANP32A and B proteins support activity of a human-adapted IAV polymerase in human cells ( Long et al . , 2016; Sugiyama et al . , 2015; Watanabe et al . , 2014 ) . Using CRISPR/Cas9 , we generated human eHAP1 cells that lacked expression of both human ANP32A and ANP32B protein ( Staller et al . in review ) . In WT eHAP1 cells , human-adapted IAV polymerase ( PB2 627K ) , derived from an H5N1 virus A/turkey/England/50-92/1991 ( 50-92 ) , was active , whereas the WT avian polymerase ( PB2 627E ) was not . Exogenous expression of C-terminally FLAG-tagged chANP32A could rescue the activity of avian IAV polymerase whereas expression of chANP32B-FLAG , which naturally lacks the 33 amino acid insertion , did not ( Figure 2a ) . In double knockout cells , neither human-adapted nor avian-origin polymerase were active . Expression of chANP32A-FLAG rescued activity of both polymerases but expression of chicken ANP32B-FLAG rescued neither , despite confirmation of robust expression by western blot ( Figure 2b and c ) . This suggests that chicken ANP32B is not functional for IAV polymerase and that the IAV polymerase activity relies on ANP32A in chicken cells . To confirm this in chicken cells , we used CRISPR/Cas9 gene editing to generate chicken DF-1 cells which lacked chANP32B but retained chANP32A expression ( Figure 2—figure supplement 1 ) . Wild type DF-1 cells had mRNAs for chANP32A , B and E ( Figure 2—figure supplement 1 ) and supported activity of avian IAV polymerase bearing either PB2 627E or 627K . Overexpression of chANP32B-FLAG did not affect activity ( Figure 2d ) . DF-1 bKO cells also supported activity of both polymerases and again , exogenous expression of chANP32B had no effect . Since chicken cells lacking expression of chANP32B did not demonstrate any loss of IAV polymerase activity compared to WT , this implied that chANP32B is not functional for IAV polymerase and that IAV polymerase uses solely ANP32 family member A in chicken cells . To investigate the function of ANP32A in chicken cells we utilised a cell type that is more amenable to genome editing and clonal growth . Primordial germ cells ( PGCs ) are the lineage restricted stem cells which form the gametes in the adult animal . PGCs from the chicken embryo can be easily isolated and cultured indefinitely in defined medium conditions ( van de Lavoir et al . , 2006; Whyte et al . , 2015 ) . Chicken PGCs can be edited using artificial sequence-specific nucleases and subsequently used to generate genome edited offspring ( Park et al . , 2014; Oishi et al . , 2016 ) . Under appropriate in vitro conditions PGCs can acquire pluripotency and be subsequently differentiated into multiple cell types ( Matsui et al . , 1992; Shim et al . , 1997; Shamblott et al . , 1998; Park and Han , 2000 ) . Chicken PGC cells were genome edited using CRISPR/Cas9 and a single guide RNA which generated chANP32A knock-out cells ( aKO ) containing a biallelic deletion of 8 nucleotides in exon 1 . PGCs lacking the 33 amino acid insertion in chANP32A were generated using a pair of guide RNAs to remove exon five resulting in chicken cells with a mammalian-like ANP32A ( Δ33 ) ( Figure 3a ) . The precise deletions were confirmed by Sanger sequence analysis of subcloned PCR products from genomic DNA , and both found to be homozygous at both alleles ( Figure 3—figure supplement 1 ) . We differentiated the edited chicken PGCs into fibroblast-like cells using serum induction with the aim of generating cell lines to test avian IAV polymerase activity ( Figure 3—figure supplement 2 ) . The predicted alterations of ANP32A protein in these cells were confirmed by western blot analysis of the PGC-derived fibroblast cells ( Figure 3b ) . WT , Δ33 , and aKO and PGC-derived cell lines were tested for the functional effects of alteration or loss of chANP32A expression on IAV polymerase activity measured by reconstituted minigenome assay . Both avian ( PB2 627E ) and human-adapted polymerase ( PB2 627K ) were active in WT fibroblast cells ( Figure 3c ) . Removal of the 33 amino acids from ANP32A resulted in restriction of the 627E polymerase but not the 627K polymerase , mirroring the avian IAV polymerase phenotype observed in mammalian cells ( Long et al . , 2016 ) . Both polymerases were restricted in cells lacking chANP32A ( aKO ) . Expression of exogenous chANP32A in Δ33 and aKO cells rescued avian IAV polymerase activity ( Figure 3d & e ) demonstrating the specificity of the genetic alterations . The lack of polymerase activity in the aKO PGC cell line supports the hypothesis that , in the absence of chANP32A , the remaining ANP family members including chANP32B or chANP32E could not support IAV polymerase activity in chicken cells , even though ANP32B and E mRNAs were readily detected in both DF-1 and PGC cells ( Figure 2—figure supplement 1 and Figure 3—figure supplement 1 ) . ANP32 proteins share a common domain organisation in which an N terminal domain consisting of 5 consecutive leucine rich repeats ( LRR 1–5 ) is followed by a cap and central domain and a C terminal low complexity acidic region ( LCAR ) . In avian ANP32A proteins ( except some flightless birds ) a sequence duplication , derived in part from nucleotides that encode 27 amino acids ( 149-175 ) , has resulted in an additional exon and an insertion of up to 33 amino acids between the central domain and the LCAR ( Figure 4a ) . We previously showed that insertion of the 33 amino acids from the central domain of chANP32A into the equivalent region of the human ANP32A or huANP32B proteins conferred the ability to rescue the activity of a restricted avian IAV polymerase in human cells . The equivalent 33 amino acid insertion into chANP32B ( chANP32B33 ) did not support avian IAV polymerase activity ( Figure 4b ) . In order to ascertain the domains of chANP32B that rendered it non-functional for IAV polymerase activity , we generated chimeric constructs between human and chicken ANP32B . To measure the rescue of avian IAV polymerase in human 293 T cells , all chimeric constructs had the 33 amino acid sequence derived from chANP32A inserted between the LRR and LCAR domains . Western blot analysis and immunofluorescence confirmed that all chimeric constructs were expressed and localised to the cell nucleus as for the wild type ANP32 proteins . ( Figure 4b and Figure 4—figure supplement 1 ) . Swapping the LCAR domain of chANP32B into huANP32B33 did not prevent the rescue of avian IAV polymerase ( huANP32B33LCAR ) . Introduction of the central domain of chANP32B into huANP32B ( huANP32B33CENT ) significantly reduced rescue efficiency and swapping the LRR domain of chANP32B ( huANP32B33LRR ) rendered the protein non-functional to avian IAV polymerase ( Figure 4b ) . By sequential swapping of each LRR repeat , the 5th LRR of chANP32B was found to be the domain that prevented rescue of avian IAV polymerase ( Figure 4b ) . The fifth LRR contains five amino acid differences between human and chicken ANP32B , highlighted on the crystal structure of huANP32A , plus an additional one difference to chANP32A ( Figure 4d and Figure 4—figure supplement 2 ) . Swapping chANP32B’s fifth LRR into chANP32A also prevented rescue of avian IAV polymerase activity in human cells ( chANP32ALRR5 ) ( Figure 4c ) . Introduction of the single amino acid changes derived from the chANP32B LRR5 sequence into chANP32A revealed that mutations N129I and D130N significantly reduced the ability of chANP32A to rescue avian IAV polymerase activity in human cells ( Figure 4c ) . Minigenome assays with co-expressed chANP32A or chANP32AN129I in aKO chicken fibroblast cells confirmed that the 129I mutation significantly reduced the ability of chANP32A to support avian-origin ( PB2 627E ) or human-adapted ( PB2 627K ) IAV polymerase activity ( Figure 4e ) . As chANP32A KO PGC-derived fibroblast cells did not support of IAV polymerase despite expressing chANP32B , we were able to use these cells to understand in more detail the sequences in chANP32A required for IAV polymerase activity . The results above showed that the 33 amino acid insertion , fifth LRR and central domain are important for the ability of chANP32A to support function of avian IAV polymerase . We performed the minigenome assay in aKO cells with polymerases containing either PB2 627E and 627K with co-expression of further chANP32 mutants including: chANP32A in which the 27 amino acids in the central domain preceding the 33 amino acid insertion were scrambled ( chANP32Ascr149-175 ) or chANP32A with the 33 amino acid insertion scrambled ( chANP32Ascr176-208 ) ( Figure 5a ) . Both mutants were expressed and localised to the nucleus ( Figure 5c and Figure 4—figure supplement 1 ) . The first mutant , chANP32Ascr149-175 , did not support either PB2 627E or 627K polymerase , suggesting the sequence of the central domain is important for function of IAV polymerase . The second mutant , chANP32A scr176-208 , only supported PB2 627K function , confirming that the sequence of the 33 amino acid insertion , not just the extended length is required for avian IAV polymerase ( PB2 627E ) ( Figure 5b ) . An interaction between ANP32A and IAV polymerase was demonstrated previously that is dependent on the presence of all three polymerase subunits ( Mistry et al . in preparation & Baker et al . , 2018; Domingues and Hale , 2017 ) . To examine the interaction between IAV polymerase and chANP32 proteins we employed a split luciferase complementation assay as a quantitative measure of binding ( Munier et al . , 2013; Cassonnet et al . , 2011 ) . The C-terminus of the PB1 subunit of avian origin IAV polymerase was fused with one half of gaussia luciferase ( PB1luc1 ) and the C-terminus of chicken ANP32A or B with the second half ( chANP32Aluc2 and chANP32Bluc2 ) ( Figure 6a ) . Reconstitution of PB1luc1 , PB2 and PA together with chANP32Aluc2 in human 293 T cells gave a strong Normalised Luciferase Ratio ( NLR ) ( Figure 6—figure supplement 1 ) with polymerases containing either PB2 627E or 627K ( Figure 6b ) . Luciferase complementation was significantly less between polymerase and chANP32Bluc2 , and even insertion of the 33 amino acids from chANP32A did not restore the signal ( chANP32B33luc2 ) ( Figure 6b ) . When chANP32A carried the single N129I mutation ( chANP32AN129Iluc2 ) , luciferase complementation was reduced 22-fold for PB2 627E polymerase and 52-fold forPB2 627K polymerase ( Figure 6c and Figure 6—figure supplement 1 ) . These results suggest that the loss of support of polymerase function by chANP32AN129I was due to a disruption of binding to IAV polymerase . ANP32A proteins bind to histones as part of their role in chromatin regulation ( Reilly et al . , 2014 ) . To measure if the mutation N129I had any effect on this cellular interaction , we generated expression plasmids that encoded human histone four with luc1 fused to the C-terminus ( H4luc1 ) and histone 3 . 1 with luc2 fused to the C terminus ( H3luc2 ) . As expected , H4luc1 and H3luc2 generated a strong NLR , reflecting their interaction in the nucleosome ( Luger et al . , 1997 ) . The ability of chANP32A to bind histone four was not impaired by mutation N129I , suggesting chANP32N129I was not altered in this cellular role , despite abrogation of its support of IAV polymerase ( Figure 6d ) . The data above suggest that chANP32B cannot substitute for chANP32A in support of IAV polymerase in chicken cells . Since chicken cells that completely lack expression of chANP32A show no polymerase activity in the minigenome assay , they might be refractory to IAV infection . Multi-cycle growth kinetics of recombinant influenza A viruses were measured in WT and aKO PGC-derived fibroblast cells ( Figure 7 ) . To ensure robust infection , recombinant viruses were generated carrying H1N1 vaccine strain PR8 haemagglutinin ( HA ) , neuraminidase ( NA ) and M genes; this also mitigated the risks of working with avian influenza viruses with novel antigenicity . Infectious titres of recombinant virus with internal genes of avian H5N1 virus 50–92 were not detected in the chicken cells lacking ANP32A infected at low MOI ( Figure 7a ) . At higher MOI virus titres were significantly reduced compared to WT chicken cells , almost 325-fold less at 8 hr post infection and 16-fold less by 24 hr ( Figure 7c ) . Similarly , at low MOI , a recombinant virus with internal genes from the H7N9 virus A/Anhui/1/2013 , had limited virus growth in aKO cells but replicated efficiently in WT fibroblasts ( Figure 7b ) . At the higher MOI , peak viral titres were 1365-fold less than in WT cells at 8 hr post infection and 100-fold less by 24 hr ( Figure 7d ) . Since virus growth in aKO cells was observed at the higher MOI , we sequenced the PB2 gene of virus progeny to determine if this replication was due to adaptation in PB2 . At 24 hr post infection the sequence of the PB2 gene from virus in supernatants of cells infected at high MOI was determined . Virus recovered from aKO and WT cells was found to be identical and contained no sequence changes compared with the inoculum , suggesting adaptation in PB2 was not required for the low level of replication seen in the aKO cells . In conclusion , PGC derived fibroblast cells lacking chANP32A were resilient to IAV replication , particularly at lower multiplicities of infection .
We show that avian origin IAV polymerases rely exclusively on chicken ANP32A family member for their replication , because they are unable to co-opt chicken ANP32B . We found avian ANP32B proteins formed a separate phylogenetic group from other ANP32Bs ( Figure 1 and Figure 1—figure supplement 1 ) . Synteny demonstrated that an avian ANP32B homologue was present in coelacanth , amphibians and non-placental mammals as these loci were identical to the ANP32B locus in birds ( Figure 1—figure supplement 2 ) . A functional avian ANP32B homologue has been lost in placental mammals although a very small part of an ANP32 gene remained in humans ( Figure 1—figure supplement 2 ) . Human ANP32C is an intronless gene that is most closely related to ANP32A ( Reilly et al . , 2014 ) and is unrelated to the avian ANP32B clade . There was no evidence found of a mammalian ANP32B homologue in birds . Chicken ANP32B could not support influenza polymerase function due to an amino acid difference in LRR5 at residue 129 that adversely affected the interaction between chANP32B and influenza polymerase ( Figure 6 ) . Other avian ANP32B proteins , including those of duck and turkey , carry isoleucine at residue 129 suggesting that our findings may also be applicable to other avian hosts ( Figure 4—figure supplement 2 ) . The replacement of the exposed polar residue , asparagine ( N129 ) with the hydrophobic isoleucine ( I ) may have led to the disruption of a key electrostatic interaction between ANP32A and the virus polymerase complex . In addition to the residue 129I , the central domain ( amino acids 141–175 ) of chANP32B also contributed to its poor efficiency at rescuing avian IAV polymerase function in human cells ( Figure 4 ) . This , together with the observation that scrambling amino acids 149–175 in chANP32A prevented both human-adapted and avian IAV polymerase function ( Figure 4 ) suggests that LRR5 and the central domain of ANP32A are crucial to IAV polymerase function . Our finding that chANP32B is non-functional for IAV polymerase was recently corroborated by another research group; this preliminary work also found that this phenotype mapped to residues 129I and 130N of chANP32B ( Zhang et al . , 2019 ) . The observation that scrambling the 33-amino acid insertion prevented avian IAV polymerase rescue ( Figure 5 ) is consistent with results from by Domingues and Hale and Baker and colleagues which showed that the SUMO Interaction motif ( SIM ) -like sequence present in the 33 amino acid insertion ( VLSLV ) , was required for strong binding to both 627E and 627K polymerase and its deletion or mutation decreased its ability to support avian IAV polymerase activity in human cells ( Baker et al . , 2018; Domingues and Hale , 2017 ) . Understanding the domains important to binding and function may help us understand the mechanism by which ANP32A or B support IAV polymerase which is still not fully elucidated ( Sugiyama et al . , 2015 ) . Chicken cells lacking ANP32A did not support activity of avian or human-adapted IAV polymerase in minigenome assay ( Figure 2 ) . However , at higher MOI , virus replication was observed in aKO cells , although still significantly lower than in WT PGC cells ( Figure 7c&d ) . This implies some IAV polymerase function , albeit inefficient , in the absence of chANP32A in the context of virus infection . Other viral products present during virus infection such as NEP may partly compensate for the block in replication in cells that lack chANP32A . Indeed NEP expression has been reported to rescue avian polymerase replication in human cells ( Mänz et al . , 2012 ) . Nonetheless the significantly reduced level of virus replication observed in the chicken cells that lack chANP32A in vitro implies that in vivo , chickens that do not express ANP32A or express altered protein may be resilient to infection by IAV . It will be pertinent to investigate whether IAV can evolve to replicate in chicken cells that lack or express a mutated ANP32A although no adaptation of PB2 gene was observed here . The discrepancy between the lack of polymerase activity in the minigenome assay in absence of ANP32A yet limited replication observed using infectious virus in the same cells may ultimately reveal interesting insights about how ANP32A supports polymerase . The use of the PGC-derived chicken cells to investigate a host factor essential for virus raises the possibility of generating genome-edited chicken models resistant or resilient to infection . Chicken PGCs can be efficiently genome-edited to generate specific haplotypes ( Idoko-Akoh et al . , 2018 ) . Our novel method of chicken PGC differentiation into fibroblast-like cells enabled robust testing of a defined genotype , and will permit future investigation of other host genetic factors identified through forward genetic screens and suspected to play important roles in virus infections ( Smith et al . , 2015; Wang et al . , 2014 ) . We demonstrated that a mutated chANP32A was able to bind histone 4 , suggesting this cellular role may not be affected by amino acid change at 129 . However , there are many roles attributed to ANP32A proteins in the cell , such as embryogenesis , and disruption of these functions may limit the ability to generate healthy gene-edited animals ( Reilly et al . , 2014; Reilly et al . , 2011 ) . In summary , we provide evidence that specific domains of ANP32 proteins are important for the function of IAV polymerases and describe a lack of redundancy in the involvement of ANP32 family members to support IAV polymerase complex in chicken cells that is determined by the variation in ANP32 protein sequences . These data may aid in the design of novel small molecule inhibitors that disrupt the ANP32-polymerase interface and form the basis of a potential pathway for the generation of influenza virus resilient animals .
The GFP+ PGCs used in the experiments were obtained by crossing the Roslin Green ( ubiquitous GFP ) line of transgenic chickens with a flock of commercial Hyline layer hens maintained at the Roslin Institute to produce heterozygous fertile eggs for PGC derivations ( Pettersen et al . , 2004 ) . Commercial and transgenic chicken lines were maintained and bred under UK Home Office License . All experiments were performed in accordance with relevant UK Home Office guidelines and regulations . The experimental protocol and studies were reviewed by the Roslin Institute Animal Welfare and Ethical Review Board ( AWERB ) Committee . Chickens for egg production were maintained under the HO code of practice ( ISBN 9781474112390 ) . ANP32A guide RNAs ( gRNA ) were designed using CHOPCHOP gRNA web tool ( http://chopchop . cbu . uib . no/ ) ( Montague et al . , 2014; Labun et al . , 2016 ) . gRNA 5’-CGGCCATGGACATGAAGAAA-3’ targeting ANP32A exon1 , and gRNAs: 5’-AGCTGGAAGCAATATGTACT-3’ and 5’-CATTCCCCTCGCTCCTTCAA-3’ targeting either side of exon 5 ( Δ33 PGC cells ) were cloned into pSpCas9 ( BB ) −2A-Puro ( pX459 v2 . 0; a gift from Dr . Feng Zhang ) using Materials and methods described by previously ( Ran et al . , 2013a ) . For DF-1 ANP32B gRNAs , the guides 5’-TTCAGATGATGGGAAGATCG-3’ and 5’-GGTTCTCAAAATCTGAAGAG-3’ were cloned into the double ‘nickase’ vectors pSpCas9n ( BB ) −2A-GFP ( pX461 ) and pSpCas9n ( BB ) −2A-Puro ( pX462 ) respectively ( Ran et al . , 2013a ; Ran et al . , 2013b ) . Gaussia luc1 and luc2 were generated by gene synthesis ( GeneArt , ThermoFisher ) using the sequence previously described ( Cassonnet et al . , 2011 ) . Homo sapiens Histone 4 ( NP_003533 . 1 ) and 3 . 1 ( NP_003520 . 1 ) were generated gene synthesis ( GeneArt , ThermoFisher ) . Luc1 or luc2 were added to the C-termini of ANP32 , PB1 , H4 or H3 . 1 using the linker sequence , AAAGGGGSGGGGS , by overlapping PCR . The 33 amino acid insertion was added to huANP32B after residue 173 and to chANP32B after residue 181 ( preserving an acid region before SIM motif [Domingues and Hale , 2017] ) . The LRR ( amino acids 1–149 ) , central domain ( amino acids 150–175 ) or LCAR ( amino acids 176–262 ) from chANP32B were swapped into huANP32B33 to generate chimeric constructs . ANP32 constructs were made by overlapping PCR or by gene synthesis ( GeneArt , ThermoFisher ) with either a FLAG tagged fused to the C-terminus with a GSG linker or to mCherry with a GSGGGSGG linker . Human embryonic kidney ( 293T ) ( ATCC ) and Madin-Darby canine kidney ( MDCK ) cells ( ATCC ) were maintained in cell culture media ( Dulbecco’s modified Eagle’s medium ( DMEM; Invitrogen ) supplemented with 10% fetal calf serum ( FCS ) ( Biosera ) , 1% non-essential amino acids ( NEAA ) and with 1% penicillin-streptomycin ( Invitrogen ) ) and maintained at 37°C in a 5% CO2 atmosphere . Human eHAP1 cells ( Horizon Discovery ) were cultured in Iscove’s Modified Dulbecco’s Medium ( IMDM ) supplemented with 10% fetal bovine serum ( FBS ) , 1% NEAAs , and 1% penicillin/streptomycin . Chicken fibroblast ( DF-1 ) ( ATCC ) cells were maintained in DF-1 cell culture media ( DMEM supplemented with 10% FCS , 5% tryptose phosphate broth ( Sigma-Aldrich ) , 1% NEAAs and 1% penicillin-streptomycin and maintained at 39°C in a 5% CO2 atmosphere . Cell line authentication: DF-1 , eHAP1 and 293 T cells were authenticated by mRNA analysis confirming the relevant species . All continuous cell lines were routinely screened for mycoplasma contamination and were mycoplasma free . PGCs were derived and cultured in FAOT medium as previously described ( Whyte et al . , 2015 ) . PGCs were transiently transfected with 1 . 5 µg of PX459 V2 . 0 vector using Lipofectamine 2000 ( Invitrogen ) and treated with puromycin as previously described ( Idoko-Akoh et al . , 2018 ) . Subsequently , single cell cultures of puromycin-resistant cells were established to generate clonal populations for downstream experiments as previously described in Idoko-Akoh et al . ( 2018 ) . To identify an ANP32A Δ33 PGC cell line , PCR products were directly sequenced using PCR primers to analyse mutation genotypes of isolated single cell clones . To identify an ANP32A KO PGC cell line , PCR products were cloned into pGEM-T Easy vector ( Promega ) and sequenced using T7 promoter forward primer by Sanger sequencing . DF-1 cells were transfected with the described CRISPR/Cas9 constructs using Lipofectamine 2000 ( Invitrogen ) and subject to puromycin selection . Single cell clones were expanded and analysed by PCR of genomic DNA and Sanger sequencing using primers ( 5’-TTTTTGCTTACATCTGAGGGC-3’ , 5’-CCTCCGCAGTTATCAGGTTAGT-3’ ) for ANP32A exon1 , ( 5’-GCTCCCTGGTCTGCTAGTTAT-3’ , 5’-GGTCTACGCAACCACACATAC-3’ ) for ANP32A exon five and ( 5’-CCCTTAAGGTGAGCACAGGG-3’ , 5’-AACATAGCACCACTCCCAGC-3’ ) for ANP32B exon2 . eHAP1 dKO cells were generated as described ( Staller et al . in review ) . PGCs were cultured in 500 µl of high calcium FAOT medium containing 1 . 8 mM CaCl2 in fibronectin-coated wells ( 24-well plate ) for 48 hr ( Figure 3—figure supplement 2 ) ( Whyte et al . , 2015 ) . Subsequently , PGCs were transferred into PGC fibroblast medium and then refreshed every 48 hr by removing and replacing with 300 µl of PGC fibroblast cell culture medium . Adherent fibroblast-like cells were observed within 72 hr . Cells were then refed every two days and split 1:4 every four days . PGC fibroblast cell cultures were expanded to 85–90% confluency in 24-well plates before using for transfection , infection or western blot analysis . PGC fibroblast cells were maintained in cell culture media ( Knockout DMEM ( 10829018 , Gibco ) with 10% ES grade FBS ( 16141061 , Invitrogen ) , 1% chicken serum ( Biosera ) , 0 . 1% 100xNEAA ( Gibco ) , 0 . 1% Pyruvate ( 11360070 , Gibco ) , 0 . 1% 100xGlutamax ( Gibco: 35050–038 ) , 0 . 5 mg ml−1 ovotransferin ( C7786 , Sigma ) ) and 1% penicillin-streptomycin at 37°C with 5% CO2 . Recombinant influenza A PR8 ( A/PR/8/34 ( H1N1 ) ) 3:5 reassortant virus ( PR8 HA , NA and M genes with PB1 , PB2 , PA , NP and NS genes from A/Anhui/1/13 ( H7N9 ) was generated by reverse genetics at The Pirbright Institute , UK . Reverse genetics virus rescue was performed by transfection of Human Embryonic Kidney ( HEK ) 293 T cells ( ATCC ) with eight bi-directional pHW2000 plasmids containing the appropriate influenza A virus segments and co-culture in MDCK cells ( ATCC ) with addition of 2 µg ml−1 of TPCK treated Trypsin ( Sigma-Aldrich ) . Rescued viruses were passaged once in embryonated hen’s eggs to generate working stocks . Recombinant PR8 3:5 reassortant 50–92 ( A/turkey/England/50-92/1991 ( H5N1 ) was described previously [Long et al . , 2013] ) . Virus was diluted in Knockout DMEM and incubated on PGC fibroblast cells for 1 hr at 37°C ( MOI as indicated in the relevant figure legends ) after which inoculum was removed and cells washed with PBS followed by MES buffer ( pH 4 , 37°C ) for five mins and a further PBS wash . Infection media ( Knockout DMEM ( 10829018 , Gibco ) , 0 . 14% BSA and 1 μg ml−1 TPCK trypsin ( Sigma-Aldrich ) ) wash added and cells were incubated at 37°C . To ensure residual virus was removed , a 0 hr time point was taken . Cell supernatants were harvested and stored at −80°C . Infectious titres were determined by plaque assay on MDCK cells . vRNA extraction from cell supernatants was performed using QIAamp Viral RNA Mini Kit ( Qiagen 52906 ) . First strand synthesis was performed using SuperScript IV reverse transcriptase with primer 5’-GCAGGTCAAATATATTCAATATGG-3’ . cDNA was amplified using KOD Hot Start DNA polymerase ( Merck 71086 ) using primers 5’-GCAGGTCAAATATATTCAATATGG-3’ and 5’-GGTCGTTTTTAAACAATTCGAC-3’ and the PCR product was sequenced by Sanger sequencing . Influenza polymerase activity was measured by use of a minigenome reporter which contains the firefly luciferase gene flanked by the non-coding regions of the influenza NS gene segment , transcribed from a species-specific polI plasmid with a mouse terminator sequence . The human and chicken polI minigenomes ( pHOM1-Firefly and pCOM1-Firefly ) are described previously ( Moncorgé et al . , 2013 ) . pCAGGS expression plasmids encoding each polymerase component and NP for 50–92 ( H5N1 A/Turkey/England/50–92/91 ) are described previously ( Long et al . , 2013 ) . To measure influenza polymerase activity , 293 T cells were transfected in 48-well plates with pCAGGS plasmids encoding the PB1 ( 20 ng ) , PB2 ( 20 ng ) , PA ( 10 ng ) and NP ( 40 ng ) proteins , together with 20 ng species-specific minigenome reporter , either Empty pCAGGS or pCAGGS expressing ANP32 ( 50 ng ) and , as an internal control , 10 ng Renilla luciferase expression plasmid ( pCAGGS-Renilla ) , using Lipofectamine 3000 transfection reagent ( Invitrogen ) according to manufacturers’ instructions . DF-1 and PGC fibroblast cells were transfected as 293 T cells but with twice the concentration of DNA . Cells were incubated at 37°C . 20–24 hr after transfection , cells were lysed with 50 μl of passive lysis buffer ( Promega ) , and firefly and Renilla luciferase bioluminescence was measured using a Dual-luciferase system ( Promega ) with a FLUOstar Omega plate reader ( BMG Labtech ) . 293 T cells were transfected with PB1luc1 ( 25 ng ) , either PB2 627E or PB2 627K ( 25 ng ) , PA ( 12 . 5 ng ) and chANP32Aluc2 , chANP32AN129Iluc2 , chANP32Bluc2 or chANP32B33luc2 ( 12 . 5 ng ) . For split luciferase assays measuring histone interaction , 50 ng of either chANP32Aluc2 , chANP32AN129Iluc2 , H4luc1 or H3luc2 were transfected into 293 T cells . Control samples assessed the interaction between H4 or PB1luc1 and an untagged luc2 construct or the appropriate ANP32Aluc2 construct and an untagged luc1 construct . All other components transfected into control samples remained consistent with those transfected in with the interacting proteins of interest . 24 hr after transfection , cells were lysed in 50 ul Renilla lysis buffer ( Promega ) for one hour at room temperature . Gaussia luciferase activity was then measured from 10 ul of lysate using the Renilla luciferase kit ( Promega ) with a FLUOstar Omega plate reader ( BMG Labtech ) . Normalised luminescence ratios were calculated by dividing the luminescence measured from the interacting partners by the sum of the interaction measured from the two controls for each sample ( Figure 6—figure supplement 1 ) as previously described ( Cassonnet et al . , 2011 ) . For analysis of PGC derived fibroblasts ( Figure 3b ) , at least 300 , 000 cells were lysed in 60 µl of 1X RIPA lysis buffer ( sc-24948 , Santa Cruz Biotechnology ) according to the manufacturer’s instruction . Protein concentration was determined using the Bradford method with the Quick Start Bradford Protein Assay Kit ( #5000202 , BIORAD ) according to the manufacturer’s instruction ( Bradford , 1976 ) Denaturing electrophoresis and western blotting were performed using the NuPAGE electrophoresis system ( Invitrogen ) following the manufacturer’s protocol . For all other Western blots , cells were lysed in lysis buffer ( 50 mM Tris-HCl pH 7 . 8 ( Sigma Aldrich ) , 100 mM NaCl , 50 mM KCl and 0 . 5% Triton X-100 ( Sigma Aldrich ) , supplemented with cOmplete EDTA free Protease inhibitor cocktail tablet ( Roche ) ) and prepared in Laemmli 2 × buffer ( Sigma-Aldrich ) . Cell proteins were resolved by SDS–PAGE using Mini-PROTEAN TGX Precast Gels ( Bio-Rad ) . Immunoblotting was carried out using the following primary antibodies: rabbit α-ANP32A ( Sigma-Aldrich AV40203 ) , mouse α-β-actin ( Sigma-Aldrich A2228 ) , mouse α-FLAG ( F1804 , Sigma-Aldrich ) , mouse α-Lamin B1 ( MAB5492 , Merck ) , mouse α-PCNA ( sc-25280 , Santa Cruz ) , rabbit α-Histone 3 ( AB1791 , Abcam ) , rabbit α-vinculin ( AB129002 , Abcam ) , rabbit α-Gaussia Luc ( E80235 , NEB ) , rabbit α-PB1 ( PA5-34914 , Invitrogen ) and rabbit α-PB2 ( GTX125926 , GeneTex ) . The following secondary antibodies were used: goat anti-rabbit HRP ( CST #7074 ) , anti-mouse HRP ( CST #7076 ) , goat α-mouse AlexaFluor-568 ( A11031 , Invitrogen ) , sheep α-rabbit HRP ( AP510P , Merck ) and goat α-mouse HRP ( STAR117P , AbD Serotec ) . Protein bands were visualised by chemiluminescence ( ECL +western blotting substrate , Pierce ) using a FUSION-FX imaging system ( Vilber Lourmat ) . Total RNA from PGC fibroblast and DF-1 cells were extracted using an RNeasy mini kit ( Qiagen ) , following manufacturer’s instructions . During extraction of RNA , RNeasy columns were treated with RNase-Free DNase ( Qiagen ) . RNA samples were quantified using a Nanodrop Spectrophotometer ( Thermo Scientific ) . Equal concentrations of RNA were subject to first strand synthesis using RevertAid ( Thermo Scientific ) with Oligo ( dT ) ( Thermo Scientific ) . This product was then quantified with Fast SYBR Green Master Mix ( Thermo Scientific ) using the following sequence-specific primer pairs: RS17 , ( 5’-ACACCCGTCTGGGCAACGACT-3’ and 5’-CCCGCTGGATGCGCTTCATCA-3’ ) , RPL30 ( 5’-CCAACAACTGTCCTGCTTT-3’ and 5’-GAGTCACCTGGGTCAATAA-3’ ) , chANP32A ( 5′-GTTTGCAACTGAGGCTAAGC-3′ and 5’-CAACTGTAGGTCATACGAAGGC-3’ ) , chANP32B ( 5′- GGTGGCCTTGAAGTTCTAGC-3′ , and 5’-ATGAGCATCGTCACCTCGC-3’ ) , chANP32E ( 5’- GAACTAGAGTTTCTTAGCATGG-3’ and 5’- TCTCTCTGCAAGGACCTCCAG-3’ ) . Real-time quantitative PCR analysis was performed ( Applied Biosystems ViiA 7 Real-Time PCR System ) . All work with infectious agents was conducted in biosafety level two facilities , approved by the Health and Safety Executive of the UK and in accordance with local rules , at Imperial College London , UK . ANP32 sequences were downloaded from Ensembl ( Gene Trees ENSGT00940000153254 and ENSGT00940000154305 . ) Amino acid sequences were aligned using MUSCLE ( Edgar , 2004 ) and the maximum likelihood tree was constructed using RAxML-HPC2 v . 8 . 2 . 10 ( Stamatakis , 2014 ) ( GTRGAMMA model , 100 bootstraps ) on XSEDE run on CIPRES ( Miller et al . , 2010 ) . Mapmodulin from Drosophila melanogaster was used as an outgroup . Statistical analysis was performed using GraphPad Prism v . 7 . Sequencing data was analysed using Geneious Sorfware R6 . Image analysis was done using Image J and Microsoft Office 2016 . | The influenza A virus pandemic of 1918 killed more people than the armed conflicts of World War 1 . Like all other pandemic and seasonal influenza , this virus originated from bird viruses . In fact , avian influenza viruses continually threaten to spark new outbreaks in humans , but pandemics do not occur often . This is because these viruses must undergo several adaptations before they can replicate in and spread between people . Viruses make new copies of themselves using the molecular machinery of the cells that they invade . The proteins that make up this machinery are often slightly different in different species , and so a virus that can replicate in cells of one species might not be able to do so when it invades a cell from another species . In 2016 , researchers discovered that species differences in a cell protein called ANP32A pose a key barrier that avian influenza viruses have to overcome . Now , Long et al . – including some of the researchers involved in the 2016 study – show that the avian influenza virus cannot replicate in chicken cells that lack ANP32A . Exploring closely related versions of the genes that produce ANP32A and its relative ANP32B in different species revealed the region of the protein that the virus relies on to support its replication . Long et al . speculate that by making a few small changes to the ANP32A gene in chickens , it might be possible to generate a gene-edited chicken that is resilient to influenza . Close contact with poultry has led to hundreds of cases of ‘bird ‘flu’ in South East Asia , many of which have been fatal . Moreover , if avian influenza viruses mutate further in an infected person , a new pandemic could begin . Stopping influenza viruses from replicating in chickens would prevent people from being exposed to these dangerous viruses , whilst also improving the welfare of the chickens . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2019 | Species specific differences in use of ANP32 proteins by influenza A virus |
Stable adherence to epithelial surfaces is required for colonization by diverse host-associated microbes . Successful attachment of pathogenic microbes to host cells via adhesin molecules is also the first step in many devastating infections . Despite the primacy of epithelial adherence in establishing host-microbe associations , the evolutionary processes that shape this crucial interface remain enigmatic . Carcinoembryonic antigen-related cell adhesion molecules ( CEACAMs ) encompass a multifunctional family of vertebrate cell surface proteins which are recurrent targets of bacterial adhesins at epithelial barriers . Here , we show that multiple members of the primate CEACAM family exhibit evidence of repeated natural selection at protein surfaces targeted by bacteria , consistent with pathogen-driven evolution . Divergence of CEACAM proteins between even closely related great apes is sufficient to control molecular interactions with a range of bacterial adhesins . Phylogenetic analyses further reveal that repeated gene conversion of CEACAM extracellular domains during primate divergence plays a key role in limiting bacterial adhesin host tropism . Moreover , we demonstrate that gene conversion has continued to shape CEACAM diversity within human populations , with abundant human CEACAM1 variants mediating evasion of adhesins from pathogenic Neisseria . Together this work reveals a mechanism by which gene conversion shapes first contact between microbes and animal hosts .
Epithelial surfaces are typically the initial point of contact between metazoans and microbes ( Brown and Clarke , 2017 ) . As such , host factors at this barrier play an important role in facilitating or deterring microbial colonization . Bacterial attachment to epithelial surfaces is often mediated by a broad class of surface proteins termed adhesins ( Kline et al . , 2009 ) . In addition to permitting the growth and colonization of commensal microbes , adhesins are also key virulence factors for many pathogenic bacteria . Adhesin-mediated adherence to host cells is often required for other downstream processes including biofilm formation , epithelial invasion , and the delivery of toxic effectors into host cells ( Kline et al . , 2009; Sadarangani et al . , 2011; Figure 1 ) . Microbial adherence can also trigger epithelial cell signaling cascades , further shaping host responses to resident and invasive microbes . Despite the fundamental importance of epithelial adherence for bacterial colonization and infectious disease pathogenesis , the dynamics of these interactions between host surface proteins and bacterial adhesions over evolutionary timescales remain a mystery . Theory predicts that exploitation of host proteins by pathogens places a significant burden on host populations , driving selection for beneficial mutations that limit microbial invasion or virulence . From a microbial perspective , host defenses can also pose an existential threat resulting in reciprocal adaptation to enhance colonization , growth , and transmission ( Aleru and Barber , 2020; Brockhurst et al . , 2014; Hamilton et al . , 1990; Van Valen , 1973 ) . However , pathogens hijack many host factors not directly involved in immunity , possibly limiting their adaptive potential in response to pathogen interaction . For example , epithelial surface proteins are not only essential for interacting with the environment but also serve crucial cellular and physiological functions including barrier maintenance , cell-cell communication , as well as coordinating host physiological and developmental pathways ( Kuespert et al . , 2006 ) . It therefore remains unclear the extent to which such proteins are able to adapt in the face of pathogen antagonism . Bacterial adhesins interact with a wide range of molecules present on host epithelial surfaces ( Chatterjee et al . , 2021 ) . One important target of bacterial adhesins on vertebrate epithelia are the carcinoembryonic antigen-related cell adhesion molecule ( CEACAM ) family of proteins ( Gray-Owen and Blumberg , 2006 ) . There have been multiple independent expansions of this gene family across mammals ( Kammerer and Zimmermann , 2010; Pavlopoulou and Scorilas , 2014 ) and the human genome encodes 12 functional CEACAM genes as well as several pseudogenes ( Gray-Owen and Blumberg , 2006; Kammerer and Zimmermann , 2010 ) . Collectively , CEACAMs are expressed on nearly all vertebrate epithelial surfaces including the microbe-rich surfaces of the urogenital , respiratory , and gastrointestinal tracts ( Tchoupa et al . , 2014 ) . Epithelial CEACAMs play a variety of roles in cell adhesion as well as intra- and intercellular signaling ( Gray-Owen and Blumberg , 2006; Kuespert et al . , 2006; Tchoupa et al . , 2014 ) . A subset of CEACAMs are also expressed on other cell types , including T-cells and neutrophils where they play important roles in immune signaling and pathogen recognition . CEACAMs typically consist of an extracellular N-terminal IgV-like domain ( also termed the N-domain ) , a variable number of IgC-like domains , and either a GPI anchor or a transmembrane and cytoplasmic domains . The characteristics of this cytoplasmic domain in turn influences the functional properties of CEACAM proteins . Extracellular protein-protein interactions involving CEACAMs have been shown to primarily occur through the extracellular N-domain ( Kuespert et al . , 2007; Markel et al . , 2004 ) . While the functions of many CEACAM proteins remain obscure , mammalian CEACAM1 , CEACAM5 ( also known as CEA ) , and CEACAM6 have been shown to contribute to immunoregulation , cell-cycle progression , and development ( Gray-Owen and Blumberg , 2006; Kuespert et al . , 2006; Tchoupa et al . , 2014 ) . These CEACAMs , along with CEACAM3 , are also notable as recurrent targets of bacterial adhesins ( Gray-Owen and Blumberg , 2006; Tchoupa et al . , 2014 ) . A growing number of bacterial genera have been found to target these CEACAM proteins to promote epithelial adherence and host colonization , including Neisseria , Haemophilus , Escherichia , Fusobacterium , Streptococcus , and Helicobacter ( Brewer et al . , 2019; Gray-Owen and Blumberg , 2006; Javaheri et al . , 2016; Königer et al . , 2016; van Sorge et al . , 2021 ) . Additionally , several viruses have been reported to bind CEACAMs , including human cytomegalovirus , influenza A , murine coronavirus , and Middle East respiratory syndrome coronavirus ( Chan et al . , 2016; Hemmila et al . , 2004; Macmaniman et al . , 2014; Rahman et al . , 2021 ) . While many of the bacteria reported to target CEACAMs are able to colonize the host as benign commensals , they are also capable of causing serious infection especially in young children . Such infections may prove severely detrimental to host fitness even when not fatal . Among bacteria , a number of structurally unrelated CEACAM-binding adhesin proteins have been described . This diversity of structures suggests that CEACAM recognition has arisen independently multiple times during bacterial evolution . These structurally diverse proteins include the integral outer membrane Opa proteins in Neisseria ( Fox et al . , 2014 ) , the immunoglobulin-type β protein from group B Streptococcus ( van Sorge et al . , 2021 ) , as well as other trimeric and globular protein domains ( Bonsor et al . , 2018; Conners et al . , 2008 ) . Bacterial CEACAM recognition can lead to several distinct outcomes ( Figure 1 ) . First , adherence to epithelial CEACAMs can provide a stable habitat to support bacterial growth and proliferation . In mice , for example , expression of human CEACAM1 is sufficient to establish stable colonization by otherwise human-restricted strains of Neisseria meningitidis ( Johswich et al . , 2013 ) . Second , CEACAM binding may facilitate bacterial dissemination through the host epithelium ( Wang et al . , 1998 ) . Third , in the case of the bacterium Helicobacter pylori , CEACAM-adhesin interactions promote the translocation of virulence factors into host cells via the type 4 secretion system leading to severe gastritis and stomach ulcers in humans ( Javaheri et al . , 2016; Königer et al . , 2016 ) . Finally , bacterial adhesins can potentiate CEACAM-mediated signaling cascades to manipulate cellular functions , including preventing immune cell activation ( Gur et al . , 2019a; Gur et al . , 2019b; Sadarangani et al . , 2011 ) , increasing cellular adhesion to prevent shedding of infected cells ( Muenzner et al . , 2016; Muenzner et al . , 2010 ) , and activation of apoptosis ( N’Guessan et al . , 2007 ) . Interactions between adhesins and CEACAMs , particularly bacterial immunoglobulins that appear to mimic CEACAMs , are predicted to further disrupt endogenous CEACAM adhesion and signaling functions ( Macmaniman et al . , 2014; Moonens , 2018; van Sorge et al . , 2021 ) . Previous work has reported multiple instances of gene gain and loss as well as high levels of sequence divergence among a subset of CEACAM genes ( Adrian et al . , 2019; Rhesus Macaque Genome Sequencing and Analysis Consortium et al . , 2007; Kammerer and Zimmermann , 2010; Pavlopoulou and Scorilas , 2014 ) . These findings , coupled with the observation that many CEACAM-binding bacteria possess a narrow host range , suggest that host genetic variation may be a major determinant of bacterial colonization . In the case of CEACAM3 , which is expressed exclusively in neutrophils and aids in destruction of CEACAM-binding bacteria , there is compelling evidence that residues at the interface of adhesin binding are evolving rapidly in a manner consistent with positive selection ( Adrian et al . , 2019 ) . Unlike many other characterized mammalian CEACAMs , CEACAM3 appears to be a dedicated immune protein acting as a decoy receptor for CEACAM targeting bacteria ( Bonsignore et al . , 2019 ) . The evolution of epithelial CEACAMs not dedicated to immune defense and the consequences of their evolution for microbial interactions remains unclear .
To assess patterns of primate CEACAM gene evolution , we compiled sequences of human CEACAM orthologs present in publicly available genome databases ( Figure 2—source data 1A-C ) . Nineteen representative species were analyzed including four New World monkeys , ten Old World monkeys , and five hominid species . Some orthologs of human CEACAMs were not identified in a subset of primate genomes , likely due to losses or gains of specific CEACAMs along different lineages or incomplete genome assembly . With the exception of CEACAM3 , for which additional exons annotated in Old World monkeys were included ( detailed in Materials and methods ) , only genomic sequences that aligned to annotated human exons were used for subsequent phylogenetic analyses . To determine if primate CEACAMs have been subject to positive selection , protein-coding sequences were analyzed using the PAML NS sites package ( Yang , 2007 ) . This program uses a maximum likelihood framework to estimate the rate of evolution of each gene or codon , expressed as the ratio of normalized nonsynonymous ( dN ) to synonymous ( dS ) nucleotide substitutions ( dN/dS or ω ) , under different models of evolution . An excess of nonsynonymous substitutions relative to synonymous substitutions between orthologs can suggest that beneficial mutations have been repeatedly fixed by positive selection . A comparison of models that allow and disallow sites evolving under positive selection ( ω > 1 ) can determine the likelihood that a particular protein-coding sequence has been evolving under positive selection . We found that eight of the twelve primate CEACAM paralogs in our dataset possess genetic signals of positive selection ( p-value ≤ 0 . 05; Figure 2—source data 1D ) including CEACAM1 , CEACAM3 , CEACAM5 , and CEACAM6 which have previously been shown to interact with bacterial adhesins ( Gray-Owen and Blumberg , 2006 ) . In addition , we also identified elevated ω values for CEACAM7 , CEACAM8 , CEACAM18 , and CEACAM20 . To identify specific amino acid positions that contribute to signatures of positive selection , we analyzed CEACAM sequences using the Bayes empirical Bayes analysis as implemented in the PAML NS sites package , as well as the programs FUBAR and MEME from the HyPhy software package ( Figure 2—source data 1E ) . To control for the potential impact of recombination on these inferences , we used the program GARD to identify potential breakpoints in our datasets and build phylogenies of gene segments based on predicted breakpoints . We then performed phylogenetic analyses using these GARD-informed phylogenies . Our analyses collectively revealed that sites with elevated ω were concentrated in the N-domain of many CEACAM proteins ( Figure 2A; Figure 2—figure supplement 1 ) . Sites under positive selection in CEACAM18 and CEACAM20 were more dispersed throughout the protein , not localizing to a specific domain . The statistical support for positive selection of CEACAM18 and CEACAM20 in primates was also modest compared to that for other CEACAM proteins . We next sought to determine the functional impact of divergence at rapidly evolving sites in the CEACAM N-domain . Residues that contribute to protein-protein interactions have been extensively annotated for CEACAM1 , involving both host factors and bacterial adhesins . Overlaying sites under positive selection with known adhesin and host protein-binding sites ( Figure 2—source data 1F ) revealed extensive overlap between all three categories ( Figure 2B ) and demonstrates that sites with elevated ω tend to cluster on the protein-binding surface . Mapping rapidly evolving CEACAM1 residues onto a co-crystal structure of human CEACAM1 bound to the HopQ adhesin from H . pylori strain G27 ( Moonens , 2018 ) , we confirmed that multiple sites fall along the binding interface of the two proteins ( Figure 2C ) . In summary , these results demonstrate that multiple primate CEACAM orthologs exhibit signatures of repeated positive selection within the N-domain which facilitates bacterial and host protein interactions . To assess how rapid divergence of primate CEACAMs influences recognition by bacterial adhesins , we focused on CEACAM1 which is widely expressed across different cell types ( Gray-Owen and Blumberg , 2006 ) and has numerous well-documented microbial interactions ( Figure 2—source data 1F ) . Recombinant GFP-tagged CEACAM1 N-domain proteins from a panel of primate species were expressed and purified from mammalian cells ( see Materials and methods ) . Previous studies have demonstrated that the CEACAM N-domain is both necessary and sufficient to mediate interactions with bacterial adhesins ( Javaheri et al . , 2016; Kuespert et al . , 2007; Markel et al . , 2004 ) . We focused our experiments on CEACAM1 binding to two distinct classes of bacterial adhesins: HopQ encoded by H . pylori and the Opa family adhesins expressed by Neisseria species . The HopQ adhesin is a H . pylori-specific outer membrane protein that appears to be universally encoded by H . pylori strains and whose interaction with human CEACAM1 has been well characterized ( Bonsor et al . , 2018; Javaheri et al . , 2016; Königer et al . , 2016; Moonens , 2018 ) . For our assays we used the common H . pylori laboratory strains G27 ( Baltrus et al . , 2009 ) , J99 ( Alm et al . , 1999 ) , and Tx30a ( ATCC 51932 ) , which have previously been confirmed to bind human CEACAM1 ( Javaheri et al . , 2016 ) . The HopQ proteins encoded by these strains encompass the two major divisions of HopQ diversity , termed Type I and Type II ( Cao and Cover , 2002; Javaheri et al . , 2016 ) . Strains G27 and J99 both encode a single copy of a Type I HopQ adhesin , while Tx30a encodes a Type II HopQ adhesin . All strains include extensive divergence in the CEACAM1-binding region ( Bonsor et al . , 2018; Moonens , 2018 ) . Opa proteins are a highly diverse class of adhesins encoded by Neisseria species that are structurally distinct from the HopQ adhesin ( Bonsor et al . , 2018; Fox et al . , 2014; Moonens , 2018; Sadarangani et al . , 2011 ) . For our study we tested CEACAM1 binding to Opa52 and Opa74 , which despite their limited sequence identity are both known to bind human CEACAM1 . While Opa74 is only known to recognize human CEACAM1 , Opa52 also binds CEACAM3 , CEACAM5 , and CEACAM6 ( Roth et al . , 2013 ) . Because Neisseria species typically encode multiple unique phase-variable Opa variants , individual Opa genes from Neisseria gonorrhoeae were cloned and expressed heterologously in K12 Escherichia coli , which does not bind to CEACAM proteins . To assess pairwise interactions between primate CEACAMs and bacterial adhesins , we incubated recombinant CEACAM1 N-domain proteins with individual bacterial strains . Bacterial cells were washed , pelleted , and the presence of bound CEACAM1 protein was assessed by western blot . We observe that all bacterial strains tested bind to the human CEACAM1 N-domain , consistent with previous studies ( Figure 3A ) . Incubation of H . pylori strain G27 with GFP alone fails to yield detectable signal , confirming that binding is CEACAM-dependent ( Figure 3B ) . Furthermore , a Δhopq mutant of strain G27 does not exhibit significant CEACAM1 binding , consistent with previous reports that HopQ is the sole CEACAM-binding adhesin present in these strains ( Figure 3—figure supplement 1 ) . Examining non-human CEACAM1 bacterial binding , the chimpanzee CEACAM1 N-domain , which differs from the human protein at four amino acid positions , binds to all adhesin-expressing strains except Opa74 . Gorilla CEACAM1 , which differs from the human N-domain at five sites ( three non-overlapping with chimpanzee ) , is also unable to bind Opa74 but does bind H . pylori strains and Opa52 . Orangutan CEACAM1 is unable to interact with any bacterial strain tested , nor do baboon and squirrel monkey . We noted that despite the limited species divergence between bonobos and chimpanzees , bonobo CEACAM1 does not bind any of the tested bacterial strains ( Figure 3A ) . Previous studies have found the results of CEACAM-binding assays to be consistent between western blotting and by flow cytometry ( Adrian et al . , 2019; Javaheri et al . , 2016; Königer et al . , 2016; Kuespert et al . , 2007 ) . We confirmed this for our system with H . pylori strain G27 , using flow cytometry to detect specific binding of GFP-tagged CEACAMs on the bacterial cell surface ( Figure 3B ) . These results demonstrate that CEACAM1 N-domain divergence between closely related primate species , even within the great apes , determines bacterial recognition in an adhesin-specific manner . The inability of H . pylori strains or N . gonorrhoeae adhesins to bind bonobo CEACAM1 was surprising given bonobo’s close phylogenetic relationship to both humans and chimpanzees . While archaic humans are believed to have diverged from our primate relatives at least 5 million years ago , the major divergence between chimpanzees and bonobos occurred only 1–2 million years ago ( Prado-Martinez et al . , 2013 ) . Closer inspection revealed that the bonobo CEACAM1 N-domain sequence is unusually divergent from that of both humans and chimpanzees , while other regions of the coding sequence show higher degrees of identity ( Figure 4—figure supplement 1A ) . To investigate bonobo CEACAM1 evolution further , we first validated the bonobo CEACAM1 N-domain sequence present in our bonobo reference genome through comparison of assemblies and sequencing reads from multiple bonobo individuals as well as through direct Sanger sequencing of the CEACAM1 N-domain from bonobo genomic DNA ( see Materials and methods ) . Having confirmed the identity of the bonobo CEACAM1 reference sequence , we compared this gene to sequences from other hominids . Relative to its orthologs in humans and chimpanzees , bonobo CEACAM1 differs at nearly 20% of sites in the N-domain whereas humans and chimpanzees differ at only about 4% of sites . In contrast , outside of the N-domain bonobo CEACAM1 diverges from humans and chimpanzees at approximately 2% of sites , while human and chimpanzee CEACAM1 differ at around 1% of sites . We also noted that the number of divergent sites between bonobo and human in the N-domain ( 18 residues ) is nearly identical to the number of divergent sites between bonobo and chimpanzee ( 20 residues ) , despite the closer phylogenetic relationship between bonobos and chimpanzees . In fact , the divergence between the bonobo and chimpanzee CEACAM1 N-domains is greater than that between chimpanzee and the earliest diverging member of the hominid clade , orangutan ( 81% versus 83% amino acid identity , respectively ) . A comparison of N-domain sequences for CEACAM5 , another rapidly evolving CEACAM , further highlights the extreme divergence of bonobo CEACAM1 . Between human CEACAM5 and the bonobo and chimpanzee CEACAM5 sequences , there are only ten and nine amino acid changes respectively , while bonobo and chimpanzee differ at only five sites along the entire length of the N-domain ( Figure 4—figure supplement 1B ) . The degree of divergence within the N-domain of bonobo CEACAM1 suggests processes other than sequential accumulation of single nucleotide mutations could be responsible . One mechanism by which this could occur is through gene conversion , a form of homologous recombination in which genetic material from one location replaces sequence in a non-homologous location , often with substantial sequence similarity ( Chen et al . , 2007 ) . Gene conversion can provide an important source of genetic novelty and a mechanism that can accelerate adaptation ( Bittihn and Tsimring , 2017; Daugherty and Zanders , 2019; Gendreau et al . , 2021 ) . To determine if inter-locus recombination has shaped the evolution of CEACAM genes in primates , we looked for evidence of discordance between species and gene trees . Gene-species tree discordance can be an indication of multiple evolutionary processes , including a history of gene conversion between paralogs . In a maximum likelihood-based phylogeny of full-length CEACAM-coding sequences , clades containing single CEACAM paralogs were inferred with robust statistical support ( Figure 4A , Figure 4—figure supplement 2 ) . In general , the relationships between CEACAM homologs are inferred with high confidence and reflected species relationships as expected for the divergence of orthologous-coding sequences . To determine if there have been domain-specific instances of gene conversion , we constructed phylogenetic trees of individual CEACAM domains . Typically , we expect paralogs to form clearly defined clades reflecting species divergence . This is the pattern we observe for full-length CEACAM-coding sequences , indicating that overall the paralogs have remained distinct since their initial duplication and have steadily diverged between species . Specific CEACAM domain sequences generally follow this pattern ( Figure 4B , Figure 4—figure supplements 3–6 ) . However , the N-domains of CEACAM1 , CEACAM3 , CEACAM5 , and CEACAM6 deviate strikingly from this norm and form a single monophyletic group ( hereafter called CCM1356 ) , albeit one with low bootstrap support ( Figure 4B , Figure 4—figure supplements 3 and 4 ) . Within the CCM1356 clade we observe that rather than clustering by paralog , N-domains are split into subclades representing the three major primate lineages ( Figure 4C , Figure 4—figure supplement 4 ) . In general , the close phylogenetic relationship of sequences within these clades is well supported . This topology suggests that these CEACAM N-domains are more similar to paralogous domains within the same species or primate lineage than they are to their respective orthologs across species . Several well-supported nodes provide further evidence that gene conversion is driving concerted evolution within the CCM1356 clade ( Figure 4C ) . Certain pairs of N-domains , such as CEACAM3 and CEACAM1 in gorilla and CEACAM1 and CEACAM5 in orangutan , form monophyletic groups with strong bootstrap support . As these relationships are not observed for the other domains of these CEACAM proteins , this suggests conversion events affecting only the N-domains of these CEACAMs occurred in these species . New World monkeys provide the most striking phylogenetic evidence of gene conversion among primates . For each of the four New World monkey species examined , the N-domains of CEACAM1 , CEACAM5 , and CEACAM6 are all more closely related within species than to their orthologs in other species , suggesting gene conversion has independently acted on the N-domains of these three CEACAMs at least four times within this single clade ( Figure 4C ) . These findings are consistent with N-domains of CEACAMs 1 , 3 , 5 , and 6 undergoing widespread concerted evolution , likely facilitated by gene conversion . To further test for evidence of gene conversion acting on primate CEACAM family members , we applied the GARD algorithm from the HyPhy software package . GARD detects topological changes between trees inferred from segments of a gene alignment , assesses the likelihood they are consistent with recombination , and identifies potential breakpoints . Consistent with our phylogenetic examination of CEACAM homologs , GARD detects strong evidence of recombination for CEACAM1 , CEACAM3 , CEACAM5 , and CEACAM6 ( Figure 4D , Figure 2—figure supplement 1A ) . In all cases , breakpoints were identified at the C-terminus of the N-domain or in immediately adjacent IgC domains . This pattern is consistent with repeated N-domain gene conversion between CCM1356 paralogs and is also in line with our phylogenetic reconstructions of CEACAM IgC domains ( Figure 4—figure supplement 5 ) . In addition to CEACAM1 , CEACAM3 , CEACAM5 , and CEACAM6 , GARD also indicates a recombination breakpoint for CEACAM7 that would encompass the N-domain . While we do not detect discordance in our N-domain gene tree that implicates gene conversion involving CEACAM7 , there is a single instance in the IgC domain tree of a gorilla CEACAM5 IgC domain grouping more closely with homologs of the IgC domain of CEACAM7 ( Figure 4—figure supplement 5 ) . A breakpoint in this region is also consistent with CEACAMs with rapid N-domain evolution being involved in gene conversion events as well as previous analyses ( Zid and Drouin , 2013 ) . Together these results support a model in which gene conversion between rapidly diverging CEACAMs has contributed to N-domain diversification during primate evolution . Phylogenetic analyses confirm that the bonobo CEACAM1 N-domain is not closely related to other primate CEACAM1 sequences but fail to strongly support its relationship to any other single CEACAM N-domain . Reasoning that the extant bonobo CEACAM1 gene may have arisen from multiple iterative recombination events , we performed a BLAST search of genomes on the NCBI database using base pairs 103–303 of the bonobo CEACAM1 sequence ( corresponding to resides 1–67 of the N-domain ) as our query . Human and chimpanzee are roughly 86% identical to bonobo CEACAM1 in this region versus 99% identical ( a single nucleotide change ) in the remaining 120 base pairs ( Figure 4—figure supplement 1 ) . This search identifies orangutan CEACAM3 as the closest match . While the similarity between the first 120 bp of bonobo CEACAM1 and orangutan CEACAM3 is striking and the final third of the nucleotide sequence is nearly identical to human and chimpanzee CEACAM1 , other segments of bonobo CEACAM1 are still quite divergent from all other N-domain sequences ( Figure 5A ) . A BLAST search of this region in bonobo CEACAM1 ( base pairs 221–380 ) indicates the greatest similarity is with the analogous region from rhesus macaque CEACAM6 . However , the increased similarity of macaque CEACAM6 in this region compared to other CEACAMs is marginal ( Figure 5A ) . The extreme divergence of the bonobo CEACAM1 N-domain from other CEACAM1 homologs in even its closest relatives could indicate that this particular sequence has been evolving independently of other N-domain alleles for a long period of time as a result of balancing selection . This has been observed for other genes involved in host-pathogen conflicts , most notably major histocompatibility complex alleles ( Meyer et al . , 2018 ) . In this case , we might expect to identify alleles similar to bonobo CEACAM1 currently circulating in other hominid populations , and likewise alleles similar to CEACAM1 sequences observed in humans and chimpanzees may be found in the larger bonobo population . In a search of human genetic variation data available through the International Genome Sample Resource ( IGSR ) accessed through the Ensembl webserver ( https://www . ensembl . org ) , there is no evidence for any alleles with similarity to bonobo CEACAM1 circulating within human populations . Searching population data from the Great Apes Genome Project ( Prado-Martinez et al . , 2013 ) , alleles similar to bonobo CEACAM1 are not found for chimpanzees , gorillas , or orangutans . Likewise , CEACAM1 alleles similar to those found in humans and chimpanzees are not observed in any of the bonobo genomes from the same dataset . Given the information at hand , it is difficult to precisely determine the series of mutational events that produced the bonobo CEACAM1 allele or determine the likely origin point of this allele in the diversification of hominids . However , these results are consistent with multiple independent instances of gene conversion giving rise to bonobo CEACAM1 , with subsequent fixation of this haplotype in bonobo populations since their divergence from chimpanzees over the last million years . Given the large number of residue changes between human and bonobo CEACAM1 , we next sought to determine if a subset of rapidly evolving sites are sufficient to either impair or restore recognition by bacterial adhesins . To test this , we generated CEACAM1 N-domain proteins in which a subset of residues between humans and bonobos were swapped . We focused on sites that are identical in humans and chimpanzees but differ in bonobos and which exhibit high ω across primates , resulting in a total of five tested sites ( Figure 5A and B ) . Of these residues we chose to mutate adjacent amino acids 27–29 as a single group . This patch of sites is highly variable among the rapidly evolving CEACAMs , particularly CEACAM1 , CEACAM3 , and CEACAM5 ( Figure 5—figure supplement 1 ) . None of the ‘humanized’ mutants in the bonobo CEACAM1 background were sufficient to confer binding ( Figure 5C ) . In contrast , introduction of bonobo residue 44 into human CEACAM1 ( mutation Q44L ) prevents binding by H . pylori and Opa-expressing strains , while introduction of bonobo variable sites 27–29 abolishes binding to Opa74 ( Figure 5C ) . Mutation G51Q has no appreciable impact on binding by H . pylori strain G27 or Neisseria Opa74 , but blocks binding by H . pylori strain Tx30a and reduces binding to J99 and Opa52 . Collectively these results reveal that multiple single positions in human CEACAM1 exhibiting signatures of positive selection are sufficient to impair recognition by multiple bacterial adhesins . Moreover , these findings also demonstrate how instances of gene conversion between CEACAM paralogs could serve as large-effect adaptive mutations mediating microbial evasion . Pervasive evidence of positive selection acting on CEACAMs in primates raises the question as to whether CEACAM variants that evade pathogen recognition are currently segregating in human populations . To explore the existence of human CEACAM variants and their consequences for bacterial interactions , we queried human single nucleotide polymorphism ( SNP ) and haplotype data for rapidly evolving CEACAM paralogs available from the IGSR accessed through the Ensembl genome browser ( see Materials and methods ) . We found that variation in the N-domains of CEACAM6 , CEACAM7 , and CEACAM8 predominantly consists of polymorphisms not shared with other CEACAM proteins and found on isolated haplotypes . In contrast , CEACAM1 , CEACAM3 , and CEACAM5 N-domain variation is composed primarily of extended haplotypes ( Figure 6—figure supplements 1–4 ) . Furthermore , these extended haplotypes increase similarity between CEACAM1 , CEACAM3 , and CEACAM5 , consistent with possible gene conversion events . Indeed , some haplotypes not only have changes at nonsynonymous sites that increase similarity with these CEACAMs , but also include shared synonymous changes . These observations suggest that gene conversion among CEACAMs has occurred relatively recently and may be ongoing in human populations . A search of polymorphisms for CEACAM1 in human populations reveals three high-frequency nonsynonymous variants within the N-domain: Q1K ( rs8111171 ) , A49V ( rs8110904 ) , and Q89H ( rs8111468 ) ( Figure 6A ) . The haplotype containing all three alternative alleles is the most frequent non-reference CEACAM1 haplotype annotated , occurring in 14% of the human population overall and in up to 43% of individuals in African populations ( Figure 6A ) . In total , nearly 17% of all sequenced individuals carry at least one of these high-frequency SNPs ( Figure 6—figure supplement 2 ) . Of the three variants , A49V and Q89H both lie within regions of CEACAM1 known to interact with bacterial adhesins suggesting they may alter bacterial adherence ( Figure 6B ) . To determine if these high-frequency CEACAM1 polymorphisms affect bacterial recognition , we generated recombinant CEACAM1 N-domain variant proteins for use in our adhesin-binding assays . None of the variants are able to abolish CEACAM1 binding to our panel of H . pylori strains ( Figure 6C ) . In contrast , Neisseria Opa-expressing strains exhibit highly variable recognition of multiple human CEACAM1 variants . The Q1K mutation alone has no impact on binding , while A49V abolishes recognition by Opa74 , and variant Q89H abrogates binding to both Opa52 and Opa74 ( Figure 6C ) . Combinatorial CEACAM1 variants reveal that these mutations behave in a dominant manner , with Q89H dominant over A49V ( Figure 6C ) . Together these results demonstrate that high-frequency human polymorphisms in CEACAM1 are sufficient to impair binding by specific classes of bacterial adhesins present in human pathogens . These findings further suggest that high-frequency CEACAM variants could alter human colonization or infection by pathogenic Neisseria , including causative agents of gonorrhea and meningitis .
Our investigation of species-specific bacterial adherence to CEACAM1 revealed an unforeseen example of extreme genetic divergence within the great apes . The bonobo CEACAM1 gene could represent a rapid succession of single residue changes combined with multiple recombination events arising in bonobos under strong selection and/or a population bottleneck . Alternatively , this allele may be ancient and have been subject to balancing selection or incomplete lineage sorting in ancestral hominid populations . We also considered that the source of the bonobo CEACAM1 sequence may not be from functional CEACAM genes , but a pseudogenized CEACAM sequence or a pregnancy-specific glycoprotein , a family of proteins closely related to CEACAMs . However , a BLAST search of relevant NCBI databases ( see Materials and methods ) fails to identify any new genomic regions in bonobos or other primates with greater sequence identity than what had already been found . While there are multiple possible explanations for the highly divergent nature of bonobo CEACAM1 , absent further evidence the origin of this particular allele remains obscure . What is clear from the example of bonobo CEACAM1 , however , is the extent to which gene conversion can rapidly generate diversity between closely related species and the impact of such variation on interactions with microbes . During the course of investigating the origin of the bonobo CEACAM1 sequence , we discovered evidence that gene conversion has shaped the evolution of many CEACAMs across primates , primarily impacting homologous proteins targeted by bacteria . While we identify several instances of likely gene conversion , results from phylogenetic analyses probably represent an underestimate of the true number of recombination events that have occurred among rapidly evolving CEACAMs in primates . Repeated episodes of gene conversion can obscure past instances of recombination and hinder their identification by gene-species tree discordance . GARD analyses and recombination detection programs in general also tend to miss many recombination events ( Bay and Bielawski , 2011; Kosakovsky Pond et al . , 2006 ) . One particularly interesting example in orangutans implicates multiple conversion events impacting CEACAM1 , CEACAM5 , and CEACAM8 ( Figure 5—figure supplement 1 ) . Phylogenetic analyses indicate a species-specific conversion event between CEACAM1 and CEACAM5 in orangutans . Prior to the CEACAM1-CEACAM5 conversion , however , residues 29–64 in either CEACAM1 or CEACAM5 were likely replaced by the homologous sequence from CEACAM8 . Evidence for this event includes not only multiple nonsynonymous substitutions shared with orangutan CEACAM8 , but a shared synonymous substitution in both orangutan CEACAM1 and CEACAM5 only observed in hominid CEACAM8 homologs . Despite this evidence , neither our phylogenetic analyses nor GARD analyses suggest CEACAM8 has been involved in gene conversion . Like CEACAM7 , the involvement of CEACAM8 in intra-paralog gene conversion is consistent with CEACAMs with rapid N-domain evolution participating in gene conversion events . Overall , the rapid shuffling of genetic variation among CEACAM genes that we observe could greatly augment the potential for host adaptation in the face of microbial antagonism . It has been suggested that gene conversion between CEACAM paralogs preserves the ability of CEACAM3 to effectively mimic bacterially antagonized CEACAMs and thereby maintain its function as a decoy receptor ( Zid and Drouin , 2013; Zimmermann , 2019 ) . Indeed , our results and those of Adrian et al . , 2019 , support the importance of maintaining the similarity of CEACAM3 to other adhesin-binding CEACAMs in apes and Old World monkeys . However , our findings suggest that gene conversion does not only serve to maintain CEACAM3’s mimicry function . There is no evidence that New World monkeys encode a CEACAM3 homolog , yet within this group gene conversion appears to be rampant between CEACAM1 , CEACAM5 , and CEACAM6 ( Figure 4C ) . Additionally , we observe multiple conversion events in hominids that do not involve CEACAM3 ( Figure 4C , Figure 5—figure supplement 1 , and Figure 6—figure supplement 4 ) . Though we cannot rule out the possibility that New World monkey CEACAM1 , CEACAM5 , or CEACAM6 may have evolved a mimicry function similar to CEACAM3 , our observations in hominids suggest additional drivers for pervasive gene conversion within the CCM1356 group . High sequence similarity and close chromosomal proximity , such as that seen for the N-domains of the CCM1356 group ( Figure 2—figure supplement 1B , Figure 4—figure supplement 3 ) , could result in periodic interconversion of paralogous sequences , which in many cases may be functionally neutral ( Bittihn and Tsimring , 2017; Zid and Drouin , 2013 ) . However , we show that the common CEACAM1 variant , Q89H , which matches CEACAM5 at this position , abolishes binding to at least two Opa proteins ( Figure 6C ) . Human CEACAM1 variant A49V , which reduces binding to Opa74 , also matches the equivalent residue in CEACAM5 as well as CEACAM3 . These examples illustrate the potential benefit of gene conversion between epithelial CEACAM paralogs beyond CEACAM3 mimicry . Given that gene conversion within this family appears to be enriched specifically among CEACAMs recognized by bacteria , we propose that gene conversion among epithelial CEACAMs reflects a general mechanism of pathogen evasion ( Figure 7 ) . Such a mechanism allows beneficial sets of mutations to spread , whether between decoys and targets or among antagonized epithelial CEACAMs , more rapidly than conversion of residues through independent mutational events ( Bittihn and Tsimring , 2017 ) . In addition , recombination provides decoys the ability to gain binding to CEACAM antagonists through exchanges from epithelial CEACAMs as observed for CEACAM3 ( Adrian et al . , 2019 ) . Finally , the interchangeability of CEACAM domains could further accelerate the exchange and spread of beneficial mutations in host genomes ( Bittihn and Tsimring , 2017 ) . In these ways gene conversion provides an important mechanism by which the host can keep pace with rapidly evolving pathogenic microbes . Our study examined the influence of CEACAM1 diversity on bacterial binding using adhesins from H . pylori and N . gonorrhoeae as models . However , a diverse range of pathogenic and commensal microbes are known to bind host CEACAMs , and it is unlikely that natural selection across primates has been driven by any single pathogen . What is clear from these results is the major influence that CEACAM divergence and polymorphisms can have on recognition by structurally unrelated bacterial adhesins , and moreover the impact that such diversity could have on shaping future disease outbreaks . It is also notable that recombination between paralogs contributes to the immense diversity of Opa genes Neisseria ( Sadarangani et al . , 2011 ) . In this regard CEACAM-Opa interactions could reflect a general mechanism by which gene conversion between paralogs contributes to reciprocal adaptation in both hosts and microbes . In addition to exploring the role of CEACAM gene conversion among primates , we provide evidence that this process continues to shape CEACAM diversity within human populations . The three human CEACAM1 variants we test in our adhesin-binding assay are part of a group of related CEACAM1 haplotypes that increase sequence similarity to CEACAM3 and/or CEACAM5 ( Figure 6—figure supplement 1 ) . Extended haplotypes that increase similarly to CEACAM1 at both synonymous and nonsynonymous positions in the N-domain are also found for CEACAM3 and CEACAM5 in humans ( Figure 6—figure supplement 3 and Figure 6—figure supplement 4 ) . Indeed , haplotypes consisting of variants of putative recombination events are the most common non-reference alleles for CEACAM1 , CEACAM3 , and CEACAM5 ( Figure 6 , Figure 6—figure supplements 1–4 ) . Variant sites in these proteins tend to lie along the protein-binding interface of the N-domain and often impact residues known to influence adhesin recognition . The relationships between these different CEACAM haplotypes appears to be complex , as many different combinations of partial variant haplotypes exist for each CEACAM paralog . The haplotype structures we observe suggest these CEACAM variants are the result of one or more conversion events between paralogous sequences , likely followed by further recombination with the major CEACAM allele . Important questions remain regarding the rapid evolution of a subset of primate CEACAM proteins . Among these questions is why CEACAM7 and CEACAM8 show similar patterns of evolution to bacterially antagonized CEACAMs despite no known instances of bacterial antagonism . The simplest explanation is that CEACAM7 and CEACAM8 are themselves the targets of as yet unidentified pathogen antagonists ( Sintsova et al . , 2015 ) . Alternatively , their rapid evolution may reflect pressure to maintain binding with rapidly evolving bacterially antagonized CEACAMs ( Gray-Owen and Blumberg , 2006; Skubitz and Skubitz , 2008 ) could merely be a result of their genomic proximity to rapidly evolving CEACAMs prone to gene conversion ( Zid and Drouin , 2013 ) or could be the result of some as yet unknown evolutionary pressures . Another intriguing aspect of rapid CEACAM evolution is the impact rapid divergence might have beyond interactions with pathogenic microbes . Given the extensive overlap of CEACAM-binding sites among unrelated bacterial adhesins , the ramifications of rapid CEACAM evolution likely extend beyond the adhesins of pathogens to those of commensal and beneficial microbes as well . For commensal microbes that rely on these interaction surfaces , CEACAM evolution could significantly alter their ability to colonize the host . The impact of CEACAM divergence on composition of the host microbiome and/or the evolution of commensal strains warrants further investigation . Studies of other ‘housekeeping’ proteins targeted by pathogens have found that sites under positive selection typically do not overlap with sites involved in essential host functions ( Barber and Elde , 2014; Demogines et al . , 2013 ) . This is clearly not the case for CEACAMs , where we observed extensive overlap between sites involved in host protein interactions , sites targeted by bacterial adhesins , and sites undergoing rapid evolution ( Figure 2B ) . How CEACAMs are able to rapidly evolve while maintaining their other essential host protein interactions remains a mystery . Future studies on CEACAM protein functions , interaction networks , and pathogen antagonism will likely clarify these outstanding questions regarding rapidly evolving CEACAMs . Collectively our study provides evidence that repeated adaptation among primate CEACAMs has shaped host-specific cell adherence by diverse pathogenic bacteria . We find that over half of the CEACAM paralogs found in humans display signatures of positive selection across the primate lineage , localized primarily to the extracellular N-domain . We further discovered that rapid evolution of CEACAM N-domains has been facilitated by extensive ‘shuffling’ of sequences between a subset of CEACAM paralogs likely through repeated gene conversion . The diversification of primate CEACAM N-domain sequences has likely had significant consequences for interactions between primates and bacteria . Consistent with observations across other primate species , we also provide evidence that gene conversion events may impact bacterial pathogen recognition of CEACAMs in contemporary human populations . Together this work reveals how dynamic evolutionary processes have shaped bacterial-host cell associations with consequences for infectious disease susceptibility . | Trillions of bacteria live in and on the human body . Most of them are harmless but some can cause serious infections . To grow in or on the body , bacteria often attach to proteins on the surface of cells that make up the lining of tissues like the gut or the throat . In some cases , bacteria use these proteins to invade the cells causing an infection . Genetic mutations in the genes encoding these proteins that protect against infection are more likely to be passed on to future generations . This may lead to rapid spread of these beneficial genes in a population . A family of proteins called CEACAMs are frequent targets of infection-causing bacteria . These proteins have been shown to play a role in cancer progression . But they also play many helpful roles in the body , including helping transmit messages between cells , aiding cell growth , and helping the immune system recognize pathogens . Scientists are not sure if these multi-tasking CEACAM proteins can evolve to evade bacteria without affecting their other roles . Baker et al . show that CEACAM proteins targeted by bacteria have undergone rapid evolution in primates . In the experiments , human genes encoding CEACAMs were compared with equivalent genes from 19 different primates . Baker et al . found the changes in human and primate CEACAMs often occur through a process called gene conversion . Gene conversion occurs when DNA sections are copied and pasted from one gene to another . Using laboratory experiments , they showed that some of these changes enabled CEACAM proteins to prevent certain harmful bacteria from binding . The experiments suggest that some versions of CEACAM genes may protect humans or other primates against bacterial infections . Studies in natural populations are needed to test if this is the case . Learning more about how CEACAM proteins evolve and what they do may help scientists better understand the role they play in cancer and help improve cancer care . Studying CEACAM evolution may also help scientists understand how bacteria and other pathogens drive protein evolution in the body . | [
"Abstract",
"Introduction",
"Results",
"Discussion"
] | [
"evolutionary",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2022 | Evolution of host-microbe cell adherence by receptor domain shuffling |
The prediction of the structures of proteins without detectable sequence similarity to any protein of known structure remains an outstanding scientific challenge . Here we report significant progress in this area . We first describe de novo blind structure predictions of unprecendented accuracy we made for two proteins in large families in the recent CASP11 blind test of protein structure prediction methods by incorporating residue–residue co-evolution information in the Rosetta structure prediction program . We then describe the use of this method to generate structure models for 58 of the 121 large protein families in prokaryotes for which three-dimensional structures are not available . These models , which are posted online for public access , provide structural information for the over 400 , 000 proteins belonging to the 58 families and suggest hypotheses about mechanism for the subset for which the function is known , and hypotheses about function for the remainder .
Despite substantial efforts over decades , high-resolution structure prediction is currently limited to proteins that have homologs of known structure , or small proteins where thorough sampling of the conformational space is possible ( <100 residues; even in this case , predictions can be very inaccurate ) . For roughly 41% of protein families , there is currently no member with known structure ( Kamisetty et al . , 2013 ) . While high-resolution ab initio structure prediction has remained a challenge , considerable success has been achieved in generating high-accuracy models when sparse experimental data are available to constrain the space of conformations to be sampled . This additional information , in combination with a reasonably accurate energy function , has enabled the determination of high-resolution structures for much larger proteins ( Raman et al . , 2009; DiMaio et al . , 2011; Lange et al . , 2012 ) . Recent work has shown that residue–residue contacts can be accurately inferred from co-evolution patterns in sequences of related proteins ( Marks et al . , 2011; Morcos et al . , 2011; Hopf et al . , 2012; Jones et al . , 2012; Marks et al . , 2012; Nugent and Jones , 2012; Sułkowska et al . , 2012; Kamisetty et al . , 2013 ) . While early approaches estimated these restraints by inverting a covariance matrix ( Marks et al . , 2011; Morcos et al . , 2011; Jones et al . , 2012 ) , subsequent studies have shown that a pseudo-likelihood ( PLM ) -based approach ( Balakrishnan et al . , 2011 ) results in more accurate predictions ( Ekeberg et al . , 2013; Kamisetty et al . , 2013 ) . Distance restraints derived from such predictions have been used to model a wide range of unknown protein structures ( Hayat et al . , 2014; Wickles et al . , 2014; Abriata , 2015; Antala et al . , 2015; Hopf et al . , 2015; Tian et al . , 2015 ) and protein–protein complexes ( Ovchinnikov et al . , 2014; Hopf et al . , 2014 ) . However , while the generated structures often recapitulate the fold of the target protein , it has not been clear whether such methods can yield high-accuracy models of complex protein structures .
In the recent CASP11 ( 11th critical assessment of techniques for protein structure prediction ) blind test of protein structure prediction methods , we predicted the structures of proteins from large families with no representatives of known structure by integrating co-evolution derived contact information from GREMLIN ( Kamisetty et al . , 2013 ) into the Rosetta structure prediction methodology ( Simons et al . , 1999; Rohl et al . , 2004; Raman et al . , 2009 ) . Starting from an extended polypeptide chain , Monte Carlo + Minimization searches through conformations with local structure consistent with the local sequence were carried out , optimizing first a low-resolution energy function favoring hydrophobic burial and backbone hydrogen bonding , and second a detailed all atom energy function describing hydrogen bonding and electrostatic interactions , van der Waals interactions , and solvation ( Das and Baker , 2008 ) . In the first phase , sampling was carried out in internal coordinates ( the backbone torsion angles ) , and hence , to avoid loss of sampling efficiency by early formation of contacts between residues distant along the sequence , predicted contact information was first added for residues close along the chain and subsequently for residues with increasing sequence separation . The contact information was implemented through residue–residue distance restraints whose strength and shape were functions of the strength of the evolutionary covariance between the residues ( see ‘Materials and methods’ ) . Large numbers of independent trajectories were carried out using the Rosetta@Home distributed computing project , and the lowest energy ( Rosetta all atom energy plus evolutionary restraint fit ) models were recombined and further optimized using a new iterative version ( see ‘Materials and methods’ ) of the RosettaCM hybridization protocol ( Song et al . , 2013; Kim et al . , 2014 ) . The five lowest energy structures were submitted as predictions to the CASP organizers . When several months later the actual structures of these proteins were revealed , the predictions were found to be considerably more accurate than any previous predictions made in the 20 years of CASP experiments for proteins over 100 amino acids that lack homologs of known structure . Two particularly striking examples are shown in Figure 1; the prediction for the complex 256 residue structure of T0806 is 3 . 6 Cα-RMSD from the crystal structure ( 2 . 9 Å over 223 residues ) , and the prediction for the 108 residue T0824 is 4 . 2 Cα-RMSD from the crystal structure ( 2 . 7 Å over 77 residues ) . The models accurately recapitulate the complex topologies of the proteins . Due to time restraints , the calculations could not be run to convergence during CASP; with additional sampling , the lowest energy model for T0806 has an RMSD of 2 . 1 Å over 245 residues to the experimentally determined structure . Both the co-evolution derived contacts and the new iterative hybridization protocol were critical to obtaining higher accuracy models: Rosetta calculations without constraints failed to converge ( data not shown ) , and the ROBETTA server models generated without the hybridization step were considerably less accurate ( 11 . 6 vs 3 . 6 Cα-RMSD for T0806 and 14 . 0 vs 4 . 2 Cα-RMSD for T0824 ) . 10 . 7554/eLife . 09248 . 003Figure 1 . Accurate blind structure prediction of CASP11 targets T0806 and T0824 . ( A ) Location of the most strongly co-evolving residue pairs . Lines connect residue pairs with normalized coupling strength greater than 1 . 0; yellow , distance less than 5 Å; orange , distance less than 10 Å and red , greater than 10 Å in the models . ( B ) CASP11 submitted models , colored from N to C terminus ( blue to red ) . ( C ) X-ray crystal structures . For T0806 , the Cα RMSD over the full-length protein is 3 . 6 Å and 2 . 9 Å over 223 aligned residues . For T0824; the Cα RMSD over the full-length protein is 4 . 2 Å and 2 . 7 Å over 77 aligned residues . For statistics on all five models submitted during CASP , see Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09248 . 00310 . 7554/eLife . 09248 . 032Figure 1—source data 1 . The Cκ-RMSD and GDT-TS calculations are over the full-length sequence . The total GREMLIN score for the model is reported . The most accurate models have the best GREMLIN score . DOI: http://dx . doi . org/10 . 7554/eLife . 09248 . 032 Having found that protein structures can accurately be modeled using co-evolution information , we set out to build models for representatives of large protein families in bacteria with no detectable structural homologs . To facilitate evaluation of such models , we developed a length-independent measure of the fit between a set of predicted contacts and a model: the ratio of the total GREMLIN score of the model to the score expected if it were the native structure ( Rc , see ‘Materials and methods’ ) . We chose to focus on families with at least 4× ( protein length ) sequences to ensure that the predicted contacts have high accuracy ( Kamisetty et al . , 2013; at 4L sequences , the top 1 . 5L contacts are on average 50% correct ) . Families with detectable structure homologs were excluded using a sensitive sequence search method ( HHsearch [Söding , 2005] ) . For computational efficiency , an initial scan was done using a single sequence , excluding families where the top hit had an e-value of 1 or greater to any protein of known structure . We identified 100 families satisfying these criteria in Escherichia coli ( Gram-negative , Proteobacteria ) , and an additional 22 , 5 , and 4 families in Bacillus subtilis ( Gram-positive , Firmicutes bacterium ) , Halobacterium salinarum ( Euryarchaeota ) , and Sulfolobus solfataricus ( Crenarchaeota ) , respectively ( Supplementary file 1 ) . For each of these top families , we carried out a more sensitive profile–profile sequence search against the Protein Data Bank ( PDB ) using HHsearch ( Söding , 2005 ) and the fold recognition method SPARKS-X ( Yang et al . , 2011 ) . We eliminated families if the top HHsearch hit had an E-value less than 1E-04 and was consistent with GREMLIN contacts . An alternative approach to structure modeling using predicted contacts is to search for weak fold recognition matches to known protein structures and determine if any of the hits fit the predicted contacts . This approach is not very effective for the families identified as described above; for only 4 of the 122 families with HHsearch E-values greater than 1E-04 did one of the top ten hits from HHsearch or SPARKS-X match the predicted contacts ( have Rc values greater than 0 . 6 ) . Many of the families we identified with no homologs of known structure are transmembrane ( TM ) proteins . To evaluate the accuracy of our co-evolution-based structure prediction method on TM proteins , we tested it on a benchmark of 13 TM proteins with recently determined structures . Rather than evaluating the lowest energy five models as in the case of the CASP experiment , we instead selected the most central ( see ‘Materials and methods’ ) low-energy model and eliminated positions not converged within the lowest energy models or not constrained by contact information ( see ‘Materials and methods’ ) . As shown in Table 1 , for the 11 of the 13 proteins for which the structure prediction calculations converged , the RMSD of the predicted structure to the experimentally determined structure over the converged and constrained residues was below 4 . 0 Å ( the RMSDs over the structurally aligned regions were all below 2 . 8 Å ) . Features such as kinked , discontinuous , and re-entrant helices as well as coils within the bilayer that complicate approaches to membrane protein structure prediction that assume the accuracy of a TM helix prediction were all recovered correctly ( for example , the re-entrant helices of aquaporin; the power of fragment-based approaches to model such features was noted in Nugent and Jones , 2012 ) . 10 . 7554/eLife . 09248 . 006Table 1 . Transmembrane protein benchmarkDOI: http://dx . doi . org/10 . 7554/eLife . 09248 . 006PDBNameSeq/lenFull proteinConvergedAlignedrmsdLengthrmsdLengthrmsdLength4HE8_H ( 3 . 3 ) NADH-quinone oxidoreductase subunit 817 . 34 . 92692 . 11832 . 22341SOR_A ( N/A ) Aquaporin-026 . 22 . 72212 . 11882 . 02004Q2E_A ( 3 . 4 ) Phosphatidate cytidylyltransferase18 . 65 . 42623 . 51762 . 81784HTT_A ( 6 . 8 ) Sec-independent protein translocase protein14 . 63 . 92251 . 81242 . 41814P6V_E ( 3 . 5 ) Na ( + ) -translocating NADH-quinone reductase subunit D14 . 35 . 01941 . 4492 . 81554J72_A ( 3 . 3 ) Phospho-N-acetylmuramoyl-pentapeptide-transferase19 . 96 . 63233 . 12512 . 42373V5U_A ( 1 . 9 ) Sodium/Calcium exchanger10 . 23 . 92973 . 72842 . 32454PGS_A ( 2 . 5 ) Uncharacterized protein YetJ15 . 43 . 52072 . 71752 . 21834QTN_A ( 2 . 8 ) Vitamin B3 transporter PnuC9 . 04 . 22023 . 01552 . 81784OD4_A ( 3 . 3 ) 4-hydroxybenzoate octaprenyltransferase22 . 83 . 92753 . 42422 . 82314O6M_A ( 1 . 9 ) CDP-alcohol phosphotransferase13 . 34 . 11884 . 01652 . 31594WD8_A ( 2 . 3 ) Bestrophin domain protein5 . 94N/A268Not converged4F35_A ( 3 . 2 ) Transporter , NadC family14 . 5N/A434Not convergedColumn 1 , PDB code ( resolution of the crystal structure ) ; column 2 , protein name; column 3 , sequences per length , after filtering to reduce the redundancy to 90%; column 4 , RMSD of predicted structure to native structure; column 5 , length of native structure modeled; column 6 , RMSD over converged and constrained region; column 7 , length of converged and constrained region; column 8 , RMSD over TM-align structural alignment; column 9 , length of structurally aligned region . We built models for representatives of the 121 families with unknown structures using the Rosetta co-evolution-guided structure prediction protocol , eliminating from the lowest energy structures the non-converged and non-constrained residues . The calculations converged for 58 of the 121 proteins ( Table 2 ) . Four targets had Rc values less than 0 . 7; these targets contain clusters of contacts that may be involved in homo-oligomeric formation . The models are very different from those generated using traditional profile search and threading methods: with the exception of five targets with TMscore of 0 . 5 ( Table 2 , columns 7–8 ) , the structural similarity of the Rosetta models to the top ranked models generated by HHsearch/SPARK-X is very low . The intractability of modeling these families using profile–profile/fold recognition methods is reiterated by the very low similarity between the models that best fit the contacts produced by HHsearch and SPARKS-X ( Table 2 , column 9; Supplementary file 2 ) . 10 . 7554/eLife . 09248 . 007Table 2 . Comparison of fold recognition and Rosetta models for large protein familiesDOI: http://dx . doi . org/10 . 7554/eLife . 09248 . 007Known functionRcTMscoreName#seqEvHHSPMM_HHM_SPHH_SPWECH: O-acetyltransferase ( YiaH ) 24 , 750−2 . 40 . 00 . 10 . 90 . 10 . 20 . 1SATP: Succinate-acetateproton symporter ( YaaH ) 2298−2 . 10 . 40 . 51 . 10 . 30 . 30 . 8LSPA: Lipoprotein signal peptidase8156−2 . 00 . 20 . 11 . 00 . 20 . 30 . 3YADH: ABC-type multidrug transport permease42 , 626−2 . 00 . 10 . 10 . 70 . 30 . 20 . 2YEBZ: Putative copper export protein4067−2 . 00 . 10 . 10 . 80 . 20 . 30 . 2CRCB: Fluoride ion exporter7829−1 . 80 . 20 . 31 . 00 . 20 . 20 . 3LPTG: Lipopolysaccharide export system permease8101−1 . 80 . 00 . 10 . 90 . 10 . 10 . 2FTSW: Lipid II flippase14 , 900−1 . 70 . 00 . 11 . 00 . 10 . 20 . 2RFAL: O-antigen ligase13 , 535−1 . 70 . 20 . 10 . 90 . 30 . 20 . 2CCMB: Heme exporter protein B2433−1 . 60 . 10 . 10 . 70 . 20 . 20 . 2MLAE: ABC transporter permease for lipid asymmetry7662−1 . 40 . 00 . 10 . 90 . 10 . 20 . 3SULP: Sulfate permease6647−1 . 20 . 10 . 00 . 80 . 20 . 20 . 2TOLQ: Biopolymer transport protein9256−1 . 20 . 10 . 10 . 70 . 20 . 20 . 2LGT: Prolipoprotein diacylglyceryl transferase8121−1 . 10 . 10 . 21 . 00 . 20 . 30 . 3Q97UR7: N-methylhydantoinase B ( HyuB-3 ) 4491−1 . 00 . 10 . 11 . 10 . 10 . 10 . 1YGAZ: putative L-valine exporter6435−1 . 00 . 10 . 20 . 90 . 20 . 30 . 2CCMC: Heme exporter protein C5965−0 . 80 . 10 . 11 . 10 . 20 . 20 . 2YEDZ: Sulfoxide reductase heme-binding subunit2247−0 . 70 . 20 . 21 . 00 . 20 . 30 . 3YIAM: TRAP transporter small permease protein10 , 715−0 . 70 . 10 . 21 . 10 . 30 . 30 . 2TTDA: Tartrate dehydratase , alpha subunit4238−0 . 60 . 00 . 11 . 20 . 10 . 10 . 1UPPP: Undecaprenyl pyrophosphate phosphatase7842−0 . 60 . 00 . 11 . 00 . 20 . 20 . 2PLSY: Probable glycerol-3-phosphate acyltransferase6112−0 . 40 . 10 . 21 . 10 . 20 . 40 . 2FLIL: Flagellar protein2690−0 . 30 . 70 . 50 . 80 . 50 . 40 . 9CYDB: Cytochrome bd oxidase 268640 . 00 . 10 . 11 . 00 . 20 . 20 . 1CYDA: Cytochrome bd oxidase 162000 . 10 . 00 . 11 . 20 . 10 . 20 . 2MOTA: Motility protein A , flagellar motor proton conductor47340 . 30 . 10 . 10 . 90 . 10 . 10 . 2SLYB: Outer membrane lipoprotein18600 . 30 . 10 . 20 . 80 . 20 . 20 . 1MRED: Rod shape-determining protein15460 . 60 . 50 . 50 . 80 . 50 . 40 . 6ZUPT: Zinc transporter10 , 5170 . 60 . 10 . 10 . 80 . 20 . 10 . 2YOHK: Putative effector of murein hydrolase LrgB39412 . 30 . 20 . 10 . 90 . 40 . 20 . 2PRSW: Membrane proteinase25005 . 30 . 20 . 20 . 90 . 30 . 30 . 7DDG: Lipid A biosynthesis palmitoleoyl acyltransferase94305 . 80 . 40 . 11 . 00 . 40 . 20 . 2Unknown functionRcTMscoreName#seqEvHHSPMM_HHM_SPHH_SPYQFA: UPF0073 inner membrane protein7596−2 . 60 . 10 . 41 . 10 . 20 . 50 . 3YCED: Uncharacterized protein1604−2 . 50 . 10 . 20 . 90 . 20 . 20 . 2YPHA: Inner membrane protein2986−2 . 20 . 10 . 41 . 00 . 20 . 30 . 2YADS: UPF0126 inner membrane protein5222−1 . 90 . 10 . 10 . 90 . 20 . 30 . 2YHHN: Uncharacterized membrane protein2529−1 . 90 . 10 . 20 . 90 . 20 . 30 . 2YIDH: Inner membrane protein1041−1 . 90 . 10 . 20 . 60 . 30 . 30 . 2YITE: UPF0750 membrane protein8326−1 . 70 . 10 . 10 . 90 . 20 . 30 . 3HDED: Acid resistance membrane protein2885−0 . 60 . 10 . 20 . 80 . 20 . 20 . 2YFIP: DTW domain-containing protein3100−1 . 50 . 20 . 20 . 90 . 20 . 20 . 1YPJD: ABC-type uncharacterized permease6180−1 . 40 . 20 . 20 . 90 . 20 . 30 . 2YJFL: UPF0719 inner membrane protein1581−1 . 30 . 10 . 10 . 70 . 20 . 30 . 3YTEJ: Uncharacterized membrane protein5733−1 . 20 . 10 . 11 . 00 . 20 . 20 . 2YIHY: UPF0761 membrane protein10 , 144−0 . 90 . 10 . 10 . 90 . 10 . 20 . 2YQAA: Inner membrane protein2187−0 . 90 . 10 . 31 . 00 . 20 . 40 . 3YHID: Uncharacterized protein4416−0 . 70 . 20 . 21 . 00 . 20 . 10 . 2YLOU: Uncharacterized protein3738−0 . 70 . 40 . 50 . 90 . 30 . 30 . 8YGDD: UPF0382 inner membrane protein3025−0 . 60 . 50 . 31 . 00 . 30 . 20 . 4YJCH: Inner membrane protein1307−0 . 50 . 30 . 20 . 80 . 40 . 20 . 2YFCA: UPF0721 transmembrane protein18 , 8460 . 00 . 10 . 11 . 00 . 20 . 30 . 2YOHJ: Putative effector of murein hydrolase36080 . 40 . 20 . 30 . 50 . 30 . 40 . 6YHHQ: Inner membrane protein33980 . 70 . 40 . 21 . 00 . 40 . 30 . 2YAII: UPF0178 protein31440 . 80 . 60 . 71 . 10 . 50 . 50 . 4YUXK: Predicted thiol-disulfide oxidoreductase18811 . 30 . 30 . 31 . 10 . 30 . 30 . 5YICC: UPF0701 protein42931 . 50 . 10 . 11 . 00 . 10 . 10 . 1YEIH: UPF0324 inner membrane protein48634 . 20 . 30 . 20 . 90 . 40 . 50 . 7RARD: Putative chloramphenical resistance permease74 , 5076 . 30 . 10 . 11 . 00 . 30 . 30 . 2Column 2: number of unique proteins in family; Column 3: negative log10 of E-value of top match found in HHsearch profile–profile search of PDB; Columns 4–6: fit to predicted contacts ( Rc value ) of best fitting of top 10 HHsearch hits ( column 4 ) , of best fitting of top 10 SPARKS-X hits ( column 5 ) , and Rosetta model ( column 6 ) . Native structures have Rc values ranging from 0 . 7 to 1 . 2 ( Figure 17 ) . Columns 7–9: structural similarity ( TMscore ) between Rosetta model ( M ) and best fitting HHsearch model , between Rosetta model and best fitting SPARKS-X model , and between best fitting HHsearch and SPARKS-X models . The Rosetta models fit the contacts as well as expected for native structures and are very different from best fitting HHsearch and SPARKS-X models . For RARD and YEIH , the HHsearch E-value is less than 1E-04 , the recommended threshold for inclusion in the same Pfam clan ( Xu and Dunbrack , 2012 ) , but the fit with the co-evolutionary contacts was very poor ( Rc < 0 . 3 ) ; these two cases are discussed in sections below . For FLIL and YAII , the Rc values for very weak HHSearch and SPARKS-X hits ( E-values worse than 0 . 1 ) are greater than 0 . 6 but the contacts constrain only a portion of the structure . Based on the benchmark , we expect that our monomeric protein models should be within 4 . 0 Å RMSD of the actual structure . Provided there are not large conformational changes upon docking , protein–protein complexes can be accurately assembled from crystal structures or comparative models of the constituent monomers using GREMLIN contact predictions ( Ovchinnikov et al . , 2014 ) . Thus , the models of complexes we provide in this article are likely to be fairly accurate if the monomeric subunits are predicted accurately , but there is clearly more room for error in our more complex multi-subunit predictions . The models are available at ( External Database: http://gremlin . bakerlab . org/structures/ ) . The biological implications of all of these structures cannot be explored in a single paper; here , we describe functional insights obtained from a subset of the models . These insights derive in part from the distribution of evolutionarily conserved residues in the models , as conserved sequence motifs tend to mark functional sites in structures ( Zuckerkandl and Pauling , 1965; Villar and Kauvar , 1994; Pei and Grishin , 2001; Muth et al . , 2012 ) . As is evident in Figure 2 , the conserved residues cluster quite strongly in the predicted structures . We describe first , hypotheses on mechanism for proteins of known function , and second , hypotheses on function for proteins with currently unknown function . In the following sections , the predictions are grouped by known biological functions assigned by Clusters of Orthologous Groups ( Galperin et al . , 2015 ) . Hopf et al . ( 2012 ) also used co-evolution information to guide membrane protein structure prediction and function assignment; we compare to their conclusions in the two cases common to both studies . 10 . 7554/eLife . 09248 . 008Figure 2 . Conserved residues tend to cluster in the predicted structures . Residue conservations from multiple sequence alignments were mapped to predicted structures using Al2Co ( Pei and Grishin , 2001 ) and are colored in rainbow from blue ( variable ) to red ( conserved ) . The most conserved residues ( red or orange ) , displayed as spheres to highlight their positions , tend to line interaction surfaces and indicate potential functional sites . DOI: http://dx . doi . org/10 . 7554/eLife . 09248 . 008
The models presented in this article for 58 large protein families which cannot be accurately modeled using comparative modeling or fold recognition methods cover a significant fraction of the prokaryotic sequences for which structural information was previously unavailable . Each of these families have thousands of members ( Table 2 , column 2 ) , hence these models have quite broad impact . The analyses of a small subset of these structures provided here only begins to uncover the wealth of information relating to function they contain . In addition to the new structure-based interpretation of existing sequence conservation and mutational data the models enable , they illustrate the complex transformations occurring in membrane protein evolution: for example , the changes in YeiH and RarD structural element connectivity compared to previously known structures . With the advent of sensitive sequence profile–profile comparison methods , much of protein structure modeling has been reduced to sequence alignment , and indeed for functional interpretation often much can be learned simply by draping the query sequence on a homologous structure; in contrast , as illustrated in the examples above , in the co-evolution-guided de novo structure prediction case , structure modeling is critical to functional insight . Large-scale genome sequencing is having an unanticipated impact on protein structure modeling , enabling accurate protein structure and protein complex modeling using co-evolution-based predicted contacts . The importance of this approach to structural biology over the next decade will depend on the balance between two opposing trends: as more organisms are sequenced , the number of protein families with sufficient sequences for accurate modeling will increase , but as more structures are determined , there are fewer families for which accurate models cannot be produced by reliable comparative modeling methods . An increase in the number of eukaryotes sequenced—for example , by projects such as the recent Tara Ocean expedition ( Bork et al . , 2015; Sunagawa et al . , 2015 ) —would make it possible to accurately model a large number of eukaryote-specific protein families of considerable biological interest . Because of the comparative difficulty in experimental structure determination , it is likely that co-evolution-based prediction will continue to have the most impact for membrane proteins . In this article , we present models for half of the large protein families in prokaryotes which do not currently have structures . The value of a comparable number of structures of eukaryotic protein families may justify the investment in genome sequencing of a diverse set of ∼400 simple eukaryotes . For proteins not belonging to sufficiently large or diverse families but for which functional selections have been developed , it should be possible to develop experimental sequence covariation data sets by library generation , functional selection , and next generation sequencing . Significant resources were invested in the Protein Structure Initiative ( PSI ) , with the initial goal ‘to make the three-dimensional , atomic-level structures of most proteins easily obtainable from knowledge of their corresponding DNA sequences ( Burley et al . , 2008 ) ’ . It is notable that structure models can now be generated for exactly the original class of proteins targeted by the PSI—large protein families without any available information—but at a small fraction of the cost .
Protein-coding genes were extracted from the E . coli ( AUP000000625 ) , B . subtilis ( AUP000001570 ) , H . salinarum ( AUP000000554 ) , S . solfataricus ( AUP000001974 ) reference genomes in the UniProt proteome database ( UniProt Consortium , 2014 ) . Each protein from these proteomes was scanned against the PDB using HHsearch ( -ssm 0 , from HHsuite v . 2 . 0 . 15; [Remmert et al . , 2012] ) to identify proteins with no homologs of known structures ( e-value of the top hit >1 ) . We used two versions of the PDB database , one from 01 January , 2012 and one from 31 January , 2015 . For the subset that had no hits in 2015 , a multiple sequence alignment ( MSA ) was generated using Jackhmmer ( -E 1E-20 -N 8 , [Eddy , 2009] ) and the uniref90 database ( Suzek et al . , 2007 ) from January , 2015 . The alignments were filtered using HHfilter ( -id 90 -cov 75 ) , and positions that had more than 75% gaps were removed . To reduce redundancy , we constructed hidden Markov models ( HMMs ) using HHmake from each MSA and clustered the HMMs based on HHΔ ( Kamisetty et al . , 2013 ) , a measure of HMM–HMM similarity . Families were assigned to the same cluster if the HHΔ was less than 0 . 5 . The shortest E . coli protein was selected in each cluster; if no E . coli protein was in the cluster , a representative from B . subtilis , H . salinrum , or S . solfataricus was selected . Families for which the ( number of sequences ) / ( length of representative protein ) were greater than four were selected for modeling as described below . If the GREMLIN-predicted contacts ( see below ) were sparse and primarily between residues close along the linear sequence , the alignment was regenerated at an e-value 1E-40 cutoff , and the GREMLIN calculation repeated . If this resulted in too few sequences , the family was discarded ( this eliminated six families ) . TM protein domains that had a hit ( e-value < 1E-20 ) in 2015 , but no hit ( e-value > 1 ) in 2012 were selected for the TM benchmark . Alignments were created using the UniProt sequence associated with the PDB , and trimmed at the N and C termini to match the crystal structure . We also include aquaporin ( PDB: 1SOR_A ) to test our protocol in modeling reentrant helices . GREMLIN ( v2 . 01 ) was used to learn a global statistical model of the sequences in large families using pseudolikelihood optimization ( Balakrishnan et al . , 2011 ) . We previously reported that the accuracy of contact prediction using the residue–residue coupling values obtained from the model-fitting procedure is dependent on the number of sequences per length and the relative score ( Kamisetty et al . , 2013 ) . To account for these dependencies , we constructed a model ( Figure 16 ) that estimates the probability of being in contact using a pdb30 data set from PISCES ( resolution limited to 2 . 5A or better , from 04 January , 2014; [Wang and Dunbrack , 2003] ) , with length of at least 100 residues . MSAs were generated for each of the 10 , 358 pdb chains using HHblits ( -n 8 -e 1E-20 -maxfilt ∞ -neffmax 20 -nodiff -realign_max ∞ ) , and HHfilter ( -id 90 -cov 75 ) in the HHsuite ( Remmert et al . , 2012 ) . The 3392 pdb chains with more than 10 sequences per length were subsampled to create MSAs with varying number of sequences , which were used to estimate probability of contact ( Figure 16A ) . CCMPRED v0 . 1 , a parallel implementation of GREMLIN ( Seemayer et al . , 2014 ) , was used for the subsampled alignments . For CCMPRED , the default maximum number of iterations was modified to 100 to ensure convergence . The remaining 7047 pdb chains with less than 10L sequences were saved as a test set ( Figure 16B , Figure 17 ) . The top 3L/2 scores of residue pairs with sequence separation 3 or greater were normalized by rescaling the range so that the minimal value is 0 . 5 and average value is 1 . 0 . Contact prediction accuracy was found to be a simple function of the residue–residue normalized coupling value , the number of sequences , the length ( Figure 16—figure supplement 1 ) , and the sequence separation as shown in Figure 16 . The sigmoidal fit to these observed frequencies was used to estimate the probability of each contact being formed in the native structure . 10 . 7554/eLife . 09248 . 022Figure 16 . Dependence of the accuracy of predicted contacts on the normalized GREMLIN score ( sco ) , the effective number of sequences ( seq ) , the length ( len ) , and the sequence separation ( sep ) . Contacts are defined based on amino acid specific Cβ-Cβ distance cutoffs as described in SI Table 3 in Kamisetty et al . ( 2013 ) . ( A ) Observed vs predicted accuracies over a large data set of proteins of known structure with deep alignments ( Supplementary file 3 ) , sub sampled to different extents ( seq/√ ( len ) = 4 ( red ) , 8 ( green ) , 15 ( purple ) , 32 ( cyan ) , and 96 ( orange ) ) . Circles represent observed contact prediction accuracies , solid lines , a fit to a sigmoid function of the normalized coupling value , the number of sequences , the length , and the sequence separation ( see Figure 16—figure supplement 1 and Figure 16—figure supplement 2 ) . ( B ) Observed vs predicted accuracies in an independent data set of variable length alignments for 7047 pdb chains ( Supplementary file 3 ) , using maximum number of sequences obtained with HHblits as opposed to subsampling a large alignment . Circles again represent observed contact prediction accuracies; solid lines , the predicted accuracy using the model obtained by fitting to the data in ( A ) . The contact prediction accuracy is correctly modeled for the independent data set , justifying its use on the unknown cases described in this article . The Equation use to calculate P ( contact|sco , seq , len , sep ) is P ( contact|sco , seq , len , sep ) ≈ 0 . 89 ( 1–P ( contact|sep ) ) 1+exp ( −0 . 58 ( seqlen ) 0 . 50 ( sco−5 . 46 ( seqlen ) −0 . 53 ) ) +P ( contact|sep ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09248 . 02210 . 7554/eLife . 09248 . 023Figure 16—figure supplement 1 . Contact prediction accuracy is better correlated with ( #sequences/sqrt ( length ) ) than with ( #sequences/length ) . Accuracy is computed for the top 3L/2 GREMLIN predictions , with sequence separation ≥3 , based on Cβ-Cβ amino acid specific distance as described in SI Table 3 in Kamisetty et al . ( 2013 ) . The number of sequences after reducing the redundancy to 80% is shown . A set of 7047 pdb chains ( see Supplemental file 3 ) was divided into two groups by length ( less than 150 and greater than 400 ) . ( A ) Larger proteins with similar number of sequence were less accurate then the smaller proteins . ( B ) #Sequences/length as often used does not accurately account for length dependence . There is a clear separation between the blue and green distributions . ( C ) #Sequences/√length better accounts for the length dependency . The blue and green distributions overlap . DOI: http://dx . doi . org/10 . 7554/eLife . 09248 . 02310 . 7554/eLife . 09248 . 005Figure 17 . The Rc metric used to assess fit of predicted contacts to a model . The expected total GREMLIN score if the structure was native was estimated by summing sco*P ( contact|sco , seq , len , sep ) over all contacts with sep ≥6 . To evaluate the fit of a particular model to a predicted contact set , we take the ratio of the actual total GREMLIN score of the model to the expected total score computed as above; we refer to this ratio of observed and expected contact scores as ‘Rc’ . Blue line: the distribution of Rc in native structures with 4L–10L sequences; Red line: distribution of Rc after randomly reassigning contact predictions to structures . Rc values less than 0 . 7 are very infrequently observed for native structures; we use this value as a cutoff to evaluate the fit of a predicted contact set to a model . DOI: http://dx . doi . org/10 . 7554/eLife . 09248 . 005 To evaluate the significance of a match between predicted contacts and a model , we determined the expected total GREMLIN score over all contacts with sequence separation of 6 or greater using P ( contact ) . To evaluate the fit of a particular model to a predicted contact set , we take the ratio of the actual total GREMLIN score of the model to the expected total score computed as above; we refer to this ratio of observed and expected contact scores as ‘Rc’ throughout the text . As shown in Figure 17 , Rc ranges from 0 . 7 to 1 . 2 for native proteins , and from 0 to 0 . 3 when contact maps and structures are randomly paired . The Rc was evaluated over the shortest overlap of the two lengths ( contact map length vs pdb length ) . For homo-oligomer complexes , the Rc score includes all chains across all bio units . Residue-pair-specific distance restraints for use in the Rosetta structure prediction calculations were generated based on the normalized GREMLIN scores . Distance restraints were implemented as sigmoidal functions of the form: restraint ( d ) = weight ( 1 + exp ( −slope ( d − cutoff ) ) + intercept ) , where d is the distance between the constrained Cβ atoms ( Cα in the case of glycine ) , the distance cutoffs and slopes are amino acid pair specific ( SI Table 3 in Kamisetty et al . , 2013 ) , and the weight is the normalized Gremlin score multiplied by three to give the contact restraints roughly the same total dynamic range as the Rosetta energy . These distance restraints supplement the Rosetta energy function; the combination ensures the sampling of physically realistic structures consistent with the contact predictions . For TM proteins , the Rosetta energy function was modified to reflect the exposure of non-polar residues in the membrane-spanning regions: the lazaridis-karplus solvation energy term weight was set to zero ( fa_sol = 0 . 00 ) , and to compensate for the short range repulsion implicit in the solvation model , the Lennard-Jones repulsive and attractive terms were given equal weights . We found this simple approach was equally effective and considerably less computationally intensive than the RosettaMembrane approach , which requires estimating the TM region for energy evaluation . The Rosetta ab initio protocol ( Simons et al . , 1999; Rohl et al . , 2004 ) was used to generate 10 , 000 independent models guided by the covariance-derived restraints . For the benchmark set , fragments database from 2011 was used; for aquaporin , the fragments were filtered to remove any homologs with e-value < 1 . After the generation of fragments for aquaporin , we examined the PDB files that contributed the most to the fragment set and verified that they did not contain aquaporin-like structures . The models generated by Rosetta ab initio were refined with an iterative version of the RosettaCM ( Song et al . , 2013 ) hybridization protocol used to refine models generated with contact information in CASP10 ( Kim et al . , 2014 ) . In each iteration , 20 models are produced by recombination and minimization . In addition to the recombination of secondary structure chunks in the input models , fragment insertion was allowed in all positions . Iterations were continued until the procedure converged . For a 200-residue protein , the average runtime to produce a single model is about 30 min for RosettaAB and about 20 min for RosettaCM . If in the initial Rosetta ab initio calculations , the top 10 models selected by restraint score converged ( average pairwise TMscore [Zhang and Skolnick , 2004] > 0 . 8 ) , the top five models were input directly into the iterative RosettaCM hybridization protocol . If models converged over substructures ( average pairwise TMscore between 0 . 5 and 0 . 8 ) , the top 10 models were first expanded by recombination to a population of 1000 structures , and the top five models were input into the RosettaCM hybridization protocol . If the Rosetta ab initio calculations did not converge ( average pairwise TMscore <0 . 5 ) , we carried out an additional 10 , 000 Rosetta ab initio trajectories; if the top models did not converge , we considered the structure of the protein not accurately predictable using our approach . 15 of the 121 families were eliminated at this stage . We also eliminated families for which the models generated by the hybridization protocol did not satisfy the predicted contacts; 37 additional families were eliminated at this step . Proteins over 400 amino acids for which there was little convergence of the lowest energy generated models were parsed into multiple domains ( <200 residues ) guided by the predicted contact information keeping overlaps of at least 50 residues between each domain , and Rosetta ab initio was used to generate models for each domain separately . If the overlapping regions in each domain converged during modeling , these were used to assemble the full model , otherwise the domains were trimmed to converged residues and docked using RosettaDock ( Chaudhury et al . , 2011 ) . If models converged overall in the Rosetta ab initio calculations but specific sets of contact restraints were consistently violated , we explored the possibility that the violations correspond to interactions between monomers in a homo-oligomer . To test for oligomeric contacts , docking was performed between two copies of the model using RosettaDock guided by the co-evolution-derived constraints . We developed a simple measure of convergence and contact violation after the hybridization protocol to trim regions with higher chance of being in error . The top five percent of the models were selected based on the sum of the Rosetta all atom energy and the contact restraint score and superimposed using THESEUS v3 . 1 ( Theobald and Wuttke , 2006 ) . The mean square deviation of the Cα coordinates of each residue was computed , and after smoothing with a Gaussian spanning three residues before and after the central residue , residues with MSD > 2 Å2 were trimmed . We also eliminated residues in regions in which there were either very few contact restraints , or the majority of the restraints were violated . For the benchmark set , the model closest to the average of the lowest energy 5% models was selected , and the RMSD to the native structure was computed over ( 1 ) the full length of the protein , ( 2 ) the converged and constrained residues , and ( 3 ) the residues structurally aligned using TM-align . The latter alignments are longer and more accurate , but selection of the subset of residues requires knowledge of the native structure; this is not the case for ( 2 ) . Our models provide starting templates to model any member of these families using comparative modeling , which requires relatively little computer time . To evaluate the protein space our 58 models cover , we carried out Jackhmmer search ( -E 1E-20 -N 8 , [Eddy , 2009] ) with uniref100 database ( Suzek et al . , 2007 ) from January , 2015 and eliminated identical sequences and sequences aligned over less than 75% of the protein; the number of remaining sequences for each of the 58 families is listed in Table 2 . To compare our method to the EvFold method , we submitted the alignments in our membrane protein benchmark to the EvFold web server . We compared the accuracy of the best of the 50 models generated by the server to our single selected model for each family in Tables 3 , 4 . The Rosetta models are considerably more accurate , but the EvFold server is orders of magnitude faster . We compare to the server results rather than to the results in the previously published Evfold paper as this would be unfair since there were fewer available sequences at the time the paper was written and the contact prediction method ( mfDCA ) was somewhat less accurate . For the server comparison , we chose to predict contacts using the PLM option as this is very similar to GREMLIN . 10 . 7554/eLife . 09248 . 025Table 3 . Comparison of methods on CASP11 targetsDOI: http://dx . doi . org/10 . 7554/eLife . 09248 . 025BAKER*Jones-UCL*Evfold-web serverTargetsCα-RMSDGDT-TSCα-RMSDGDT_TSCα-RMSDGDT-TST08063 . 660 . 46 . 834 . 38 . 230 . 0T08244 . 255 . 39 . 241 . 48 . 132 . 6*Full-length Cα-RMSD and GDT-TS calculation based on the best of five models submitted to CASP11 from BAKER and Jones-UCL groups . For Evfold , the values for best of 50 models generated by the web server are reported , sorted by full-length Cα-RMSD . For the comparison , the alignments used during CASP11 were provided as input to the Evfold-web server , with PLM option selected . For T0824 , the minimal number of sequence limit was set to 0 to allow Evfold-web server to run . PLM , pseudo-likelihood . 10 . 7554/eLife . 09248 . 026Table 4 . Comparison of methods on transmembrane benchmark setDOI: http://dx . doi . org/10 . 7554/eLife . 09248 . 026BAKEREvfold-web serverTargetsCα-RMSDGDT-TSCα-RMSDGDT-TS4HE8_H4 . 954 . 55 . 350 . 31SOR_A ( aquaporin ) 2 . 769 . 76 . 144 . 54Q2E_A5 . 445 . 612 . 921 . 74HTT_A3 . 960 . 66 . 441 . 84P6V_E5 . 056 . 67 . 431 . 84J72_A6 . 667 . 112 . 933 . 83V5U_A3 . 958 . 84 . 647 . 14PGS_A3 . 566 . 34 . 648 . 14QTN_A4 . 259 . 64 . 951 . 44OD4_A3 . 955 . 64 . 153 . 44O6M_A4 . 164 . 011 . 233 . 0The Cα-RMSD and GDT-TS calculations are over the full sequence . For Evfold web server results , we report the best Cα-RMSD of 50 models returned . For the comparison , the alignments we used were provided as input to the Evfold-web server , and the pseudo-likelihood method was selected . | Proteins are long chains made up of small building blocks called amino acids . These chains fold up in various ways to form the three-dimensional structures that proteins need to be able work properly . Therefore , to understand how a protein works it is important to determine its structure , but this is very challenging . It is possible to predict the structure of a protein with high accuracy if previous experiments have revealed the structure of a similar protein . However , for almost half of all known families of proteins , there are currently no members whose structures have been solved . The three-dimensional shape of a protein is determined by interactions between various amino acids . During evolution , the structure and activity of proteins often remain the same across species , even if the amino acid sequences have changed . This is because pairs of amino acids that interact with each other tend to ‘co-evolve’; that is , if one amino acid changes , then the second amino acid also changes in order to accommodate it . By identifying these pairs of co-evolving amino acids , it is possible to work out which amino acids are close to each other in the three-dimensional structure of the protein . This information can be used to generate a structural model of a protein using computational methods . Now , Ovchinnikov et al . developed a new method to predict the structures of proteins that combines data on the co-evolution of amino acids , as identified by GREMLIN with the structural prediction software called Rosetta . A community-wide experiment called CASP—which tests different methods of protein prediction—showed that , in two cases , this new method works much better than anything previously used to predict the structures of proteins . Ovchinnikov et al . then used this method to make models for proteins belonging to 58 different protein families with currently unknown structures . These predictions were found to be highly accurate and the protein families each have thousands of members , so Ovchinnikov et al . 's findings are expected to be useful to researchers in a wide variety of research areas . A future challenge is to extend these methods to the many protein families that have hundreds rather than thousands of members . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | Large-scale determination of previously unsolved protein structures using evolutionary information |
Malaria elimination strategies require a thorough understanding of parasite transmission from human to mosquito . A clinical model to induce gametocytes to understand their dynamics and evaluate transmission-blocking interventions ( TBI ) is currently unavailable . Here , we explore the use of the well-established Controlled Human Malaria Infection model ( CHMI ) to induce gametocyte carriage with different antimalarial drug regimens . In a single centre , open-label randomised trial , healthy malaria-naive participants ( aged 18–35 years ) were infected with Plasmodium falciparum by bites of infected Anopheles mosquitoes . Participants were randomly allocated to four different treatment arms ( n = 4 per arm ) comprising low-dose ( LD ) piperaquine ( PIP ) or sulfadoxine-pyrimethamine ( SP ) , followed by a curative regimen upon recrudescence . Male and female gametocyte densities were determined by molecular assays . Mature gametocytes were observed in all participants ( 16/16 , 100% ) . Gametocytes appeared 8 . 5–12 days after the first detection of asexual parasites . Peak gametocyte densities and gametocyte burden was highest in the LD-PIP/SP arm , and associated with the preceding asexual parasite biomass ( p=0 . 026 ) . Male gametocytes had a mean estimated circulation time of 2 . 7 days ( 95% CI 1 . 5–3 . 9 ) compared to 5 . 1 days ( 95% CI 4 . 1–6 . 1 ) for female gametocytes . Exploratory mosquito feeding assays showed successful sporadic mosquito infections . There were no serious adverse events or significant differences in the occurrence and severity of adverse events between study arms ( p=0 . 49 and p=0 . 28 ) . The early appearance of gametocytes indicates gametocyte commitment during the first wave of asexual parasites emerging from the liver . Treatment by LD-PIP followed by a curative SP regimen , results in the highest gametocyte densities and the largest number of gametocyte-positive days . This model can be used to evaluate the effect of drugs and vaccines on gametocyte dynamics , and lays the foundation for fulfilling the critical unmet need to evaluate transmission-blocking interventions against falciparum malaria for downstream selection and clinical development . Funded by PATH Malaria Vaccine Initiative ( MVI ) . NCT02836002 .
Malaria , a disease caused by Plasmodium parasites , continues to be a public health burden . Despite a reduction in the malaria case incidence of ~40% , and mortality by 62% over the last decade , malaria caused ~429 , 000 deaths in 2015 ( World Health Organization , 2016 ) . Apart from the direct health implications , malaria is a substantial contributor to ongoing poverty in affected countries . Recently , the spread of artemisinin-resistant parasites has emerged as a global health concern . Both the recent gains in malaria control and concerns about artemisinin resistance have stimulated programs to eliminate malaria ( World Health Organization , 2016 ) . Novel interventions may support malaria elimination efforts in endemic settings ( Griffin et al . , 2010 ) that are further dependent on political and financial commitments to maximize coverage with currently available interventions and improve surveillance systems to optimize disease notification and treatment ( Moonen et al . , 2010 ) . A major challenge to eliminating malaria is its highly efficient transmission by Anopheles mosquitoes . Transmission to mosquitoes starts when a small proportion of asexual parasites commit to form male and female gametocytes . It is currently unclear what stimulates gametocyte commitment and when gametocyte commitment first occurs ( Nilsson et al . , 2015 ) . Upon commitment , maturation of gametocytes takes place predominantly in the bone marrow , and requires 7 days ( range 4–12 ) of development . ( Eichner et al . , 2001 ) Subsequently , mature gametocytes ( parasites that are not associated with clinical disease ) appear in the peripheral blood , where they may circulate for an average of 6 days ( Eichner et al . , 2001; Bousema et al . , 2010 ) . During this period , blood-feeding Anophelines may ingest gametocytes where , after a sporogonic development phase , sporozoites reach the mosquito salivary gland rendering the mosquito infectious to humans upon its next bite . Early work based on the microscopic evaluation of experimental P . falciparum infection ( malariatherapy ) studies reported that gametocytes may make their appearance in small numbers around 10 days following the first day of fever ( Shute and Maryon , 1951; Ciuca et al . , 1937 ) . The renewed focus on malaria elimination requires a thorough understanding of malaria transmission dynamics - when mature male and female gametocytes are first produced upon infection and how long they circulate in peripheral blood ( Sinden , 2017 ) . These parameters are difficult to measure in naturally acquired infections where frequent super-infections , immunity and other factors dictate parasite and gametocyte dynamics ( Bousema and Drakeley , 2011 ) . Interventions that specifically aim to reduce gametocyte development , circulation time or infectivity are highly desirable in the context of malaria elimination and require effective models for the early clinical evaluation . The controlled human malaria infection ( CHMI ) model allows the induction of parasitemia under highly standardized conditions and plays an important role in the assessment of safety and efficacy of novel antimalarial drugs and vaccines ( Sauerwein et al . , 2011 ) . Preliminary evidence for the induction of female gametocytes in CHMI studies with blood stage inoculum was recently demonstrated using piperaquine monotherapy ( Pasay et al . , 2016; Farid et al . , 2017 ) . In this study , we aimed to develop a CHMI transmission model to induce gametocyte carriage after mosquito bite infection . The primary objective of the current trial was to safely induce gametocytemia in study participants by the use of different ( sub ) curative drug regimens based on sulfadoxine-pyrimethamine ( Bousema and Drakeley , 2011; Butcher , 1997 ) and piperaquine ( Adjalley et al . , 2011 ) .
From a total of 49 screened candidate participants , 16 volunteers were included in a first cohort and randomly assigned to four study arms prior to challenge ( Figure 1 ) . After observed transient liver enzyme elevations in the first cohort , the study was temporarily put on hold and the already initiated infections in the second cohort of 13 participants were abrogated by curative treatment on day 3 post challenge . The hold was lifted after reviewing safety data . Participants from the first cohort completed all study visits , and form the basis of the current manuscript; their baseline characteristics are shown in Table 1 . After exposure to bites of a standard protocol of five P . falciparum infected mosquitoes , all participants developed parasitemia on days 6 . 5–12 post-challenge; peak parasite densities ranged from 1050 to 63113 Pf/mL ( Figure 2; Figure 2—figure supplement 1; Table 2; Supplementary file 1 ) . Due to asexual recrudescence in seven of the eight participants after a subcurative treatment ( T1 ) with LD-PIP , a curative treatment ( T2 ) had to be administered before day 21 post challenge . The median period between T1 and T2 was 9 . 1 ( range of 7 . 7–11 . 7 ) , 10 . 0 ( range of 9 . 2–10 . 2 ) , 4 . 7 ( range of 2–10 . 7 ) , and 2 . 5 ( range of 1 . 5–5 . 0 ) days for study arms LD-SP/SP , LD-SP/PIP , LD-PIP/PIP , and LD-PIP/SP , respectively . In participants receiving a subcurative LD-SP as T1 , no recrudescent infection occurred and T2 was initiated on day 21 per protocol . One participant from treatment arm LD-PIP/PIP developed asexual recrudescence after T2 , and received end treatment with atovaquone/proguanil on day 36 . The remaining participants did not develop recrudescent infections after T2 , and were treated with atovaquone/proguanil on day 42 as per protocol . All participants also developed gametocytemia as determined by Pfs25 qRT-PCR ( Figure 2; Figure 3A; Figure 2—figure supplement 1 ) . Gametocytes were first detected 8 . 5–12 days after the initial peak of asexual parasites with no statistically significant difference in time to gametocyte appearance between study arms ( p=0 . 26 ) ( Table 2 ) . The median peak density of gametocytes was 83 gametocytes/mL ( range 11–1285 ) when all study participants were considered . Peak gametocyte densities were higher in the study arm randomised to LD-PIP/SP , with a median of 627 gametocytes/mL ( range of 199–1285 ) , compared to 38 gametocytes/mL ( range of 11–368 ) , 30 gametocytes/mL ( range of 13–101 ) , 83 gametocytes/mL ( range of 46–99 ) , for arms LD-SP/SP , LD-SP/PIP , and LD-PIP/PIP , respectively ( Figure 2; Figure 2—figure supplement 1; Table 2 ) . Thirteen ( 81% , 13/16 ) participants showed gametocytes on at least 5 consecutive days . The mean number of consecutive gametocyte-positive days was 24 . 5 ( range of 17–25 ) for the LD-PIP/SP arm and was higher than for other arms ( Table 2; Figure 2 ) . Using multi-level logistic regression ( random effect for within-group variation ) , we estimated that the average proportion of days that individuals tested positive for gametocytes was 27 . 4% ( LD-SP/SP ) , 35 . 9% ( LD-SP/PIP ) , 51 . 4% ( LD-PIP/PIP ) , and 48 . 3% ( LD-PIP/SP ) ( Table 2 ) . The LD-PIP/PIP and LD-PIP/SP arms ( i . e . those receiving ‘low dose PIP’ ) each had significantly higher average proportions of gametocyte-positive days than both arms LD-SP/SP and LD-SP/PIP ( posterior probability 90 . 8% and 86 . 1% , respectively; 81 . 1% joint probability of arms LD-PIP/PIP and LD-PIP/SP both being higher than both LD-SP/SP and LD-SP/PIP ) . Furthermore , the area under the curve ( AUC ) for gametocyte density showed a statistically significant difference between arms ( p=0 . 04 ) . The LD-PIP/SP arm had a significantly higher gametocyte load ( area under the curve ) than each of the other three treatment arms ( 94 . 4% posterior probability of being the highest; Figure 3B ) . After correction for the asexual AUC , the probabilities of the gametocyte AUC in the LD-PIP/SP arm being higher than the other three decreased to 97 . 2% , 96 . 3% , and 96 . 2% ( from 99 , 1% , 98 . 9% , and 95 . 4% ) , and the probability of LD-PIP/SP being higher than all other study arms decreased to 94 . 0% . Both female and male gametocytes were detected in 14/16 ( 88% ) participants ( Figure 4; Figure 4—figure supplement 1 ) . Gametocyte sex-ratio’s and circulation times have to be interpreted with caution since they rely on two separate qRT-PCR assays with differences in assay sensitivity ( Figure 5; Supplementary file 2 , 3 ) . On average 2 . 5 times as many female gametocytes were observed compared to male gametocytes per measured time-point ( Figure 4; mean ratio 2 . 5 ( SD = 2 . 5 ) ) . Combining all treatment arms , the best estimate of gametocyte half-life was 5 . 1 days ( 95% CI 4 . 1–6 . 1 ) for female gametocytes and 2 . 7 days ( 95% CI 1 . 5–3 . 9 ) for male gametocytes ( Figure 4—figure supplement 2 ) . Gametocytes are produced from their asexual progenitors , and hence asexual parasite kinetics and gametocyte kinetics are related . The AUC of asexual parasitemia was statistically significantly associated with the AUC of gametocytemia ( r2 = 0 . 31 , p=0 . 026 ) , as shown in Figure 3C . The mean time-window between the first asexual parasites and the first appearance of gametocytes was 10 . 6 ( SD = 0 . 65 ) days , see Table 2 . Membrane feeding experiments were performed as an exploratory objective , and confirmed infectivity of gametocytes in three mosquitoes from three study arms on days 25 ( LD-PIP/SP and LD-SP/SP arms ) and 31 ( LD-SP/PIP arm ) post-infection . Mean gametocyte densities at those time-points were 106 gametocytes/mL ( SD = 175 ) , and 28 gametocytes/mL ( SD = 47 ) , respectively . Expressed as a proportion of all examined mosquitoes , 0 . 0002% ( 3/14400 ) of mosquitoes became infected in these exploratory assessments . Possible and probable related adverse events after challenge infection are shown in Figure 6 and Table 3 . The most frequently reported adverse events were fatigue , malaise , headache , fever , nausea , and chills . Grade three adverse events were reported in 14/16 ( 88% ) participants , and were predominated by headache ( n = 8 ) , chills ( n = 6 ) , and nausea ( n = 5 ) . All possible and probable related adverse events resolved by the end of study . No serious adverse events occurred . The median number of adverse events was 20 . 5 per individual; the median number of adverse events with a grade three severity score was 1 . 5 per individual . There was no evidence for a difference between study arms in the occurrence of adverse events ( p=0 . 49 ) or grade three adverse events ( p=0 . 28 ) . Laboratory abnormalities during the study are shown in Table 4 . Most prevalent abnormalities were elevated transaminases ( ALT/AST ) ( n = 16 ) , decreased lymphocytes ( n = 15 ) , decreased neutrophils ( n = 13 ) , and decreased platelets ( n = 12 ) . The only grade three laboratory abnormalities were elevated ALT ( n = 8 ) , and elevated AST ( n = 7 ) . 16/16 ( 100% ) volunteers showed mild to severe ALT/AST elevations . 5/16 ( 31% ) mild ( grade 1 ) ; 3/16 ( 19% ) moderate ( grade 2 ) , and 8/16 ( 50% ) severe ( grade 3 ) ( up to 25 x ULN ) ALT/AST elevations . These derangements were transient , and returned to baseline values within the normal range before the end of the study . A detailed overview of these liver function test derangements can be found in the supporting information ( Figure 6—figure supplement 1 ) . These unexpected safety findings were reported to the Safety Monitoring Committee ( SMC ) and CCMO , and thoroughly reviewed .
Here , we present a CHMI model to induce mature gametocytes after mosquito bite infection in malaria-naive study participants . The timing of the first appearance of gametocytes suggests that a fraction of the first wave of asexual parasites commit to the production of male and female gametocytes . With the use of antimalarial drugs that attenuates asexual stage infections but leave ( developing ) gametocytes unaffected , we determined biologically plausible half-lives of male and female gametocytes , and show preliminary evidence of the potential of this model to complete the lifecycle of malaria in mosquito feeding assays . Malaria elimination efforts require a thorough understanding of the transmissibility of infections . Gametocyte commitment occurs for a fraction of asexual parasites under regulation of the transcription factor AP2-G with the entire progeny of a sexually committed schizont forming either male or female gametocytes ( Kafsack et al . , 2014 ) . Our findings , based on novel sex-specific gametocyte qRT-PCR , confirm earlier work from malariatherapy studies where gametocytes were first detected by microscopy at 9–11 days after asexual parasites ( Ciuca et al . , 1937; Shute and Maryon , 1951 ) . These data indicate very early gametocyte commitment and are in line with our earlier observations that Pfs16 mRNA , the earliest gametocyte transcript , is detectable at the moment of peak parasitemia in CHMI models ( Schneider et al . , 2004 ) . This timing is highly relevant for understanding gametocyte transmission biology . The circulation of mature gametocytes has not been reported in previous CHMI trials using curative regimens of chloroquine , artemether-lumefantrine , or atovaqoune-proguanil , and our data illustrate the differential impact of antimalarial drugs on developing gametocytes . Once treatment is initiated , gametocyte production ceases abruptly ( in the case of artemisinins ) , remains unaffected , or may even be stimulated under drug pressure as suggested for sulfadoxine-pyrimethamine and piperaquine ( Bousema and Drakeley , 2011; Butcher , 1997; Adjalley et al . , 2011 ) . In our study , we aimed for a protracted low density of asexual parasitemia demonstrating that early abrogation of asexual infections by both sulfadoxine-pyrimethamine and piperaquine permits successful mature gametocyte development . SP has long been associated with a rapid appearance of gametocytes that is too early to be explained by de novo gametocyte production upon drug pressure and has thus been hypothesized to reflect an efflux of sequestered gametocytes upon treatment ( Butcher , 1997 ) . Evidence for the permissiveness of piperaquine to ( developing ) gametocytes is more recent ( Pasay et al . , 2016; Farid et al . , 2017; Adjalley et al . , 2011 ) . In the current study , group sizes are limited and comparisons between treatment arms have to be interpreted with caution . CHMI studies are logistically challenging and the number of volunteers that can be monitored to ensure participant safety is an important consideration when defining the study size . Our sample size calculation was based on the optimistic assumption that the vast majority of volunteers would develop mature gametocytes; an assumption that was supported by the current data . With our limited study size , our findings indicate that none of the study drugs prevented the appearance of gametocytes after treatment , thereby suggesting limited or no effect of PIP and SP on developing or mature gametocytes ( Bolscher et al . , 2015 ) . We hypothesized that slow acting drugs may promote the development of gametocytes ( Méndez et al . , 2002 ) , potentially via microvesicles that are derived from infected erythrocytes ( Nilsson et al . , 2015 ) and differences between drug regimens in the rate at which asexual parasites are cleared upon T1 and T2 would result in different gametocyte dynamics . Although our findings indicate highest gametocyte concentrations in the LD-PIP/SP arm , more observations and thus additional studies are needed to allow the construction of a model that allows a quantification of gametocyte commitment at different time-points during the study ( e . g . prior to T1 , during the phase of parasite recrudescence and following T2 ) . One hypothesis would be similar gametocyte commitment in all arms after T1 but a more rapid release of gametocytes that accumulated in the bone marrow between T1 and T2 . We present the novel evidence that both male and female gametocytes appear early , upon infection . Our findings suggest an earlier appearance of female gametocytes ( 18 . 8 days ( SD 1 . 8 ) compared to male gametocytes 20 . 3 days ( SD 1 . 2 ) ) and a longer circulation time of female gametocytes that is in line with previous estimates from naturally infected individuals ( Bousema et al . , 2010; Ciuca et al . , 1937 ) . Whilst both male and female gametocytes are consistently detected at densities of 0 . 1 gametocyte/µL ( Stone et al . , 2017 ) , the highly abundant Pfs25 mRNA makes the female gametocyte qRT-PCR more sensitive than the male PfMGET qRT-PCR . Differences in gametocyte dynamics between male and female gametocytes should therefore be interpreted with caution . Gametocyte densities remained below the threshold of detection by microscopy throughout the study period and were strongly associated with the preceding densities of the asexual progenitors . Participants in the LD-PIP/SP study arm showed the highest gametocytes densities , and a mean female/male sex ratio of 4 . 1 ( SD = 5 . 1 ) , in line with gametocyte sex-ratios in natural infections ( ~3 to 5 females to one male ) ( Ciuca et al . , 1937; Delves et al . , 2013 ) . We confirmed the infectivity of gametocytes in three mosquitoes from three study arms . The very low rate of infected mosquito corroborates observations from naturally acquired infections where mosquito infection becomes highly unlikely below 1000–10 , 000 gametocytes/mL ( Gonçalves et al . , 2016 ) . The sporadic mosquito infections thus demonstrate that mature gametocytes in sex-ratios supportive of mosquito infections can be achieved in CHMI transmission models . Studies on the evaluation of TBIs will need a further optimized protocol aimed to achieve higher gametocyte densities by increasing duration and load of the asexual parasite burden . For the evaluation of gametocytocidal interventions in the CHMI transmission model , gametocyte densities should be sufficiently high to quantify an intervention-associated reduction in gametocyte appearance or gametocyte half-life . For the evaluation of interventions that reduce the transmissibility of gametocytes , higher mosquito infections should be achieved at proportions that allow the detection of meaningful reductions in mosquito infection rates in experimental arms . Low infectivity in membrane feeding assays may be overcome by achievement of higher gametocyte densities in the model , and the use of gametocyte concentration methods ( Reuling et al . , 2017 ) , or by direct skin feeding assays ( Bousema et al . , 2012 ) . In line with recent findings , we observed recrudescent infections in 7/8 participants treated with LD-PIP ( Pasay et al . , 2016 ) . Recrudescent infections were not observed in arms that first received LD-SP , suggesting that this dose , although 1/3 of the standard curative dose of sulfadoxine-pyrimethamine , is curative at asexual parasite densities observed in our participants . It has been hypothesized that the prolonged parasitemia under drug pressure increases gametocyte commitment ( WWARN Gametocyte Study Group , 2016 ) . The duration of parasite multiplication between T1 and T2 was relatively short in this study ( 2–5 days ) for subjects with recrudescent infections , and the contribution of drug pressure may thus have been limited . The current findings suggest that further lowering the SP dose may be considered to prolong asexual parasite exposure . The liver enzyme elevations found in our study led to a structured risk analysis , and review by independent experts . Transient , asymptomatic liver function test ( LFT ) derangements have been reported in volunteers in previous CHMI studies , and are likely to be related to the asexual stage parasitemia , and subsequent treatment . Detailed studies on gametocyte biology and dynamics , and the early development of novel drugs and vaccines that target malaria transmission ( TBIs ) are currently restricted to in vitro assays , such as drug sensitivity assays , and standard membrane feeding assays ( SMFA ) ( Bousema and Drakeley , 2011; Wells et al . , 2009 ) . Recently , a humanized mice model has been developed to investigate P . falciparum sexual commitment that could , therefore , bridge in vitro assays to in vivo animal studies that take into account drug metabolism and gametocyte sequestration ( Duffier et al . , 2016 ) . Also , an experimental Plasmodium vivax transmission model in human has been reported ( Griffin et al . , 2016 ) . However , mechanisms underlying P . falciparum gametocytogenesis and dynamics have never been addressed in a controlled clean system in humans . Here , we present a novel CHMI transmission model for P . falciparum that can be used to study gametocyte biology and dynamics providing novel insights and tools in malaria transmission and elimination efforts . The dynamics of gametocyte commitment , maturation , sex ratio , and sequestration found in our model reflect parasite dynamics found in naturally acquired infections , although parasite densities are much lower than in many endemic settings . This model can be used to evaluate the effect of drugs and vaccines on gametocyte dynamics and sex ratios . With its current performance , the CHMI transmission model may allow testing of vaccination strategies that reduce the production of gametocytes from their asexual progenitors or accelerate their clearance from the blood stream ( Stone et al . , 2016 ) , and the testing of gametocytocidal drugs ( White , 2013 ) . To allow testing of sterilizing effect of drugs on circulating gametocytes ( White et al . , 2014 ) or the effect of antibodies that interfere with gametocyte fertilisation inside the mosquito gut ( Stone et al . , 2016 ) , the model needs to be optimized to achieve considerably higher mosquito infection rates . The current work lays the foundation for fulfilling the critical unmet need to evaluate transmission-blocking interventions against falciparum malaria for downstream selection and clinical development .
This single centre , open-label randomised trial was conducted at the Radboud university medical center ( Radboudumc ) , Nijmegen , the Netherlands . Healthy malaria-naive male and female participants aged 18–35 years were recruited from June until November 2016 . Screening included physical examination , electrocardiography ( ECG ) , hematology and biochemistry parameters and serology for human immunodeficiency virus ( HIV ) , hepatitis B and C , and asexual stages of P . falciparum . Informed consent was provided by all participants at screening visit . The central committee for research involving human subjects ( CCMO ) , and the Western Institutional Review Board ( WIRB ) approved the protocol for this study ( NL56659 . 091 . 16 ) . The trial was conducted according to the principles outlined in the Declaration of Helsinki and Good Clinical Practice standards , and registered at ClinicalTrials . gov , identifier NCT02836002 ( Supplementary file 4; Reporting Standard 1 ) . A total of 16 participants were included in the analysis of this study . After inclusion , study participants were randomly allocated to one of the four different treatment arms ( n = 4 per group ) with low-dose ( LD ) of either piperaquine ( PIP ) or sulfadoxine-pyrimethamine ( SP ) , followed by curative regimen of piperaquine or sulfadoxine-pyrimethamine upon recrudescence; ( i ) LD-SP/SP , ( ii ) LD-SP/PIP , ( iii ) LD-PIP/SP , or ( iv ) LD-PIP/SP . Randomisation was done by a computer-generated random number table ( Microsoft Excel 2007 , Redmond , WA ) . All study participants were subjected to a standard CHMI with five female Anopheles stephensi mosquitoes infected with the P . falciparum strain 3D7 ( Sauerwein et al . , 2011; Cheng et al . , 1997 ) . P . falciparum 3D7 asexual and sexual blood stages were cultured in a semi-automated culture system and used to infect mosquitoes by standard membrane feeding as described previously ( Ponnudurai et al . , 1986; Ponnudurai et al . , 1989 ) . The 3D7 lineage that was used in the current study is based on a 3D7 bank described in detail in Cheng et al . ( 1997 ) . To examine molecular markers of drug resistance , we used available Illumina whole genome sequencing data ( https://www . ebi . ac . uk/ena/data/view/PRJEB12838 ) ; aligning reads to the P . falciparum reference genome v3 ( plasmoDB ) with bowtie2 ( sourceforge ) and obtaining consensus sequences for dhps and dhfr genes with samtools . No mutations were identified in the dhfr gene; the only detected mutation was dhps A437G which , by itself , is not associated with sulfadoxine-pyrimethamine resistance ( Staedke et al . , 2004 ) . Plasmepsin II/III duplication events are associated with piperaquine resistance ( Witkowski et al . , 2017 ) but were not observed although the sequence similarities with neighboring genes Plasmepsin I and IV suggest that unambiguous quantification may require more specific gene targeting . Importantly , piperaquine sensitivity of our 3D7 lineage was previously confirmed by in vivo experiments ( Pasay et al . , 2016 ) . We conclude that the lineage used was sensitive to both sulfadoxine-pyrimethamine and piperaquine . Participants were monitored twice daily on an outpatient basis from day 6 after exposure to infected mosquitoes until malaria parasites were detected at a density of ≥5000 parasites per milliliter ( Pf/mL ) by qPCR or a positive thick blood smear , upon which they were treated with a subcurative dose of 500 mg/25 mg sulfadoxine-pyrimethamine ( Roche , Boulogne-billancourt , FR ) or 480 mg of piperaquine phosphate ( PCI Pharma Services , Tredegar , UK ) . After the first treatment ( T1 ) , participants continued to visit the study center twice daily for another 4 days to monitor the initial clearance of parasitemia by qPCR , after which they were monitored once a day for recrudescence . On day 21 or upon parasite density reaching ≥1500 Pf/mL , participants received a second treatment ( T2 ) , consisting of 1000 mg/50 mg sulfadoxine-pyrimethamine or 960 mg of piperaquine phosphate . After the second treatment , participants were monitored daily for 3 days , then three times a week until final treatment with atovaquone/proguanil ( Malarone ) on day 42 . Adverse events were recorded , and blood sampling was performed to monitor parasitemia and blood safety parameters . Symptoms of malaria were treated with acetaminophen up to 4000 mg daily , and nausea with metoclopramide up to 30 mg daily , if necessary . Parasite density was determined by quantitative PCR ( qPCR ) targeting the multicopy 18S rRNA gene ( Hermsen et al . , 2001 ) ; samples collected in the morning were processed immediately , evening samples 12 hr later . Thick blood smears were taken during evening visits , double-read and considered positive if two or more parasites were detected in 0 . 5µ µL ( Laurens et al . , 2012 ) . The presence of gametocytes was monitored in samples from day 7 . 5 after challenge until end of study by quantitative reverse-transcriptase PCR ( qRT-PCR ) targeting female-specific Pfs25 mRNA and male specific PfMGET ( Pf3D7_1469900 ) and using sex-specific trendlines ( Stone et al . , 2017; Pett et al . , 2016 ) . All samples with an estimated gametocyte density ≥5 gametocytes per mL ( gametocytes/mL ) were considered gametocyte positive . The duration of gametocyte carriage as an indicator of stable gametocyemia was defined as the maximum number of consecutive days with detectable gametocytemia above the threshold for detection . Direct Membrane Feedings Assays ( DMFA ) were performed as exploratory measures on days 21 , 25 and 31 post-infection with ~300 mosquitoes per feed per participant ( total of ~14 , 400 mosquitoes ) ( Bousema et al . , 2013; Lensen et al . , 1998; Ouédraogo AL et al . , 2013 ) . Mosquito infection status was determined on day 12 by circumsporozoite ( CSP ) ELISA ( Stone et al . , 2015 ) followed by qPCR confirmation of mosquitoes where the OD exceeded the mean +3 standard deviations of control mosquitoes ( Graumans et al . , 2017 ) . Adverse events were recorded and graded by the research physician as mild ( easily tolerated , grade 1 ) , moderate ( interfering with daily activity , grade 2 ) or severe ( preventing daily activity , grade 3 ) , and in the case of fever as mild ( 38 . 0–38 . 4°C ) , moderate ( 38 . 5–38 . 9 , °C ) or severe ( ≥39°C ) . Safety blood tests were performed daily , including full blood counts , LDH and highly sensitive troponin-T . Biochemistry tests including liver function test were assessed at screening , inclusion , 2 days after every treatment and at the end of study , and on additional days if considered relevant for clinical decision-making . For the quantification of the P . falciparum Pfs25 transcript levels total NA was RQ1 DNaseI treated according to the manufacturer’s protocol . 2 µL of DNaseI-treated material was run in a total volume of 25 µL of TaqMan RNA-to-Ct qRT-PCR reaction mixture ( Applied Biosystems , Foster City , California ) . For the quantification of the P . falciparum male gametocyte enriched transcript ( PfMGET ) , cDNA was synthesized from Total NA with the High Capacity cDNA Reverse Transcription Kit ( Applied Biosystems ) . Samples were added in a 1: one ratio to the mastermix . 2 µL of cDNA was run in a total volume of 20 µL making use of the GoTaq qPCR Master Mix ( Promega , Madison , Wisconsin ) . Male P . falciparum gametocytes were quantified using a standard curve of serially diluted StageV male gametocytes from the transgenic PfDynGFP/P47mCherry line ( Lasonder et al . , 2016 ) . Detailed information on the validation and performance characteristics of the assays can be found in the supporting materials ( Figure 5; Supplementary file 2 , 3; Figure 5—figure supplement 1 , 2 ) . The primary study outcomes were the frequency and magnitude of adverse events , and the prevalence of gametocytes by Pfs25 qRT-PCR . The prevalence of gametocytes is the presence of female gametocytes as measured by qRT-PCR targeting female-specific Pfs25 mRNA at any of the twice daily measurements from day 6 . Secondary outcomes were the peak density and time-point of peak density of male and female gametocytes , the AUC of gametocyte density , and assessment of the dynamics of gametocyte commitment , maturation and sex-ratio . The AUC of gametocyte density represents the total gametocyte exposure over time ( gametocyte load ) . Assessment of gametocyte infectivity to Anopheles stephensi mosquitoes by DMFA was an exploratory study endpoint . The sample size was calculated based on preliminary data that > 95% of the participants would develop gametocytemia . Conservatively , we considered the approach unsuitable for gametocyte induction if <50% of individuals developed mature gametocytes . We , therefore , powered the trial to estimate a 90% confidence interval around the proportion of gametocytaemic individuals that excludes 50% . If eight individuals ( allowing for one dropout per arm ) , and 6/7 or 7/7 of these individuals become gametocytaemic , we would be able to estimate this proportion with a lower limit of the 90% Wilson confidence interval ≥54 . 8% ( the lower limit of the 95% confidence interval being 48 . 7% ) . Differences between study arms were assessed by comparing mean values using a one-way ANOVA or non-parametric equivalents . To further identify which study arm ( s ) potentially deviated from others , we jointly estimated the differences between all four arms in a Bayesian framework ( standard linear regression model , no mixed effects ) , using Hamiltonian Monte Carlo as implemented in the R package rstanarm , and using an uninformative ( uniform ) prior for the explained variation ( R^2 ) ( see R codes used in Source code 1 ) ( Team SD , 2016 ) . For discrete variables ( e . g . the number of positive assays ) , the chi-squared test or Fisher’s exact test was used ( two-tailed ) . The total number of adverse events and total number of grade three adverse events were calculated per individual and compared by non-parametric Kruskal Wallis test . A previously developed model was used to estimate gametocyte half-life for female and male gametocytes separately ( Bousema et al . , 2010 ) . For this analysis , gametocyte observations were included from 12 days after the last detection of asexual parasites until the end of study . This was based on the gametocyte sequestration time of 10–12 days in this study , and the assumption that the number of newly released gametocytes would thus be minimal in this observation period . All model fittings were carried out using the PROC NLMIXED procedure in SAS ( Version 9 , SAS Institute Inc ) and included no covariates other than time ( see Source code 2 for SAS code ) . The AUC was computed by GraphPad Prism 5 ( USA ) with the ( X2-X1 ) * ( Y1 +Y2 ) /2 formula ( X = days post challenge; Y = gametocytes per mL ( ≥5 gametocytes/mL ) ) as used repeatedly for each adjacent pair of points defining the curve; the total AUC was used . | The parasite that causes malaria , named Plasmodium falciparum , has a life cycle that involves both humans and mosquitoes . Starting in the saliva of female Anopheles mosquitoes , it enters a person’s bloodstream when the insects feed . It then moves to the person’s liver , where it infects liver cells and matures into a stage known as schizonts . The schizonts then divide to form thousands of so-called merozoites , which burst out of the liver cells and into the bloodstream . The merozoites infect red blood cells , producing more schizonts and yet more merozoites , which continue the infection . To complete its life cycle , the parasite must return to a mosquito . Some of the parasites in the person’s blood transform into male and female cells called gametocytes that are taken up by a mosquito when it feeds on that person . Inside the mosquito , male and female parasites reproduce to create the next generation of parasites . The new parasites then move to the mosquito’s salivary glands , ready to begin another infection . Stopping the parasite being transmitted from humans to mosquitoes will stop the spread of malaria in the population . Yet it has proven difficult to study this part of the life cycle from natural infections . Here , Reuling et al . report a new method for generating gametocytes in human volunteers that will enable closer study of the biology of malaria transmission . The method is developed using the Controlled Human Malaria Infection ( CHMI ) model . Healthy volunteers without a history of malaria are bitten by mosquitoes infected with malaria parasites . Shortly afterwards , the volunteers are given a drug treatment to control and reduce their symptoms . The gametocytes form during this phase of the infection . At the end of the experiment , all the volunteers receive a final treatment that completely cures the infection . Reuling et al . recruited 16 volunteers and assigned them to four groups at random . Each group received a different drug regime . Roughly a week after the mosquito bites , all participants showed malaria parasites in their blood , and between 8 . 5 and 12 days later , mature gametocytes started to appear . This early appearance suggests that the parasites start to transform into gametocytes when they first emerge from the liver . The experiment also revealed that female gametocytes stay in the blood for a longer period than their male counterparts . These results are proof of principle for a new way to investigate malaria infection . The new model provides a controlled method for studying P . falciparum gametocytes in people . In the future , it could help to test the impact of drugs and vaccines on gametocytes . Understanding more about these parasites’ biology could lead to treatments that block malaria transmission . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"medicine",
"microbiology",
"and",
"infectious",
"disease"
] | 2018 | A randomized feasibility trial comparing four antimalarial drug regimens to induce Plasmodium falciparum gametocytemia in the controlled human malaria infection model |
The different segments of the nephron and glomerulus in the kidney balance the processes of water homeostasis , solute recovery , blood filtration , and metabolite excretion . When segment function is disrupted , a range of pathological features are presented . Little is known about nephron patterning during embryogenesis . In this study , we demonstrate that the early nephron is patterned by a gradient in β-catenin activity along the axis of the nephron tubule . By modifying β-catenin activity , we force cells within nephrons to differentiate according to the imposed β-catenin activity level , thereby causing spatial shifts in nephron segments . The β-catenin signalling gradient interacts with the BMP pathway which , through PTEN/PI3K/AKT signalling , antagonises β-catenin activity and promotes segment identities associated with low β-catenin activity . β-catenin activity and PI3K signalling also integrate with Notch signalling to control segmentation: modulating β-catenin activity or PI3K rescues segment identities normally lost by inhibition of Notch . Our data therefore identifies a molecular network for nephron patterning .
All adult vertebrates depend on renal nephrons accurately performing a diverse set of roles that protect and maintain blood homeostasis . The diversity of roles that the nephron performs is reflected in the range of symptoms of abnormal nephron function and the consequent diseases range from , for example , Bartter syndrome ( abnormal water and salt loss ) , acidosis , and rickets to essential hypertension ( Simon et al . , 1996; Schedl , 2007 ) . In spite of the importance of proper nephron function and segmentation , little is known regarding how the nephron patterns and how different specialised segments form during nephrogenesis . Thus far , only a handful of genes have been shown to control the development of specific nephron segments of which none have been explicitly connected to an explanation for how patterning is regulated along the whole length of the nephron . Nephrons form during embryonic development , when Wnt9b , secreted by the ureteric bud , activates a canonical β-catenin-mediated pathway in a population of overlying Six2+ mesenchymal nephron progenitor-cells ( Kobayashi et al . , 2008; Karner et al . , 2011; Park et al . , 2012; Das et al . , 2013 ) . In the canonical WNT pathway , WNT signalling results in the destabilisation of the GSK-3β/CK1α/AXIN2/APC complex and prevents the normal tagging of cytosolic β-catenin for degradation . Stabilised β-catenin translocates to the nucleus and together with members of the TCF family of transcription factors controls the expression of a wide range of target genes ( Clevers and Nusse , 2012 ) . One of these β-catenin target genes , Wnt4 , triggers a mesenchymal-to-epithelial transition ( MET ) ( Stark et al . , 1994; Kispert et al . , 1998 ) in the nephron progenitor cells and these reconfigure into an epithelial renal vesicle ( RV ) . Following the MET , the RV becomes polarised and connects to the ureteric bud at its distal end ( Georgas et al . , 2009 ) , and during subsequent developmental steps , the nephron starts to display additional signs of pattern-formation along its proximal–distal axis . Several distinct cell-populations form and these produce the different segments of the adult nephron ( Saxen , 1987 ) ; a Wt1+ cell population gives rise to proximal structures , a Jag1+ population to the medial part and Lgr5+ cells generate the distal nephron segments ( Armstrong et al . , 1993; Cheng et al . , 2003; Chen and Al-Awqati , 2005; Cheng et al . , 2007; Kreidberg , 2010; Barker et al . , 2012 ) . These segments are in turn further subdivided into functionally specialised portions , which express specific combinations of transmembrane transporters/channels for salts , glucose , and metals ( Raciti et al . , 2008 ) . How the differentiation of these segments is regulated remains unknown . The initiation of the nephron MET is driven by β-catenin signalling ( Kobayashi et al . , 2008; Karner et al . , 2011; Park et al . , 2012 ) , but the Wnt4 driven MET is most likely mediated by the non-canonical Ca2+–NFAT pathway ( Burn et al . , 2011; Tanigawa et al . , 2011 ) . It remains uncertain by what mechanism and at what precise stage the Six2+ cells or the RV develop distinct nephron segment lineages ( Lindstrom et al . , 2013 ) . Post-MET , Wnt9b acts via the planar cell polarity pathway and controls the orientation of cell division and the elongation of collecting tubules ( Karner et al . , 2009 ) . Wnt7b also has a role as it controls the development of the medulla and papilla of the kidney ( Yu et al . , 2009 ) . Notch signalling has previously been identified as being important for the formation of the proximal tubule ( Cheng et al . , 2003 , 2007 ) . Notch2−/− nephrons form no proximal tubules or glomeruli ( Cheng et al . , 2007 ) . However , ectopic expression of the intracellular and active Notch1-domain ( N1ICD ) in nephrons blocks glomerular development ( Cheng et al . , 2003 , 2007; Boyle et al . , 2011 ) . N1ICD expression in Six2+ cells can actually substitute for Wnt9b and trigger nephron induction and MET ( Boyle et al . , 2011 ) . Whether Notch or Wnt is important for the initial patterning of the nephron immediately post-MET remains to be determined . Using in vivo and ex vivo techniques we demonstrate that a gradient of β-catenin activity , along the proximal–distal nephron axis , controls the differentiation of segment-specific cell fates . We further investigate how β-catenin activity is prevented in the proximal and medial segments and show that BMP/PTEN/PI3K signalling in the medial nephron actively promotes the medial segment identity whilst blocking β-catenin activity . In addition , we show that modulating β-catenin or PI3K activity partially rescues the nephron segment defect phenotypes associated with the loss of Notch function . Our findings provide a model where multiple signalling pathways are integrated to control nephron segment-identity specification .
Regulation of β-catenin activity is essential for nephron induction and MET ( Davies and Garrod , 1995; Kuure et al . , 2007; Park et al . , 2007 ) . To determine whether β-catenin is involved in post-MET stages of nephron development , we tracked its activity in embryonic kidney organ cultures using a β-catenin signalling reporter mouse strain ( TCF/Lef::H2B-GFP; Ferrer-Vaquer et al . , 2010 ) . Confocal and time-lapse microscopy indicated strong activity of the reporter in the ureteric bud as described before ( Ferrer-Vaquer et al . , 2010; Burn et al . , 2011 ) . Importantly , we also detected different GFP intensities , reporting β-catenin activity , along the proximal–distal axis of the nephron . The nomenclature we use to describe the domains of the proximal–distal axis is as defined by the GenitoUrinary Development Molecular Anatomy Project ( gudmap . org ) for S-shaped body nephrons . Confocal z-projections ( Figures 1A , 3D rendering Video 1 ) of whole nephrons at an early stage of development show the signal being highest at the distal end of the nephron , where it connects to the ureteric bud , and gradually decreasing towards the proximal end . Time-lapse imaging of developing TCF/Lef::H2B-GFP expressing nephrons showed that the different GFP signal intensities propagated in a distal-to-proximal direction over time alongside the normal nephron growth and segmentation ( Figure 1—figure supplement 1A and Video 2 ) . Confocal imaging confirmed different GFP intensities in nephrons at later stages: S-shaped bodies ( Figure 1B and Figure 1—figure supplement 1B ) and more mature nephrons ( data not shown ) , and we consistently found that the podocytes and their precursors at the extreme proximal end of the nephrons were almost completely devoid of β-catenin activity ( Figure 1A , B , Figure 1—figure supplement 1B; Video 1 ) . We quantified the TCF/Lef::H2B-GFP signal in cells located in the distal , medial , and proximal segments of nephrons and plotted their intensities against their position . The segments were defined with antibodies for Jag1 ( medial segment; Chen and Al-Awqati , 2005; Georgas et al . , 2009 ) , Cdh1 ( distal segment; Cho et al . , 1998 ) , and by morphology . The TCF/Lef::H2B-GFP signal intensities showed an exponentially decreasing gradient ( R2 = 0 . 999; n = 11 nephrons ) suggestive of a single source and first-order decay of the activating signal ( Figure 1C ) . Jag1 immunofluorescent intensity data were used to indicate positions within nephrons . The GFP intensities measured in the distal segments differed from those in proximal segments by a minimum of a 15-fold difference to a maximum of a 72-fold difference ( mean = 39-fold difference; n = 10 nephrons; Figure 1C ) . 10 . 7554/eLife . 04000 . 003Figure 1 . β-catenin activity levels form a reversed gradient along the axis of the nephron . ( A–B ) TCF/Lef::H2B-GFP expression in nephrons: ( A ) late renal vesicle/early comma-shaped body nephron , ( B ) S-shaped body nephron , lines: white–nephron axis , purple–ureteric bud , green–distal nephron , red–medial nephron , blue–proximal nephron/glomerular precursors . Heat-maps display signal intensity in different nephron segments . ( C ) Quantification of nuclear H2B-GFP and cell-membrane Jag1 antibody stain signal-intensity along the proximal–distal axis . Error bars represent SEM of pixels representative of 10 µm segments . Right-hand side graph shows mean values for segments , as identified by H2B-GFP and Jag1 profiles ( n = 11 nephrons ) , error bars indicate SEM . p-values derived from t-tests . ( D ) Antibody stains against total β-catenin and phosphorylated β-catenin in S-shaped body nephron—Jag1 marking the medial segment . White dashed line indicating nephron axis . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 00310 . 7554/eLife . 04000 . 004Figure 1—figure supplement 1 . β-catenin reporter and antibody data show different β-catenin activity levels along the axis of the nephron . ( A ) TCF/Lef:H2B-GFP time-lapse data of a nephron developing through post-MET Pretubular Aggregate ( PTA ) , Renal Vesicle ( RV ) , Comma-shaped ( CB ) , and S-shaped ( SB ) stages . These data are shown as time-lapse in Video 2 . ( B ) S-shaped body nephron from Figure 1B shown at four different brightness settings that represent a doubling in brightness for each field going left to right . This clearly shows that all segments of the nephron are positive for the TCF/Lef:H2B-GFP reporter but follow a visual gradient . Green arrows point to positive cells in the dimmest portion of the nephron and an area of the kidney that even in at the brightest settings show up as negative . ( C ) Quantification of antibody stain for β-catenin phosphorylated at Ser33/Ser33/Thr41 in the distal , medial , and proximal segments ( nine measurements per nephron in five nephrons ) . Error bars represent SEM , p-values derived from t-tests . ( D–F ) Lef1 and Ccnd1 ( CycD1 ) antibody stains in S-shaped body nephrons; Pax2 is used as a structural marker against all nuclei within the nephron , Jag1 is used to detect the medial segment , Wt1 is used to detect the developing podocytes in the proximal segment . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 00410 . 7554/eLife . 04000 . 007Video 1 . 3D reconstructions of nephrons . 3D reconstruction of renal vesicle ( top ) , S-shaped body ( middle ) , and more mature nephron ( bottom ) . Nephrons are positive for TCF/Lef::H2B-GFP , Jag1-red , Cdh1-blue ( left panel ) and the TCF/Lef::H2B-GFP reporter is shown in a heat-map overlay ( right panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 00710 . 7554/eLife . 04000 . 008Video 2 . Time-lapse capture of nephron . The nephron is the same as shown in Figure 1—figure supplement 1A ( from a TCF/Lef::H2B-GFP reporter kidney ) is shown during the earliest stages of nephron development . Segments and stages are annotated based on the brightfield channel ( not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 008 To further assess the observed gradient in β-catenin activity , we analysed the expression and activity of the β-catenin protein directly . Using pan-β-catenin antibodies , we found β-catenin to be expressed throughout the developing nephrons ( Figure 1D ) . Antibodies for β-catenin that was already tagged for degradation ( phosphorylated at Ser33/Ser37/Thr41 ) , most strongly labelled the proximal domain ( twofold higher compared to medial and distal , p = 6 . 2 × 10−5 and 1 . 2 × 10−4 ) ( Figure 1—figure supplement 1C ) . Target genes of β-catenin ( Lef1 and Ccnd1 ) were also expressed along the gradient ( Figure 1—figure supplement 1D–F ) . The distribution of β-catenin protein and the expression of direct β-catenin target genes support the existence of the activity gradient . To investigate whether the β-catenin activity gradient has functional implications for the development of the nephron proximal–distal axis , we used inhibitors of GSK-3β ( CHIR99021 ) and Tankyrase ( IWR1 ) to positively and negatively modify β-catenin signalling ex vivo in whole organ cultures . We extensively characterised these small molecule inhibitors to ensure that they had their expected effect on the β-catenin signalling pathway; these data can be found in Figure 2—figure supplements 1–4 and Videos 3–4 and are briefly described here . First , we used the β-catenin reporter TCF/Lef::H2B-GFP and qRT-PCR analyses and confirmed that the inhibitors acted as expected ( Figure 2—figure supplement 1 ) . Second , because maximal activation of β-catenin has previously been suggested to be incompatible with MET ( Kuure et al . , 2007; Park et al . , 2007 , 2012 ) , we identified CHIR concentrations that induced nephron differentiation ( PAX2 and PAX8 expression ) but still permitted epithelialisation ( CDH1 expression; Figure 2—figure supplement 2 ) . Increased β-catenin signalling induced ectopic nephrons to form at the periphery of the kidneys and large nephron structures developed within the ureteric tree ( Figure 2—figure supplements 1 , 2 ) . Third , we demonstrated that these ectopic structures , just like the tree-bound structures , were derived from Six2+ nephron progenitors by fate-mapping these using a Six2-CreGFP ( Dolt et al . , 2013 ) and a Rosa26tdRFP Cre reporter mouse model ( Luche et al . , 2007 ) ( Figure 2—figure supplement 3 and Video 3 ) . Fourth , since pharmacological inhibitors can have multiple targets , we confirmed their specificity by combining activators and inhibitors of the WNT-pathway . Pax8-Cre lineage tracing and immunofluorescent analyses allowed us to show that by blocking β-catenin signalling downstream of GSK-3β ( by administering ICG001 which binds CBP and prevents β-catenin/CBP interaction ) , but not upstream ( IWR1 ) , CHIR-induced ectopic nephron formation was inhibited ( Figure 2—figure supplement 4A and Video 4 ) . Fifth , additional inhibitors against components of the β-catenin/Canonical WNT-pathway also confirmed our findings: ( BIO ( GSK3β inhibitor ) induced effects similar to CHIR; salinomycin ( LRP6 inhibitor ) induced effects similar to IWR1 ( Gandhirajan et al . , 2010; Lu et al . , 2011 ) ; Figure 2—figure supplement 4B ) . 10 . 7554/eLife . 04000 . 009Video 3 . Time-lapse capture of Six2GFPCre/Rosa26tdRFP kidneys . Kidneys cultured in control medium and CHIR medium . Timing and conditions as shown in videos . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 00910 . 7554/eLife . 04000 . 010Video 4 . Time-lapse capture of Pax8Cre YFPlox-stop kidneys . Kidneys cultured in control medium , IWR1 and CHIR medium , or ICG001 and CHIR medium . Timing and conditions as shown in videos . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 010 Using these inhibitors we modulated β-catenin signalling in Wt1+/GFP nephrons , where GFP labels the proximal cell population , allowing maturing podocytes within glomeruli to be recognized by their bright concentrated GFP signal ( Figure 2A , Video 5 ) . The Wt1+/GFP also labels the nephron progenitor cells and the whole of the pre-tubular aggregate during the MET , but at lower intensities , in accordance with the known expression pattern of Wt1 ( Kreidberg , 2010 ) . Time-lapse analysis showed that , by decreasing β-catenin activity , we favoured the differentiation of the proximal cell identity as was seen by an increase in maturation rate of glomeruli; in contrast , increased β-catenin activity blocked proximal identity formation and the formation of glomeruli ( Figure 2A–C ) . Podxl antibody staining ( Figure 2D ) , qRT-PCR analysis of additional markers for proximal differentiation ( Nphs2 , Synpo , Nphs1 , Podxl ) , and Wt1+/GFP reporter kidneys ( Figure 2—figure supplement 5A , B ) confirmed that these markers were expressed earlier when β-catenin activity was decreased and that they were repressed when β-catenin activity was increased . We consistently noticed a small number of individual cells in CHIR conditions that remained positive for Wt1 or Podxl ( Figure 2A , D ) . Releasing these cells from increased β-catenin activity , by removing CHIR , allowed them to resume their development ( Figure 2E ) . The proximal identity almost fully recovered and glomerular structures formed at almost the same size as those found in controls , show that the ectopically increased β-catenin activity had been actively suppressing the proximal identity ( Figure 2—figure supplement 5C ) . In contrast , removing IWR1 did not result in the reversal of the phenotype ( Figure 2—figure supplement 5D ) . 10 . 7554/eLife . 04000 . 011Figure 2 . Pharmacological modulation of β-catenin signalling alters proximal segment development . ( A ) Time-lapse analysis of treated Wt1+/GFP kidneys—same as shown in Video 5 . ( B ) Quantification of mean time taken for first glomeruli to mature to crescent-shaped stage where glomeruli are tightly packed and exhibit a bright signal . ( C ) Mean number of mature glomeruli after 3800 min of culture . ( D ) Kidneys stained for podocyte marker Podxl , β-catenin target Lef1 , and epithelial marker Cdh1 . Arrowheads indicating structures positive for Podxl . ( E ) The proximal identity resumed its formation when CHIR was removed after 48 hr—white arrowheads indicate larger Podxl positive structures , blue arrowheads indicate very small Podxl positive structures , dashed line separates ectopic nephron zone ( e . n . z . ) nephrons from those inside the ureteric tree ( UB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 01110 . 7554/eLife . 04000 . 012Figure 2—figure supplement 1 . β-catenin signalling is altered in response to pharmacological inhibitors . ( A–B ) TCF/Lef:H2B-GFP kidneys displaying low levels of β-catenin signalling in IWR1 conditions whereas strong activation of β-catenin signalling is detected in CHIR conditions . Typical nephrons within respective conditions are indicated with yellow arrowheads and nephrons captured at 60× and stained Jag1 and Cdh1 are shown in ( B ) . ( C ) Signal intensity of TCF/Lef:H2B-GFP β-catenin reporter is increased and decreased in the tips of the ureteric bud in response to CHIR and IWR1 , respectively . ( D ) qRT-PCR of known β-catenin target genes in response to IWR1 and CHIR treatment . Some β-catenin target genes responded as predicted ( Axin2 , Lef1 ) , others did not ( c-Myc and Ccnd1 ) as is expected from cell type-specific β-catenin targets ( Sansom et al . , 2005; Sansom et al . , 2007 ) —Axin2 and Lef1 data is also shown in Figure 4D to put them into the context of other expression changes . ( E ) Effects of different β-catenin activity levels on kidneys morphogenesis and nephron-formation . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 01210 . 7554/eLife . 04000 . 013Figure 2—figure supplement 2 . ‘Just-right’ β-catenin signalling levels drive MET . ( A ) Response of β-catenin target gene ( Lef1 ) , induction markers ( Pax2 and Pax8 ) , and epithelialisation marker ( Cdh1 ) to different CHIR concentrations . Inserts showing all channels for kidneys stained for Jag1 , Cdh1 , and indicated marker . As in previous studies , strong activation of β-catenin disrupted the normal MET . Arrowheads indicate ectopic nephrons . All error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 01310 . 7554/eLife . 04000 . 014Figure 2—figure supplement 3 . Ectopic nephrons form from Six2 expressing progenitors . Six2GFPCre with conditional RFP lineage tracing highlighting all nephron progenitors cell ( GFP+ ) and all nephron lineages ( RFP+ ) . Kidneys cultured in CHIR show ectopic RFP+ structures forming from previous GFP+ nephron progenitors . Dashed line indicates separation of ectopic nephrons from the ureteric bud tree and endogenous ‘tree-bound’ nephrons . Data also presented in Video 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 01410 . 7554/eLife . 04000 . 015Figure 2—figure supplement 4 . Co-inhibition experiments confirm specificity of pharmacological inhibitors . ( A ) Co-inhibition of GSK-3β ( CHIR ) and Tankyrase ( IWR1 ) or CBP ( ICG001 ) . ICG001 blocks TCF interacting with the CBP co-activator ( Emami et al . , 2004 ) downstream of ( CHIR ) . CHIR effects are not blocked by IWR1 inhibiting Tankyrase upstream of GSK-3β but are blocked by ICG001 . Dashed line separates ectopic nephrons from those adjacent to the ureteric bud . Error bars indicate SEM . Arrowheads indicate ectopic nephrons . ( B ) 7 . 5 µM BIO ( GSK-3β inhibitor ) mimics 1 . 5 µM CHIR and ectopic nephrons form and the UB ( ureteric tree ) branches is as in ( A ) . Arrowheads indicate ectopic nephrons , dashed line separates ectopic nephron zone ( e . n . z . ) nephrons from those inside the ureteric tree ( UB ) . ( B ) 100 nM salinomycin ( LRP6 inhibitor ) treated samples show UB branching similar to that caused by IWR1 . The effect on nephrons is also similar . Antibody stains and scale bars as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 01510 . 7554/eLife . 04000 . 016Figure 2—figure supplement 5 . The proximal cell-identity is promoted by decreased β-catenin signalling . ( A ) qRT-PCR data for genes indicative of terminally differentiated glomerular cells . ( B ) Wt1+/GFP kidneys stained for Pax8 and Cdh1–arrowheads indicating structures positive for GFP . ( C ) Size of glomerular structures in rescued nephrons as shown in Figure 2E . All error bars indicate SEM . ( D ) Kidney development in kidneys cultured for a full 96 hr in IWR1 or 48 hr in IWR1 followed by 48 hr in control medium . Antibody stains as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 01610 . 7554/eLife . 04000 . 017Video 5 . Time-lapse capture of Wt1+/GFP kidneys . Kidneys cultured in control conditions or treated with IWR1 or CHIR . Timing , conditions , and scale as specified . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 017 Distal cells express the epithelial stem-cell marker and β-catenin target Lgr5 ( Barker et al . , 2012 ) . To test whether the β-catenin gradient also controls the formation of the distal identity , we monitored Lgr5-EGFP expression in kidneys at different β-catenin activity levels using the Lgr5+/EGFP-IRES-CreERT2 mouse model ( Barker et al . , 2012 ) . Lgr5 was expressed as previously described ( Barker et al . , 2012 ) , although we primarily detected strong Lgr5-EGFP expression in a subsection of the distal domain adjacent to the medial segment ( Figure 3A ) . Lgr5 expression levels were increased at high levels of β-catenin activity . When β-catenin activity was decreased , the expression domain extended longer although the actual GFP signal was at lower levels compared to controls and samples where β-catenin activity was increased ( Figure 3A ) . We analysed the dynamics of Lgr5 expression under different β-catenin activity levels using time-lapse imaging of Lgr5+/EGFP-IRES-CreERT2 kidneys . The number of Lgr5 positive nephrons increased significantly at higher β-catenin activity levels ( 3 . 0×; p = 0 . 05 ) compared to controls or kidneys with decreased β-catenin activity ( Figure 3B , C ) . However , in samples with decreased β-catenin activity , Lgr5 positive nephrons emerged at an earlier time-point and these nephrons again appeared to be more elongated , but the GFP signal was lower compared to controls ( Figure 3B and Video 6 ) . Compared to controls , the actual number of Lgr5 positive nephrons remained unchanged in IWR1 treated samples ( Figure 3C ) . Lgr5 is also an R-spondin receptor and mediates β-catenin signalling ( de Lau et al . , 2011 ) . To test whether Lgr5 was functionally modulating the β-catenin signalling gradient , and thereby controlling nephron patterning , we intercrossed the Lgr5+/EGFP-IRES-CreERT2 knockin mice to homozygosity and analysed these for segmentation defects . Homozygous mutants displayed no obvious phenotype in the developing kidney ( Figure 3—figure supplement 1 and Video 7 ) . 10 . 7554/eLife . 04000 . 005Figure 3 . Pharmacological modulation of β-catenin signalling alters distal segment development . ( A ) Lgr5-EGFP expression in treated nephrons with segmentation markers . ( B ) Time-lapse analysis of treated Lgr5+/EGFP-IRES-CreERT2 kidneys–arrowheads indicate developing nephrons , red-dashed line indicates ureteric bud ( UB ) . ( C ) Mean number of Lgr5-EGFP positive nephrons per kidney . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 00510 . 7554/eLife . 04000 . 006Figure 3—figure supplement 1 . Lgr5 expression domain in heterozygous and homozygous Lgr5+/EGFP-IRES-CreERT2 kidneys . Cultured Lgr5+/EGFP-IRES-CreERT2 kidneys stained for segmentation marker Jag1 and epithelial marker Cdh1 . ( A and C ) Lgr5+/EGFP-IRES-CreERT2 heterozygous kidneys and kidneys and ( B and D ) Lgr5 EGFP-IRES-CreERT2/EGFP-IRES-CreERT2 homozygous kidneys and nephrons display normal segmentation . Homozygous tissue displays stronger EGFP expression as expected but nephrons appear morphologically normal . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 00610 . 7554/eLife . 04000 . 018Video 6 . Time-lapse capture of Lgr5+/EGFP-IRES-CreERT2 kidneys . Kidneys cultured in control conditions or treated with IWR1 or CHIR . Timing , conditions , and scale as specified . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 01810 . 7554/eLife . 04000 . 019Video 7 . Time-lapse capture of Lgr5+/EGFP-IRES-CreERT2 and Lgr5 EGFP-IRES-CreERT2/EGFP-IRES-CreERT2 kidneys . Kidneys cultured in control conditions over 6 days . Timing and scale as specified . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 019 Collectively , these data confirm that the cell populations along the proximal–distal axis of the nephrons are responsive to changes in β-catenin signalling and respond positively and negatively as would be predicted from the proposed β-catenin activity gradient . Having shown that the proximal and distal cell populations are both affected by increased β-catenin activity and the proximal is affected by decreased β-catenin activity , we hypothesized that the different levels of β-catenin signalling within the gradient specify the positions of identities in the nephron . If so , modulating β-catenin signalling might impose abnormal distal or proximal identities along the axis of the developing nephron ( Figure 4A ) . We immunostained inhibitor treated nephrons with markers for proximal , medial , and distal identities ( Wt1 , Jag1 , Cdh1 ) and measured the size of the domains in which each was expressed ( Figure 4B ) . Increasing β-catenin did mildly increase total nephron lengths ( 1 . 17× , p = 0 . 017 ) compared to controls ( Figure 4C ) , but the distal segment was significantly longer ( 1 . 8× , p = 0 . 0003 ) and the proximal segment was severely reduced ( 0 . 2× , p = 1 . 5 × 10−6 ) . The medial segment showed no significant change in length ( 1 . 3× , p = 0 . 067 ) . These data confirm that both the distal and proximal nephron segments respond to increased β-catenin signalling according to our hypothesis , but the medial remained unchanged in size . Decreasing β-catenin signalling , on the other hand , resulted in much elongated nephrons ( Figure 4B , C: 1 . 7× , p = 1 . 8 × 10−14 ) , as we had previously noticed ( Figure 3A ) . Here , the length of the proximal segment was unchanged compared to controls ( 1 . 0× , p = 0 . 93 ) , but as we showed above , this segment was more mature in appearance and gene expression ( Figure 2 ) . The elongation of the nephron was primarily due to the increases in both the distal ( 2 . 5× , p = 0 . 0002 ) and the medial segments ( 1 . 7× , p = 0 . 0007 ) and the nephrons appeared thinner . Although nephron identities responded to increased β-catenin as our model predicted , the large morphological changes obscured any subtle changes in segmentation in those nephrons with decreased β-catenin activity . To address this , we tested whether a gradual increase in β-catenin activity , which only mildly shifted activity levels away from the normal , would give a gradual decrease in proximal identity as would be expected if identities are β-catenin dosage-dependent . This was observed in nephrons that we exposed to different incremental concentrations of CHIR ( Figure 4—figure supplement 1 ) . qRT-PCR analysis of RNA from whole treated kidneys confirmed that a mild increase in β-catenin activity promoted expression of most distal segment genes ( Wnt4 , Pax8 , Fgf8 , Lhx1 , Lgr5 ) ( Figure 4D ) thus mirroring the effect of decreasing β-catenin activity; surprisingly , Pax2 did not respond similarly to Pax8 . Medial segment control genes ( Jag1 , Dll1 , Heyl , and Irx2 ) did not increase in response to CHIR as distal genes did . At the highest β-catenin activity levels used [6 µM] , nephron formation was reduced ( Figure 2—figure supplements 1–2 ) and the expression profile thus dropped as would be expected ( Figure 4D ) . To test if the observed phenotypes were due to a direct effect on the nephrogenic lineage or an indirect effect via the ureteric bud , we isolated metanephric mesenchyme away from the ureteric bud , induced it to form nephrons with spinal cord ( Grobstein , 1953 , 1955 ) , and modulated β-catenin activity therein . Comparable to the phenotype observed in whole kidney rudiments , we found that inhibition of β-catenin activity favoured patterning towards the proximal fate , whereas increasing its activity had the opposite effect ( Figure 4E ) . 10 . 7554/eLife . 04000 . 020Figure 4 . Shifts in positional identity by altered β-catenin activity . ( A ) Model of predicted changes in segmentation if the gradient of β-catenin activity specifies positional identities in the nephron . Nephrons depicted as spheres representing renal vesicle stage . Dashed line indicates nephron segments . Gradient bar indicates β-catenin activity . ( B ) Antibody stains against segment specific markers in nephrons with different β-catenin signalling conditions . ( C ) Proximal , medial , and distal nephron domain-sizes in Control , CHIR , and IWR1 treated kidneys . Mean values and SEMs indicated within bars on graph . ( D ) qRT-PCR analysis of markers for nephron induction displayed as a heat-map with information displayed in figure . The RNA was isolated after 48 hr of culture from whole kidneys . ( E ) Antibody stains on nephrons developed in isolated mesenchyme . ( F and G ) Apc+/1638N Ctnnb1Y654/E654 kidneys where Ctnnb1Y654 is the wild-type allele . Kidneys characterised using anti-Wt1 , Jag1 , Podxl , Lam , and Cdh1 . Arrowheads and boxed area indicate ectopic nephrons in E–F . Dashed lines separate metanephric and mesonephric regions in E–F . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 02010 . 7554/eLife . 04000 . 021Figure 4—figure supplement 1 . Gradual shifts in positional identity by gentle changes in β-catenin activity . ( A ) Typical nephrons for their conditions as cultured at incremental CHIR dosages ( 0 µM , 0 . 50 µM , 0 . 75 µM , 1 µM , 1 . 25 µM , and 1 . 5 µM ) . Dashed lines indicate the axis and lengths of nephron segments . Kidneys stained for Wt1 , Jag1 , and Cdh1 . UB–ureteric bud . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 02110 . 7554/eLife . 04000 . 022Figure 4—figure supplement 2 . β-catenin activity dosage-dependent phenotypes in series of Apc and Ctnnb1 models . Kidneys characterised using anti-Wt1 , Jag1 , Podxl , Lam , and Cdh1 . ( A ) Ctnnb1Y654/E654 and Ctnnb1E654/E654 , ( B ) Apc+/1572T Ctnnb1Y654/Y654 and Apc +/1572T Ctnnb1Y654/E654 . Ctnnb1Y654 is the wild-type allele . Dashed lines separate metanephric and mesonephric regions . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 022 Previous genetic approaches to investigate the role of β-catenin during nephron formation have been hampered by the problem that when β-catenin is maximally activated in nephron progenitors MET is concurrently blocked ( Kuure et al . , 2007; Park et al . , 2007 ) . Our data , however , suggest that MET is only incompatible with very high levels of β-catenin activity ( Figure 4D and Figure 2—figure supplements 1–2 ) , therefore identifying a dose–response correlation between β-catenin activity and renal MET . To confirm this , and to test if nephron segmentation defects could also be generated by genetic means , we made use of Apc and Ctnnb1 ( the gene encoding β-catenin ) mutants that only result in mild increases in β-catenin activity . The Ctnnb1E654 model ( van Veelen et al . , 2011 ) expresses a Y654E mutation in the endogenous β-catenin gene ( Ctnnb1 ) leading to an increased propensity to signal , whereas the Apc1572T and Apc1638N models ( Fodde et al . , 1994; Gaspar et al . , 2009 ) carry endogenous Apc-truncations associated with low or moderate activation of β-catenin signalling , respectively . Such hypomorphic alleles by themselves or in combinations can provide a series of increasing β-catenin activity levels ( Kielman et al . , 2002; Buchert et al . , 2010 ) . We generated different combinations of these alleles that display increasing levels of β-catenin activity in order of increasing β-catenin activity: Ctnnb1+/E654 , Ctnnb1E654/E654 , Apc+/1572T Ctnnb1+/E654 , and Apc+/1638N Ctnnb1+/E654 ( Gaspar et al . , 2009; van Veelen et al . , 2011 ) . We cultured E12 . 5 kidney rudiments for 5 days and stained them for different compartments of the kidneys with antibodies against Wt1 , Jag1 , Podxl , Lam , and Cdh1 . Ctnnb1+/E654 and Ctnnb1E654/E654 kidneys ( Figure 4—figure supplement 2 ) were indistinguishable from wild-type ( Ctnnb1+/+ ) kidneys ( data not shown ) . Kidneys from Apc+/1572T Ctnnb1+/+ and Apc+/1572 Ctnnb1+/E654 embryos had normal nephrons ( Figure 4—figure supplement 2 ) , though some kidneys developed supernumerary ureteric buds ( data not shown ) . The mutants with the highest level of β-catenin activity , Apc+/1638N Ctnnb1+/E654 displayed the most severe phenotype . The kidneys from such embryos showed effects ranging from those with severely reduced and morphologically abnormal branching and growth ( Figure 4F , G ) to those with more normal ureteric bud trees . Significantly , a small number of nephrons formed ectopically away from the ureteric bud in several kidneys . The ectopic nephrons were found to be positive for Jag1 , Cdh1 , and laminin but lacked podocytes , as shown by the absence of Wt1 and Podxl signal ( Figure 4F , G right panels , respectively ) , in accordance with our pharmacological data presented above . A few endogenous nephrons within the ureteric bud tree managed to produce proximal domains ( Figure 4F , G ) . The ectopic nephrons lacking proximal domains together with the absence of β-catenin signalling in this end of the nephrons ( Figure 1 ) and our inhibitor experiments , strongly suggest that increased β-catenin activity disrupted normal patterning of the developing nephron and blocked the development of the most proximal end . The canonical Wnt signalling pathway is known to have a direct effect on cell proliferation ( Davidson and Niehrs , 2010 ) . We tested if this was part of the mechanism via which β-catenin controls nephron patterning . Neither increasing nor decreasing β-catenin signalling levels changed the number of mitotic nuclei per nephron ( Figure 5A , B ) nor did blocking cell proliferation with methotrexate ( MTX; Chabner and Young , 1973 ) affect the patterning of the nephron . Nephrons that formed in MTX were much smaller compared to control nephrons but expressed Cdh1 , Wt1 , and Jag1 correctly ( Figure 5C ) , and the ratio of the size of the medial and proximal segment was unchanged ( Figure 5D–F ) . Moreover , nephrons treated simultaneously with MTX and either CHIR or IWR produced nephrons with the same segmentation changes as caused by IWR1 or CHIR on their own ( Figure 5—figure supplement 1A ) , in spite of the nephrons now being smaller as a result of the MTX treatment . Finally , MTX treatment did not affect the existence of the β-catenin activity gradient as observed with the TCF/Lef::H2B-GFP reporter . Moreover , co-inhibition with MTX and either IWR1 or CHIR still decreased and increased the reporter signal , respectively ( Figure 5—figure supplement 1B–D ) . 10 . 7554/eLife . 04000 . 023Figure 5 . Modulating β-catenin activity shifts positional identities along the nephron without altering proliferation . ( A ) Proliferation in TCF/Lef::H2B-GFP expressing nephrons treated with CHIR and IWR1 . Nephron axis–dashed white line . Phosphorylated Histone 3 used as a marker for mitotic cells . ( B ) Quantification of mitotic nuclei per nephron . ( C ) Effect of Methotrexate ( MTX ) on nephron development and patterning . Nephrons outlined with white dashed line . ( D–F ) Measurement of Jag1+ and Wt1+ segment sizes of nephrons in control , 250 nM Methotrexate ( MTX ) , and 500 nM MTX conditions . All error bars indicate SEM . p-values generated using Student's t test . Scale bars and antibodies as indicated on fields . UB–ureteric bud . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 02310 . 7554/eLife . 04000 . 024Figure 5—figure supplement 1 . Modulating β-catenin activity shifts positional identities along the nephron regardless of proliferation levels . ( A ) CHIR and IWR1 alter segmentation similarly with or without MTX . Dashed-white line indicates nephron axis . ( B–C ) Inhibiting proliferation does not block the formation of a GFP-gradient in TCF/Lef:H2B-GFP expressing nephrons . Nephrons outlined with white dashed line . Segment labelled on images . Red dots indicate examples of nuclei that were quantified for graph ( C ) . Error bars indicate SEM . p-values were calculated using t-tests . ( D ) Signal intensity of TCF/Lef:H2B-GFP β-catenin reporter is decreased and increased with IWR1 and CHIR also , when MTX is added . Nephrons outlined with white dashed line . Scale bars and antibodies as indicated on fields . UB–ureteric bud . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 02410 . 7554/eLife . 04000 . 025Figure 5—figure supplement 2 . Modulating β-catenin activity shifts positional identities along the nephron regardless of apoptosis levels . ( A ) Treated kidneys stained with TUNEL-assay to detect apoptotic cells . ( B ) Treated kidneys ( live ) stained with Annexin V assay to detect apoptotic cells . 6-CF marks proximal tubules and PT and as specified . Scale bars and antibodies as indicated on fields . UB–ureteric bud , N–Nephron . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 025 Treatment with neither IWR1 nor CHIR altered apoptosis in the nephrons ( Figure 5—figure supplement 2 ) . Only very low levels of apoptosis were detected , primarily at the periphery of the kidneys , in line with previous data ( Foley and Bard , 2002 ) . CHIR treated kidneys did display some apoptotic nuclei , but these were located in the mesenchymal progenitor population surrounding ureteric bud tips . Collectively , these data indicate that changes in proliferation and apoptosis are not the driving force behind the β-catenin signalling gradient or the patterning of the nephron . In the intestine BMP negatively regulates β-catenin signalling by inhibiting the PTEN/PI3K/AKT pathway to maintain high levels of active GSK3β ( He et al . , 2004 ) . BMP2/4/7 are expressed in the nephron ( Oxburgh et al . , 2011 ) and the BMP/SMAD pathway is active ( Blank et al . , 2008; Brown et al . , 2013 ) ; albeit with an unknown function . We confirmed BMP pathway activity in the medial segment of the nephron using pSMAD1/5/8 reporter BRE-LacZ ( Blank et al . , 2008 ) ( Figure 6A ) . This closely correlated with the Jag1+ domain ( Figure 6A ) . BMP activity was further confirmed using antibodies specific for pSMAD1/5/8 ( Figure 6B ) . 10 . 7554/eLife . 04000 . 026Figure 6 . The β-catenin activity gradient is modified by changes to BMP and PI3K signalling . ( A ) BRE-LacZ pSMAD reporter shows strong labelling in medial segment; co-stained for Wt1 , Jag1 , Cdh1 . ( B ) pSMAD1/5/8 specific antibody stain . Lines and labelling in A–B indicate different segments . ( C ) pPTEN and pAKT levels in the nephron after inhibition of BMPR with 4 µM LDN-193189 or PI3K with 20 µM Ly294002 . White arrowheads and arrows indicate staining in ureteric bud and nephrons , respectively . ( D ) Time-lapse of single TCF/Lef::H2B-GFP positive nephrons developing from induction stage through S-shaped body stage . Nephron axis in red . Nephrons treated as specified . CM -cap mesenchyme , UB–ureteric bud . ( E ) Quantification of TCF/Lef::H2B-GFP intensities at different positions along the proximal–distal axis at S-shaped body stage/22 hr . Multiple nephrons used as indicated on graph . Error bars indicate SEM . Average lengths of nephrons at 22 hr for each condition are indicated on the graph . ( F ) Inhibition of BMPR or PI3K alters the medial segment negatively and positively , respectively . ( G ) qRT-PCR data of segment-specific markers and β-catenin target genes displayed as a heat-map with gene information displayed in figure . ( H ) p-β-catenin Ser552 localisation in S-shaped nephron . Antibody stains and scale bars as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 02610 . 7554/eLife . 04000 . 027Figure 6—figure supplement 1 . Rescue/reversals experiments for kidneys treated with 20 µM Ly294002 or 4 µM LDN193189 . Kidneys were cultured for 96 hr in the inhibitors or for 48 hr in inhibitor followed by another 48 hr in control medium . Kidneys were stained for Wt1 , Jag1 , and Cdh1 to display nephron formation and overall morphology . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 02710 . 7554/eLife . 04000 . 028Figure 6—figure supplement 2 . Changes to BMP and PI3K signalling alters nephron segmentation and β-catenin signalling . ( A ) Data from time-lapse captured TCF/Lef:H2B-GFP nephrons treated with LDN-193189 or Ly294002 or as controls . The percentage frequency is plotted against pixel intensity values . The GFP intensity was measured for all pixels throughout the nephron at 22 hr of culture . Ly294002 treated nephrons had fewer bright nuclei whilst LDN-193189 and control nephrons were not distinguishable at this stage with this method . Error bars indicate SEM . ( B ) Whole TCF/Lef:H2B-GFP kidneys treated with LDN-193189 , Ly294002 , CHIR , and combinations thereof . Yellow and red dashed circles indicate ureteric bud tree-bound endogenous and ectopic nephrons , respectively . ( C ) Inhibition of BMPR , PI3K , and activation of β-catenin predictably alters the medial and proximal segments . Dashed line indicates separation between ectopic and endogenous ureteric bud tree-bound nephron structures . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 028 To test if BMP signalling could inhibit β-catenin signalling via the PTEN pathway as it does in the intestine ( He et al . , 2004 ) we first investigated if BMP can signal via PTEN in the kidney . To do this we blocked BMP signalling at the BMP receptor ( BMPR ) level with inhibitor LDN-193189 ( Boergermann et al . , 2010 ) and PI3K with inhibitor Ly294002 ( Vlahos et al . , 1994; Gharbi et al . , 2007 ) . Inhibiting PI3K is equivalent to further activating any existing BMP/PTEN signalling ( He et al . , 2004 ) . If BMP signals via PTEN , the expected outcome of BMPR inhibition would be a rise in the levels of phosphorylated PTEN ( inactive ) and phosphorylated AKT ( active ) ( He et al . , 2004 ) , which was indeed found in nephrons and the ureteric bud ( Figure 6C ) . We tested whether the inhibitors had reversible effects by releasing the kidneys from inhibition after 48 hr of treatment . A moderate recovery towards the control phenotype was observed after removal of the inhibitors ( Figure 6—figure supplement 1 ) . To test if BMP and PI3K signalling in the nephron affects β-catenin activity , we treated TCF/Lef::H2B-GFP kidneys with both inhibitors . If the system operates as in the intestine , BMPR inhibition should prevent BMP/PTEN signalling from antagonising β-catenin activity thus resulting in increased β-catenin reporter activity ( He et al . , 2004 ) . Inhibition of PI3K should do the opposite . The total β-catenin reporter activity as measured throughout nephrons , 22 hr after nephron formation , was indistinguishable between controls and BMPR inhibited nephrons , but PI3K inhibition reduced β-catenin reporter activity levels as expected ( Figure 6—figure supplement 2A ) . The time-lapse data did however visually show higher and lower TCF/Lef::H2B-GFP activity in the BMPR and PI3K inhibited nephrons , respectively , as we had anticipated ( Videos 8 , 9 top part ) . Thus , we monitored the β-catenin reporter of each nephron over the whole 22 hr time-span and measured the activity along the whole distal-to-proximal axis ( Figure 6D , E ) . Nephrons were randomly picked during the whole time-span of the kidney being monitored ( 96 hr ) and the averages were determined for each treatment . This showed that the high to medium β-catenin reporter activity extended ectopically into the middle portion of the nephron when BMPR was inhibited and low levels were detected when PI3K was blocked ( Figure 6D , E and Video 8 , top part ) . 10 . 7554/eLife . 04000 . 029Video 8 . Time-lapse capture TCF/Lef::H2B-GFP kidneys . Kidneys cultured in control medium , 4 µM LDN-193189 , 20 µM Ly294003 , 1 . 5 µM CHIR , 4 µM LDN-193189 , and 1 . 5 µM CHIR , or 20 µM Ly294003 and 1 . 5 µM CHIR . Timing and scales are as specified . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 02910 . 7554/eLife . 04000 . 030Video 9 . Time-lapse capture TCF/Lef::H2B-GFP kidneys . Kidneys cultured in control medium , 4 µM LDN-193189 , 20 µM Ly294003 , 1 . 5 µM CHIR , 4 µM LDN-193189 , and 1 . 5 µM CHIR , or 20 µM Ly294003 and 1 . 5 µM CHIR . Timing and scales are as specified . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 030 Moreover , when BMPR or PI3K were inhibited , nephrons elongated at different rates compared to controls; 11 . 2 µm/hr ( BMPR ) , 10 . 6 µm/hr ( PI3K ) , and 8 . 4 µm/hr ( control ) ( data not shown ) . Given the observed increase of the β-catenin reporter in response to blocking BMPR , by inhibiting BMPR and simultaneously activating β-catenin with CHIR , this should synergistically expand β-catenin signalling in the medial nephron segment where BMP activity was detected . The ectopic and tree-bound nephrons that formed in these dual-inhibitor conditions displayed very high β-catenin activity throughout the nephron structures ( Figure 6—figure supplement 2B and Videos 8 , 9 , lower part ) . Interestingly , co-inhibition of PI3K and activation of β-catenin also positively affected β-catenin reporter activity; however , the dynamics of this was clearly different from that caused by inhibition of BMPR and activation of β-catenin with CHIR ( Videos 8 , 9 , lower part ) . To test if segmentation was altered by inhibition of BMPR or PI3K , we stained kidneys for Wt1 , Jag1 , and Cdh1 . Activating β-catenin and inhibiting BMPR at the same time completely removed the Wt1+ cells , strongly reduced the Jag1+ cells , and formed nephrons made of distal regions ( Figure 6—figure supplement 2C ) . Activating β-catenin and inhibiting PI3K at the same time made nephrons form large medial/Jag1+ segments with few WT1+ proximal cells present ( Figure 6—figure supplement 2C ) . These data suggest that PI3K inhibition positively promotes a medial fate . Inhibiting PI3K on its own strongly promoted the development of the Jag1+ segment and inhibiting BMPR had a slight negative effect on the Jag1+ segment as expected , the latter perhaps partially hidden by the overall increase in nephron length ( Figure 6F and Figure 6—figure supplement 2C ) . qRT-PCR analyses on segment-specific genes confirmed that inhibiting PI3K had a positive effect on the medial domain; medial markers Dll1 , Jag1 , Irx2 , and Fgf8 increased , as did distal marker Lhx1 ( Figure 6G ) . Inhibition of BMPR positively affected distal markers Bmp2 and Lgr5 expression and also some medial markers Heyl and Irx2 ( Figure 6G ) . Pax2 and Wnt4 were both down-regulated . Co-inhibiting BMPR and PI3K resulted in nephrons highly similar to those in just PI3K-inhibitor conditions confirming that PI3K is downstream of BMPR and that PI3K can positively affect the medial segment independently of BMPR/SMAD ( data not shown ) . Since both β-catenin signalling and nephron patterning was modulated via the PI3K pathway , we tested whether AKT positively regulates β-catenin signalling via phosphorylation of β-catenin at Ser552 , as demonstrated elsewhere ( Fang et al . , 2007 ) . β-catenin phosphorylated at Ser552 was localised primarily to the apical surfaces of the nephrons in the distal and medial segments ( Figure 6H ) . In a subset of cells , the phosphorylated Ser552 β-catenin was detected at lateral and basal surfaces . 80% of such cells were found in the distal and medial segments ( 45/56 positive cells counted in five nephrons ) and the remaining 20% in proximal cells . Unlike phosphorylated PTEN and phosphorylated AKT , which responded to PI3K and BMPR inhibition , phosphorylated Ser552 β-catenin did not change on inhibition of either ( data not shown ) . These data indicate that the positional identities in the nephron can be modified through BMP mediated PI3K/AKT signalling and that simulated increased BMP signalling ( via inhibition of PI3K ) leads to reduced β-catenin signalling , whereas simulated decreased BMP signalling ( via inhibition of BMPR ) results in the opposite effect . To date , the only other major pathway known to control nephron patterning is the Notch-signalling pathway ( Cheng et al . , 2007 ) . Notch-signalling specifies the proximal and medial identity , but it can also block the adjacent glomerular identity when over-activated in these cells ( Cheng et al . , 2007; Boyle et al . , 2011 ) . In many tissues , Notch and Wnt pathways act both in synergy and in opposition ( Andersson et al . , 2011 ) , and Notch has also been reported to antagonise the PTEN/PI3K/AKT pathway in clear-cell renal-cell carcinoma ( Liu et al . , 2013 ) . Our data already pointed to a negative correlation between PI3K activity and the expression of Notch ligands Dll1 and Jag1 . Furthermore , we showed activity of the BMP pathway in the segment affected when Notch is absent . Because we have shown that decreased β-catenin activity enhances the formation of the glomerular segment and PI3K inhibition promotes the medial segment , we tested if inhibition of β-catenin or PI3K could rescue these respective identities when Notch activity is lost . We first inhibited Notch using the γ-secretase DAPT as previously published ( Cheng et al . , 2003 ) . This resulted in a loss of proximal and medial structures and glomerular precursors ( positive for Podxl and Lotus tetragonolobus lectin ( LTL ) , Figure 7A ) and these effects mimic those phenotypes seen in Notch2−/− animals ( Cheng et al . , 2007 ) . Removal of the inhibitor led to a partial recovery of proximal segments ( Figure 7—figure supplement 1 ) . Co-inhibition of β-catenin and Notch-signalling resulted in the reappearance of nephrons positive for LTL and Podxl that displayed glomerular structures and medial and proximal development ( Figure 7A ) . The Wt1+/GFP reporter kidneys in time-lapse ( Figure 7B ) and antibody stains that identify these segments ( Figure 7C ) confirmed that by decreasing β-catenin activity in Notch-inhibited kidneys we partially rescued the absence of normal Notch-signalling . The inhibition of β-catenin did not restore the expression of key Notch target genes ( Jag1 , Dll1 , Heyl , Hey1 ) ( Figure 7D ) . Co-inhibition of PI3K and Notch also resulted in a partial rescue ( Figure 7—figure supplement 2A and Videos 10 , 11 ) . Wt1+ cells were not evident but Jag1 was again expressed , indicating that PI3K exerts an effect specific to the medial segment ( Figure 7—figure supplement 2A ) . We assessed how inhibition of Notch , with or without inhibition of β-catenin and PI3K signalling , affected β-catenin activity using the β-catenin reporter TCF/Lef::H2B-GFP . Inhibition of Notch resulted in stunted nephrons without Wt1+ or Jag1+ cells with medium to high level of β-catenin activity throughout ( Figure 7—figure supplement 2B–C; Video 11 ) . Co-inhibition of Notch and β-catenin signalling strongly decreased the β-catenin activity and Wt1+ proximal cells were again present . Co-inhibition of Notch and PI3K increased the size of the nephrons and Jag1+ cells were again present and the β-catenin reporter was slightly decreased . Triple inhibition of Notch , PI3K , and β-catenin signalling did not result in an improved rescue ( data not shown ) . Together , these data show that proximal and medial nephron cells require Notch signalling and simultaneously need β-catenin and PI3K signalling to be kept at low levels , respectively . 10 . 7554/eLife . 04000 . 031Figure 7 . Altered β-catenin activity rescues the loss of Notch . ( A ) Kidneys treated with DAPT and DAPT/IWR1 and stained for LTL , β-laminin , Cdh1 , and Podxl–arrowheads indicate LTL-positive nephrons , inserts show magnified nephrons with Podxl staining for podocytes , yellow line outlines nephron tubules . ( B ) Time-lapse analysis of Wt1+/GFP kidneys treated with DAPT and DAPT + IWR1–arrowheads show GFPHIGH structures in developing proximal segments , red dashed line indicates ureteric bud positions ( UB ) . ( C ) Structures positive for Jag1 and Wt1 in treated kidneys–arrowheads indicating double-positive structures . ( D ) qRT-PCR data for Notch target genes ( Jag1 , Dll1 , Heyl , Hey1 ) . All error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 03110 . 7554/eLife . 04000 . 032Figure 7—figure supplement 1 . Rescue/reversals experiments for kidneys treated with 2 µM DAPT . Kidneys were cultured for 96 hr in DAPT or for 48 hr in DAPT followed by another 48 hr in control medium . Kidneys were stained for Wt1 , Jag1 , and Cdh1 to display nephron formation and overall morphology . Antibody stains indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 03210 . 7554/eLife . 04000 . 033Figure 7—figure supplement 2 . Altered β-catenin activity or PI3K signalling rescues the loss of Notch . Data from kidneys treated with 2 µM DAPT and in combination with 20 µM Ly294002 or 2 µM IWR1 . ( A–B ) TCF/Lef:H2B-GFP nephrons stained for WT1 , JAG1 , and CDH1 . WT1 and JAG1 stains are not optimally compatible with the PFA fixation required to preserve the GFP signal . Thus , wild-type kidneys were also treated and stained for WT1 JAG1 and CDH1—shown in ( C ) . Jag1+ structures indicated with arrowheads . Stains , scales , and treatments as specified . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 03310 . 7554/eLife . 04000 . 034Video 10 . Time-lapse capture of TCF/Lef::H2B-GFP kidneys . Kidneys in this video were cultured in control medium or 20 µM Ly294003 . Timing and scales are as specified . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 03410 . 7554/eLife . 04000 . 035Video 11 . Time-lapse capture of TCF/Lef::H2B-GFP kidneys . Kidneys in this video were cultured in 2 µM DAPT or 20 µM Ly294003 with 2 µM DAPT . Timing and scales are as specified . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 035
A major outstanding question in kidney development has been to understand how the nephrons are patterned . It has been shown that Wnt via canonical β-catenin signalling induces nephrons to form and through non-canonical pathways regulate the nephron mesenchymal-to-epithelial transition as well as subsequent tubule elongation ( Davies and Garrod , 1995; Kuure et al . , 2007; Park et al . , 2007; Karner et al . , 2009; Burn et al . , 2011; Tanigawa et al . , 2011 ) . Our data now show that β-catenin activity is essential throughout nephron development by regulating the patterning of the nephron through interactions with the BMP , PTEN/PI3K , and Notch pathways ( Figure 8 ) . 10 . 7554/eLife . 04000 . 036Figure 8 . Model for molecular pathways interacting to control the patterning of the nephron . β-catenin activity is necessary to determine a distal cell identity but must be excluded from the proximal nephron . BMP/PTEN/PI3K antagonises β-catenin activity in the medial segment and positively promotes a medial fate and Notch ligand expression . Notch is essential for medial and proximal development . Glomerular progenitor cells strongly inhibit β-catenin function , possibly via a WT1-dependent mechanism as in Sertoli cells ( Chang et al . , 2008 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04000 . 036 Our data combined suggest a model where a gradient of β-catenin activity specifies positional identities along the nephron axis . Additional signal transduction pathways modulate local β-catenin activity levels . Its action has a differential effect on nephron segments by controlling differentiation , maturation , and their size . While cell proliferation and apoptosis are likely to be essential in other aspects of nephrogenesis , our data exclude a direct role in the initial patterning mechanism . With the current data , it is difficult to identify key downstream targets and that way describe the next level of mechanism of this process , as many targets are also widely used as segmentation markers . Using a variety of methods , our combined data strongly suggest the existence of a gradient in β-catenin activity along the proximal–distal axis of the developing nephron . First , the TCF/Lef::H2B-GFP activity reporter showed a clearly recognisable visual gradient which we confirmed by quantifying the reporter signal along the nephron axis . Reporter activity is high at the distal end of the nephron , where it connects to the ureteric bud , and gradually reduces in the proximal direction , with the developing glomerulus at the extreme proximal end being devoid of β-catenin activity ( Figure 1A–C ) . Second , while antibody staining for pan β-catenin showed equal expression of the protein throughout the post-MET nephron , staining with phospho-specific antibodies ( indicative of protein targeted for break-down ) was greatly enhanced in the proximal end of nephrons , the same segment where the reporter is least active . This also indicates that the gradient is an activity gradient , not an expression gradient . Third , the pharmacological and genetic modulation of the β-catenin signal result in a phenotype that can be predicted from the reporter and antibody data . In isolation , none of these analyses would prove the existence of a β-catenin activity gradient . Reporters of the type used here can sometimes give unreliable data , however , the model we used was found to give a very reproducible activity patterns that overlaps with many other β-catenin activity reporters ( Ferrer-Vaquer et al . , 2010 ) . Phospho-specific antibodies are powerful tools to monitor activity of signal transduction pathways , but their use can be difficult to optimize , especially in the kidney organ culture system . Maybe surprisingly , our antibody data do not show nuclear β-catenin as would be expected for active signalling . It should be realized , however , that this is the same for the signal in the ureteric bud which is generally accepted to have active β-catenin signalling ( Bridgewater et al . , 2008; Marose et al . , 2008 ) . This could indicate technical limitations of the whole-mount staining , but it should also be noted that in many other developing systems it can be difficult to detect nuclear β-catenin . Even in colorectal cancer it was shown that in cells from the same tumours , all assumed to have the same initiating loss of APC or oncogenic activation of β-catenin , dynamic localisation changes between nucleus and cytoplasm can be found ( Fodde and Brabletz , 2007 ) . Similarly , expression of a constitutively activated β-catenin mutant in mice throughout the intestine results in different responses in different parts of the intestine , including different localization ( varying between nuclear , cytoplasmic , or membranous; Leedham et al . , 2013 ) . Finally , a functional role for β-catenin in the patterning process as shown by our pharmacological and genetic data does not prove that this is dependent on an activity gradient . However , combined , these data strongly suggests that this gradient is real and functional in the patterning of the nephron . The reporter signal showed an exponential decrease , suggesting first-order decay of a single source from the distal end . The proximal domain ( Wt1+ ) , which is normally characterized by low or absent β-catenin activity , does not form when β-catenin activity is increased , whereas it forms quicker when the activity of the pathway is decreased ( Figure 2A–D ) . In the other domains of the developing nephron ( Lgr5+ and Jag1+ ) , cells also respond to the level of β-catenin activity that is forced on them ( Figure 3A , B , 4C , D ) . Increasing β-catenin activity had a positive effect on the distal domain ( Figures 3B , 4C ) as expected . Whilst the effect of decreasing β-catenin signalling was clear on the proximal segment it led to a more complex effect in the other segments . Lgr5 became expressed at earlier time-points , albeit at lower levels ( both in terms of GFP fluorescence in the Lgr5+/EGFP-IRES-CreERT2 model and also mRNA levels analysed by qRT-PCR ) compared to controls and when β-catenin signalling was increased ( Figure 3B ) . The tubules also elongated excessively ( Figure 4C ) . We speculate that the reason for this is that decreasing β-catenin signalling might have resulted in the whole of the nephron maturing faster and therefore elongating more , however , decreasing β-catenin signalling particularly favoured the proximal domain . Since proliferation was not affected by either increasing or decreasing β-catenin signalling , the changes in tubular morphology and length must have been caused by another mechanism . The tubules appeared thinner , suggesting cellular rearrangements as a likely cause . In addition to using genetic models we also utilised pharmacological inhibitors to enable us to fine-tune the β-catenin activity dosage through titration of the concentration of the compounds ( Davies , 2009 ) . This made it possible to inhibit as well as activate the pathway and , crucially , to achieve temporal control in the experimental system . Not only did this approach exclude the possibility that the observed phenotypes were due to cellular toxicity , but by washing away the β-catenin activator we showed that proximal nephron structures could recover and reform if the sustained β-catenin signalling , imposed on them by CHIR , was removed ( Figure 2E ) . Genetic models for activation of β-catenin are available , but inhibiting the pathway is not possible with current mouse models . The series of different Apc and Ctnnb1 mouse models that we employed have previously been used to show that β-catenin activity levels are selected for by different types of cancer , and these ectopic increases in signalling activity can also lead to a range of developmental defects ( Kielman et al . , 2002; Gaspar et al . , 2009; Buchert et al . , 2010; Bakker et al . , 2013 ) . Our data indicate that increasing levels of β-catenin activity through Apc and Ctnnb1 mutations lead to successively more severe kidney phenotypes , but only at the highest activity–dose achieved in this study did we detect ectopic nephron development ( Figure 4F , G ) . It has been known for a long time that LiCl , which is an inhibitor of GSK3β , can induce nephron formation but it also blocks MET at high concentrations ( Davies and Garrod , 1995 ) . This has since been confirmed by other , more specific , GSK3β inhibitors as well as genetic activation of β-catenin ( Kuure et al . , 2007; Park et al . , 2007 ) . Therefore , the appearance of the epithelialized , though abnormal , nephrons in the Apc+/1638N Ctnnb1+/E654 mutants and the ectopic nephrons in the inhibitor studies indicate that β-catenin activation has dosage-dependent effects on the developing kidney . The absence of proximal markers in these ectopic nephrons correlated with the lack of reporter activity as well as the phosphorylation state of β-catenin in that part of the nephron . It is important to note that some nephrons within the ureteric bud trees of Apc+/1638N Ctnnb1+/E654 mutant animals still managed to produce proximal domains . Whether this indicates redundancy or an ability to overcome increased β-catenin levels through ureteric bud-derived signals remains to be determined . Using cultures of spinal cord-induced metanephric mesenchymes , we also show that the observed phenotypes are not an indirect effect of changes in the ureteric bud but are a direct effect in the mesenchyme . Cell proliferation could be a mechanism through which β-catenin controls the patterning of the nephron . However , we found that proliferation along the nephron axis does not change in response to stimulating or inhibiting β-catenin activity , nor does blocking cell proliferation affect the patterning ( Figure 5C ) , suggesting that proliferation does not play a mechanistic role in this patterning process . Even in nephrons which were treated with MTX and which were significantly smaller compared to controls did quantification of the TCF/Lef::H2B-GFP activity reporter show that there was a gradient in β-catenin signalling as also seen in normal nephrons . Similarly , when nephrons were treated with MTX it was still possible to induce segmentation changes by adding either IWR1 or CHIR . This implies that the underlying mechanisms behind nephron segmentation can adequately adapt to reduced cell numbers and an overall reduction in nephron size . We also found that apoptosis in nephrons was not altered by either CHIR or IWR1 . Together , these findings would therefore suggest that increasing and decreasing β-catenin signalling changes segmentation by forcing cells to change their identity and that this is the most likely mechanism underlying the alterations to segmentation . We also investigated how β-catenin signalling interacts with other pathways in the nephron and how the β-catenin activity gradient is limited from expanding proximally ( Figure 6 ) . Our data indicate that , similar to the case in intestinal stem cells ( He et al . , 2004 ) , β-catenin activity is negatively regulated by BMP/PTEN/PI3K signalling in the medial domain of the nephron . This domain is normally positive for a BMP/pSMAD reporter and pSMAD staining . This link between BMP and β-catenin is likely to reflect a normal function of BMP in this nephron segment since inhibition of BMPR and PI3K provoked opposite effects on the medial segment and BMPR inhibition increased levels of both phosphorylated PTEN and AKT . Furthermore , inhibition of PI3K on its own made the medial segment visually grow much larger and Jag1 and other medial markers were expressed more strongly . The significance of PI3K signalling in connection with Wnt/β-catenin pathway is strongly debated ( Ng et al . , 2009 ) , but the PI3K pathway has not only been shown to alter Wnt signalling in the gut but also in embryonic stem cells ( Paling et al . , 2004 ) . We tested whether β-catenin was phosphorylated at Ser552 as it has been reported to be by AKT ( Fang et al . , 2007 ) . Based on our data , we expect low levels of PI3K/AKT signalling in the medial segment , and consequently , we would therefore anticipate that β-catenin would only be phosphorylated at Ser552 at very low levels if at all . We did detect phosphorylated β-catenin at the apical surfaces and lateral surfaces mainly within the distal and medial domain but never within the nucleus ( Figure 6H ) . A number of cells displayed stronger β-catenin staining which also extended to the lateral and basal surfaces . We currently do not know the significance of this and the staining pattern did not change in response to either BMPR inhibition of PI3K inhibition . It will be important to explore precisely how PI3K and BMP/pSMAD signalling regulate the Wnt/β-catenin pathway , in the nephron and elsewhere . This should be the subject of future studies . To date , one of the best characterised signalling pathways that have been implicated in nephron patterning is the Notch pathway ( Cheng et al . , 2007; Boyle et al . , 2011 ) . As Wnt and Notch signalling are tightly linked during development ( Hayward et al . , 2008 ) , we examined if the role of Notch signalling in nephron pattern formation is linked to the β-catenin activity gradient . We found that inhibition of Notch resulted in nephrons with medium to high levels of β-catenin activity throughout . By co-inhibiting β-catenin signalling in Notch-inhibited kidneys we partially reversed the latter's patterning phenotype . This was also true of co-inhibition of PI3K and Notch . In common to inhibition of β-catenin and PI3K was that they both decreased β-catenin signalling in nephrons and positively affected the development of the proximal-most and medial segments , respectively . Inhibition of β-catenin signalling in Notch-inhibited nephrons reduced the level of β-catenin signalling and rescued the proximal-most cells whilst co-inhibition of Notch and PI3K partially rescued the medial segment . It is therefore tempting to speculate that the mechanism for these rescues is by reduction of the level of β-catenin signalling . The mechanism of the rescue needs to be further studied , particularly in light of that the expression of known Notch target genes was not rescued by the β-catenin inhibitor ( Figure 7D ) . Wnt ligands have long been known to act as one of the classic morphogens that drive patterning during development ( Zecca et al . , 1996 ) . To be effective as a gradient , the cells within the gradient should be able to respond to the different concentrations of extracellular Wnt ligand . In previous studies , a gradient of recombinant XWnt8 has been shown to be able to drive anterior–posterior neural patterning in dissociated Xenopus embryos with a corresponding gradient of β-catenin activity ( Kiecker and Niehrs , 2001 ) . Similarly , a gradient in nuclear β-catenin was described in the patterning of the segmentation clock that controls somitogenesis ( Aulehla et al . , 2008 ) . Although this study analysed different β-catenin levels , in contrast to our approach , the authors focused on the extreme ends of the gradient ( ‘off’ vs ‘maximal’ ) , and it was concluded that an unidentified co-factor was essential for interpreting the Wnt gradient . We show that in the nephron variations in the levels of β-catenin activity within the observed β-catenin gradient are directly responsible for correct patterning . Whilst the phenotypic effect of the β-catenin activity gradient in patterning the nephron is obvious , the mechanistic rationale for this is unclear . It is difficult to see how the present molecular/biochemical model for β-catenin signalling would allow the existence of β-catenin dosage-dependent differential gene expression . The model predicts that once β-catenin is stabilized , it will translocate to the nucleus and activate its target genes ( Clevers and Nusse , 2012 ) . The data presented here , together with the ‘just-right’ signalling model for the role of APC mutations and β-catenin activity in cancer ( Albuquerque et al . , 2002; Gaspar et al . , 2009 ) clearly show that β-catenin signalling is not a binary process and suggest additional levels of control of the β-catenin transcriptional output . Further , biochemical analysis of β-catenin function will be needed to elucidate this mechanism . At present we do not know which Wnt ( assuming it is a Wnt ) it is that drives the gradient or how β-catenin activity is antagonised in the proximal cells . For the Wnts , none of the published Wnt knockout mouse models display a phenotype that would suggest a clear role in patterning of the nephron , although both Wnt9b and Wnt7b , when deleted , alter nephron morphogenesis . The Wnt7b knockout throughout the embryo proper or in the ureteric bud was described to lack a medulla due to disturbances of the plane of cell division , resulting in a failure of the Loop of Henle to elongate ( Yu et al . , 2009 ) . As this phenotype was mainly analysed at later stages of kidney development , it would require the inclusion of the TCF/Lef::H2B-GFP reporter and analysis of appropriate markers to determine if this phenotype is linked to the patterning defects we describe here . A conditional Wnt9b knockout in post-MET nephrons resulted in disturbed planar cell polarity as well but resulted in expansion of tubule diameter instead of the Loop of Henle defect ( Karner et al . , 2009 ) . Wnt9b is known to regulate planar cell polarity in the nephron , and the canonical pathway and non-canonical pathways are widely believed to be mutually competitive and inhibitory ( Grumolato et al . , 2010 ) . Although competition is known to occur mainly at the receptor level , it could be imagined that competition for intracellular downstream targets have the consequence that when decreasing the β-catenin activity using IWR1 , there is shift from the canonical pathway towards the planar cell polarity pathway resulting in the observed nephron elongation . Again , extensive further analysis of the Wnt9b model is required to demonstrate involvement in the processes we describe here . We analysed Lgr5 as a potential modifier of Wnt signalling establishing the β-catenin gradient . The nephrons forming in Lgr5+/EGFP-IRES-CreERT2 homozygotes were morphologically indistinguishable to heterozygotes or control animals ( Figure 3—figure supplement 1 ) . This is maybe not surprising as Lgr5 knockouts have previously been shown to be without phenotype in the intestine where it is an important marker of intestinal stem cells ( de Lau et al . , 2011 ) . A second question is how a single source of Wnt could establish a β-catenin signalling gradient within a morphologically convoluted tissue . It is plausible that the responsible Wnt forms an intraluminal gradient which would therefore potentially form a gradient irrespective of the actual nephron shape . It is also possible that all the nephron cells are exposed to a homogenous level of Wnt ligand and that they by cell-specific means modulate this to specific levels via , for example , Notch or BMP signalling . In the proximal domain , the β-catenin activity levels are even lower than in the medial domain , but no BMP activity is detected there . A plausible candidate for preventing β-catenin activity in this segment would be Wt1 , as in in Sertoli cells Wt1 was shown to limit β-catenin activation ( Chang et al . , 2008 ) . Whilst additional Wnt and other knockout studies , either conventional or conditional , might provide new clues to extend our data into a yet more complete genetic pathway , the rate of nephron formation makes this a particularly challenging process . Since patterned S-shaped body nephrons form from mesenchymal progenitors within a 24 hr time-span , to conditionally knock out Wnts , it would be necessary to identify potential Cre-driving genes that are expressed at the very earliest stages of nephrogenesis but which are not active prior to nephron formation , since deletion at that time-point is likely to block the process . At present we do not know of any Cre driver that would be able to do this , so this might be a long-term goal .
Animals were kept at facilities at the MRC Human Genetics Unit and the University of Edinburgh ( UK ) ; Columbia University , New York ( USA ) ; Maine Medical Center , Maine ( USA ) ; the Beatson Institute , Glasgow ( UK ) ; and the Erasmus Medical Centre , Rotterdam ( The Netherlands ) . All animal experiments and animal use were carried out according to regulations specified by the Home Office ( MRC/UoE ) , approved by the Animal Ethics Committee and carried out in accordance with Dutch and international legislation ( Erasmus ) and conducted under PHS guidelines and approved by the relevant Institutional Animal Care and Use Committees ( Columbia/Maine ) . All animal experiments were performed under Project Licenses 60/3788 and 60/4473 . Outbred CD1 animals were obtained from Charles Rivers ( UK ) . Embryos were generated through timed mating , with noon of the day a vaginal plug was found considered E0 . 5 . TCF/Lef:H2B-EGFP ( Tg ( TCF/Lef1-HIST1H2BB/EGFP ) 61Hadj ) ( Ferrer-Vaquer et al . , 2010 ) were crossed with 129/SvEV and Wt1+/GFP ( Wt1tm1Nhsn ) ( Hosen et al . , 2007 ) mice were crossed with CD1 . Pax8+/Cre ( Pax8tm1 ( cre ) Mbu ) mice ( Bouchard et al . , 2004 ) were crossed to Rosa26eYFP/eYFP ( Gt ( ROSA ) 26Sortm1 ( EYFP ) Cos ) animals ( Srinivas et al . , 2001 ) . Six2+/GCiP ( Dolt et al . , 2013 ) mice were crossed with Rosa26tdRFP ( Gt ( ROSA ) 26Sortm1Hjf ) ( Luche et al . , 2007 ) . Ctnnb1E654 ( Ctnnb1tm1 . 2Wvv ) ( van Veelen et al . , 2011 ) , Apc+/1572T ( Apctm2Rfo ) ( Gaspar et al . , 2009 ) , and Apc+/1638N ( Apctm1Rak ) ( Fodde et al . , 1994 ) were intercrossed as required . Lgr5+/EGFP-IRES-CreERT2 ( B6 . 129P2-Lgr5tm1 ( cre/ERT2 ) Cle/J ) ( Barker et al . , 2012 ) were crossed with C57BL/6 ( Harlan ) or intercrossed . BRE-Hspa1a-LacZ were described before ( Blank et al . , 2008 ) . Kidneys were dissected from E12 . 5 embryos and cultured at 37°C with 5% CO2 on 0 . 4 µm PET Transwell membranes ( Corning , New York , NY ) . The culture medium consisted of MEM ( SIGMA , UK M5650 ) , 10% FCS , and 1% Pen/Strep . Isolated mesenchyme was induced with spinal cord as previously described ( Davies , 1994; Davies and Garrod , 1995 ) . Media was changed to contain IWR1 or CHIR after 48 hr and cultures kept for an additional 48 hr . Isolated Six2+ cells were pelleted for 5 min at 800×g and induced with spinal cord for 24 hr . The cultures were treated with CHIR or IWR1 . CHIR99021 ( University of Dundee , UK ) , ICG001 ( Enzo LifeSciences , UK ) , IWR1 ( TOCRIS ) , salinomycin ( SIGMA ) , BIO ( TOCRIS , UK ) , Ly294002 ( TOCRIS ) , LDN-193189 ( STEMGENT , Cambridge , MA ) , methotrexate ( SIGMA ) , were used as specified in the text . For TaqMan experiments , samples were placed in RNALater ( Ambion ) . A minimum of three kidneys were used per replicate of each condition and all experiments were performed in triplicates . RNA was isolated from cultured kidneys using RNAeasy micro kits ( QIAGEN , The Netherlands ) . cDNA was made using SuperScript II ( Invitrogen , Carlsbad , CA ) and random primers ( Promega , Madison , WI ) . Assays were done with the Universal Probe Library ( Roche ) and were designed using the Universal Probe Library Assay Design Center ( http://www . roche-applied-science . com/sis/rtpcr/upl/index . jsp ? id=UP030000 ) ; primer and probe details can be found in Supplementary file 1 . All TaqMan assays used a mouse GAPDH internal control ( Roche ) . Heat-maps were generated using the matrix2png interface ( http://www . chibi . ubc . ca/matrix2png/bin/matrix2png . cgi ) . GSK-3β , glycogen synthase kinase 3 beta; WNT , wingless-related MMTV integration site; Cited1 , Cbp/p300-interacting transactivator with Glu/Asp-rich carboxy-terminal domain 1; Wt1 , Wilms' tumour 1 homologue; Six2 , sine oculis-related homeobox 2; Jag1 , jagged 1; Ctnnb1 , catenin ( cadherin associated protein ) beta 1 ( β-catenin ) ; PTEN , phosphatase and tensin homolog; PI3K , phosphatidylinositide 3-kinase; AKT , thymoma viral proto-oncogene 1; BMP , bone morphogenetic protein; Apc , adenomatosis polyposis coli; Lgr5 , leucine rich repeat containing G protein coupled receptor 5; Lhx1 , LIM homeobox protein 1; Fgf8 , fibroblast growth factor 8; Pax8 , paired box gene 8; UB , ureteric bud; Pod . , podocyte; Pt . , parietal epithelium . | The main function of the kidney is to filter blood to remove waste and regulate the amount of water and salt in the body . Structures in the kidney—called nephrons—do much of this work and blood is filtered in a part of each nephron called the glomerulus . The substances filtered out of the blood move into a series of ‘tubules’ , another part of the nephrons , from where water and soluble substances are reabsorbed or excreted as the body requires . If the nephrons do not work correctly , it can lead to a wide range of health problems—from abnormal water and salt loss to dangerously high blood pressure . For organs and tissues to develop in an embryo , signalling pathways help cells to communicate with each other . These pathways control what type of cells the embryonic cells become and also help neighbouring cells work together to form specialised structures with particular functions . Much is unknown about how the nephron develops , including how its different structures coordinate their development with each other so that they form in the right position in the nephron . A protein called beta-catenin was already known to play an important role in the signalling pathways that trigger the earliest stages of nephron formation . Lindström et al . further investigated how this protein helps the nephron to develop by using a wide range of techniques , including growing genetically altered mouse kidneys in culture and capturing images of the developing nephrons with time-lapse microscopy . The combined results reveal that the levels of beta-catenin activity coordinate the development of the different structures in the nephron . The beta-catenin protein is not equally active in all parts of the nephron; instead , it forms a gradient of different activity levels . The highest levels of beta-catenin activity occur in the tubules at the furthest end of the developing nephron; this activity gradually decreases along the length of the nephron , and the glomerulus itself lacks beta-catenin activity altogether . Experimentally manipulating the levels of beta-catenin at different points along the nephron caused those cells to take on the wrong identity , causing parts of the nephron to form in the wrong place . Lindström et al . were also able to establish that the signalling pathway controlled by beta-catenin activity interacts with three other well-known signalling pathways as part of a network that controls nephron development . More research is required to find out which signal activates beta-catenin in the first place and from where in the kidney this signal comes . It also remains to be discovered how a particular cell in the tubule interprets the exact activities of the different signals to give the cell its specific identity for that place in the nephron . A better understanding of these sorts of processes will eventually help build new kidneys for people with kidney failure . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology"
] | 2015 | Integrated β-catenin, BMP, PTEN, and Notch signalling patterns the nephron |
The staining of neurons with silver began in the 1800s , but until now the great resolving power of the laser scanning confocal microscope has not been utilized to capture the in-focus and three-dimensional cytoarchitecture of metal-impregnated cells . Here , we demonstrate how spectral confocal microscopy , typically reserved for fluorescent imaging , can be used to visualize metal-labeled tissues . This imaging does not involve the reflectance of metal particles , but rather the excitation of silver ( or gold ) nanoparticles and their putative surface plasmon resonance . To induce such resonance , silver or gold particles were excited with visible-wavelength laser lines ( 561 or 640 nm ) , and the maximal emission signal was collected at a shorter wavelength ( i . e . , higher energy state ) . Because the surface plasmon resonances of noble metal nanoparticles offer a superior optical signal and do not photobleach , our novel protocol holds enormous promise of a rebirth and further development of silver- and gold-based cell labeling protocols .
In the late 1880s , Santiago Ramón y Cajal began characterizing the cytoarchitecture of different types of brain neurons in a variety of animal species by employing the Golgi stain or ‘black reaction’ method . For reasons still unknown , only a random subset of neurons became labeled in their entirety , leading to our understanding that the nervous system is largely composed of individual cells separated by synapses . Although the Golgi method , which involves the deposition of silver , allowed for unprecedented insight into the morphological features of neurons , subsequent methods were developed so that specific neurons of interest could be characterized . For example , targeted neurons were filled with cobalt or nickel ions , and later intensified through the deposition of silver ( Pitman et al . , 1972; Tyrer and Bell , 1974; Mesce et al . , 1993a ) . Eventually , the popularity of metal-filling and silver intensification methods gave way to the development and use of fluorescent compounds to label neurons ( Stretton and Kravitz , 1968; Stewart , 1981; Mesce et al . , 1993b; Vitzthum et al . , 1996 ) . Furthermore , fluorescent specimens imaged with the laser scanning confocal microscope ( LSCM ) , made available in the late 1980s ( Amos et al . , 1987 ) , provided vastly improved results—in-focus 3D reconstructions of individual neurons were now easily obtained . With repeated laser imaging , however , a number of fluorophores in wide use showed susceptibility to photobleaching , thus lessening the ability of fluorescently labeled samples to be archived and repeatedly imaged over time . The use of silver and gold particles for studies of cell and molecular biology , however , has had a recent resurgence because of the ability of metals to resolve small structural features in a variety of tissues . Enzyme metallographic technology , for example , enables the maximal sensitivity and resolution of cancer-related gene and protein expression patterns with either in situ hybridization or immunocytochemical techniques ( Downs-Kelly et al . , 2005; Turzhitsky et al . , 2014 ) . To date , these techniques and other modern methods have required conventional brightfield microscopy , involving the unwanted capture of out-of-focus light , and time-consuming reconstruction of serial histological sections or optical image planes . It is clear that enabling the imaging of silver or gold particles with the LSCM would be of enormous benefit to neuroscientists and cell biologists alike . To this end , we discovered a novel strategy that permits the imaging of metal-impregnated cells with the LSCM . Silver- or gold-labeled neurons were imaged by using a method whereby the laser-induced photon emission , from any given sample , was captured at its higher energy state ( shorter wavelength ) as compared to the laser wavelength used to excite the metal-impregnated samples . This phenomenon can be used in the presence of high background fluorescence , and contrasts with conventional fluorescence microscopy whereby the emitted photons are of a longer wavelength ( i . e . lower energy state ) . Our results are consistent with the hypothesis that the energy emitted after laser excitation is based on the phenomenon of silver ( or gold ) local surface plasmon resonance ( Willets and Van Duyne , 2007 ) and not metallic reflectance or fluorescence . Thus , the unexpected pairing of decades-old anatomical methods with newer confocal imaging technology is poised to unlock new information residing in a myriad of archived histological specimens . Furthermore , silver-impregnated preparations should retain their high-quality image for a century or more . For example , the original Golgi preparations created by Sanchez y Sanchez and Cajal in 1916 were recently beautifully reimaged using conventional brightfield microscopy ( Strausfeld , 2012 ) . We predict that an enormous reservoir of Golgi and other silver-stained nervous systems , from a wide range of vertebrate and invertebrate animals , can now be mined and easily imaged in their entirety . With the advent of enzyme metallography , these samples too can be visualized with unparalleled detail and reconstructed in three dimensions . Note , the imaging of silver-intensified cobalt-filled neurons with the LSCM was reported earlier in abstract form ( Thompson et al . , 2011; 2014 ) .
To appreciate the enhanced image quality produced by our new protocol , histological preparations of insect neurons ( labeled with cobalt and intensified with silver ) were imaged in multiple ways . First , a conventional brightfield image of a labeled neuron within a female abdominal ganglion of a grasshopper is presented . The cell body , neurites , axon and fine branching pattern can be seen ( Figure 1a ) . However , this photomicrograph underscores the basic problem of brightfield microscopy when trying to illustrate the 3D structure of an entire neuron . First , the depth of focus is short relative to the size of the neuron . Second , the out-of-focus branches obscure the in-focus portions of the neuron . Thus , studies of cobalt-filled neurons are usually presented as camera Lucida tracings in which the cell’s processes at all depths of focus are traced by hand onto paper ( not shown ) ; such drawings do not retain depth information . Stacked images from brightfield microscopy offer the promise of improvement ( Figure 1b ) , but they too often lack the detail obtained using the LSCM . 10 . 7554/eLife . 09388 . 003Figure 1 . Image of a cobalt-filled , silver-intensified motoneuron , viewed with traditional microscopic imaging methods . View of an insect ( grasshopper ) ganglion containing a motoneuron labeled by retrograde transport of cobaltous chloride , which was placed in a well surrounding a transected lateral nerve . The cobalt-filled motoneuron subsequently received a deposition of silver using the Timm’s silver intensification method Tyrer and Bell , 1974 . All images presented here were taken using a 10× objective lens . ( a ) Brightfield image of the cobalt-filled silver-intensified motoneuon . ( b ) Stacked transmitted-light images acquired using the Nikon A1 spectral confocal microscope . ( c-e ) Confocal images obtained using appropriate settings for imaging specific fluorescent dyes; excitation and detection wavelengths were optimized for the specific fluorophore under consideration ( listed in caption ) . ( c ) Alexa Fluor 488 setting: Sample was excited with a 40 mW argon multiline laser at 488 nm and emission was detected at 525/50 nm ( center wavelength/bandwidth ( FWHM ) ) . ( d ) TRITC setting: Sample was excited with a 40 mW 561-nm diode laser and emission was detcted at 595/50 nm . ( e ) Cy5 setting: Sample was excited with a 10 mW 638 nm diode laser and emison was detected at 700/75 nm . ( f ) Imaging in the reflectance mode . Poor-quality images were obtained when the motoneuron was excited with a 561-nm laser line , and the emission was detected in a region over the same wavelength ( using a notch filter ) . Only a portion of the neuronal arbors was visible and in focus . Scale bars = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 09388 . 003 We next imaged the same locust ganglion with the LSCM adjusted to the manufacturer’s standard recommended wavelength settings . Such settings are typically based on laser excitation and emission wavelengths to detect fluorophores used in neuroanatomy , such as Alexa Fluor 488 , Texas Red ( TRITC ) , and Cy5 . Based on these parameters , we observed strong autofluorescence from the ganglion while the silver-labeled neurons revealed a collective dark blur ( Figure 1c–e ) . Metals do , however , have the ability to reflect light . The reflectance mode of the LSCM has been utilized previously , especially in the diagnosis of skin cancers ( Calzavara-Pinton et al . , 2008 ) . We thus examined our samples using this mode of operation whereby the excitation and emission wavelengths of the labeled neurons were relatively matched . Although some visualization of silver-impregnated neurons was observed , the reflective properties of the metal-impregnated neurons were poor , likely due to the thickness of the preparation and the refractive-index mismatch throughout the sample ( Figure 1f ) . We employed a Nikon A1 Spectral Confocal Microscope to obtain all the confocal images presented here . The microscope was equipped with 4 standard photomultiplier tube ( PMT ) detectors and an array of 32 PMTs for spectral detection . When the settings of the LSCM were adjusted so that the laser excitation was of a longer wavelength compared to the shorter wavelengths selected for the detection of emitted photons , spectacular 3D electronic images of cobalt/silver impregnated neurons were produced; importantly , in-focus single optical planes could be stacked and rotated for three-dimensional renderings ( Figure 2a , b ) . [Note: Figure 2—figure supplement 1 shows details of post-capture image processing to remove the perineurial sheath . A video of the neurons rotating in 3D is also provided ( Video 1 ) . ] Thus , silver-impregnated preparations imaged in this new way had all the positive attributes of fluorescently labeled preparations for which the confocal microscope has been so predominantly useful . Additionally , images could be obtained from older archived preparations; for example , the specimen illustrated in Figure 3 was more than 25 years old . As fluorophores always emit light at a lower energy state ( i . e . longer wavelength ) than their excitation laser light , the settings we used ( i . e . detection at shorter wavelengths ) would not be applied during the standard operation of the LSCM . 10 . 7554/eLife . 09388 . 004Figure 2 . Silver-impregnated motoneurons imaged with the new protocol and the LSCM . Cobalt-backfilled and silver-intensified neurons , within two different insect ganglia , imaged with the LSCM . Images include selected regions of interest ( i . e . somata and neural processes ) ; signal from the ganglion’s perineurial sheath was removed ( in FIJI ) . For additional information on image processing and the optical removal of the sheath , see the ‘Materials and methods’ and Figure 2—figure supplement 1 . ( a-aiii ) For these images , a 561-nm excitation laser line was used and the photic emission was obtained from 490–530 nm . ( ai , aii ) Rotations of the neuronal image revealed the three-dimensional cytoarchitectual features of the neuronal arbors in both the lateral and dorsal-ventral planes . ( aiii ) A higher magnification of the left-side region of the ganglion ( at 9 o’clock ) highlights the level of detail observed among the fine branches visible using our new protocol . ( b-bii ) A different insect ganglion , showing a single cobalt-filled and silver intensified motoneuron . ( b ) Fine morphological details of an insect motoneuron and its ramifying branches are shown here . Laser excitation and emission collection were the same as in ( a ) . Note that the neurite connecting the soma to its branches was not visible due to loss of tracer not image capture . ( bi , bii ) Lateral and dorsal-ventral rotations , respectively , demonstrating the ability of cells to be rendered in three-dimensions for analysis of arbor patterning . Scale bars = 100 µm . See video of neuronal rotations in the supplementary data section . LSCM: Laser scanning confocal microscope . DOI: http://dx . doi . org/10 . 7554/eLife . 09388 . 00410 . 7554/eLife . 09388 . 005Figure 2—figure supplement 1 . Image processing of volume renderings and removal of perineurial sheath . ( a ) Raw image resulting from blind unmixing of spectral data . ( b ) Image of rendering after perineurial sheath is removed in FIJI . Note that signal from outside of ganglion obscures motoneuron branches in 3D and orthogonal views . ( c ) Image of rendering containing only signal from selected regions of interest ( ROI ) . Scale bars = 100 µmDOI: http://dx . doi . org/10 . 7554/eLife . 09388 . 00510 . 7554/eLife . 09388 . 006Video 1 . Volume rendering of silver-impregnated locust motoneurons . Cobalt-filled and silver-intensified neurons within an insect ganglion . Image was processed as described in the ‘Materials and methods’ . Imaris software was used to generate a volume rendering and the video was exported as an AVI file . Handbrake software was used to convert the video to MP4 format . Video depicts same motoneurons as presented in Figure 2a . DOI: http://dx . doi . org/10 . 7554/eLife . 09388 . 00610 . 7554/eLife . 09388 . 007Figure 3 . Optimal excitation and emission wavelengths to image silver-impregnated cells with the LSCM . Multiple images of a cobalt-filled silver-intensified stretch receptor organ , from a hawkmoth , were compared across different laser excitation wavelengths using the LSCM and compared to a brightfield image . ( a ) Standard brightfield image of the insect stretch receptor organ in the abdomen . The arrow points to the cell body of the sensory neuron , which is attached to a modified muscle fiber; the arrowhead points to efferent motor terminals that regulate the tension of the fiber . Other silver-impregnated fine nerves are out of focus in the background . ( b-e ) An excitation laser line of 405 nm ( b ) , 488 nm ( c ) , 561 nm ( d ) , or 640 nm ( e ) was used to determine optimal wavelengths to obtain best signal-to-noise ratio images . Emissions are plotted to the right and color coded with neuronal image . With the 405 laser line , light was collected from 400 nm to 590 nm in 6 nm bins; this entire spectrum was used to generate the image depicted ( b ) . The other images ( c-e ) were formed by collecting the spectrum from 430–740 nm ( in 10 nm bins ) . The associated graphs indicate the signal-to-noise ratio calculated by determining the ratio of the intensity of emitted signal from a region of interest ( ROI ) divided by the intensity of the background . Insets within the graphs indicate the raw intensity values for the signal ( black ) and background ( gray ) . In each trace the excitation laser wavelength is indicated as a vertical dashed line . Note that at the 561-nm laser excitation , the image quality is best . The nucleus of the stretch receptor is visible as are the efferent terminals on the muscle-like fiber; the previously out-of-focus fine nerve fibers are now in focus . Scale bars = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 09388 . 007 How important is the specific laser excitation wavelength to achieving image quality ? We addressed this question by exciting silver-impregnated samples at different wavelengths ( 405 , 488 , 561 , or 638 nm; exact laser line indicated within the graphs ) ( Figure 3 ) . We then measured the emissions over a three-decade range of wavelengths with a spectral detector ( Figure 3 ) . To compare the signal intensity across images and imaging types , we used a selection of four images from the scan and measured the emission intensity of both the region of interest ( ROI ) and the background ( Figure 3 ) . By taking a ratio of the intensity of the ROI and the background , we were able to calculate relative intensity and compare data across samples . In addition , all samples compared were acquired using a similar laser power setting ( 90% for the 561nm excitation , 100% for all other excitation wavelengths ) , pinhole size ( 1 . 2AU ) , and gain ( 120 for 561 nm , and 115 for all other excitations ) . Each of the anatomical images shown ( Figure 3b–e ) is a collection of the entire spectrum of wavelengths to the left and right of the excitation wavelength . We found that 561-nm and 640-nm excitation laser lines were required to produce strong emissions and high signal-to-noise images . The overall intensity of the shorter wavelength emissions with the 561-nm laser line was stronger ( Figure 3 ) , and hence , it became the laser setting of choice for the remainder of our studies . This is consistent with the peak absorption of silver and gold nanoparticles at ~550 nm ( Link and El-Sayed , 2003 ) . Preparations were then stimulated with the 561-nm laser line to enable us to determine the optimal wavelength to use for detection of the emitted photons ( Figure 4 ) . We detected signal at 10 nm intervals along the range from 440 nm to 730 nm . Images represent groupings of 40 nm intervals . The images in the frames labeled 440–480 and 490–530 ( Figure 4c , ci ) showed a markedly stronger signal and low background compared to the other detection ranges ( Figure 4cii–v ) . The graph at the bottom of the figure also showed the strong enhancement of the signal-to-noise ratio for its corresponding image that occurred in the 440–480 and 490–530 nm range of detection . It should be noted , however , that within the 440–480 nm range the high peak on the graph was not due to an abundance of signal , but rather the low noise ( see graph inset ) . The second peak in the graph occurred where we had slightly higher noise but a much higher signal , resulting in greater imaging success when the signal was collected from 490–530 nm . With the short-wavelength detection setting of 490–530 nm , and a laser excitation setting of 561 nm , a bright neuron against a dark and low noise background emerged . 10 . 7554/eLife . 09388 . 008Figure 4 . Optimal collection wavelengths to image silver-impregnated cells with the LSCM . Cobalt-filled and silver-intensified single insect motoneuron was imaged over a variety of different emission wavelengths ( i . e . that which was collected ) . ( a ) Insect ganglion preparation imaged with a 561-nm excitation laser line and photic output collected over the full spectral range ( 430–750 nm ) . Note that only the soma is faintly visible . ( b ) Same cell imaged under standard brightfield conditions , where the cell body and numerous fine branches were now visible , but deeper branches were out of focus . ( c-cv ) Collected emissions from the 561-nm excited samples were parsed into 40 nm bandwidths . Clearly , the optimal ranges for signal maximization were in the shortest wavelengths , approximately 440–530 nm ( see c ) . These same data are also shown graphically below the images , highlighting the strongly enhanced contrast of the signal at optimal wavelengths . It is noteworthy that at 440–480 nm the signal is relatively low ( see inset ) , thus photic collection at 490–530 nm appears to be more optimal for image quality . Scale bars = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 09388 . 008 Can preparations that contain other metals than silver also be imaged with the LSCM ? To examine this possibility , we looked at a number of histological preparations that had used different metals and impregnation techniques to characterize neuronal morphologies . Some neurons were: 1 ) filled with cobalt but had no silver intensification; 2 ) stained with silver but contained no cobalt ( e . g . Golgi , reduced silver and Gallyas stains ) ( Gallyas et al . , 1980 ) ; 3 ) stained with gold ( Bodian’s stain ) . Our findings indicate that silver or another noble metal such as gold was required . Cobalt alone produced poor-quality images , regardless of the excitation laser selected for use ( Figure 5 ) . Golgi stains of different brain regions from different species were also imaged , for example , the insect optic lobe ( Figure 5a ) , mammalian pyramidal cell ( Figure 5b ) , and mouse Purkinje cells ( Figure 5c ) , all of which yielded high quality confocal images ( excitation laser of 561 nm and detection of 430–740 nm ) . In addition , the Gallyas-stained sample ( Figure 5d ) , the amino-cupric silver stained one ( Figure 5f ) , and the gold-based Bodian’s stain ( Figure 5e ) all produced quality neural images . The graph to the right of each set of images shows a plot that , essentially , indicates the signal-to-noise ratio for each sample . For images shown in Figure 5a–f , the best ratio in all samples occurred at 450 nm , and a secondary peak was located just above 500 nm . The insets of each plot show overall intensity of the signal in black and the background signal in gray . From these plots it is clear , however , that the high signal intensity just to the left of the excitation wavelength decreases to a lower value near 430 nm . Because of the lower signal in the 440–480 nm range , it is more reliable to detect the signal in the range of 490–530 nm . 10 . 7554/eLife . 09388 . 009Figure 5 . Examination of various metal-based stained neurons , across species , imaged with the LSCM . ( a-f ) Silver- and gold-impregnated tissues were imaged using standard brightfield microscopy ( left ) and compared to LSCM images excited with a 561-nm laser line ( right ) . Normalized intensity graphs ( signal to noise ratio ) are shown to the right of each confocal image . Insets provide the raw intensity data for the detected signal ( black ) and detected background noise ( gray ) . The dashed line indicates the wavelength of the excitation laser line that was used ( 561 nm ) . Preparations were as follows: ( a ) Golgi stain of a honeybee optic lobe . Note the presence of what appear to be dendritic spines ( arrowhead ) ; ( b ) mammalian pyramidal neuron; ( c ) mouse Purkinje neuron; ( d ) Gallyas stain of a monkey brain; ( e ) Bodian’s stain of a mammalian motor end plate; ( f ) amino-cupric-silver stain of a mouse brain . In ( g ) , insect neurons were backfilled with cobalt but not subsequently silver intensified . Without silver , the emission intensity profile is flat and the confocal image shows only faint arbor-free somata . Scale bars = 100 µm except for one bar in the far-right confocal image in ( a ) bar = 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 09388 . 009 In summary , the most straightforward and optimal way to implement the new methods communicated here is to image the metal-impregnated sample with the LSCM excitation laser set to 561 nm ( or common 543 nm and 555 nm laser lines ) and collect photons in the range of 490–530 nm .
In this study , we showed that the LSCM can be utilized to capture photons for image formation emitted by silver- or gold-impregnated neuronal specimens . Our analysis identified two constraints on imaging . One was that the microscope’s longer wavelength laser lines ( 561 nm or 640 nm ) were required for excitation . The second was that wavelength detection settings needed to be shorter , in the range of 440–530 nm . This relationship between stimulating wavelength and collecting wavelength contrasted with those involving fluorescent dyes . For example , fluorophores excited with a 561-nm laser line emit photons with wavelengths in the range of 570–610 nm , due to energy loss during the fluorescence energy transfer ( Stokes Law ) . In fact , many LSCMs utilize filter assemblies that routinely filter out emission wavelengths shorter than the excitation laser line , thus preventing the type of capture we have outlined here . The explanation , we believe , for both the presence of photic emissions from our silver and gold preparations and for the fact that they are of higher energy than the excitation wavelength , is the phenomenon of surface plasmon resonance . Plasmons are coherently oscillating free electrons on the surface of noble metal nanoparticles , such as silver and gold ( Willets and Van Duyne , 2007 ) . When the vector of incident light matches the oscillations of plasmons , there is an interaction or resonance , thus the term surface plasmon resonance ( Willets and Van Duyne , 2007 ) . Different sizes and shapes of plasmonic silver and gold nanoparticles , and their spacing , will produce different resonances that can vary in wavelength across the visible light range and the infrared . Photons can either be scattered in all directions by plasmonic nanoparticles or absorbed and transferred into heat or luminescence ( Zhang et al . , 2007; Rycenga et al . , 2011 ) . Of the noble metals and copper , known to exhibit plasmon resonance in the visible light range , silver is the most efficient for light capture . For example , a single nanoparticle can interact with 10 times more light than is physically incident upon it; thus , a silver nanoparticle is more efficient than any similarly sized chromophore ( Rycenga et al . , 2011; Evanoff and Chumanov , 2005 ) . Excitation and emission photons are not related in a one-for-one ratio , but rather , the excitation laser line ( longer wavelength ) provides the energy input necessary to sustain the plasmonic light-emitting state of the metallic nanoparticles . The optical changes are large because of the strong interactions of light with noble metals ( Evanoff and Chumanov , 2005 ) . While silver and gold are known to support surface plasmons at visible wavelengths , cobalt emits inefficiently in the 300 nm range—outside the collection wavelength setting available on the LSCM we used ( Kaminskienė et al . , 2013 ) . Thus , the hypothesis that our silver- and gold-impregnated samples can be imaged as a consequence of surface plasmon resonance is supported by the fact that cobalt alone did not result in an emission in the visible light range ( Figure 5g ) . Furthermore , the wavelength observed for silver and cobalt/silver nanoparticle plasmonic emissions is in the range of 440–530 nm ( Evanoff and Chumanov , 2005 ) , providing further support for our hypothesis . Metallic nanoparticles of silver or gold have been exploited for use in medicine and biotechnology , in part , because they are nontoxic , do not photobleach , and are of relatively low cost ( Barnes et al . , 2003; Iwasaki et al . , 2006; Lal et al . , 2007; Jain et al . , 2008; Javier , 2008; Hamidi and Oskuei , 2014; Eustis and El-Sayed , 2006 ) . The use of plasmon-related technology is rapidly expanding in the biological fields and for other applications ( Downs-Kelly et al . , 2005; Fan et al . , 2014 ) . Metal nanoparticles are wavelength-tunable for enhanced optical contrast , they can be tagged with antibodies , and can serve as molecular sensors ( Turzhitsky et al . , 2014 ) . One particularly interesting application involves cell labeling with nanoparticles coupled to fluorescent dyes ( Koyama and Tophyama , 2013 ) . Significant plasmonic enhancement of fluorescent dyes is observed , apparently the result of increased fluorophore stability permitting longer probing times with higher excitation energy; noble metal nanoparticle plasmons can enhance the signaling of fluorophores by two orders of magnitude or more creating exceptionally bright labels ( Jain et al . , 2008; Demchenko , 2013 ) . It has also been shown that fluorophore-conjugated antibodies can be used to colabel antigens in cobalt-filled Timm’s silver-intensified nervous system preparations ( Moos , 1993 ) , indicating that such preparations and others could be imaged with our new method and simultaneously with traditional ones appropriate for the emissions of the fluorescent antibody . Silver emissions in our study revealed fine in-focus 1 µm or smaller fibers associated with insect neurons in wholemount ( see Figure 4ci ) , which compared favorably to LSCM images of wholemounted insect neurons labeled with the Cy-5 fluorophore , shown previously to yield a high signal-to-background ratio ( Mesce et al . , 1993b ) . In contrast to most fluorophores , however , silver and gold nanoparticles do not photobleach and have higher extinction coefficients—they can also be imaged with super-resolution microscopy methods and electron beams that surpass the ca . 200 nm diffraction resolution limit of a conventional LSCM ( Zhang et al . , 2015 ) . Recently , local plasmon resonance and enhanced dark-field illumination , based on wavelength modulation , has resolved the coordinates of 80 nm silver nanoparticles with a 2 . 9 nm precision ( Zhang et al . , 2015 ) . Because silver and gold nanoparticle plasmon resonance can generate a high signal to noise ratio ( i . e . is bright ) , silver-labeled specimens like the ones we imaged have the potential to reveal structures at the nanometer scale when imaged with super-resolution microscopy methods . Although we have yet to determine the sizes and shapes of the silver nanoparticles deposited on the neurons we imaged , spectral and theoretical consideration should reveal such information ( Verbruggen et al . , 2013 ) . By chemically controlling the sizes and shapes of silver or gold nanoparticles deposited on neuronal surfaces , the intensity and spectral characteristics of such cells could be customized and enhanced ( Evanoff and Chumanov , 2005 ) . Other investigators have shown that nanoparticles of silver are themselves stabilized by the presence of cobalt . It has been noted that oxidation favors CoO over Ag2O , so that bimetallic nanoparticles with cobalt and silver in proximity would be expected to have reduced silver degradation , thus permitting more environmentally stable silver plasmonic behavior ( Saravanan et al . , 2011; Sachan et al . , 2012 ) . Note that cobalt was used to backfill the insect neurons in this study . The cobalt was first precipitated as CoS that then served as the nucleus for silver deposition in the Timm's silver intensification procedure ( see Methods ) . In traditional Golgi staining , black microcrystals of silver chromate form in stained cells , and these too have impressive longevity in excess of 90 years . Our method , which does not rely upon fluorescent dyes , holds enormous potential for stimulating a reexamination of archived preparations , whether of Golgi-stained or cobalt-silver labeled nervous systems . By using the cobalt-silver intensification technique in combination with our LSCM protocols , new preparations can also be generated and imaged with ease , precision , and with great detail in three dimensions . Such preparations are essentially permanent , and the information gathered from them increases the data available for characterizing neurons as individuals or as members of classes for comparative studies , adding to emerging neural information banks ( Parekh and Ascoli , 2013 ) . This archivability also promises to benefit clinical research and disease-related diagnostic techniques . Finally , just as plasmon resonance can explain the continued intensity of the red ( use of silver nanoparticles ) and yellow ( gold nanoparticles ) colors found in centuries-old medieval stained glass and other works of art ( Sciau , 2012 ) , metal-impregnated neurons too will likely never fade , neither in their information content nor in their intrinsic beauty .
Brightfield images were taken using either a Nikon Eclipse E800 microscope , which allowed for extended depth of field capabilities using Nikon Elements ( v . 4 ) software or with a Nikon Eclipse E400 that used ACT 1 software . Confocal images were taken with a Nikon A1 spectral confocal microscope employing settings that achieved optimal Nyquist resolution . Even though there was no danger of photobleaching the sample , laser power was adjusted to avoid over saturation of the image during data acquisition . Spectral confocal imaging facilitates the capture of the up-converted signal , but it is not required for our new protocol as the plasmon resonance signal can be captured at wavelengths shorter than the 561-nm laser line excitation used in this study . The reflective and fluorescence default modes on a conventional non-spectral LSCM are suboptimal as they will catch only the tails of the peak of the signal . Although the excitation laser lines of 561 and 640 nm are standard on many LSCMs , the optical filter sets to capture light shorter than the excitation wavelength are not standard . If the LSCM were not tunable to detect up-conversion , a spectral detector or optical setup would need to be procured from the manufacturer of a given confocal microscope . A blind deconvolution algorithm in Fiji , an open source platform for image analysis ( Schindelin et al . , 2012 ) , was used where indicated in figure legends to demonstrate further the ability of this method to show fine fibers . Additional spectral separation was performed using the Nikon Elements software on images where indicated within the figure legends . Where side views of confocal preparations were seen , these images were rotated using the software package Imaris . For the 3D rendering in Figure 2 , image stacks were opened in Nikon Elements software and plasmon resonance channels were unmixed and the resulting channels selected and extracted from the spectral image . Additionally , one channel composed of autofluorescence ( longer wavelength ) from the ganglion was extracted to provide spatial context to the volume rendering . Each channel was exported in TIFF format and scaled to 8-bit depth . Images demonstrating plasmon resonance were opened in FIJI . Tools for ‘oval’ , ‘rectangle’ or ‘freehand’ were used to select ROI in the images . The command ‘fill’ or ‘clear outside’ were then utilized to keep signal within the selected ROI or to discard signal outside of the ROI . Criteria for selection of a ROI included signal from the perineurial sheath as well as other signals that obscured motoneuron branches in 3D and orthogonal views . This method was done for all optical sections in the stack as well as for each extracted channel , autofluorescence channel excluded . All extracted channels were then merged into a single multichannel TIFF image . The resulting image was then pseudocolored in Elements . The autofluorescence channel and plasmon resonance channels were assigned unique color LUTS . For Figure 2 , plasmon resonance channels were pseudocolored yellow and the autofluorescence channel was pseudocolored blue . Images were imported into Imaris software , and 3D renderings were created using the Surpass view . Orthogonal images were obtained using the ‘Snapshot’ feature . Animations were generated in Imaris by inserting a key frame and a corresponding 360 degree horizontal rotation ( Mode = Maximum intensity projection , Rendering Quality = 1 . 000 , Frame= no ) . Orthogonal images were saved in TIFF format and 3D animations were saved as AVI files . Some renderings were later converted to MP4 file format for presentation using Handbrake software . Note: results can be achieved by blind unmixing spectral images and using the resulting channels for volume rendering and orthogonal images using Nikon Elements . 3D renderings can then be generated using Nikon Elements and/or FIJI software . To determine intensity values within confocal images , a selection of four images from the stack were used . A ROI with maximal labeling was chosen for measurements from four images as was a region within the sample background . A subset of four images was used in an attempt to keep intensity values comparable from sample to sample . In cases where the same preparation was imaged using different settings , an effort was made to make sure the same frames were used to obtain intensity values allowing for a more direct comparison . The labeled neuronal preparations shown in Figures 1 , 2 and 4 were obtained from grasshopper ( Schistocerca americana ) neurons located in abdominal ganglia and were backfilled with cobalt chloride and processed with the Timm’s silver intensification procedure ( see references below for protocols and recipes ) . The stretch receptor organ shown in Figure 3 was obtained from the hawkmoth , Manduca sexta , and filled with cobalt chloride and silver intensified as mentioned above . Additional preparations using other stains were loaned to us or purchased . NeuroScience Associates ( Knoxville , TN ) , provided us with the amino cupric stain of mouse brain tissue ( Figure 5f ) ; the Golgi stain of mouse Purkinje cells , representing a standard protocol ( Ranjan and Mallick , 2010 ) ( Figure 5c ) ; and the Gallyas stain of primate tissue , prepared according to previously described methods ( Braak and Braak , 1991 ) ( Figure 5d ) . The mammalian cerebral pyramidal neuron ( Figure 5b ) was a Golgi stained specimen purchased in 1969 from Ward’s Sciences ( West Henrietta , NY ) . The Bodian’s gold-based stain ( Figure 5e ) of motor end plates was purchased from Carolina Biological , Burlington , NC . The insect optic lobe ( Figure 5a ) was stained with Golgi and provided by Dr . Susan Fahrbach ( Wake Forest University ) prepared according to methods described previously ( Farris et al . , 2001 ) . We provided the cobalt-labeled slide ( Figure 5g ) . | A fresh slice of brain tissue has a fairly uniform appearance , even when viewed under a microscope . To study the neurons and other cells in the brain , scientists must therefore first prepare tissue samples using methods that make it easier to see certain kinds of cells , or particular features of them . One method that has been available for over a century is to use metal particles to stain some of the cells . For example , when the Spanish anatomist Santiago Ramón y Cajal investigated how brain cells – or neurons – are organized in the brain in the late 1880s , he made the cells visible by staining them with silver . This silver staining technique , called the Golgi method , bears the name of Camillo Golgi who first discovered it . Both Golgi and Ramón y Cajal are considered to be the pioneers of neuroscience , and shared the Nobel Prize in Physiology or Medicine in 1906 . Over the years , silver staining was superseded by the use of fluorescent probes . Light travels in the form of waves , and different colors of light have different wavelengths ( the distance between the peaks of the wave ) . Shining light of one specific color onto a fluorescent probe causes it to emit light of a longer wavelength . By detecting this emitted light , it is possible to visualize structures that contain the probes . In the late 1980s , the invention of the laser-scanning confocal microscope allowed highly detailed three-dimensional reconstructions of individual neurons to be obtained using these fluorescent labels . Unfortunately , the lifespan of fluorescent probes is limited by the fact that their fluorescence decreases with repeated use , in a process called photobleaching . Traditional silver stains avoid this problem , but standard confocal microscopy cannot obtain good images from metal-stained cells . Now , Thompson , Harley et al . have overcome this problem by using the confocal microscope in a new way to detect emitted light with shorter wavelengths than the light that was initially absorbed ( rather than the longer wavelength light normally detected ) . This protocol produced highly detailed three-dimensional images of individual metal-stained neurons that had been impregnated with silver or gold particles . The short wavelength light is thought to result from the activity of free electrons called plasmons that are present on the surface of small metal particles ( nanoparticles ) that are about one millionth of a centimeter in size . When plasmons absorb radiation of a specific wavelength , they vibrate rapidly and emit their excess energy in the form of light . Medieval craftsmen unknowingly exploited this same phenomenon when they added silver and gold particles to molten stained glass , producing windows with vivid red and yellow colors that are still vibrant today . A return to metal-based staining of brain tissue could produce similar longevity for today’s tissue samples . Equally , the procedure developed by Thompson , Harley et al . opens up the possibility of revisiting archived material with the tools of modern confocal microscopy . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"tools",
"and",
"resources",
"neuroscience"
] | 2015 | Plasmon resonance and the imaging of metal-impregnated neurons with the laser scanning confocal microscope |
Presynaptic terminals release neurotransmitters spontaneously in a manner that can be regulated by Ca2+ . However , the mechanisms underlying this regulation are poorly understood because the inherent stochasticity and low probability of spontaneous fusion events has curtailed their visualization at individual release sites . Here , using pH-sensitive optical probes targeted to synaptic vesicles , we visualized single spontaneous fusion events and found that they are retrieved extremely rapidly with faster re-acidification kinetics than their action potential-evoked counterparts . These fusion events were coupled to postsynaptic NMDA receptor-driven Ca2+ signals , and at elevated Ca2+ concentrations there was an increase in the number of vesicles that would undergo fusion . Furthermore , spontaneous vesicle fusion propensity in a synapse was Ca2+-dependent but regulated autonomously: independent of evoked fusion probability at the same synapse . Taken together , these results expand classical quantal analysis to incorporate endocytic and exocytic phases of single fusion events and uncover autonomous regulation of spontaneous fusion .
Synaptic terminals release neurotransmitters either spontaneously or in response to presynaptic action potentials ( APs ) ( Fatt and Katz , 1952 ) . In addition to the well-established role of AP-evoked neurotransmitter release in information transfer and processing , a growing number of studies assign a key role for spontaneous release in synaptic homeostasis and plasticity ( Sutton and Schuman , 2005; Kavalali et al . , 2011; Hawkins , 2013 ) . Recent work indicates that these two modes of neurotransmission are largely independent in terms of their presynaptic regulation as well as postsynaptic signaling consequences ( Sara et al . , 2005; Sutton et al . , 2006 , 2007; Atasoy et al . , 2008; Melom et al . , 2013; Nosyreva et al . , 2013; Wierda and Sorensen , 2014 ) . However , the molecular mechanisms that underlie the segregation of the two forms of release are only beginning to be elucidated ( Hua et al . , 2011; Pang et al . , 2011; Ramirez et al . , 2012; Bal et al . , 2013; Zhou et al . , 2013; Wang et al . , 2014 ) . There is strong evidence that synaptic vesicles recycle at rest in the absence of presynaptic APs and take up exogenous probes such as FM dyes , antibodies or horseradish peroxidase ( Ryan et al . , 1997; Murthy and Stevens , 1998; Sara et al . , 2005; Peng et al . , 2012; Kavalali and Jorgensen , 2014 ) . Studies also suggest that endocytic mechanisms mediating synaptic vesicle retrieval after spontaneous fusion diverge from those that trigger endocytosis after AP-evoked exocytosis ( Chung et al . , 2010; Peng et al . , 2012; Meng et al . , 2013 ) . However , visualizing single spontaneous vesicle fusion and retrieval events has been technically difficult as the stochastic nature and low probability of spontaneous fusion requires long-term imaging with high temporal resolution , which typically gives rise to significant photobleaching and potential photodamage . Earlier attempts at detecting spontaneous synaptic vesicle exo-endocytosis using capacitance measurements heavily relied on signal averaging and was confounded by susceptibility to contamination by capacitance changes unrelated to synaptic vesicle exocytosis ( Sun et al . , 2002; Yamashita et al . , 2005 ) . The current lack of insight into single synaptic vesicle retrieval leaves open the question of whether spontaneous synaptic vesicle exocytosis is tightly and temporally coupled to vesicle retrieval . In this study , we used the vesicular glutamate transporter or the vesicle protein synaptophysin as carriers for luminal pH-sensitive fluorescent probes and optimized imaging conditions to minimize photobleaching without compromising our ability to detect a majority of spontaneous synaptic vesicle fusion events . Our optical recording conditions were similar to our earlier work where we characterized single AP-evoked fusion events with a median probability of 0 . 2 ( Leitz and Kavalali , 2011 ) indicating that these settings enable visualization of release from single boutons ( Murthy et al . , 1997 ) . Under these conditions , we found that single synaptic vesicles are retrieved extremely rapidly ( <370 ms ) after spontaneous fusion indicating the fluorescence decay of individual events was dominated by vesicle re-acidification . These spontaneous fusion events were coupled to postsynaptic N-methyl-D-aspartate ( NMDA ) receptor-driven Ca2+ signals as supported by their temporal and spatial juxtaposition to D ( − ) -2-Amino-5-phosphonopentanoic acid ( AP-5 ) -sensitive fluorescence signals originating from a Ca2+ indicator targeted to postsynaptic densities . Surprisingly , we uncovered a significant fraction of putative multiquantal events that increased in prevalence at elevated Ca2+ concentrations . Furthermore , we could not detect appreciable correlation between the propensities of evoked and spontaneous fusion events at increasing Ca2+ concentrations . These experiments demonstrated that spontaneous fusion propensity in a given synapse is regulated autonomously and independently of evoked release probability . Taken together , these results expand classical quantal analysis ( Del Castillo and Katz , 1954; Boyd and Martin , 1956 ) to incorporate exocytic and endocytic phases of single fusion events and provide insight into the properties and regulation of single spontaneous fusion events in relation to their AP-evoked counterparts that originate from the same synaptic bouton .
To identify spontaneous fusion events , we first imaged synapses expressing vGlut-pHluorin at 8 Hz to minimize photobleaching and identified small , rapid increases in fluorescence ( Figure 1A–D ) in the presence of the voltage-gated Na+ channel blocker , tetrodotoxin ( TTX ) . These increases in fluorescence were distinguishable from noise and could be fit with a single Gaussian curve of mean amplitude of 623 ± 122 a . u ( Figure 1B , C ) and average ΔF/F of 3 . 8 ± 0 . 93% . These events decayed rapidly with a weighted average decay time constant ( heretofore referred to as ‘decay time’ ) of 0 . 29 ± 0 . 017 s . However , the decay times of these events were distributed according to a non-geometric β-function with a non-geometric average fluorescence decay time of 0 . 37 s and median decay time of 0 . 28 s ( Figure 1D ) . Moreover , in contrast to our earlier observations ( Leitz and Kavalali , 2011 ) , the fluorescence decay of these spontaneous events proceeded immediately following fluorescence increase without any observable dwell time . To confirm that these rapidly decaying increases in fluorescence were due to genuine spontaneous vesicle fusion , we sought to manipulate the rate of decay by addition of 50 mM Tris–HCl in the extracellular solution . Indeed , equiosmolar substitution of extracellular NaCl with 50 mM Tris–HCl slowed vesicle reacidification to an average decay time of 0 . 55 s with median decay time 0 . 32 s ( Figure 1E , F ) indicating that the decay phase of these events was dominated by vesicle re-acidification . We then compared these spontaneous fluorescence increases to those evoked by single action potential stimulation in the absence of TTX ( Figure 1G–I ) . We found that fusion events evoked by stimulation had a mean fluorescence amplitude of 641 ± 171 a . u . ( Figure 1H ) , similar to spontaneous fluorescence increases . However , these events decayed more slowly with average decay time of 0 . 83 s and median decay time of 0 . 83 s ( Figure 1I ) . Finally , to evaluate whether we were able to visualize the entirety of the fluorescent signal originating from spontaneous vesicle fusion events , we incubated neurons with the vacuolar ATPase inhibitor , folimycin ( 80 nM ) , and measured spontaneous increases in fluorescence ( Figure 1J , K ) . In the presence of folimycin , increases in fluorescence take on a staircase-like waveform with mean amplitude of 609 ± 195 a . u . ( Figure 1K ) , similar to that of spontaneous fusion events observed in the absence of folimycin . There were no significant differences in the mean amplitude between spontaneous increases in fluorescence with or without folimycin and those that were evoked by stimulation ( ANOVA p value >0 . 5 ) . However , when comparing amplitude distributions there was a significant difference between spontaneous events in the absence of folimycin and both spontaneous events in the presence of folimycin and action-potential evoked single vesicle fusion events ( KStest < 0 . 05 Dmax = 0 . 08 at bin 450 a . u . ) ( Figure 1L ) . This result suggests that while we were able to acquire the entire fluorescence waveform of the majority of spontaneous fusion events , there remains a small population of low amplitude events that we were unable to detect due to the rapid decay of fluorescence signals . Regardless , our results still indicate a clear divergence in the reacidification rate between vesicles that recycle at rest and those that fuse in response to action potentials . 10 . 7554/eLife . 03658 . 003Figure 1 . Spontaneous increases in vGlut1-pHluorin fluorescence decay rapidly . vGlut1-pHluorin was expressed via lentiviral infection in dissociated hippocampal cultures and neurons were imaged at 16–19 days in vitro . ( A ) Example image of vGlut1-pHluorin expression in NH4Cl ( 20 mM ) . Arrows indicate putative synapses . Scale bar is 5 μm . ( B ) Images were recorded in the presence of TTX , AP-5 and CNQX and spontaneous increases in fluorescence were observed . Example traces of spontaneous increases in fluorescence ( 3 left traces ) and the average of all events in this experiment ( right trace ) are shown . Raw data is in grey , black trace is the moving average of three points , and the blue trace is a fit to a first order decay . ( C ) The amplitudes of spontaneous increases in fluorescence are distinguishable from noise . White bars are the amplitude change of a random section of the fluorescence recording , while successful events are in grey ( for detection criteria see ‘Materials and methods’ ) . Blue dashed line is a Gaussian fit centered at 0 a . u . and standard deviation of 212 a . u . ( χ2 = 0 . 9 ) . Red dashed line is a Gaussian distribution fit to the data with mean of 623 ± 122 a . u . and CV = 0 . 17 ( n = 1178 events from four coverslips prepared from two cultures ) . ( D ) Distribution of decay times of spontaneous increases in fluorescence ( grey ) have a non-geometric ( Beta-function ) average of 0 . 37 s with upper bound = 7 and lower bound = 2 ( n = 1178 events from four coverslips prepared from two cultures ) . Negative amplitude events from the same puncta ( white ) have a non-geometric ( Beta-function ) average of 0 . 20 s ( n = 98 events from the same four coverslips ) . ( E ) Example traces of spontaneous increases in fluorescence in the presence of high ( 50 mM ) Tris-buffered extracellular solution . ( F ) The resulting spontaneous increases in fluorescence were slower to decay with average decay time of 0 . 55 s with upper bound = 3 . 8 and lower bound = 2 . 1 ( n = 1212 events from three coverslips generated from two cultures . ) Inset shows cumulative probability distribution of decay time in cells with extracellular solution containing HEPES ( Control; squares ) and 50 mM Tris ( circles ) ( KStest p < 0 . 001 with Dmax = 0 . 13 at 0 . 5 s ) . ( G ) Example traces of increases in fluorescence in response to single action-potential stimulation delivered at 0 . 05 Hz . ( H ) Amplitudes of stimulated increases in fluorescence are distinguishable from noise ( white bars and blue Gaussian distribution ) and could be fit with a Gaussian distribution centered at 641 ± 117 a . u . and CV = 0 . 26 ( red dashed line; χ2 = 0 . 94 ) . ( I ) Increases in fluorescence due to single-action potential stimulation delivered at 0 . 05 Hz were slower to decay with a non-geometric average of 0 . 83 s with upper bound = 4 . 4 and lower bound 3 . 6 and median decay time of 0 . 83 s ( n = 694 events from four coverslips generated from two cultures ) . ( J ) Example traces of spontaneous increases in fluorescence in the presence of folimycin ( 80 nM ) . With inhibition of the vATPase , vesicles cannot be reacidified and therefore increases in fluorescence do not decay ( n = 445 events from four coverslips generated from two cultures ) . ( K ) Amplitude distribution of spontaneous increases in fluorescence in the presence of folimycin are also distinguishable from noise and can be fit with a Gaussian curve with mean amplitude of 609 ± 195 a . u . and CV = 0 . 32 ( χ2 = 0 . 87 ) . ( L ) Cumulative probability histogram of amplitudes of spontaneous increases in fluorescence in 2 mM extracellular Ca2+ in the absence ( open square ) and presence of folimycin ( grey square ) , and amplitudes of fluorescence increases evoked by stimulation ( black square ) . The amplitude distributions of false positive events in 0 mM extracellular Ca2+ and folimycin ( open diamond ) and negative amplitude distributions of 0 mM extracellular Ca2+ false positive events ( open triangle ) were significantly smaller than amplitudes of putative spontaneous events with or without folimycin ( KStest p < 0 . 05 for both false positive and negative amplitude ) or stimulation-evoked fusion events ( KStest p < 0 . 05 again for both conditions ) , but were not significantly different from one another ( KStest p > 0 . 25 between false positives and negative amplitude distributions ) . The only significant difference between all distributions was between spontaneous increases in fluorescence with and without folimycin ( KStest p < 0 . 05 Dmax = 0 . 08 at bin 450 a . u . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03658 . 003 Finally , to assess the contribution of false positives to our analysis , we employed two approaches . First , we used the same detection criteria to identify potential ‘hits’ with a negative amplitude ( white bars in Figure 1D ) , that is , ‘events’ that were less than −2 times the standard deviation of the points prior in 2 mM Ca2+ . Setting our detection criteria for negative amplitudes revealed a false ‘hit’ rate of 0 . 065 per bouton per minute , suggesting that false hits contribute a negligible component to our measurements . Second , we analyzed the positive and negative amplitudes of transient fluorescence fluctuations seen in 0 mM Ca2+ where genuine event frequency as verified by the presence of folimycin ( leading to staircase increases in fluorescence ) is extremely low ( 29 events in 10 min from 200 puncta from four coverslips , that is ∼0 . 01 events/bouton/min compared to ∼0 . 5 events/bouton/min in 2 mM Ca2+ ) . Despite the presence of folimycin these false positive events decayed back to baseline and had a significantly lower fluorescence amplitude distribution ( KStest < 0 . 05; Figure 1L ) . When these 0 mM Ca2+ experiments were analyzed for a negative amplitude , the absolute value of these negative false events were not significantly different from the positive amplitude false events ( KStest > 0 . 3; Figure 1L ) . Taken together these data suggest that our detection criteria reliably identify events due to genuine spontaneous vesicle fusion with a low level of false positive hits ( ∼10% per recording ) . To investigate the relationship between these fast spontaneous increases in presynaptic fluorescence to postsynaptic receptor activation , we moved to a dual-color system utilizing the red-shifted pHluorin variant , pHTomato , fused to the presynaptic vesicle protein synaptophysin ( SypHTomato ) ( Li and Tsien , 2012 ) and the green fluorescent Ca2+ indicator , GCaMP5K , fused to the post-synaptic density protein 95 ( PSD-95-GCaMP5K ) ( Akerboom et al . , 2012 ) . Additionally , we excluded extracellular Mg2+ to allow Ca2+ influx through postsynaptic NMDA receptors . In this setting we can verify presynaptic vesicle fusion events in the red channel using the resulting Ca2+ influx in the green channel as a coincidence detector of a successful presynaptic fusion event ( Figure 2 ) . However , this system is not without caveats: first , because we are now monitoring two wavelengths our temporal resolution decreased from >8 Hz to ∼4 Hz; second , SypHTomato results in an elevated surface expression level compared to vGlut-pHluorin that required a post-hoc decay correction to compensate for photobleaching ( See ‘Materials and methods’ ) . Despite these issues , spontaneous increases in SypHTomato fluorescence were still detected and were distinguishable from noise with mean amplitude of 303 ± 92 a . u . ( Figure 2C ) and ΔF/F of 1 . 9 ± 0 . 67% . In 2 mM extracellular Ca2+ approximately 69% of spontaneous increases in SypHTomato elicited a PSD-95-GCaMP5K signal within ±1 frame , increasing to 83% between −1 to +5 frames . We include this −1 group because in our system we obtain only one time point for each cycle of imaging ( i . e . one frame of red channel and one frame of green channel are both assigned the same time point ) . Therefore , fusion events that may occur halfway through an imaging cycle would appear shifted by −1 frame . Spontaneous increases in fluorescence that were correlated ( within ±1 frame ) with Ca2+ signals were distinguishable from noise with mean amplitude of 170 ± 50 a . u . ( Figure 2D ) and ΔF/F of 2 . 2 ± 1 . 3% . We next compared the amplitudes of these spontaneous fusion events with increases in fluorescence elicited by single action potential stimulation ( Figure 2E , F ) . Stimulation resulted in events that were distinguishable from noise with mean amplitude of 345 ± 95 a . u . ( Figure 2F ) . Although both spontaneous and AP-evoked fluorescence increases were distinguishable from noise , the small amplitude of these events combined with the rapid decay times made precise measurements of individual decay times difficult . Therefore we averaged all events within an experiment to estimate the overall average decay times ( Figure 2G ) . We found that , as with vGlut-pHluorin , spontaneous fluorescence increases in SypHTomato decayed faster than those evoked by stimulation with average decay times of 0 . 5 ± 0 . 2 s and 1 . 0 ± 1 . 5 s , respectively . To confirm that we were able to observe the full fluorescence increase of spontaneous fusion events with our reduced temporal resolution , we incubated neurons in folimycin and observed stepwise increases in SypHTomato fluorescence . In the same experiments to verify that the increases in the GCaMP5K fluorescence were due to Ca2+ influx through NMDA receptors , we perfused an extracellular solution containing the NMDA receptor blocker AP-5 ( Figure 2H–S ) . We were able to identify spontaneous fusion events according to stepwise increases in SypHTomato fluorescence alone without relying on the coincidence with GCaMP5K mediated Ca2+ signals ( Figure 2H–S ) . Under these conditions , we observed spontaneous fusion events with mean amplitude of 316 ± 64 . 9 a . u ( Figure 2L ) similar to both those in the absence of folimycin and those due to stimulation . Events in the same synapses in the presence of AP-5 were also similar with mean amplitude of 270 ± 65 a . u . ( Figure 2M ) . The GCaMP5K signal in the absence of AP-5 had mean amplitude of 153 ± 57 . 6 a . u . ( Figure 2R ) , similar to without folimycin , but in the presence of AP-5 , the GCaMP signal amplitude decreased to the baseline noise , 13 . 2 ± 56 . 0 a . u . ( Figure 2S ) . Together these data complement our observations using vGlut1-pHluorin and show that spontaneous vesicle fusion events are coupled to postsynaptic NMDA receptor-driven Ca2+ signals and , following exocytosis , these vesicles are retrieved and re-acidified on a much faster time course than their AP-evoked counterparts . 10 . 7554/eLife . 03658 . 004Figure 2 . Dual color imaging shows that spontaneous fusion events are coupled to postsynaptic NMDA receptor-driven Ca2+ signals . ( A ) Example images of SypHTomato and PSD-95-GCaMP5K expression . Arrows indicate putative synapses . Scale bar is 5 μm . ( B ) Example traces of SypHTomato ( raw data in red , a moving average of three points in black and a fit of the decay time t in blue ) and PSD-95-GCaMP5K ( raw data in green , a moving average of three points in black and a fit of the decay time in blue ) . Because spontaneous increases in fluorescence were very small , we averaged events for each experiment representative average traces are shown at right . ( C ) The amplitude distribution of SypHTomato could be well fit with a Gaussian curve centered at 303 ± 92 a . u . ( χ2 = 0 . 86 ) ( n = 361 events from four coverslips generated from two cultures ) . ( D ) Amplitudes of PSD-95-GCaMP were distinguishable from noise and could be fit with a Gaussian curve with mean amplitude of 170 ± 50 ( χ2 = 0 . 98 ) ( n = 361 events from four coverslips generated from two cultures ) . ( E ) Example traces of SypHTomato fluorescence in response to single action potentials delivered at 0 . 05 Hz . Again , raw data is in red , a moving average of 3 points is in black and the decay time fit is in blue ( n = 443 events from four coverslips generated from two cultures ) . ( F ) Amplitudes of fluorescence increases evoked by action-potential stimulation could be well fit by a Gaussian curve with mean of 345 ± 95 a . u ( χ2 = 0 . 99 ) . ( G ) Averaged traces of spontaneous increases in fluorescence decayed back to baseline with decay time = 0 . 51 ± 0 . 08 s ( n = 4 ) , while events that responded to stimulation ( n = 3 ) decayed much slower t = 0 . 96 ± 0 . 06 ( Student's t-test p value <0 . 05 ) . ( H ) Example image of SypHTomato expression in the presence of folimycin . Arrows indicate putative synapses . The box is the region from which a line scan was taken ( shown in panel I ) . ( I ) Line scan of SypHTomato fluorescence . White dashed line indicates where on the corresponding trace the fluorescence step occurred , scale bar = 2 . 6 s . ( J ) Example traces of events in the presence and absence of AP-5 from the same synapses . ( K ) Average of traces from the experiment of step-wise increase in fluorescence in the presence and absence of AP-5 . ( L ) Increases in fluorescence in the presence of folimycin were separable from noise and could be fit with a Gaussian with mean amplitude of 316 ± 65 a . u . ( χ2 = 0 . 49 ) ( n = 154 events from four coverslips generated from two cultures ) . ( M ) The same synapses in the presence of AP-5 showed spontaneous increases in fluorescence that were still distinguishable from noise albeit with a slightly smaller amplitude distribution 270 ± 65 a . u . ( χ2 = 0 . 48 ) that was not significantly different from amplitudes in the presence of folimcyin ( not shown; KStest p < 0 . 01 ) ( n = 103 events from four coverslips generated from two cultures ) . ( N ) Example image of corresponding PSD-95-GCaMP5K fluorescence . White arrows indicate putative synapses . The box is the region from which a line scan was taken . ( O ) Line scan of PSD-95-GCaMP5K fluorescence signal that occurred at the same time as the above SypHTomato signal . ( P ) Example traces of events in the presence and absence of AP-5 . ( Q ) Average of traces in the presence and absence of AP-5 . In the presence of AP-5 , entry of Ca2+ into the postsynaptic terminal is prevented and thus there is no GCaMP5K signal . ( R ) PSD-95-GCaMP5K signals were separable from noise and could be fit with a Gaussian curve with mean amplitude of 153 ± 58 a . u . ( χ2 = 0 . 80 ) ( n = 154 events from four coverslips generated from two cultures ) . ( S ) In the presence of AP-5 , Ca2+ is prevented from entering the postsynaptic terminal and results in no detectable GCaMP5K events ( n = 103 events from four coverslips generated from two cultures . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03658 . 004 Using presynaptic imaging similar to that used in this study , we have previously shown that increasing extracellular Ca2+ increased synaptic vesicle fusion probability and increased the likelihood of multivesicular fusion events , which is observed as an increase in fluorescence amplitude ( Leitz and Kavalali , 2011 ) . To assess if this scenario also applied to vesicles that fuse spontaneously , we measured the amplitude of spontaneous increases in vGlut-pHluorin fluorescence in 2 , 4 and 8 mM extracellular Ca2+ ( Figure 3A–C ) . Surprisingly , we found that fluorescence amplitude distributions even in 2 mM extracellular Ca2+ did not fit well to a single Gaussian curve ( D'Agostino-Pearson omnibus K2 normality test p < 0 . 02 ) , nor could the sum of two Gaussian curves with means at integer multiples ( i . e . a quantal distribution of amplitudes indicative of two or more vesicles undergoing fusion simultaneously ) account for the distribution of amplitudes ( Figure 1B red lines ) . Instead , amplitude distributions in 2 and 4 mM extracellular Ca2+ ( Figure 3D , E , respectively ) were best fit by the sum of two Gaussian curves at 1 quantal mean ( 1q = 623 ± 91 a . u . , a slightly smaller standard deviation than in Figure 1 ) and 1 . 3 times the first mean ( 1 . 3q = 834 ± 111 a . u . ) ( Figure 3D , E black lines ) . Fluorescence amplitudes in 8 mM Ca2+ were best fit by the sum of three Gaussian curves distributed at 1 quantal mean ( 1q = 623 ± 91 a . u . ) , 1 . 3 times the first quantal mean ( 1 . 3q = 834 ± 111 a . u . ) and 2 times the first quantal mean ( 1246 ± 182 a . u . ) ( Figure 3F ) . This shift in fluorescence may arise from the limitations of our sampling rate and the rapid fluorescence decay time of these spontaneous events , therefore the 1 . 3q events may reflect two spontaneous events occurring close together in time but with a slight delay such that at our acquisition rate of >8 Hz , they appear to be a single event with a normalized amplitude of 1 . 3q ( Figure 3G ) . To estimate the delay between two events that would be required for such a fluorescence signal , we modeled the fluorescence decay of two hypothetical spontaneous fusion events with randomly generated amplitudes within the observed first Gaussian distribution ( 1q ) and fixed decay times of 371 ms ( the average decay time of spontaneous events in 2 mM Ca2+ ) . We used this model to generate 2000 hypothetical decay times and found that a fusion delay of 118 ms would result in an amplitude ∼1 . 3 times greater than the single quantal amplitude . We then estimated the percent of fusion events that were multivesicular ( Figure 3H ) and found that at 2 mM Ca2+ 17% of events were multivesicular , increasing to ∼30% in 4 mM Ca2+ and up to 70% of events were multivesicular in 8 mM Ca2+ . While this analysis favors preponderance of multivesicular spontaneous release at elevated Ca2+ concentrations , this shift in fluorescence amplitudes could be also due to unequal distribution of pHluorin molecules within the population of spontaneously recycling synaptic vesicles selectively mobilized at higher Ca2+ concentrations . In addition , this fluorescence shift could arise from the preferential fusion , at elevated Ca2+ conditions , of larger synaptic vesicles that contain more fluorophores . Given their cell biological complexity ( i . e . non-homogenous impact of elevated Ca2+ concentrations on vesicle populations with systematic differences ) we consider these two alternative scenarios unlikely . Finally , it is important to note that although this observed increase in event amplitude could be due to vesicle fusion from multiple synapses within our region of interest . This is more likely at elevated Ca2+ concentrations as the release probability estimates we present here at 2 mM Ca2+ are consistent with release from a single release site ( Murthy et al . , 1997 ) . 10 . 7554/eLife . 03658 . 005Figure 3 . Increasing extracellular Ca2+ increases multivesicular release . ( A–C ) Example traces of events in 2 mM ( A ) , 4 mM ( B ) and 8 mM ( C ) extracellular Ca2+ . Raw data are shown in grey , a moving average of 3 points is shown in black and the blue line indicates a fit of the decay . ( D–F ) Fluorescence amplitude distributions . White bars are noise fit by blue Gaussian distribution centered at 0 a . u . , grey bars are successful fusion events fit by multiple Gaussian curves ( black lines ) and a sum of Gaussian curves ( in red ) . ( D ) Fluorescence amplitudes in 2 mM extracellular Ca2+ were best fit by the sum of two Gaussian curves with mean amplitudes 623 ± 122 a . u . ( 1q ) and 834 ± 111 a . u . ( 1 . 3q ) ( χ2 = 0 . 88 ) . ( n = 1178 events from four coverslips generated from two cultures ) . ( E ) Fluorescence amplitudes in 4 mM extracellular Ca2+ were well fit with two similar Gaussian curves 623 ± 98 a . u . ( 1q ) and 834 ± 111 a . u . ( 1 . 3q ) ( χ2 = 0 . 70 ) ( n = 1593 events from six coverslips generated from two cultures ) . ( F ) Fluorescence amplitudes of 8 mM Ca2+ were best fit with the sum of three Gaussian curves with mean amplitudes 623 ± 91 a . u . ( 1q ) , 834 ± 111 a . u . ( 1 . 3q ) and 1246 ± 182 a . u . ( 2q ) ( χ2 = 1 ) ( n = 1110 events from three coverslips generated from two cultures ) . ( G ) Model of two hypothetical events with normalized amplitudes of 1 , decay times of 371 ms , and delay of onset of 118 ms , the same two events sampled at 120 ms results in fluorescence amplitude of 1 . 3 times a single event ( 1 . 3q ) . ( H ) The percent of events in 2 , 4 and 8 mM extracellular Ca2+ that have amplitudes greater than a single Gaussian distribution were defined as multivesicular . This suggests that there are more multivesicular events at increasing extracellular Ca2+ concentrations . DOI: http://dx . doi . org/10 . 7554/eLife . 03658 . 00510 . 7554/eLife . 03658 . 006Figure 3—figure supplement 1 . Increasing extracellular Ca2+ increases amplitude of spontaneous events detected by SypHTomato and PSD-95-GCaMP5K . ( A–C ) Fluorescence amplitude distributions of spontaneous increases in sypHTomato ( red ) were separable from noise ( white ) . Black lines are Gaussian distribution fits . ( A ) Fluorescence amplitudes of spontaneous increases in sypHTomato fluorescence in 2 mM extracellular Ca2+ . Events could be fit with a single Gaussian distribution ( black line ) with mean amplitude of 303 ± 92 a . u . ( χ2 = 0 . 86; n = 361 events from four experiments ) . ( B ) Fluorescence amplitudes of spontaneous increase in sypHTomato fluorescence in 4 mM extracellular Ca2+ . The amplitude of SypHTomato events increased in a non-quantal fashion as in the case of vGlut-pHluorin signals . Black line is the Gaussian curve due to single vesicle fusion as predicted from 2 mM Ca2+ experiments . ( n = 335 events from four experiments ) . ( C ) Fluorescence amplitudes of spontaneous increases in sypHTomato fluorescence increase further in 8 mM extracellular Ca2+ ( n = 573 events from five experiments ) ( D-F ) Fluorescence amplitudes of PSD-95-GCaMP5K ( green ) signals that corresponded with increases in sypHTomato were also separable from noise ( white ) . Black lines are Gaussian distribution fits . ( D ) Amplitude distribution of PSD-95-GCaMP5K signals that correlated with SypHTomato signals in 2 mM extracellular Ca2+ . PSD-95-GCaMP5K signals had a mean amplitude of 170 ± 50 a . u . ( n = 361 events from 4 experiments ) . ( E ) Amplitude distribution of PSD-95-GCaMP5K signals that correlated with SypHTomato signals in 4 mM extracellular Ca2+ . PSD-95-GCaMP5K signals could be well fit by the sum of two Gaussian distributions ( black dashed line ) with mean amplitudes of 191 ± 53 a . u . and 382 ± 106 a . u . ( n = 292 events from four experiments ) . ( F ) Amplitude distribution of PSD-95-GCaMP5K signals that correlated with SypHTomato signals in 8 mM extracellular Ca2+ . PSD-95-GCaMP5K signals could be well fit by the sum of three Gaussian distributions ( black dashed line ) with mean amplitudes of 208 ± 66 a . u . , 416 ± 132 a . u . and 624 ± 198 a . u . ( n = 472 events from five experiments ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03658 . 006 We also detected a similar Ca2+-dependent increase in the amplitude of spontaneous events identified in SypHTomato/PSD-95-GCaMP5K expressing neurons ( Figure 3—figure supplement 1A–F ) . In these neurons , there was an increase in GCaMP5K event amplitudes as a function of increasing extracellular Ca2+ indicative of the increased driving force of Ca2+ influx . Taken together , these data suggest that increasing extracellular Ca2+ increases the probability of two spontaneous fusion events occurring in close temporal proximity within a single synapse ( or possibly in adjacent release sites ) . We have previously shown that increasing extracellular Ca2+ increases the fluorescence decay times of vGlut1-pHluorin containing vesicles that fuse in response to stimulation ( Leitz and Kavalali , 2011 ) . Here , we wanted to determine if this property was applicable to vesicles that undergo fusion in the absence of stimulation . We increased extracellular Ca2+ to 4 mM and found that fluorescence signals decay with an average decay time of 0 . 44 s ( with upper bound = 4 . 8 and lower bound = 2 . 1 ) and median decay time of 0 . 29 s ( Figure 4A ) . In 8 mM extracellular Ca2+ , the average decay time was identical ( 0 . 44 s with upper bound = 4 . 8 and lower bound = 2 . 1 ) , and the median decay time did not appreciably change from 0 . 29 s to 0 . 28 s ( Figure 4B ) . When compared to 2 mM extracellular Ca2+ there was no significant difference in either 4 or 8 mM extracellular Ca2+ . Compared to the decay times of evoked-fusion events , decay times of spontaneous fusion events—regardless of extracellular Ca2+ concentration—were much faster ( Figure 4C ) . These data suggest that there is a fundamental difference in the kinetics of endocytosis and reacidification between vesicles that fuse spontaneously and those that fuse in response to stimulation . 10 . 7554/eLife . 03658 . 007Figure 4 . Increasing extracellular Ca2+ does not alter fluorescence decay time . ( A ) Distribution of fluorescence decay times in 4 mM extracellular Ca2+ can be fit with a beta-distribution with mean of 0 . 44 s with upper bound = 4 . 8 and lower bound = 2 . 1 ( χ2 = 0 . 9; n = 1308 events from six coverslips over three cultures ) . ( B ) Distribution of fluorescence decay times in 8 mM extracellular Ca2+ were fit with a beta-distribution with mean of 0 . 44 s with upper bound = 4 . 8 and lower bound = 2 . 1 ( χ2 = 1; n = 844 events from three coverslips from two cultures ) . ( C ) Cumulative probability histogram of decay times of spontaneous fluorescence events in 2 , 4 , and 8 mM extracellular Ca2+ showed no significant difference in decay time distributions ( e . g . KStest 2 mM Ca2+ vs 8 mM Ca2+: p > 0 . 3 with Dmax = 0 . 04 at 0 . 5 s ) . However , the decay times of all spontaneous events were significantly different from the decay times of evoked-fusion events ( for all comparisons KStest p < 0 . 01 with Dmax = 0 . 5 s ) . ( D ) Decay time did not correlate with amplitude of spontaneous events in 8 mM Ca2+ ( R2 = 0 . 0014 ) . 1q is the Gaussian mean of one event while 2q is the Gaussian mean of two simultaneous events , 1 . 3q is the amplitude calculated in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 03658 . 00710 . 7554/eLife . 03658 . 008Figure 4—figure supplement 1 . Increasing extracellular Ca2+ does not alter decay time of spontaneous increases in sypHTomato fluorescence . ( A ) Example average traces of spontaneous increases in sypHTomato fluorescence ( red ) single experiments in 2 mM ( left ) , 4 mM ( middle ) , and 8 mM ( right ) extracellular Ca2+ with decay time fit in blue . ( B ) Example average traces from the same experiments as A of PSD-95-GCaMP5K events ( green ) that correspond with sypHTomato spontaneous increases in fluorescence in 2 mM ( left ) , 4 mM ( middle ) , and 8 mM ( right ) extracellular Ca2+ with decay time fit in blue . ( C ) Decay time does not change as a function of extracellular Ca2+ , however increases in fluorescence due to stimulation are slower to decay than spontaneous increases in fluorescence ( p < 0 . 05 One-way ANOVA with Bonferroni post hoc analysis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03658 . 008 Our earlier work showed that an increase in the number of vesicles that fuse slows the fluorescence decay time of fusion events ( Leitz and Kavalali , 2011 ) . Therefore , here , we analyzed the decay times of events in 8 mM Ca2+ as a function of amplitude and found that there was no correlation in event size and fluorescence decay times ( Figure 4D ) . Furthermore , we also found a similar trend of Ca2+-independence in neurons expressing SypHTomato/PSD-95-GCaMP5K ( Figure 4—figure supplement 1A and B ) . All of these decay times , regardless of extracellular Ca2+ levels , were faster than those observed during stimulation-evoked fusion ( Figure 4—figure supplement 1C ) . It is important to note that the rate of decay is approaching the limit of our temporal resolution , which is lowered in an attempt to reduce photobleaching during long imaging episodes required to identify spontaneous fusion events . Thus , it is possible that we either cannot detect a significant change in decay times , or we are missing a subset of ultra fast decay times . Regardless , these data not only suggest that the mechanisms controlling the rate of synaptic vesicle reacidification are not the same for spontaneous and stimulation-evoked vesicle fusion but that endocytosis of synaptic vesicles released at rest is rapid even during multivesicular fusion events , unlike retrieval of vesicles released in response to stimulation . It is well established that increasing the concentration of extracellular Ca2+ increases spontaneous vesicle fusion rate at rest ( Lou et al . , 2005; Sun et al . , 2007; Xu et al . , 2009 ) . Next , we wanted to know if we see the same increase in our system . Surprisingly , we found no significant change in vesicle fusion frequency between 2 mM and 8 mM extracellular Ca2+ ( with 0 . 56 ± 0 . 02 fusion events per min in 2 mM extracellular Ca2+ and 0 . 49 ± 0 . 02 fusion events per min in 8 mM extracellular Ca2+; p value >0 . 1 ) beyond the increase in multivesicular events we reported earlier ( Leitz and Kavalali , 2011 ) ( Figure 5A ) . We then counted all events with amplitudes within the first quantal mean as a single event , and events with larger amplitudes as two events ( Figure 5B ) . Even with this criterion we did not detect a significant difference between 2 mM extracellular Ca2+ and 8 mM extracellular Ca2+ ( with 0 . 70 ± 0 . 02 and 0 . 76 ± 0 . 03 fusion events per min in 2 and 8 mM extracellular Ca2+ , respectively; p value >0 . 1 ) . However , when vesicle reacidification was buffered using 50 mM Tris–HCl ( Figure 5C ) we were able to detect a slight shift in vesicle fusion rate in 8 mM extracellular Ca2+ compared to 2 mM Ca2+ ( 0 . 42 ± 0 . 02 fusion events per min in 2 mM Ca2+ and 0 . 53 ± 0 . 02 events per min in 8 mM extracellular Ca2+; p value <0 . 05 ) . Addition of folimycin further exacerbated this effect and clearly showed a ∼2 . 7-fold increase in spontaneous fusion rate as a function of extracellular Ca2+ ( Figure 5D; 0 . 41 ± 0 . 02 events per min in 2 mM extracellular Ca2+ and 1 . 1 ± 0 . 02 events per min in 8 mM extracellular Ca2+; p value <0 . 05 ) . This increase in spontaneous fusion rate comes close to the threefold to fourfold increase we detect using postsynaptic electrophysiological recordings under the same conditions ( data not shown ) . Note that in both the Tris-buffered and folimycin-containing experiments , events were analyzed irrespective of amplitude . Taken together , these data indicate that increasing extracellular Ca2+ increases the vesicle fusion rate in our system in a manner that is detectable in the presence of folimycin . However , this increase in spontaneous fusion rate cannot be detected without increasing extracellular pH buffering or inhibiting vesicle re-acidification , indicating that elevated extracellular Ca2+ specifically increases fusion of synaptic vesicles that are retrieved and re-acidified rapidly , below the temporal resolution of our imaging protocol . This finding suggests that although elevated Ca2+ levels do not alter the kinetics of slower events detectable without the aid of altered re-acidification , Ca2+ elevation generates a new population of events that are faster in their kinetics . This observation is consistent with earlier work , which demonstrated that increasing extracellular Ca2+ facilitates the propensity of vesicle retrieval events with fast kinetics ( Ales et al . , 1999; Wu et al . , 2009 ) . 10 . 7554/eLife . 03658 . 009Figure 5 . Increasing extracellular Ca2+ increases spontaneous fusion rate . ( A ) Cumulative probability histogram of spontaneous event rate per synapse per minute in 2 mM and 8 mM extracellular Ca2+ ( n = 175 synapses from four coverslips for both conditions , KStest p = 0 . 55 with Dmax = 0 . 8 fusion events per minute per synapse , with averages of 0 . 56 ± 0 . 02 and 0 . 49 ± 0 . 02 ( Student's t-test p value = 0 . 2 ) events per minute per synapse for 2 and 8 mM Ca2+ , respectively ) . ( B ) Cumulative probability histogram of spontaneous event rate per synapse per minute in 2 mM and 8 mM extracellular Ca2+ adjusted to count large amplitude events as two vesicle fusion events ( n = 175 synapses from four coverslips for both conditions , KStest p < 0 . 05 with Dmax = 0 . 6 fusion events per minute per synapse , with averages of 0 . 70 ± 0 . 02 and 0 . 76 ± 0 . 03 ( t-test p value = 0 . 1 ) events per minute per synapse for 2 and 8 mM Ca2+ , respectively . ( C ) Cumulative probability histogram of spontaneous event rate per synapse per minute in Tris-buffered ( 50 mM ) 2 mM and 8 mM extracellular Ca2+ solutions ( n = 150 synapses from three coverslips for both conditions , KStest p < 0 . 05 with Dmax = 0 . 5 fusion events per minute per synapse , with average 0 . 4 ± 0 . 02 and 0 . 5 ± 0 . 02 ( t-test p value <0 . 05 ) fusion events per minute per synapse for 2 and 8 mM Ca2+ , respectively . ( D ) Cumulative probability histogram of spontaneous event rate per synapse per minute in 2 and 8 mM extracellular Ca2+ solution containing folimycin ( n = 175 synapses from four coverslips for 2 mM Ca2+ and 216 synapses from four coverslips for 8 mM Ca2+; KStest p < 0 . 05 Dmax = 0 . 8 fusion events per synapse per minute , with average rates of 0 . 4 ± 0 . 02 and 1 . 1 ± 0 . 02 ( p value <0 . 05 ) fusion events per minute per synapse for 2 and 8 mM Ca2+ , respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03658 . 009 Next , to evaluate the relationship between the rate of spontaneous synaptic vesicle fusion and the probability of evoked vesicle fusion in a synapse , we incubated neurons in folimycin in order to visualize all events and delivered single action potential stimulations with long inter-stimulus intervals ( 15–30 s ) . Increases in fluorescence that occurred within 1 s of a stimulation were considered to be due to evoked release ( note that we cannot then differentiate between synchronous vesicle fusion and fast asynchronous fusion ) while increases in fluorescence that did not match with a stimulation time were labeled as spontaneous vesicle fusion events ( Figure 6A ) . We found that while increasing extracellular Ca2+ concentrations increased both the overall rate of spontaneous fusion ( from 0 . 52 ± 0 . 028 events per min at 2 mM Ca2+ to 0 . 76 ± 0 . 049 events per min in 8 mM extracellular Ca2+ ) as well as the overall mean evoked fusion probability ( from 0 . 15 ± 0 . 008 in 2 mM Ca2+ to 0 . 44 ± 0 . 016 in 8 mM extracellular Ca2+ ) , at increasing Ca2+ concentrations the spontaneous vesicle fusion rate and evoked fusion probability estimates obtained from individual release sites did not show appreciable correlation ( at 2 mM Ca2+: slope = 1 . 0 , R2 = 0 . 08; at 4 mM Ca2+: slope = −0 . 14 , R2 = 0 . 006; at 8 mM Ca2+: slope = 0 . 47 , R2 = 0 . 02 ) ( Figure 6B–D ) . Although , a large fraction of the nerve terminals ( >70% ) were capable of maintaining both evoked and spontaneous neurotransmission across all Ca2+ concentrations , the rate of spontaneous transmission and the probability of successful AP-stimulated transmission are not correlated within a given release site . Furthermore , these data imply that the processes that control the kinetics of trafficking vesicles released at rest and those released in response to stimulation are distinct not only between synaptic terminals , but within a single synaptic terminal . 10 . 7554/eLife . 03658 . 010Figure 6 . Spontaneous vesicle fusion rate and stimulation-evoked fusion probability do not correlate at a given synapse . ( A ) Example traces of fusion events in the presence of folimycin . Neurons in 2 mM Ca2+ extracellular solution were stimulated with 1 AP delivered at 0 . 1 Hz while neurons in 4 and 8 mM Ca2+ were stimulated with 1 AP delivered at 0 . 033 ( 30 s inter-stimulus interval ) due to the higher probability of release . Fusion events were categorized as spontaneous or evoked by their temporal distance from the stimulation time , with events within ±1 s of stimulation selected as evoked fusion events . Note that spontaneous and evoked events are both quantal , indicated by the dashed grey line . ( B ) There is limited correlation between spontaneous fusion rate and evoked fusion probability in 2 mM Ca2+ . Distributions were best fit with a linear trend line with slope = 1 . 0 ( Pearson's correlation coefficient R2 = 0 . 08 , p value <0 . 01 and Spearman r = 0 . 16 , p value = 0 . 02; n = 200 synapses from four coverslips ) . ( C ) The correlation between spontaneous and evoked transmission decreases further as extracellular Ca2+ increases to 4 mM best fit line with slope = 0 . 1 ( Pearson's R2 = 0 . 01 , p value = 0 . 17 and Spearman r = 0 . 032 , p value = 0 . 58; n = 311 synapses from six coverslips ) . ( D ) The correlation remains low between spontaneous and evoked transmission in 8 mM extracellular Ca2+ with best fit line with slope = 0 . 5 ( Pearson's R2 = 0 . 02 , p value = 0 . 045 and Spearman r = 0 . 066 , p value = 0 . 35; n = 240 synapses from four coverslips ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03658 . 010
To visualize spontaneous exocytosis and endocytosis of single synaptic vesicles , we used a lentiviral system to express the pH-sensitive GFP ( pHluorin ) fused to the luminal domain of the vesicular glutamate transporter ( vGlut1 ) in hippocampal neurons . In an earlier study , we employed the same approach to investigate trafficking of single synaptic vesicles that fuse in response to AP stimulation where the time of stimulation can be used to identify a successful fusion event ( Leitz and Kavalali , 2011 ) . For spontaneous vesicle fusion , however , such a time stamp does not exist; therefore , to confirm our findings using vGlut1-pHluorin , we also used dual color imaging with a red-shifted pHluorin variant ( pHTomato ) fused to the luminal domain of the synaptic vesicle protein synaptophysin ( SypHTomato ) and a green Ca2+-sensitive probe ( GCaMP5K ) fused to the post-synaptic density protein 95 ( PSD-95-GCaMP5K ) . In this setting , we could monitor presynaptic vesicle fusion and postsynaptic Ca2+ entry upon NMDA receptor activation . We found that in both systems , synaptic vesicles that fuse spontaneously are retrieved and re-acidified much faster than their counterparts that fuse in response to stimulation . The rapid decay was not due to lateral diffusion of the probe because in the presence of folimycin , these events remained stable without decay and fluorescence accumulated in a stepwise fashion . It is also important to note that as these experiments were performed at room temperature , we expect that at higher , more physiological temperatures the rates of endocytosis and reacidification may be much faster than are reported here ( Renden and von Gersdorff , 2007 ) . Additionally , we emphasize that the data presented here does not imply that the postsynaptic responses arising from spontaneous and evoked vesicle fusion are necessarily different . However , earlier work from our group ( Atasoy et al . , 2008; Sara et al . , 2011 ) does suggest there is a physical segregation of postsynaptic receptor activation patterns in response to evoked and spontaneous fusion events . Unfortunately , here we cannot measure evoked quantal postsynaptic Ca2+ signals during stimulation , as in our preparation , field stimulation itself causes direct postsynaptic effects . Therefore , in this study we did not examine the kinetic properties of evoked postsynaptic quantal events . Instead , we emphasize that the retrieval and reacidification of spontaneously fusing vesicles are distinct from vesicles that fuse in response to stimulation . Unlike stimulation-evoked responses , these spontaneous increases in fluorescence decayed immediately without a detectable pause after the initial increase in fluorescence ( dwell time ) , suggesting a minimal surface residency time of the pHluorin or pHTomato probes . Taken together , these results indicate that spontaneous synaptic vesicle exocytosis and endocytosis are tightly coupled processes and that the decay phase of these transients was mainly due to vesicle re-acidification upon endocytosis ( Alabi and Tsien , 2012 ) . Although we see extremely fast endocytosis of spontaneous recycling vesicles , large probes such as antibodies against the luminal domain of synaptotagmin1 or horseradish peroxidase are known to label spontaneously endocytosing vesicles ( Sara et al . , 2005; Fredj and Burrone , 2009 ) . Therefore , spontaneously endocytosing vesicles may still form a fairly large fusion pore or may even fully collapse onto the plasma membrane—albeit without lateral dispersion of their protein components—despite being retrieved quickly . However , we cannot fully exclude the possibility that some spontaneous fusion events may form narrow fusion pores where uptake of large probes can be curtailed as seen after spontaneous fusion of peptidergic vesicles ( Vardjan et al . , 2007 ) . The rate of endocytosis and vesicle reacidification found here is considerably faster than some of the previous reports that describe endocytosis with a time constant of ∼14 s ( Granseth et al . , 2006; Balaji and Ryan , 2007 ) and vesicle reacidification rate of ∼4 s ( Atluri and Ryan , 2006 ) . The discrepancy between our work and these earlier studies may be attributed several differences . First , our previous work has indicated that in response to increasing Ca2+ concentrations as well as in boutons with higher release probability the dwell time and the decay time of single vesicle fusion events increase ( Leitz and Kavalali , 2011 ) . Therefore , we believe it is critical to select boutons with a wide range of release probabilities for single vesicle analysis . Our current work extrapolates this observation to spontaneous release events where we detect events with even faster kinetics . The vesicle re-acidification rates we estimate , on the other hand , are in agreement with studies using synaptobrevin-pHluorin ( Gandhi and Stevens , 2003 ) as well as synaptophysin-pHluorin based measurements ( Zhang et al . , 2009b ) . Furthermore , it is important to note that some of these earlier studies were performed in extracellular solution containing 25 mM HEPES while here , and in our previous work ( Leitz and Kavalali , 2011 ) , we chose a lower HEPES concentration of 10 mM as HEPES is known to have phototoxic effects at higher concentrations ( Zigler et al . , 1985 ) . Why is synaptic vesicle re-acidification after spontaneous fusion faster than vesicle re-acidification after evoked fusion ? We propose three non-mutually exclusive scenarios that can explain this difference . First , the pH buffering capacity of vesicles that endocytose after AP stimulation could be higher . Second , the function of the v-ATPase on vesicles that endocytose after AP stimulation may be slowed down since this complex is known to incorporate Ca2+ sensor proteins ( Zhang et al . , 2008 ) and was recently shown to be differentially regulated during evoked vs spontaneous fusion ( Wang et al . , 2014 ) . Finally , vesicles endocytosed during activity may rapidly trigger formation of larger vesicular structures that are expected to be slower to re-acidify due to their larger volume , consistent with findings from capacitance measurements in salamander photoreceptors ( Van Hook and Thoreson , 2012 ) as well as recent electronmicrographic analysis of endocytosis in hippocampal synapses after rapid high-pressure freeze fixation ( Watanabe et al . , 2014 ) . In this study , we also investigated the Ca2+-dependent regulation of spontaneous synaptic vesicle fusion events . At elevated Ca2+ concentrations we detected an increase in the amplitudes of fusion events consistent with exocytosis of multiple synaptic vesicles . As noted above , at our resolution we cannot exclude the possibility that the observed increase in amplitude could be due to vesicle fusion from multiple release sites within a region of interest , especially if elevated Ca2+ concentrations can activate adjacent release sites resulting in apparent multivesicular release albeit originating from two neighboring release sites . The discrepancy between the increases in mEPSC frequency we detect in electrophysiological recordings vs optical recordings of spontaneous events could indicate that some of the multivesicular events are occurring at neighboring synapses activated at elevated Ca2+ . However , our measurements of stimulation-evoked vesicle fusion probability at 2 mM Ca2+ are consistent with the dominance of single release sites in our analysis . Additionally , it is possible that increasing extracellular Ca2+ specifically promotes the spontaneous fusion of synaptic vesicles containing elevated copy number of vGlut-pHluorin endowed by either a larger physical size of the vesicle or an elevated intrinsic copy number of vGlut-pHluorin . However , in part due to the similarity between our results and the bursting activity observed previously using single-synapse recordings in hippocampal neurons ( Abenavoli et al . , 2002 ) ; we consider the increase in fluorescence amplitude as arising from the rapid successive fusion of multiple vesicles as the most cell biologically parsimonious . If this interpretation is correct , we also found that fluorescence signals originating from these multivesicular events were not slower in their rate of decay , which contrasts our earlier observations on evoked multivesicular fusion events ( Leitz and Kavalali , 2011 ) , indicating that spontaneous exocytic load ( i . e . the number of spontaneously fused vesicles on the plasma membrane ) does not significantly impact retrieval kinetics . Alternatively , even if the larger amplitude spontaneous events were due to the fusion of enlarged synaptic vesicles or synaptic vesicles containing more fluorophore , this interpretation would still imply that the size of synaptic vesicles or the number of vGlut-pHluorin molecules within a synaptic vesicle ( more membrane or more cargo , that is the exocytic load ) do not impact their retrieval kinetics at rest . In contrast to our earlier findings with vesicles that fuse in response to APs ( Leitz and Kavalali , 2011 ) , we found that the kinetics of endocytosis of spontaneous fusion events were not slowed in response to elevated extracellular Ca2+ concentrations . In all likelihood , at 8 mM Ca2+ the average kinetics of fluorescence decay became faster due to emergence of events that could only be detected following inhibition of re-acidification . Therefore , with increasing extracellular Ca2+ there was a clear increase in the rate of spontaneous fusion and the number of vesicles that undergo fusion . The relative insensitivity of vesicle retrieval kinetics to Ca2+ suggests that synaptic vesicle retrieval after spontaneous and evoked fusion is regulated via diverse mechanisms consistent with differential dependence of the two forms of endocytosis on distinct dynamin isoforms ( Chung et al . , 2010; Raimondi et al . , 2011; Meng et al . , 2013 ) and distinct postendocytic transport machineries ( Peng et al . , 2012 ) . Finally , in our system we were able to directly compare the rate of spontaneous fusion and evoked-fusion probability within a single synapse . This analysis did not reveal a significant correlation between spontaneous fusion rate and evoked fusion probability estimates from individual release sites , further supporting the notion that these two modes of neurotransmission are controlled and maintained independently ( Sara et al . , 2005; Atasoy et al . , 2008; Fredj and Burrone , 2009; Melom et al . , 2013; Peled et al . , 2014; Wang et al . , 2014 but see Groemer and Klingauf , 2007; Wilhelm et al . , 2010 ) . Importantly , at elevated Ca2+ concentrations the propensity to fuse of each of the two forms of vesicle fusion was increased ( Figure 6 ) . However , consistent with their less steep dependence on Ca2+ , spontaneous fusion events showed a milder 1 . 5–2-fold increase compared to the threefold increase detected in the propensity of evoked fusion events . Nevertheless , their lack of correlation persisted suggesting a divergence in the mechanisms that regulate Ca2+ sensitivity of evoked and spontaneous fusion events ( Xu et al . , 2009; Groffen et al . , 2010; Pang et al . , 2011; Vyleta and Smith , 2011; Ermolyuk et al . , 2013 ) . This difference in Ca2+ regulation of spontaneous and evoked fusion probability may underlie differential sensitivity of the two forms of neurotransmission of certain neuromodulators and Ca2+ signaling pathways ( Peters et al . , 2010; Ramirez and Kavalali , 2011; Bal et al . , 2013 ) . Recent studies in the Drosophila neuromuscular junction have shown that a substantial fraction of release sites carry out exclusively spontaneous or evoked neurotransmitter release ( Peled et al . , 2014; Walter et al . , 2014 also see ; Melom et al . , 2013 ) . In our measurements , we detected a substantial overlap of release sites that are capable of both forms of neurotransmission , in agreement with our earlier estimates from lower temporal resolution experiments ( Atasoy et al . , 2008 ) . This discrepancy may be consistent with the premise that in immature presynaptic release sites , spontaneous neurotransmission dominates and release gradually shifts towards evoked transmission during synapse maturation ( Polo-Parada et al . , 2001; Mozhayeva et al . , 2002; Andreae et al . , 2012 also Walter et al . , 2014 ) . Taken together , the findings we present here provide insight into the segregation of spontaneous and evoked neurotransmitter release mechanisms at the level of single synaptic vesicle fusion events . The differential regulation of spontaneous vesicle fusion suggests it has a role in neuronal signaling distinct from information transfer patterns mediated by evoked release , even within a single synapse .
Dissociated hippocampal neurons were cultured from postnatal day 0–3 Sprague Dawley rats of either sex as described previously ( Kavalali et al . , 1999 ) . At 4 days in vitro ( DIV ) , cultures were infected with lentivirus expressing vGlut-pHluorin or with SypHtomato and PSD-95-GCaMP , and experiments were conducted between 15–20 DIV when synapses reach maturity ( Mozhayeva et al . , 2002 ) . All experiments were performed at room temperature . In these experiments we relied on a lentiviral expression system . The vGlut-pHluorin construct was a generous gift from Drs Robert Edwards and Susan Voglmaier ( University of California , San Francisco ) . A modified , synaptophysin pHTomato was a generous gift from Dr Richard Tsien ( New York University Medical Center and Stanford University ) . The primers: ATATggatccggtggttctggtgtgagcaagggcgaggagaataacatggccatcatcaaggagttcatgcgcttcaag ( pHTomato . FIX . F ) and atataccggtaccagaaccacccttgtacagctcgtccatgccgccggtggagtggcggccc ( pHTomato . FIX . R ) were used to return pHTomato to the published version and attach small flexible linkers . All lentiviruses were prepared by transfection of human embryonic kidney ( HEK ) 293-T cells with the plasmid of intrest together with viral coat and packaging protein constructs ( pVSVG , pRsv-Rev , and pPRE ) using FuGENE 6 ( Promega , Madison , WI ) . 3 days after transfection , virus was harvested from HEK 293-T cell-conditioned media and added to neuronal media at 4 DIV . Single-wavelength experiments were performed using an Andor iXon Ultra 897 back-illuminated EMCCD camera ( Model no . DU-897U-CSO-#BV ) collected on a Nikon Eclipse TE2000-U microscope with a 100X Plan Fluor objective ( Nikon ) . For illumination we used a Lambda-DG4 ( Sutter instruments , Novato , CA ) with FITC filter . Images were acquired at ∼8 Hz with an exposure time of 100 ms and binning 4 ( Using Nikon Elements Ar software ) . For analysis , square regions of interest ( ROIs ) with length and width of 2 . 5 μm were generated and the resulting fluorescence values were exported to Microsoft Excel for analysis . Successful fusion events were those where the average of 3 points was greater than twice the standard deviation of 17 points ( ∼2 . 1 s ) prior provided the recordings were stable for at least 78 frames ( ∼10 s ) preceding the increase in fluorescence . Additionally , at least one additional point after the initial fluorescence increase had to be greater than twice the standard deviation of the 17 points prior; this excluded large single point increases in fluorescence from which decay times could not be determined . Noise measurements were obtained in an analogous fashion: by subtracting the average of 3 points from the mean of 17 points prior during a random period during sampling . Decay times were analyzed in Clampfit ( Molecular Devices , Sunnyvale , CA ) by fitting raw data with a single exponential decay using Levenberg–Marquardt least sum of squares minimizations . For all control experiments , the extracellular solution was a modified Tyrode's solution containing ( in mM ) : 150 NaCl , 4 KCl , 10 glucose , 10 HEPES , 2 MgCl and varying concentrations of CaCl2 ( 2 , 4 , and 8 mM ) , pH 7 . 4 ( 310 Osm ) . In experiments with folimycin ( Concanamycin A , Sigma , St . Louis , MO ) , a final concentration of 80 nM was used . For Tris-buffered experiments , solutions contained ( in mM ) : 108 NaCl , 4 KCl , 2 MgCl2 , 10 glucose and 50 Tris–HCl and 2 or 8 CaCl2 . For all Tris buffering experiments , neurons were allowed to equilibrate in extracellular solution for at least 10 min prior to imaging . We elected to use a Tris-buffered extracellular solution because it does not require continuous bubbling during perfusion; a prerequisite of bicarbonate based buffers . Moreover , it has been shown that incubation in Tris can buffer , and thus slow , vesicle reacidification ( Gandhi and Stevens , 2003; Zhang et al . , 2009a ) . All solutions were adjusted to pH 7 . 4 with NaOH and 310 Osm prior to use . To prevent network activity , postsynaptic ionotropic glutamate receptor antagonists 6-cyano-7-nitroquinoxaline-2 , 3-dione ( CNQX 10 μM; Sigma ) and D ( − ) -2-Amino-5-phosphonopentanoic acid ( AP-5 50 μM; Sigma ) were added to the experimental solutions . To prevent spontaneous action potential generation , we incubated neurons in 1 µM tetrodotoxin ( TTX ) . For simultaneous spontaneous activity and evoked stimulation experiments , TTX was omitted from the extracellular solution and neurons were stimulated using parallel bipolar electrodes ( FHC ) delivering 15–20 mA pulses with pulse width of 1 ms . At the end of each experiment , Tyrode containing 20 mM NH4Cl was added to help identify putative synaptic boutons . Due to the lower intrinsic release probability in 2 mM extracellular Ca2+ , we gave twice as many stimulations ( 20 APs delivered 15 s apart ) as in 4 and 8 mM extracellular Ca2+ ( 10 APs delivered 30 s apart ) , thus evoked fusion probability estimates in 2 mM Ca2+ are in multiples of 0 . 05 and not 0 . 1 . For dual color experiments , FITC and TRITC filters ( Chroma Technology , Bellows Falls , VT ) were inserted into the Lamda-DG . Images were collected on the same hardware as above but with increased acquisition interval of 180 ms ( using 40 ms FITC and 140 ms TRITC excitation intervals ) . The extracellular solution was the same as above , however 2 , 3-dihydroxy-6-nitro-7-sulfamoyl-benzo[f]quinoxaline-2 , 3-dione ( NBQX 10 μM; Tocris , UK ) was used instead of CNQX as it is a more specific AMPAR antagonist . AP-5 and Mg2+ were omitted from all experiments to allow Ca2+ entry through NMDARs . Full waveforms of sypHTomato and GCaMP5K fluorescence were fit with a double exponential decay time and linearized to correct for photobleaching . Corrected fluorescence waveforms were analyzed as above for fluorescence increases . Once fusion events were identified , the raw data at that time was linearly corrected according to the baseline slope from 15 s before the event . This method proved successful as in the absence of folimycin events would decay back to baseline but in the presence of folimcyin the events had a staircase fluorescence waveform . All sypHTomato and GCaMP5K events were then aligned and averaged for each experiment . Hypothetical single-vesicle fusion events were generated in Microsoft Excel with amplitudes of 623 ± 122 a . u and decaying as a single exponential with decay time of 371 ms ( as determined by the non-geometric average decay time of events in 2 mM Ca2+ fit using a Beta-function ) . 4 , 000 hypothetical fusion events were then generated with varying amplitudes using the random number generator in Microsoft Excel with constraints of a Gaussian distribution according the first Gaussian curve ( 1q = 623 ± 122 a . u ) in 2 mM Ca2+ . Random pairs of these amplitudes were then summed together with varying temporal offsets ( delays between the first and second hypothetical events ) to generate 2000 hypothetical 2-vesicle fusion events . The amplitudes of the three point moving averages were then calculated—analogous to the amplitude calculations of raw data—and the distribution of these hypothetical multivesicular events was compared to the multivesicular only component of the amplitude distribution in 8 mM extracellular Ca2+ . The most similar distribution had a temporal offset of 118 ms , which we interpret as an approximate interval between two vesicles fusing spontaneously during multivesicular fusion . Statistics were performed using Microsoft Excel or GraphPad Prism 6 . For statistical comparisons between experiments that were performed on the same population of boutons , Student's t-test was used ( 2-tailed , paired ) . In imaging experiments , n refers to the number of multiple experiments performed , with each experiment containing up to 100 regions of interest . Student's t-test ( 2-tailed , unpaired ) was used to analyze all pair wise data sets obtained from boutons under distinct conditions . The Kolmogorov–Smirnov test was used to determine differences in cumulative probability distributions . The chi-square test was used to evaluate the goodness fit of Gaussian distributions to amplitude histrograms . For analysis of multiple comparisons among synaptophysin-pHtomato experiments one-way ANOVA with Bonferroni post hoc analysis was used . For the correlation analysis , both parametric and non-parametric correlation coefficients were determined using GraphPad Prism 6 . Linear regression analysis included a test for the hypothesis that the slope = 0 , again using GraphPad Prism 6 . Finally , the D'Agostino-Pearson omnibus K2 normality test was performed in GraphPad Prism 6 . | Neurons communicate with one another at junctions called synapses . When an electrical signal known as an action potential arrives at a synapse , it causes packages called vesicles to fuse with the membrane that surrounds the neuron . The vesicles contain molecules called neurotransmitters , which are then released into the gap between the neurons . When these molecules bind to receptors on the surface of the second neuron , a copy of the action potential is generated and travels along the second neuron . The empty vesicles are then reabsorbed back into the first cell to be refilled with neurotransmitters so that the whole process can be repeated . In addition to releasing neurotransmitters in response to the arrival of an action potential , neurons sometimes release vesicles spontaneously . Such events are relatively rare and occur seemingly at random , making them difficult to study . However , by labeling a synaptic vesicle protein with a fluorescent protein , Leitz and Kavalali have constructed a system in which they can observe spontaneous vesicle fusions in single synapses in cell cultures , and follow the fate of the vesicles as they are reabsorbed back into the cell . The results reveal a number of key differences between the spontaneous events and those triggered by action potentials . Vesicles released spontaneously are retrieved and recycled much more rapidly than those that are released following the arrival of an action potential . Moreover , increases in calcium levels increase the frequency of both types of events . However , it is also clear that the calcium ions influence the two types of events independently of one another . Recent research on flies has suggested that some regions of synapses only ever release vesicles spontaneously , whereas others only ever release vesicles in response to the arrival of an action potential . The work of Leitz and Kavalali now adds to increasing evidence that the spontaneous release of neurotransmitters may have its own role in neuronal signaling that is distinct from the role played by neurotransmitters that are released in response to action potentials . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2014 | Fast retrieval and autonomous regulation of single spontaneously recycling synaptic vesicles |
Memories for recent experiences are rich in incidental detail , but with time the brain is thought to extract latent rules and structures common across past experiences . We show that over weeks following the acquisition of two distinct associative memories , neuron firing in the rat prelimbic prefrontal cortex ( mPFC ) became less selective for perceptual features unique to each association and , with an apparently different time-course , became more selective for common relational features . We further found that during exposure to a novel experimental context , memory expression and neuron selectivity for relational features immediately generalized to the new situation . These neural patterns offer a window into the network-level processes by which the mPFC develops a knowledge structure of the world that can be adaptively applied to new experiences .
Knowledge about the world is thought to involve the statistical integration of correlations within and across many individual experiences ( Ghosh and Gilboa , 2014; McClelland et al . , 1995; Preston and Eichenbaum , 2013 ) . While knowledge forms from experiences , over time memories of the experiences themselves often lose their contextual richness ( Furman et al . , 2007; Sekeres et al . , 2016; Winocur et al . , 2007 ) . According to theories of systems memory consolidation , the process by which knowledge is formed involves the gradual reorganization of networks within interconnected brain regions that include the hippocampus and neocortex ( McClelland et al . , 1995; Winocur et al . , 2010 ) . One region that seems to be particularly involved in long-term memory is the medial prefrontal cortex ( mPFC; Frankland et al . , 2004; Takashima et al . , 2006; Takehara et al . , 2003; Takehara-Nishiuchi and McNaughton , 2008 ) . The mPFC is necessary for learning new goals and behaviors when this learning depends on pre-existing knowledge ( rodents: Richards et al . , 2014; Tse et al . , 2011; Wang et al . , 2012; humans: Ghosh et al . , 2014 ) , and human imaging data indicate that the mPFC is activated when subjects use existing knowledge to encode new information ( van Kesteren et al . , 2010; Kumaran , 2013 ) , make inferences ( Zeithamova and Preston , 2010 ) , and guide decision making ( Kumaran et al . , 2009 ) . The question remains as to how this role is supported by changes in neural signaling that take place throughout learning and subsequent consolidation . Based on the hypothesis that the mPFC provides the brain with schematic knowledge—i . e . , a framework of abstract associations about environment-behavior relationships—we predict that mPFC neuron ensembles build representations of correlations common to multiple experiences over a time period that systems consolidation is known to take place; furthermore , we expect that information for these common relationships is disproportionately represented relative to context-specific information . Here we examine how two different memories with overlapping associative structures are coded by neuron populations in the mPFC of rats , and how these codes change over time . The two memories both rely on a trace eyeblink conditioning procedure , in which a neutral stimulus ( CS ) is paired with eyelid stimulation ( US ) with a stimulus-free interval between CS offset and US onset . The retrieval of this memory initially depends on the hippocampus , but over the course of two to four weeks becomes dependent on the prelimbic region of the mPFC ( Takehara et al . , 2003; Takehara-Nishiuchi et al . , 2006 ) . Over this same time , mPFC neuron ensembles develop selective firing patterns for the acquired CS-US associations and maintain stable representation thereafter , a change that takes place with or without continued conditioning ( Hattori et al . , 2014; Takehara-Nishiuchi and McNaughton , 2008 ) . These observations make trace eyeblink conditioning ideal for examining the evolution of the selectivity of mPFC ensemble activity across time . We find that the development of mPFC population codes for common task features involves bi-directional changes in selectivity for relational versus physical stimulus features over weeks after learning .
It was first important to establish that rats were capable of learning the associations of both visual and auditory CSs , while at the same time learning to discriminate CS-US trials from control trials in which the CSs were presented alone . Four rats underwent daily trace eyeblink conditioning with each session divided into two epochs ( Figure 1A; Takehara-Nishiuchi and McNaughton , 2008 ) . Within each epoch , 20 control trials were initially presented in which the auditory or visual CS was presented alone , these were then followed by 80 trials in which the CS was paired with the US . The trials were separated by an interval that was pseudorandomized across trials and ranged from 20 to 40 s . The CS-alone trials were critical to establish neural selectivity for the CS-US relational structure . Over the course of approximately 12 conditioning sessions , rats developed anticipatory blinking responses ( conditioned responses , CRs ) that peaked near the expected onset of the US; this anticipatory behavior was specific for CS-US paired trials and was not observed in CS-alone trials ( Figure 1B ) . Importantly , the increased frequency of eyeblink responses ( CR% ) was only observed during the period before US onset ( post-CS phase; Figure 1C; for CR% in individual rats , see Figure 1—figure supplement 1A ) , but not during the period before CS onset ( pre-CS phase; Figure 1—figure supplement 1B ) . Three-way repeated measures ANOVA revealed a significant phase × trial type × session interaction , F75 , 468 = 235 . 66 , p<0 . 001 ) . In CS-US paired trials , CR% during the post-CS phase became greater across sessions , but CR% during pre-CS phase did not ( follow-up two-way repeated measure ANOVA , phase × session interaction , F25 , 156 = 7 . 931 , p<0 . 001 ) . In contrast , in CS-alone trials , CR% did not change across sessions or differ between the pre- and post-CS phases ( phase × session interaction , F25 , 156 = 0 . 684 , p=0 . 867 ) . A nonparametric test also showed the significant difference in CR% across four trial types during the post-CS phase ( Friedman test , χ23 = 327 . 36 , p<0 . 001 ) . Once performance reached asymptote , CR% showed an abrupt transition between CS-alone and CS-US trials in each session ( Figure 1D ) . This discontinuity suggested that the learning process involved successful encoding of the distinct temporal context between earlier and later trials; i . e . , that CS-alone trials at the start of the trial block did not simply extinguish associations acquired on previous days . Thus , the present behavioral protocol enabled the rats to acquire two associative memories ( ACS-US and VCS-US association ) that shared a common , relational feature ( i . e . , the CS-US association ) , but differed in a discrete , physical feature ( i . e . , the sensory modality of CS ) . They also acquired the temporal context that initial trials within each epoch would not be paired with the US . 10 . 7554/eLife . 22177 . 003Figure 1 . Rats formed two associative memories that differed in incidental , physical features . ( A ) Rats were implanted with two eyelid wires to deliver mild electrical current ( US ) and record anticipatory blinking activity ( CRs ) . Single neural activity was collected from the prelimbic region of the mPFC from the first day of conditioning . Conditioning took place in two chambers , in which either a tone ( ACS ) or light ( VCS ) was presented alone ( CS alone , 20 trials ) or preceding the US by 500 ms ( CS-US , 80 trials ) . ( B ) Averaged EMG amplitude during Early ( left ) and Late learning ( right ) during CS-alone trials ( top ) and CS-US paired trials ( bottom ) . ( C ) CR expression ( CR%; ±SEM ) increased over days for both ACS-US ( red ) and VCS-US ( blue ) conditions , but not during the respective CS-alone conditions ( magenta , turquoise ) . ( D ) Within sessions in which rats reached asymptotic responding , CR% showed an abrupt transition upon the shift from the block of CS-alone trials ( magenta; turquoise ) to the block of CS-US paired trials ( red; blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 00310 . 7554/eLife . 22177 . 004Figure 1—figure supplement 1 . Behavioral performance in two blocks of trace eyeblink conditioning . ( A ) In all four rats , trials with a conditioned response ( CR% ) gradually increased during two conditions with CS-US pairings ( ACS-US paired , red; VCS-US paired , blue ) , but not during two conditions with the CS alone presentation ( ACS alone , magenta; VCS alone , turquois ) . Based on CR% and days since asymptotic responding , sessions were divided into five successive stages ( vertical lines ) . ( B ) In a few trials , EMG amplitude during a period before CS onset significantly changed from baseline . The proportion of these trials ( mean ± SEM , n = 4 rats ) did not increase across sessions or differ across four conditions . See ( A ) for the definition of line color . ( C ) In some trials , the rats engaged in an activity that greatly increased EMG amplitude before CS onset . ( D ) The frequency of these ‘hyperactive’ trials did not change across sessions or differ between four conditions ( mean ± SEM , n = 4 rats ) . See ( A ) for the definition of line color . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 004 To track neuron activity , and thereby measure the selectivity of the neural population , we extracellularly recorded action potentials of neurons in the prelimbic region of the mPFC from the first day of learning ( 10–35 neurons per day; Figure 2—figure supplement 1A–C ) . Recordings were performed with a chronically implanted microdrive ( Kloosterman et al . , 2009 ) containing 14 independently-movable , four-channel electrodes ( ‘tetrodes’ , [Wilson and McNaughton , 1993] ) . The movement of each tetrode was minimized to sample the activity of neurons located in a comparable part of the prelimbic region throughout the one-month period of daily recording . Our data , therefore , consist of some neurons sampled repeatedly across days and others sampled once ( Figure 2—figure supplement 1D ) . During learning , the neuron ensemble in the mPFC showed different firing rate changes to the auditory and visual CS ( Figure 2A; see also Figure 2—figure supplement 2 , for the remaining three rats ) , and also distinguished between the auditory CS presented alone and the auditory CS paired with the US . Differential firing to the two CSs was evident during the CS ( Figure 2B; permutation tests on the similarity of ensemble patterns for ACS alone and ACS-US paired vs . that for ACS-US paired and VCS-US paired , p=0 . 001 and 0 . 002 in two 50-ms bins during the CS ) while differential patterns between CS-alone and CS-US trials were detectable during the interval between the CS and US ( in 3 out of 8 bins , p<0 . 002 ) . By the third week after learning , at which point the behavioral expression of the CS-US association is known to depend on the mPFC ( Takehara et al . , 2003; Takehara-Nishiuchi et al . , 2006 ) , the mPFC ensemble differentiated between CS-alone trials and CS-US trials more than it differentiated between visual and auditory CS trials ( Figure 2A , Figure 2—figure supplement 2 ) . During the CS , the similarity of neuron firing patterns for two trial types with the same CS was no longer different from that for trial types with different CSs ( Figure 2B; permutation test , p=0 . 099 , 0 . 981 ) . During the CS-US interval , however , the similarity of neuron firing patterns became higher for trials with the shared stimulus relationship than for those without ( in 6 out of 8 bins , p<0 . 001 ) . Similar patterns were also observed in changes of firing rate evoked by the CS ( Figure 2—figure supplement 3 ) . Whereas during learning , the ensemble patterns of CS-evoked firings were similar between CS-alone and CS-US paired trials , they became more similar for ACS-US and VCS-US paired trials during the third post-learning week . At a finer timescale , within a single session of the third post-learning week , the ensemble activity rapidly evolved into a new , statistically uncorrelated pattern within the first ten ACS-US paired trials ( Figure 2C ) , which was tightly coupled with the time course over which the rats increased the frequency of CR expression at the beginning of ACS-US paired block ( Figure 1D ) . Upon the transition from the epoch with the ACS to another with the VCS , the ensemble similarity drastically dropped but went up again as the block of VCS-US pairings began . In contrast , during learning , the degree of ensemble differentiation across the four conditions appeared to be smaller , and it took a greater number of trials until the ensemble pattern evolved into a new pattern upon the block shift ( Figure 2—figure supplement 4 ) . Collectively , these observations suggest that initially , the mPFC ensemble encoded both the physical as well as relational features of the stimuli to a comparable degree; however , after extensive experience , the mPFC ensemble code became less sensitive to physical ( sensory ) features unique to each memory and more sensitive to their common , relational feature . 10 . 7554/eLife . 22177 . 005Figure 2 . Prefrontal ensemble codes in well-trained rats were selective for relational features . ( A ) Example behavior from a single rat . Based on the percentage of trials exhibiting a conditioned response ( CR% ) and days since asymptotic responding , sessions were divided into five successive stages ( vertical lines ) . Representative pseudocolor plots show normalized firing rate of neurons recorded from the same rat during the learning period ( gray ) and during the third week after learning ( yellow ) . Top left , ACS-US paired; top right , VCS-US paired; bottom left , ACS alone; bottom right , VCS alone . Neurons were sorted based on the ACS-induced firing rate change during ACS-US pairings from the largest increase ( Neuron #1 ) to the largest decrease ( Neuron #357 or 184 ) . During learning , ensemble activity during ACS-US pairings was similar to that during ACS alone presentations and VCS-US pairings; however , during the third post-learning week , it became more similar to that during VCS-US pairings than ACS alone presentations . White lines indicate CS onset and offset , black bars mask US artifacts . ( B ) During learning ( left ) , binned firing rates of neuron ensembles ( PV ) during the CS ( two vertical lines ) were more similar for two conditions with the same modality of CS ( ACS-US paired and ACS alone , blue , r ± 95% confidence intervals ) than two conditions with the same stimulus relationships ( ACS-US paired and VCS-US paired , green ) . Their similarity became comparable between two condition pairs during subsequent CS-US intervals . During the third post-learning week ( right ) , PV became more similar for two conditions with the shared stimulus relationship than those with the shared CS modality . * indicates p<0 . 05/10 in random permutation tests . ( C ) Trial-by-trial display of PVs during CS-US intervals in the third post-learning week . Neurons from four rats were sorted based on the firing rate during ACS-US paired trials from the highest ( Neuron #1 ) to the lowest ( Neuron #360 ) ] . A ‘template’ PV was constructed as averaged PVs across 10–80th ACS-US paired trials . The correlation coefficient between the template and PV in each trial ( ±95% confidence interval ) rapidly increased within the first ten ACS-US paired trials ( inset ) . Upon the transition from the epoch with the ACS to the next epoch with the VCS , it abruptly decreased but increased again when the VCS-US pairings began . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 00510 . 7554/eLife . 22177 . 006Figure 2—figure supplement 1 . Histology and single unit isolation . ( A ) Representative image of a tetrode tip ( white arrow ) located within the prelimbic region of medial prefrontal cortex . ( B ) Final locations of all usable tetrodes targeted at the prelimbic cortex ( black dots ) . Coordinates adapted from Paxinos and Watson ( 2007 , sixth edition ) . ( C ) During a software-assisted spike sorting process ( KlustaKwik and MClust ) , several parameters were extracted from each spike waveform recorded from four wires of a tetrode ( Ch . 1–4 ) , and two of all possible parameter combinations ( e . g . , amplitude recorded in two of four wires ) were plotted as a scatter plot . Spikes were separated into several units corresponding to spikes of each neuron based on the similarity of waveform parameters . Three examples that were isolated from one tetrode during a recording session were depicted as a set of averaged spike waveform on four wires ( Ch . 1–4 ) . The color of waveforms corresponds to the color used in the scatter plot . ( D ) Scatter plots show the amplitude of spike waveforms recorded on the channels 1 and 2 of a tetrode across three consecutive days ( a . u . , arbitrary unit ) . A few units were isolated from the recording in each day . Pseudocolor plots of binned firing rate ( middle ) show that some of these units increased ( hotter color ) or decreased ( cooler color ) their firing rate upon CS presentations . Based on the shape of spike waveforms recorded on four wires of the tetrode ( Ch . 1–4 , bottom ) , the unit 1 , 6 , and 11 appeared to be recorded from the same neuron which enhanced firing responses to the CS across three days . The unit 3 and 9 appeared to belong to the same neuron , but this neuron was not present in the recording on the third day . Other units , such as the unit 4 , 5 , and seven were recorded only in one day . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 00610 . 7554/eLife . 22177 . 007Figure 2—figure supplement 2 . Prefrontal ensemble activity during four conditions with different relational and physical stimulus features . Pseudocolor plots show normalized firing rates of neurons recorded from each of remaining three rats during learning ( left , grey ) and the third week after CR% reached asymptote ( right , yellow ) [top left , ACS-US paired; top right , VCS-US paired; bottom left , ACS alone; bottom right , VCS alone] . In each rat , neurons were sorted based on the ACS-induced firing rate during ACS-US pairings from the largest increase ( top ) to the largest decrease ( bottom ) . In all three rats , during learning ensemble firing patterns differentiated four conditions while during the third post-learning week , ensemble activity became more similar for two conditions with the same stimulus relationship ( i . e . CS-US pairings ) than those with the same CS ( i . e . , auditory or visual CS ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 00710 . 7554/eLife . 22177 . 008Figure 2—figure supplement 3 . CS-evoked firing patterns during four conditions with different relational and physical stimulus features . Pseudocolor plots show standard scores of CS-evoked firing rates of neurons recorded from four rats during learning ( left ) and the third week after CR% reached an asymptote ( right; top left , ACS-US paired; top right , VCS-US paired; bottom left , ACS alone; bottom right , VCS alone ) . Neurons were sorted based on the standard score of firing rates after CS presentations during ACS-US pairings from the largest increase ( top ) to the largest decrease ( bottom ) . CS-evoked firing patterns were similar between ACS-US paired and ACS alone conditions during learning; however , during the third post-learning week , they became similar between ACS-US paired and VCS-US paired conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 00810 . 7554/eLife . 22177 . 009Figure 2—figure supplement 4 . Trial-by-trial changes in the similarity of ensemble activity during learning . Trial-by-trial display of population firing rate vectors ( PVs ) during CS-US intervals in the learning stage . Neurons from four rats were sorted based on the firing rate during ACS-US paired trials from the highest ( Neuron #1 ) to the lowest ( Neuron #813 ) ] . A ‘template’ PV was constructed as averaged PVs across 10–80th ACS-US paired trials . The correlation coefficient between the template and a PV in each trial ( ±95% confidence interval ) gradually increased within the first twenty ACS-US paired trials ( inset ) . It abruptly decreased upon the transition from the epoch with the ACS to the next epoch with the VCS but increased again when the VCS-US pairings began . Note that the difference in the correlation coefficient between the conditions appeared to be smaller than that observed in the third post-learning week shown in Figure 2C . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 009 To unpack the observed difference in neuron selectivity between learning and post-learning periods , we tracked mPFC ensemble selectivity for relational ( CS-alone versus CS-US trials ) and physical ( ACS versus VCS trials ) features over five successive stages of learning . For these analyses , we used a machine classifier to establish the degree to which neuron populations differentiated—i . e . , could successfully distinguish—trials of each condition . Conditioning sessions were divided into five learning stages separately for each rat based on CR rate ( Figure 1—figure supplement 1A ) . The stages were: ( 1 ) Before Learning , when CR% was less than 30 , ( 2 ) Learning , during which the rate of CRs progress to an asymptotic level , and ( 3–5 ) Post 1W , 2W , and 3W , which correspond to the first , second , and third week after CR expression reached asymptote . In each stage , selectivity of the mPFC ensemble activity for relational and physical features was quantified by applying a Support Vector Machine ( SVM ) classifier to binned firing rates during intervals between the CS and US in four conditions . Better performance of the classifier reflects higher selectivity of neuron population firings for the relational and physical features that differentiate the four conditions . The classification accuracy greatly varied across five stages of learning ( Figure 3A; one-way ANOVA , F4 , 95 = 11 . 34 , p<0 . 001 ) . Examination of the specific classification errors ( the ‘confusion matrix’ , Figure 3B ) showed that in the Before Learning and Learning stages , the majority of inaccurate classifications resulted from errors in discriminating CS alone trials from CS-US paired trials . In contrast , across the post-learning weeks , the classifier made more errors in discriminating ACS trials from VCS trials , suggesting that across time prefrontal ensemble activity appeared to become more sensitive to the paired vs . unpaired trial blocks and less sensitive to stimulus modalities . This was confirmed by applying the SVM method to make binary discriminations in the relational ( CS alone vs . CS-US trial blocks ) and physical ( ACS vs . VCS trial blocks ) dimensions: classification accuracy for the relational feature significantly improved across stages ( Figure 3C; one-way ANOVA , F4 , 95 = 9 . 99 , p<0 . 001 ) ; while classification accuracy for the physical feature steadily decreased ( one-way ANOVA , F4 , 95 = 3 . 77 , p=0 . 007 ) . Notably , increased selectivity for the relational dimension took place primarily during first and second post-learning weeks ( Before Learning versus Post 2 , or 3W , Learning versus Post 1 , 2 , or 3W , Tukey HSD , p<0 . 05 ) , whereas selectivity for the physical dimension decreased during the third post-learning week ( Post 3W versus Before Learning , Tukey HSD , p<0 . 05 ) . These results suggest that the process by which the mPFC extracts knowledge about environmental structure across time , as reflected in the increased coding for stimulus relationships , may be independent from its declining sensitivity to sensory features of the stimuli . 10 . 7554/eLife . 22177 . 010Figure 3 . Prefrontal neuron ensembles became more selective for relational and less selective for physical features across learning stages . ( A ) The relative accuracy of Support Vector Machine ( SVM ) , trial type decoding ( raw accuracy minus accuracy at chance ) was separately calculated in each of five successive stages of learning . Decoding accuracy greatly fluctuated across stages . ( B ) Representative confusion matrices from SVM showing decoding accuracy between trial types . Each square in each 4 × 4 matrix shows the probability ( color ) that a trial of one type ( rows ) was classified as another type ( columns ) . Hotter colors along diagonal illustrate the high accuracy of the classifier . During the Before learning stage , most classification errors were between CS-alone and CS-US trials , illustrated by lighter blue squares within the upper left and lower right quadrants . In the later stages , errors became more common between ACS-US and VCS-US trials . ( C ) Average SVM accuracy for binary discrimination along relational ( navy ) and physical ( green ) dimensions . Curves emphasize increased information about stimulus relationships and decreased perceptional information in the ensemble . ( * indicates p<0 . 05 in comparison with Before Learning , † shows p<0 . 05 in comparison with Learning ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 01010 . 7554/eLife . 22177 . 011Figure 3—figure supplement 1 . Parameters that affected the ability of support vector machine classifier to decode conditions based on prefrontal ensemble activity . ( A ) The classifier was trained with the averaged firing rate of neuron ensembles in a series of 200 ms bins around the onset of the conditioned stimulus ( CS ) . The mean decoding accuracy ( black line ) and two standard deviations ( gray lines ) were calculated over 20 decoding runs with 150 randomly sampled neurons from 340 neurons across four rats . Regardless of bin sizes ( left , 200 ms; middle , 100 ms; right 50 ms ) , the decoding accuracy relative to chance level ( estimated by a random permutation test , p<0 . 05 ) improved upon the CS presentation ( two vertical lines indicate the onset and offset ) and remained high toward the onset of unconditioned stimulus ( black bar ) . The higher decoding accuracy , however , was achieved with the larger bin size . ( B ) With population firing vectors during the first 200-ms bin after CS offset as inputs , the decoding accuracy was higher when a larger number of neurons were included into the population firing vectors . Error bars show standard error of the mean across 20 decoding runs with randomly sampled neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 011 We next evaluated how the ensemble changes corresponded to changes of feature selectivity within single neurons . Among 2101 neurons , 65 . 1% significantly changed firing rates during the CS or CS-US interval relative to inter-trial intervals in at least one of four conditions; this proportion did not appear to change between learning and over-training periods ( Figure 4—figure supplement 1 ) . Of these rate-changing neurons , 10 . 5% showed selectivity for the relational features of the stimulus ( ‘Relational’ ) , meaning they exhibited a significantly different response pattern during CS-US paired trials compared with CS-alone trials , regardless of the modality of the CS ( alpha = 0 . 05 , random permutation test with 1000 samples; Figure 4 ) . A separate set of rate-changing neurons , 13 . 7% , were selective for the stimulus modality ( i . e . the physical features of the CS , ‘Physical’ ) , meaning firing rates differed between ACS and VCS trials , regardless of whether it was a CS-alone or CS-US paired trial . More neurons ( 17 . 6% ) were selective for both relational and physical features ( Conjunctive ) . The remaining neurons showed the same response patterns in all four conditions . 10 . 7554/eLife . 22177 . 012Figure 4 . Single prefrontal neurons showed selectivity for the different features of the memory . Representative raster plots and peri-stimulus time histograms ( 1 ms bins , smoothed with a 50 ms Hanning window ) for each of four conditions ( the presentation of auditory CS alone , magenta; pairings of the auditory CS and US , red; the presentation of visual CS alone , turquois; pairings of the visual CS and US , blue ) . Although some neurons showed the same CS-evoked firing patterns across four conditions ( Non-selective ) , others were found with firing rate changes dependent on a relational feature ( Relational , rates in CS-US paired trials differed from rates in CS-alone trials ) , a physical feature ( Physical , rates in trials with the ACS differed from those with the VCS ) , or their conjunction ( Conjunction , rates in one condition differed from the other conditions ) . Two black lines indicate CS onset and offset , and black bars mask the artifact induced by the US . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 01210 . 7554/eLife . 22177 . 013Figure 4—figure supplement 1 . The proportion of neurons responding to the CS in each learning stage . The proportion of neurons ( ±95% confidence interval ) that significantly changed their firing rate during the CS or CS-US interval in at least one of four conditions did not change during learning and over-training . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 013 To quantify learning stage-dependent changes in the feature selectivity of single neurons , we computed two independent properties , the magnitude and the consistency of differential firing rates between conditions ( Figure 5A ) . Over the five learning stages , the magnitude of differential firing for the relational dimension increased during the CS-US interval , as measured by mean ranks of a ‘Differentiation index’ ( 0-none to 1-strongest; Figure 5B , Kruskal-Wallis test , χ24 = 19 . 37 , p<0 . 001 ) . Magnitude of differential firing became significantly higher during the Post 1 and 3W stages compared with the Before Learning stage ( Wilcoxon rank sum test , p<0 . 001 , =0 . 020 , and 0 . 001 for Post 1 , 2 , and 3W , respectively ) . In contrast , the magnitude of differential firing for the physical feature did not significantly change across the stages ( χ24 = 6 . 14 , p=0 . 189 ) nor were there changes in the proportion of neurons with a high differentiation index for the relational or physical feature ( Figure 5C ) . 10 . 7554/eLife . 22177 . 014Figure 5 . Changes in neural coding over time could be decomposed into changes in differential firing magnitude vs . consistency . ( A ) Three examples of firing differentiation of a neuron between trials with the auditory conditioned stimulus ( CS , magenta ) and those with the visual CS ( blue ) . A neuron is more ‘selective’ for the physical feature , if it has a greater difference in the mean firing rate ( vertical lines ) between two conditions ( Magnitude ) and a smaller variance in trial-by-trial firing rates in each condition ( arrows; Consistency ) . ( B ) Mean ranks of the differential index between CS-alone trials and CS-US paired trials ( Relational , left ) increased during weeks after learning , whereas that between the auditory and visual CS trials ( Physical , right ) did not . ( C ) The proportion of neurons ( ±95% confidence intervals ) with high differentiation index for relational or physical feature did not change across the stages . ( D ) Across five stages , mean ranks of mutual information for the relational feature did not change . In contrast , mean ranks of mutual information for the physical feature decreased over the stages . ( E ) The proportion of neurons ( ±95% confidence intervals ) with significantly high mutual information for the physical feature showed a trend toward decreasing across stages . ( Symbols: # , ** , and *** indicate p<0 . 01 , 0 . 01 , and 0 . 001 , respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 014 Differences were also observed with respect to the consistency of differential activity between trial types , as measured by mutual information . Mean ranks of mutual information for the physical feature significantly decreased across the five learning stages ( Figure 5D , χ24 = 12 . 06 , p=0 . 017 ) , but those for the relational feature did not ( χ24 = 2 . 27 , p=0 . 686 ) . The consistency of differential firing became significantly lower during the Post 1 and 3W stages compared with the Before Learning stage ( Wilcoxon rank sum test , p=0 . 003 , 0 . 016 , and 0 . 001 for Post 1 , 2 , and 3W , respectively ) . Similarly , there was a trend toward a decline across stages in the proportion of neurons with a significant mutual information value for the physical features ( random permutation test , p<0 . 05; binomial test , Before vs . Post 3W , p<0 . 1; Figure 5E ) , while changes in this proportion were not observed for the relational feature . These findings suggest that the increase in ensemble selectivity for relational information is mediated by increases in the magnitude of differential firing , whereas reduced selectivity for perceptual information involved changes in the consistency of differential firing . Knowledge is useful when it can be applied to new situations . To examine whether relational coding generalized to a novel situation , three rats were trained for approximately 30 sessions in the paradigm described above , and were then presented with the same structure of 20 CS-alone trials and 80 CS-US trials , using only the auditory CS , in the old chamber ( Box 1 ) and in a new conditioning chamber ( Box 2; Figure 6A ) . The new chamber differed from the old chamber in the visual appearance of the walls , illumination , and texture of the floor . The rats immediately responded with CRs to the auditory CS-US pairings in the new chamber ( Figure 6B ) . Prefrontal neurons exhibited similar firing patterns during CS-US pairings in the two chambers , while also maintaining clear differentiation between the CS-US pairings and CS-alone trials within each chamber ( Figure 6C ) . Similarly , population decoding analysis revealed weak selectivity of neuron ensemble activity for the conditioning chamber but high selectivity for CS alone vs . CS-US trials ( Figure 6D , E ) : the classifier made many errors discriminating CS-US pairings in Box 1 from those in Box 2 while making few errors discriminating the CS alone trails in Box 1 and Box 2 . These data lead us to suggest that , like the neuron ensemble in the hippocampus ( Leutgeb et al . , 2005 ) , the mPFC neuron ensemble is capable of generating separate codes for CS-alone trials in two different conditioning chambers , but that it actively assimilates previously-formed codes for stimulus associations to a novel context . 10 . 7554/eLife . 22177 . 015Figure 6 . Use of existing ensemble code for CS-US relationship in a new context . ( A ) After ~30 days of conditioning three rats underwent three recording sessions in a familiar ( Epoch1/Box1 ) and novel ( Epoch2/Box2 ) environment . ( B ) Conditioned responses ( CRs ) were absent during CS-alone trials ( Box1 pink , Box2 beige ) and high during CS-US trials ( Box1 red; Box2 brown ) , with no apparent differences between familiar and novel environment . ( C ) Pseudocolor plots show normalized firing rate of 225 neurons ( sampled across three rats ) during four conditions ( top left , ACS-US in Box1; top right , ACS-US in Box2; bottom left , ACS-alone in Box1; bottom right , ACS-alone in Box 2 ) . Neurons were sorted based on ACS-evoked firing rate change during the Box1 , ACS-US pairings ( top: largest increase in neuron #1 , bottom: largest decrease in neuron # 225 ) . Ensemble activity during ACS-US pairings in Box 2 was similar to that during ACS-US in Box 1 , but not to ACS alone trials . ( D ) The confusion matrix from an SVM classifier of trial type . Most errors were misclassifications of Box1 versus Box2 ACS-US trials . ( E ) Relative decoding accuracy ( raw minus accuracy at chance ±SEM across 20 runs ) between ACS-US versus ACS alone trials and Box1 versus Box2 trials . *** indicates p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 015
Theories of systems consolidation posit that the neocortex gradually discovers common latent rules and structures from multiple past experiences and builds semantic knowledge of the external world ( McClelland et al . , 1995; O'Reilly et al . , 2014 ) . We show that , after learning , there is a gradual refinement of prefrontal neuron selectivity that may be a direct neuronal analog for this knowledge development process . Over a one-month period of repeated exposures to two similar experiences , mPFC ensemble activity gradually becomes more sensitive to their latent , relational variables; meanwhile , over time , information about perceptual , physical features of the environment is lost within this ensemble . Importantly , the selectivity for the physical features was weakened after , but not during learning , supporting the view that it underlies the mPFC’s involvement in the extraction of commonalities from previous experiences ( Richards et al . , 2014; Tse et al . , 2011; Wang et al . , 2012 ) , rather than the learning of two stimulus associations . These results speak directly to a long-standing question in the field of whether the formation of generalized , or schematized memory is merely a product of the network ‘forgetting’ incidental features . Quite to the contrary , population analyses revealed different time-courses in the development of coding for relational features compared with the loss of selectivity for physical features . Examinations of single neuron activity also revealed that the former appears to rely more on increased magnitude of firing rate differences over weeks , whereas the latter appears to involve decreased reliability of differential firing . The presently observed prefrontal neuron ensemble changes are likely to be part of the physiological basis of the mPFC’s contribution to an animal’s ability to construct , maintain , and update an associative knowledge structure used to behave adaptively in familiar and novel environments ( Ghosh and Gilboa , 2014; van Kesteren et al . , 2010; Preston and Eichenbaum , 2013; Richards et al . , 2014; Tse et al . , 2011; Wang et al . , 2012 ) . The stronger selectivity for relational over physical features is consistent with the previously reported selectivity of prefrontal neurons for rules ( Rich and Shapiro , 2009; Wallis et al . , 2001 ) or categories ( Freedman et al . , 2001 ) in well-trained animals . Our data reveal that this feature selectivity is not due to the innate inability of the mPFC ensemble to encode incidental features . In fact , during learning , the mPFC ensemble firing differentiated the physical features of the experiences to a comparable degree to the relational features ( Figures 2B and 3C , see also , Hyman et al . , 2012; Ma et al . , 2016 ) . Furthermore , although the ensemble differentiation between ACS-US and VCS-US pairings decreased , neural responses to the two CSs remained highly differentiated in the CS-alone condition ( Figure 3B ) . This observation is not consistent with a view that the weakened selectivity for the physical feature reflects a form of learned equivalence between the stimuli , arising from their equivalent associated outcome ( Honey and Hall , 1989 , 1991; Iordanova et al . , 2007; Miller and Dollard , 1941 ) . Rather , the present findings support a view of the building of a new , high-level representation that takes into account the temporal context within which the stimuli are presented . A question that remains unanswered is the type and range of experiences over which the mPFC network is capable of extracting commonalities . In the present study , both the temporal structure of the CS-US pairings and the outcome itself were common between the visual and auditory CS-US trials . Therefore , it remains unclear whether mPFC neurons only encode the predicted outcome , rather than the more abstract associative , temporal relationship between the stimuli . Results from other studies suggest that the mPFC can discriminate between situations with the same outcomes within the same environment , if only the rules for negotiating the environment differ ( Rich and Shapiro , 2009; Durstewitz et al . , 2010 ) . Thus , although the present study did not explicitly alter the outcome , it seems reasonable to suppose that behavioral savings between paradigms with overlapping rules or temporal structure , even when the outcomes or unconditioned stimuli themselves are altered , will rely on representations within the mPFC that develop over consolidation , as described presently . Is the observed change in the selectivity of prefrontal neurons a product of time passage after learning or repeated conditionings ? Neural selectivity for the CS-US relationship ( the relational feature ) is no less a product of time than experience , because it becomes strengthened with or without repeated daily conditionings after learning ( Hattori et al . , 2014; Takehara-Nishiuchi and McNaughton , 2008 ) . These observations , however , do not address whether the weakened selectivity for physical features requires continued conditioning . It is worth noting that the loss of selectivity may very well require no active process at all ( see also , Richards et al . , 2014 ) , as it could be accounted for by the lack of reinforcement over time . In that sense , the repeated exposures in the present study may be working against the weakening of selectivity that we observed . Testing this point directly requires future studies which monitor the selectivity of mPFC neurons over time with manipulations of multiple variables that include the number of different ‘CS-US’ exemplars animals are exposed to , the similarity between exemplars , the temporal proximity between exposures , and elapsed time from the exposures . When the animals were exposed to a new situation in which the same stimulus relationship took place in a novel environment , the mPFC immediately assimilated its code for the new situation to the existing , generalized code , without reverting to a code that also encodes incidental details ( Figure 6 ) . This is not due to the inability of mPFC ensemble to encode environmental features because it showed distinct firing patterns during two neutral experiences taking place in two different environments ( Figure 6D; Hyman et al . , 2012 ) . This immediate transfer of an existing abstract code may serve as a key computational basis for the assimilation of new information to a pre-existing knowledge structure ( Bartlett , 1932 ) . Experimental evidence from rodent behavioral studies and human imaging studies suggests that memory assimilation depends on the mPFC ( DeVito et al . , 2010; Richards et al . , 2014; Tse et al . , 2011; Wang et al . , 2012 ) , hippocampus ( Dusek and Eichenbaum , 1997; Iordanova et al . , 2011; Tse et al . , 2007 ) , and their interactions ( van Kesteren et al . , 2010; Kumaran et al . , 2009; Zeithamova et al . , 2012 ) . Some theories posit that mPFC-hippocampal interactions during memory assimilation may be an extension of those during memory retrieval: where the mPFC selects a memory that is the most appropriate in a current context and sends top-down signals to the hippocampus to recover its contents ( Preston and Eichenbaum , 2013 ) . From a computational perspective , the initial selection process likely relies on pattern completion of mPFC ensemble activity from cues available in a current context ( Takehara-Nishiuchi and McNaughton , 2008 ) . This would activate downstream targets , including the rhinal cortices ( Paz et al . , 2007 ) that have connections with other neocortical regions as well as the hippocampus . The former may result in the recovery of a gist-like version of previous experiences ( Insel and Takehara-Nishiuchi , 2013 ) , whereas the latter may facilitate the acquisition of new information ( Bero et al . , 2014; Preston and Eichenbaum , 2013 ) by activating neurons bearing original memories ( McKenzie et al . , 2013; Navawongse and Eichenbaum , 2013; Rajasethupathy et al . , 2015 ) . This view unites two seemingly disparate engagements of the mPFC in initial learning and in the retrieval of consolidated memory into a common computation executed by the mPFC neuron ensemble . In conclusion , our observations show the gradual development of the mPFC ensemble code for behaviorally relevant features common across multiple experiences , a process involving parallel modifications of two properties of single neuron firings across different time courses . This unique coding property of the mPFC may support its role in the formation , maintenance , and updating of associative knowledge structures that support flexible and adaptive behavior ( Ghosh and Gilboa , 2014; van Kesteren et al . , 2010; Preston and Eichenbaum , 2013; Richards et al . , 2014; Tse et al . , 2011; Wang et al . , 2012 ) .
All experiments were performed on four male Long-Evans rats ( Charles River Laboratories , St . Constant , QC , Canada ) between 16–25 weeks old at the time of surgery . Rats were housed individually in Plexiglass cages and maintained on a reversed 12 hr light/dark cycle . Water and food were available ad libitum . All methods were approved by the Animal Care and Use Committee at the University of Toronto . Tetrodes were made in-house by twisting together four 12 μm polyimide coated nichrome wires ( Sandvik , Stockholm , Sweden ) following our previous work ( Takehara-Nishiuchi and McNaughton , 2008 ) . To permit independently adjustable tetrode depths , each tetrode was housed inside a screw-operated microdrive . The complete microdrive-array consisted of a bundle of 14 microdrives , each guiding a tetrode , contained within a 3D printed plastic base ( Kloosterman et al . , 2009 ) . The Microdrive-array also enclosed the Electrode Interface Board ( EIB-54-Kopf , Neuralynx , Bozeman , MT , United States ) to which all electrodes were connected and served as the interface between the recording and stimulating electrodes and the recording system . Prior to implantation , the impedance of the nichrome tetrode wires was reduced to ~250 kOhms by electroplating them with gold . Tetrodes were then drawn inside a stainless steel cannula ( 1 . 8 mm diameter ) at the microdrive-array base and a small drop of sterilized mineral oil was added to ensure smooth movement of the tetrodes after implantation . Following guidelines set by the Institutional Animal Care Committee at the University of Toronto , all surgeries were conducted under aseptic conditions in a sterile surgical suite . For the chronic implantation of the microdrive array , rats were anesthetized with isoflurane ( 1–1 . 5% by volume in oxygen at a flow rate of 1 . 5 L/min; Halocarbon Laboratories , River Edge , NJ , United States ) and placed in a stereotaxic holder with the skull surface in the horizontal plane . All tetrodes were targeted to the prelimbic region of the medial prefrontal cortex ( PrL mPFC ) . The tetrode bundle was implanted with the same procedure as those used in our previous work ( Takehara-Nishiuchi and McNaughton , 2008 ) . A craniotomy was opened over the PrL mPFC at 3 . 2 mm anterior and 1 . 4 mm lateral to bregma and the dura matter removed . The microdrive array was then lowered at a 9 . 5° medial angle until the base made contact with the surface of the brain . The craniotomy was then sealed with Kwik-Sil ( Stoelting , Kiel , WI , United States ) and the array was held in place with self-curing dental acrylic ( Lang Dental Manufacturing , Wheeling , IL , United States ) . Immediately after the surgery , all tetrodes were lowered 1 mm into the brain . For the next 3–4 weeks , the rat was connected to the system each day to visualize the quality of activity and monitor movement of the tetrodes . Each tetrode was lowered slightly each day ( 75–125 μm ) over the course of this 3–4 week period to target tetrodes tips to the PrL mPFC at 3 . 0–4 . 0 mm ventral from the brain surface . One tetrode was positioned superficially in the cortex ( 1 mm below brain surface ) to serve as a reference electrode for single-unit activity . Once the recordings began , tetrode position was adjusted only as necessary to obtain good quality high yield recordings . This approach was necessary to sample the activity of neurons from comparable parts of the prelimbic region across five learning stages over a month . Tetrode position adjustments were only made after a given recording session providing ~24 hr for the tetrode to stabilize prior to the next recording . All rats experienced the same general experimental procedure . Beginning 3–4 weeks following microdrive array implantation , when stable single unit recordings were achieved and tetrodes were positioned within the PrL mPFC , rats were subjected to daily conditioning in the trace eyeblink conditioning paradigm . Rats were placed in a large dark rectangular box , fitted with an LED light source and speaker . Within the box rats were enclosed in a square plexiglass container ( 20 × 20 × 25 cm ) , fitted with holes on one side to enable sound-waves from the speaker to enter the enclosure . The conditioned stimulus ( CS ) was presented for 100 ms and consisted of an auditory stimulus ( 85 dB , 2 . 5 kHz pure tone ) or a visual stimulus ( white LED light blinking at 50 Hz ) . The unconditioned stimulus ( US ) was a 100 ms mild electrical shock to the eyelid ( 100 Hz square pulse , 0 . 3–2 . 0 mA ) , and the intensity carefully monitored via webcam and adjusted to ensure a proper eyeblink/head turn response ( Morrissey et al . , 2012; Tanninen et al . , 2015 ) . The timing of CS and US presentation was controlled by a microcomputer ( BasicX , Netmedia , Tucson , AZ , United States ) , and the US was generated by a stimulus isolator ( ISO-Flex , A . M . P . I . , Jerusalem , Israel ) . Daily recording sessions consisted of two epochs of conditioning , each with 100 trials , separated by a 10 min rest period . Each epoch included 20 presentations of the CS alone , followed by 80 trials in which the CS was paired with the US , separated by a stimulus-free interval of 500 ms ( see Figure 1A ) . The first epoch used only one of the two CS ( e . g . auditory CS ) , and the second epoch used the other CS ( e . g . visual CS ) , with the CS order and schedule pseudorandomized across days and across rats . This design provided four conditions for comparison: presentations of the auditory CS alone ( ACS alone ) , pairings of auditory CS and US ( ACS-US paired ) , presentations of the visual CS alone ( VCS alone ) , and pairings of VCS and US ( VCS-US paired ) . Before and after each epoch the rat was placed in a comfortable rest box separate from the conditioning box for 10 min . Upon completion of the full conditioning procedure , several animals ( n = 3 ) , underwent a similar conditioning procedure over three days in which the conditioning environment was manipulated , but the same CS was used in two epochs . The rats underwent two epochs of 20 ACS alone trials and 80 ACS-US paired trials . One epoch was run in the same conditioning chamber as the previous 30+ days of conditioning ( Box 1; a dark box with brown floors and plain walls ) , in the other epoch the conditioning took place in a box in which the visual and textile features were manipulated ( Box 2; lit box , stripped walls , white floor ) . The epoch order was pseudorandomized across rats . During the daily conditioning sessions , we simultaneously recorded action potentials from individual neurons in the prelimbic region of medial prefrontal cortex and electromyogram ( EMG ) activity from the eyelid . Action potentials were captured using the tetrode technique , which allows for recording the activity of many individual neurons per recording session ( Wilson and McNaughton , 1993 ) . Experimental rats were connected to the system through an Electrode Interface Board ( EIB-54-Kopf , Neuralynx , Bozeman , MT , United States ) contained within the microdrive array fixed to the animal’s head . The EIB was connected to a headstage ( HS-54 , Neuralynx , Bozeman , MT , United States ) , and signals were acquired through the Cheetah Data Acquisition System ( Digital Lynx and Cheetah Software , Neuralynx , Bozeman , MT , United States ) . A threshold voltage was set at 40–50 mV , and if the voltage on any channel of a tetrode exceeded this threshold , activity was collected from all four channels of the tetrode . Spiking activity of single neurons was sampled for 1 ms at 32 kHz and signals were amplified and filtered between 600–6000 Hz . EMG activity was continuously sampled at 6108 Hz and filtered between 300–3000 Hz . Behavior was analyzed with the same procedures as those used in our previous studies ( Morrissey et al . , 2012; Takehara-Nishiuchi and McNaughton , 2008; Tanninen et al . , 2015 ) . The adaptive conditioned eyeblink response ( CR ) which represents the learning of the association between the conditioned stimulus ( CS ) and unconditioned stimulus ( US ) was assessed through the analysis of electromyogram ( EMG ) activity recorded from the upper left eye-lid muscle . Each trial was assessed offline with custom codes written in Matlab ( Mathworks , Natick , MA , United States ) for the presence of a CR . The CR was defined as a significant increase in eyelid EMG amplitude immediately before US onset . Specifically , EMG activity was sampled around the presentation of the CS in each trial and the instantaneous amplitude of the signal was calculated as the absolute value of the Hilbert transform of the signal ( using the hilbert function in Matlab ) . For each trial , the average amplitude during a 300 ms period immediately before CS-presentation was defined as the Pre-Value . The averaged amplitude during a 200 ms period immediately before US-presentation was defined as the CR-Value in the post-CS phase , and the averaged amplitude during a 200 ms period around 0 . 9 s before CS onset was defined as the CR value in the pre-CS phase . A Threshold value was set as the averaged Pre-Value across trials plus two standard deviations . For a given trial , if the CR-Value exceeded the Pre-Value and the Threshold , that trial was classified as containing a CR . In some trials , the Pre-Value exceeded the Threshold value because the rats engaged in grooming , teeth grinding , or climbing immediately before CS onset ( Figure 1—figure supplement 1C ) . These trials were classified as hyperactive and discarded . The proportion of these ‘hyperactive’ trials was typically ~5% and did not change across sessions in any of the trial types ( Figure 1—figure supplement 1D ) . The ratio of trials containing a CR to the total number of valid trials within each of two conditions ( CS alone , CS-US paired ) represented the CR% for the condition for each epoch . CR% during four conditions was compared by using three-way repeated measures ANOVA with sessions , conditions , and phase as within-subjects factors as well as the Friedman test . To assess changes in neuron activity across successive stages of learning , recording sessions were divided into five stages based on the frequency of CR expression . The criteria for stages were selected based on observations of general patterns of CR acquisition and expression across many animals . In the first few days of training , rats show very few trials in which they exhibit the CR , we define this period as the Before learning stage , i . e . before the animal has begun to associate the CS with the US . Once rats begin to form this association , the percentage of trials in which they exhibit a CR rapidly increases , but it can fluctuate greatly across days . We define this period as the Learning stage . Eventually the rats reach a point in which their responding plateaus and reaches asymptote , from this point on we generally observe small fluctuations in response rate across days but rarely see large deviations . This point defines the end of the Learning stage . All days beyond this point were defined in the Post-learning week stage . To operationally define these stages we set a threshold of responding . All days prior to the rat displaying the CR in 30% of trials are defined as the Before learning stage . All days following two consecutive days of the rat displaying the CR in 60% of trials are defined in weeks as three Post learning stages . All days in between Before learning and the beginning of the Post learning stage are defined as the Learning stage . Putative single neurons were isolated offline using a specialized software package in Matlab ( KlustaKwik , author: K . D . Harris , Rutgers , The State University of New Jersey , Newark , NJ; MClust , author: D . A . Redish , University of Minnesota , Minneapolis , MN; Waveform Cutter , author: S . L . Cowen , University of Arizona , Tucson , AZ , United States ) . Both automatic spike-sorting and manual sorting were used to assign each action potential to one of the neurons recorded simultaneously on one tetrode based on the relative amplitudes on the different tetrode channels and various other waveform parameters including peak/valley amplitudes , energy , and waveform principle components . The final result was a collection of time stamps associated with each action potential from a given neuron . Only neurons with <1% of inter-spike intervals distribution falling within a 2 ms refractory period were used in the final analysis . If a neuron did not show more than 1500 spikes during the entire recording session , it was removed from further analyses due to the insufficient number of spike waveforms to confidently judge if they were spikes recorded from a real neuron or noise . An individual neuron was defined as a unit that was well isolated from raw signals recorded on a tetrode . Because we minimized the movement of tetrodes across days , some units which appeared to belong to the same neuron were recorded across a few days ( Figure 2—figure supplement 1D ) . We treated each of these units as a separate sample of a neuron . The total number of neurons is the summation of the number of isolated units across tetrodes , sessions , and rats ( Table 1 ) . Therefore , our data consist of some neurons sampled repeatedly across days and others sampled in one day . Because we were mainly interested in comparisons of ensemble selectivity across learning stages , having repeatedly sampled neurons across days was beneficial because it reduces the variability in sampled neurons across learning stages . 10 . 7554/eLife . 22177 . 016Table 1 . The number of neurons recorded from each rat during each learning stage . DOI: http://dx . doi . org/10 . 7554/eLife . 22177 . 016Before During Post 1W Post 2W Post 3W Total Rat 1 42159455680382Rat 2 64357174162184941Rat 3 78213967759523Rat 4 4284464637255Total 2268133613413602101 To examine the similarity between firing rates of a population of neurons across four conditions ( ACS-US paired , VCS-US paired , ACS alone , or VCS alone ) , we constructed four population firing rate matrices each of which contained the binned firing rate ( 50 ms ) of all recorded neurons during a 1 s period around the CS onset ( −400 to 600 ms ) in one of the conditions . For each neuron , the firing rate in each bin was divided by its maximum firing rate across four conditions . We then sorted these neurons based on their change in firing rate during the CS-US interval relative to baseline during the ACS-US paired condition . To compare CS-evoked firing patterns across four conditions , raw firing rates of each neuron were converted to standard scores by using the mean and standard deviation of firing rates during a one-second period before CS onset . To quantify the similarity of population firing rate matrices between two conditions , we calculated the Pearson correlation coefficient ( r ) between vectors of binned firing rates of two conditions that shared a relational feature ( ACS-US paired and VCS-US paired ) or a physical feature ( ACS-US paired and ACS alone ) . To test whether r values for two condition pairs were significantly different from one another , we conducted random permutation tests . Trials were randomly assigned to either of two conditions in such a manner that the relative number of trials in each condition was held constant . The r value and its difference between two condition pairs were re‐computed . This procedure was repeated 1000 times to construct sampling distributions . The difference in r values between two condition pairs was considered significant when it fell in the 0 . 0025% lower or upper tail of its corresponding distribution ( α = 0 . 05/10 , adjusted for repetition across ten 50-bins covering from 0–500 ms after CS onset ) . To examine trial-by-trial changes in ensemble similarity , we constructed population firing rate vectors which contained the firing rate of all neurons during intervals between CS offset and US onset in each of 200 trials in a session . We then defined a ‘template’ of ensemble activity for ACS-US pairings by averaging firing patterns across the 10-80th ACS-US paired trials . Pearson correlation coefficient ( r ) was calculated between the template and the firing vector of each trial . To quantify the degree of selectivity of ensemble activity for physical and relational features of conditions , we examined how accurately a machine learning algorithm , Support Vector Machine ( SVM ) classifier ( Cortes and Vapnik , 1995 ) could decode the conditions from binned firing rates of a neuron ensemble . Several studies have shown that the SVM classifier can be successful in decoding the identity of visual stimuli ( Nikolić et al . , 2009 ) , the spatial position of a visual cue ( Astrand et al . , 2014 ) , and the allocation of attention ( Tremblay et al . , 2015 ) from the activity of multiple single neuron firings . Moreover , the SVM classifier was shown to outperform several other commonly used classifiers ( Astrand et al . , 2014 ) . The SVM classifier produces a model from training data which then predicts the target values of test data given only the test data attributes . For the current study , the attributes were the normalized firing rates of a population of neurons in a trial of one of four conditions , and the target values were the condition from which they were sampled . The population firing vectors were constructed by concatenating the responses of a set of N neurons on a trial from one of four conditions . Note that the neurons were recorded in separate sessions from four rats , and thus we ignored any correlated activity between neurons . Having simultaneous recordings , however , would most likely not have changed our conclusions since we were mainly interested in comparisons of relative classification ability across four conditions based on firing rate patterns immediately after CS presentations . All algorithms were run in Matlab using the freely accessible LIBSVM library ( Chang and Lin , 2011 ) . The classifiers were trained with Radial basis function kernels . We first identified SVM parameters that maximized decoding accuracy by performing a grid search procedure ( calculating decoding performance over a range of cost and gamma SVM parameters ) for each set of training data . This was done by using a 5-fold cross-validation procedure to minimize over-fitting . In each SVM run , twenty trials of each of four conditions were randomly drawn , without replacement , from all the recorded trials and used to create a population firing rate matrix of N neurons × 80 trials . Then , in each neuron , the firing rate in each trial was divided by the maximum firing rate of the neuron across the 80 trials . Half of the trials ( 10 trials from each condition ) from each condition were then used to select the parameters with the grid search and subsequently to train the SVM classifier with these parameters . The remaining trials ( 10 trials from each condition ) were then used to test the decoding accuracy after training . The process was repeated 20 times using a different sampling of ten training and ten test trials each time . Based on the classifications , a confusion matrix was created , which indicated the proportion of classifications in which a population firing vector belonging to condition X was classified as condition Y . Our preliminary analysis used firing patterns of 340 neurons recorded during the third post-learning week to test how decoding accuracy changes depending on three parameters: ( 1 ) the bin’s temporal location relative to CS onset , ( 2 ) the size of bin used to construct population firing vectors , and ( 3 ) the number of neurons included in a population firing vector . We entered the population firing rate during a series of 200 ms bins over a 1 . 4 s period around CS onset ( 50% overlap ) to the SVM classifier ( Figure 3—figure supplement 1A ) . Decoding accuracy was significantly better than chance even prior to CS presentation , but it further increased at CS onset and remained high until US onset ( random permutation test , all data points , p<0 . 001 ) . The high decoding accuracy during CS-US intervals was consistently observed when the input was the population firing patterns with bin sizes of 100 and 50 ms; however , the overall decoding accuracy worsened with a smaller size ( Figure 3—figure supplement 1A ) . Next , we used the population firing rate during the first 200 ms after the CS offset to test how decoding accuracy changed depending on the number of neurons included in the analysis ( Figure 3—figure supplement 1B ) . Although decoding accuracy improved with a greater number of neurons included in the population firing vector , the classification with vector sizes greater than 150 neurons displayed reliably high decoding accuracy . Therefore , the main analyses were conducted with the population firing vectors during the first 200 ms time window after CS offset of 150 randomly sampled neurons . To quantify the selectivity for the relational or physical features of the conditions , the same SVM classification procedure was performed after collapsing four conditions into two conditions ( for the relational feature , CS-alone trials and CS-US paired trials; for the physical feature , trials with the ACS and those with the VCS ) . Permutation tests were performed for each SVM run using the exact same procedure as above , after assigning , for each population firing vector , a randomized condition label . This procedure , repeated 50 times , each of which generates 40 readouts , yielded the distribution of chance performance of each classifier with 2000 datasets . The raw decoding accuracy was considered as significant when it fell in the 5% upper tail of its corresponding chance performance distribution ( α = 0 . 05 ) . The relative decoding accuracy was defined as the raw decoding accuracy minus the decoding accuracy at the 5% upper tail of its corresponding chance performance distribution . To compare the change in the decoding accuracy across five learning stages , the relative decoding accuracy was calculated in 20 sets of 150 neurons randomly sampled from all recorded neurons in each stage ( Astrand et al . , 2014 ) . The relative decoding accuracy was compared across the stages by one-way ANOVA followed by a posthoc Tukey HSD test . The selectivity of firing responses of single neurons was quantified as the magnitude and consistency of firing differentiation across conditions . The magnitude of firing differentiation was quantified as a differentiation index , which compared mean firing rates during trace intervals between two conditions:Differentiation index = ( Fr1−Fr2 ) / ( Fr1+Fr2 ) where Fr1 and Fr2 are averaged firing rates during the CS-US interval across trials in two conditions . For the selectivity for relational features , Fr1 is the mean firing rate during CS-alone trials , and Fr2 was the mean firing rate during CS-US paired trials . For the selectivity for physical features , Fr1 was the mean firing rate during trials with the auditory CS , and Fr2 was the mean firing rate during trials with the visual CS . Raw differentiation indices were converted to absolute values , and these values from all neurons were compared across the five stages of learning with the Kruskal-Wallis test followed by planned pair-wise comparisons with the rank sum test . The consistency of firing differentiation was quantified as mutual information . It was computed from the joint distributions of firing rates across conditions and takes into account variances across trials within each condition:∑i , jP ( i , j ) ∗log ( P ( i , j ) P ( i ) P ( j ) ) Where P ( i , j ) is the joint probability distribution of condition ‘i’ and firing rate ‘j’ , P ( j ) is the marginal probability distribution of firing rates , averaged across conditions , and P ( i ) is the marginal distribution of firing rate in condition ‘i’ . In each neuron , the firing rate was binned into 10 bins to describe the probability distribution . To assess the significance of selectivity , permutation tests were performed for each neuron and for each combination of conditions using the exact same procedure as above , after assigning randomized condition labels to each trial . This procedure , repeated 1000 times , yielded the distribution of the chance level of mutual information values . An observed mutual information value with the correct condition labels was considered as significant when it fell in the 5% upper tail of its corresponding chance distribution . The percentage of neurons with significant mutual information was compared across the stages of learning and over-training by a binomial test . The normalized mutual information was defined as the raw mutual information minus the mean of its corresponding chance distribution , divided by the standard deviation . The values from all neurons were compared across the five stages of learning and over-training with the Kruscal-Wallis test followed by planned pair-wise comparisons with the rank sum test . Upon completion of all recordings , the location of electrodes was marked by electrolytic lesions . Rats were first injected intraperitoneally with an overdose of sodium pentobarbital . For tetrodes , 5 μA was passed through one wire of each tetrode ( positive to the electrode , negative to animal ground ) for 20 s , for LFP electrodes 20 μA was passed for 45 s . Rats were then perfused intracardially with 0 . 9% saline followed by 10% buffered formalin . The brain was removed from the skull and stored in 10% formalin for several days . For cryogenic sectioning , the tissue was infiltrated with 30% sucrose solution , frozen and sectioned in a cryostat ( Leica , Wetzlar , Germany ) at 50 μm . Sectioned tissue was stained with cresyl violet and imaged under a light microscope to locate electrode locations . Only recordings from tetrodes located in the prelimbic region of mPFC were used for single unit analysis . | Many events in our lives resemble experiences we have had before , without being identical to them . Whenever you attend a party , for example , you may well take along a gift , such as a bottle of wine or a box of chocolates , but the gift will differ on each occasion . Psychologists believe that as our memories for such events become older , the incidental details unique to each event ( such as the identity of the gift ) are mostly forgotten . However , the common underlying patterns ( what parties are like in general ) are retained . This allows us to accumulate knowledge to guide our behavior in similar situations in the future . Studies in rodents and people have shown that a region of the brain called the medial prefrontal cortex stores long-term memories about experiences . But to what extent do neurons in this region represent abstract generalized knowledge as opposed to the specific incidental details ? To find out , Morrissey et al . used hair-thin electrodes to record the activity of hundreds of cells in the medial prefrontal cortex as rats performed a learning and memory task . The rats learned that either a tone or a light signaled the delivery of a mild electric shock . Initially , cells in the medial prefrontal cortex responded differently to the tone and to the light . However , after three weeks , the cells began to show similar responses to both stimuli . The medial prefrontal cortex activity had thus transitioned from representing incidental details ( tone versus light ) to representing abstract relationships ( stimulus predicts shock ) . This may relate to how the brain extracts commonality across experiences . A lingering question is how cells in the medial prefrontal cortex become selective for abstract relationships . We know that memories are reactivated during sleep . Therefore , one possibility is that combined reactivation of different experiences selectively strengthens memories for any features common to those experiences . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | Generalizable knowledge outweighs incidental details in prefrontal ensemble code over time |
Cerebral blood flow is highly sensitive to changes in CO2/H+ where an increase in CO2/H+ causes vasodilation and increased blood flow . Tissue CO2/H+ also functions as the main stimulus for breathing by activating chemosensitive neurons that control respiratory output . Considering that CO2/H+-induced vasodilation would accelerate removal of CO2/H+ and potentially counteract the drive to breathe , we hypothesize that chemosensitive brain regions have adapted a means of preventing vascular CO2/H+-reactivity . Here , we show in rat that purinergic signaling , possibly through P2Y2/4 receptors , in the retrotrapezoid nucleus ( RTN ) maintains arteriole tone during high CO2/H+ and disruption of this mechanism decreases the CO2ventilatory response . Our discovery that CO2/H+-dependent regulation of vascular tone in the RTN is the opposite to the rest of the cerebral vascular tree is novel and fundamentally important for understanding how regulation of vascular tone is tailored to support neural function and behavior , in this case the drive to breathe .
Cerebral blood flow is highly sensitive to changes in CO2/H+ . An increase in CO2/H+ causes vasodilation and increased blood flow , which in turn facilitates removal of excess CO2/H+ . This response , known as vascular CO2 reactivity , serves to match blood flow with tissue metabolic needs ( Ainslie and Duffin , 2009 ) . Maintaining tight control of brain CO2/H+ levels is critical , as there is only a narrow range that is conducive to normal neural function . For example , a modest alkalosis of just 0 . 2 pH units can trigger seizure activity; conversely , a similar degree of acidification can inhibit cortical activity ( Schuchmann et al . , 2006 ) . Tissue CO2/H+ levels are also regulated by respiratory activity . This is accomplished by specialized subsets of neurons known as respiratory chemoreceptors that are activated by an increase in CO2/H+ ( Guyenet and Bayliss , 2015 ) . This information is then relayed to respiratory centers to enhance breathing , and consequently facilitate removal of arterial CO2 in the exhaled breath . The retrotrapezoid nucleus ( RTN ) is a region critical for respiratory chemoreception ( Guyenet and Bayliss , 2015 ) . This region contains a subset of neurons that are intrinsically sensitive to changes in CO2/H+ ( Mulkey et al . , 2004: Wang et al . , 2013 ) and relay responses to further respiratory control regions , such as the ventral respiratory complex to control breathing rate , inspiratory amplitude , active expiration and airway patency ( Guyenet and Bayliss , 2015; Silva et al . , 2016 ) . Disrupting mechanisms by which RTN neurons sense CO2/H+ abolishes ventilatory responses to CO2 and results in severe apnea ( Kumar et al . , 2015 ) . Interestingly , RTN astrocytes also support chemoreception by providing a CO2/H+-dependent purinergic drive that enhances activity of chemosensitive neurons ( Gourine et al . , 2010; Wenker et al . , 2012 ) . This function of RTN astrocytes is unique to the RTN since astrocytes elsewhere do not respond similarly to changes in pH ( Gourine et al . , 2010; Sobrinho et al . , 2014 ) . For more than a century , vascular CO2 reactivity has been assumed to be a common feature of the entire cerebrovascular tree ( Ainslie and Duffin , 2009; Roy and Sherrington , 1890 ) . However , if CO2/H+-induced vasodilation were to occur in chemosensitive regions it would accelerate removal of tissue CO2/H+ and effectively counter-regulate activity of respiratory chemoreceptors ( Xie et al . , 2006 ) . Therefore , we propose that regulation of vascular tone is specialized to support local neural network function , and specifically that a chemoreceptor region like the RTN has evolved a means of maintaining vascular tone during exposure to high CO2/H+ in a manner that supports the drive to breathe . Consistent with this , early studies showed that fast breath by breath changes in arteriole CO2 correspond with changes in pH measured at the ventral medullary surface ( Millhorn et al . , 1984 ) , suggesting tissue pH in this region is not highly buffered , possibly because blood vessels in this region do not dilate in response to CO2/H+ . Furthermore , considering that CO2/H+-evoked ATP release appears to be unique feature of RTN chemoreception ( Gourine et al . , 2010 ) and since ATP can mediate vasoconstriction in other brain regions ( Kur and Newman , 2014; Peppiatt et al . , 2006 ) , we hypothesize that CO2/H+-evoked ATP release will antagonize CO2/H+-vasodilation in the RTN , and thus prevent CO2/H+ washout , further enhancing chemoreceptor function . Consistent with this hypothesis , we find that arterioles in the RTN and cortex are differentially modulated by purinergic signaling during exposure to high CO2/H+ . Specifically , we show in vitro and in vivo that exposure to CO2/H+ caused vasoconstriction of RTN arterioles but vasodilation of cortical arterioles . The CO2/H+-response of RTN arterioles was blocked by bath application of a P2 receptor blocker ( pyridoxalphosphate-6-azophenyl-2' , 4'-disulfonic acid; PPADS ) and mimicked by a P2Y2/4 receptor agonist ( UTPγS ) but not a P2X receptor agonist ( α , β-mATP ) , suggesting mechanism ( s ) underlying this response in the RTN involve purinergic signaling and downstream activation of P2Y2 and/or P2Y4 receptors . To support the possibility that RTN vascular control contributes to respiratory behavior , we show that disruption of purinergic regulation of vascular tone or application of a vasodilator ( SNP ) to the RTN region decreased the ventilatory response to CO2 , whereas application of vasoconstrictors ( phenylephrine or U46619 ) potentiated the central chemoreflex . These results suggest for the first time that regulation of vascular tone in a respiratory chemoreceptor region is specialized to support the drive to breathe .
We initially tested our hypothesis using the brain slice preparation optimized for detecting increases or decreases in vascular tone ( see Mateials and methods ) . For these experiments , we targeted arterioles based on previously described criteria ( Mishra et al . , 2014; Filosa et al . , 2004 ) . Vessel diameter was monitored continuously during exposure to 15% CO2 ( pH = 6 . 9 ) under baseline conditions and during purinergic receptor blockade with PPADS . Consistent with our hypothesis , we found CO2/H+ differentially regulates arteriole diameter in the RTN depending on the function of purinergic receptors . For example , exposure to CO2/H+ under control conditions resulted in a vasoconstriction of −4 . 6 ± 0 . 6% ( p<0 . 0001 , N = 34 vessels ) ( Figure 1C ) ( estimated by Poiseuille’s law to decrease blood flow by ~20% ) . Further , we found that CO2/H+-induced constriction of RTN arterioles was retained in the presence of tetrodotoxin ( TTX; 0 . 5 µM ) to block neuronal action potentials ( −6 . 1 ± 1 . 6% , p=0 . 0103 , N = 6 vessels ) , thus suggesting glutamatergic CO2/H+-activated neurons ( Mulkey et al . , 2004 ) are not requisite determinants of this response . Conversely , exposure to CO2/H+ did not cause constriction of RTN arterioles during P2 receptor blockade with 5 µM PPADS ( −0 . 1 ± 0 . 9% , p=0 . 4876 , N = 8 vessels ) ( Figure 1A–C ) . We also tested effects of exogenous ATP to confirm that it functions as vasoconstrictor of RTN arterioles . Indeed , we found that exposure to ATP ( 100 µM ) resulted in a −5 . 8 ± 1 . 5% constriction ( p=0 . 0018 , N = 7 vessels ) ( Figure 1D–E ) . These results show that purinergic signaling contributes to CO2/H+-dependent control of RTN arterioles . 10 . 7554/eLife . 25232 . 003Figure 1 . CO2/H+-induced vasoconstriction of RTN arterioles is mediated by a purinergic dependent mechanism involving P2Y2/4 receptors . ( A ) trace of an RTN arteriole diameter show that increasing CO2 in the perfusion media from 5% to 15% ( balance air , in TTX ) caused vasoconstriction under baseline conditions but not in PPADS ( 5 µM ) . ( B ) example vessel image under baseline conditions and corresponding fluorescent intensity profile plots also show that exposure to high CO2 decreased vessel diameter . Profile plot scale bars: 2000 a . u . , 10 µm . ( C ) summarized results of RTN arteriole responses to CO2/H+ under baseline conditions ( N = 34 vessels ) and when P2-receptors were blocked ( 5 µM PPADS; N = 8 vessels ) , P1-receptors were blocked ( 10 µM 8-PT; N = 7 vessels ) , or ectonucleotidase activity was inhibited ( 100 µM POM1; N = 5 vessels ) . ( D ) example diameter traces show RTN arterioles constrict in response to bath application of ATP ( 100 µM ) or the selective P2Y2/4 receptor agonist UTPγS ( 0 . 5 µM ) but dilate when P1 receptors are activated by adenosine ( Ado; 1 µM ) . ( E ) summary data plotted as % diameter change in response to ATP ( N = 7 vessels ) , UTP ( N = 8 vessels ) , α , β-mATP ( 100 µM , preferential P2X agonist; N = 9 vessels ) or adenosine ( N = 9 vessels ) . ( F–G ) , immunoreactivity for P2Y2 ( F ) and P2Y4 ( G ) receptors was detected as brightly label puncta near endothelial cells ( DyLight 594 Isolectin B4 conjugate; IB4 ) , arteriole smooth muscle ( α-smooth muscle actin; αSMA ) , and astrocytes ( glial fibrillary acidic protein; GFAP ) associated with arterioles in the RTN ( N = 3 animals ) . Arrows identify receptor labeling close to endothelial or smooth muscle cells and arrowhead identifies receptor labeling of astrocyte processes . Scale bar 10 µM . Hash marks designate a difference in µm from baseline as determined by RM-one-way ANOVA and Fishers LSD test or paired t-test and asterisks identify differences in CO2/H+-induced % change under baseline conditions vs in the presence of PPADS ( C ) or ATP vs specific agonist-induced % change ( E ) ( one-way ANOVA and Fishers LSD test ) ; one symbol = p<0 . 05 , two symbols = p<0 . 01 , three symbols = p<0 . 001 , four symbols = p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 25232 . 003 Purinergic receptors are expressed by a wide variety of cell types including neurons , astrocytes , smooth muscle and endothelial cells; in the context of vascular control , the P2 receptors most commonly implicated in vasoconstriction are several members of the P2X family of ionotropic receptors and metabotropic P2Y2 , P2Y4 and P2Y6 ( Burnstock and Ralevic , 2014 ) . Since the concentration of PPADS used above to block purinergic modulation of RTN arterioles has highest affinities for P2X and P2Y2 and P2Y4 ( Ralevic and Burnstock , 1998 ) , to identify candidate P2 receptors that help maintain RTN arteriole tone during exposure to CO2/H+ , we tested effects of a selective P2Y2 and P2Y4 receptor agonist ( UTPγS ) ( Lazarowski et al . , 1996 ) and an agonist with high affinity for P2X receptors ( α , β-mATP ) ( Burnstock and Kennedy , 1985 ) . We found that bath application of UTPγS ( 0 . 5 µM ) mimicked effects of CO2/H+ by decreasing diameter of RTN arterioles ( −3 . 4 ± 0 . 6% , p=0 . 0003 , N = 8 vessels ) , whereas exposure to α , β-mATP ( 100 µM ) minimally affected arteriole tone ( p=0 . 2113 , N = 9 vessels ) ( Figure 1D–E ) . These results suggest that mechanism ( s ) underlying purinergic-dependent control of RTN arterioles during high CO2/H+ involve activation of Gq-coupled P2Y2/4 receptors . To further support this possibility , we performed immunohistochemistry using commercially available P2Y2 and P2Y4 specific antibodies in conjunction with cell-type specific markers for endothelial cells ( DyLight 594 Isolectin B4 conjugate; IB4 ) , vascular smooth muscle cells ( α-smooth muscle actin; α-SMA ) , and astrocytes ( anti-glial fibrillary acidic protein; GFAP ) . We detected P2Y2 and P2Y4 immunoreactivity in close proximity to all three cell types associated with RTN arterioles . For example , P2Y2 and P2Y4 labeling appeared as numerous intensely stained puncta near endothelial cells and smooth muscle cells and as smaller more diffuse puncta near astrocytes ( Figure 1F–G ) . The expression of these receptors together with our functional evidence suggest P2Y2/4 receptors contribute to purinergic-dependent vasoconstriction in the RTN during exposure to CO2/H+ . Considering that ATP and UTP breakdown products are known to affect vascular tone in other brain regions ( Burnstock and Ralevic , 2014 ) , we also pharmacologically manipulated P1 receptors and ectonucleotidase activity before or during exposure to CO2/H+ . We found that application of adenosine ( 1 µM ) under control conditions caused vasodilation of RTN arterioles ( 2 . 6 ± 0 . 6%; p=0 . 0027 , N = 9 vessels ) ( Figure 1D–E ) ; however , blockade of adenosine receptors with 8-phenyltheophylline ( 8-PT; 10 µM ) had negligible effects on CO2/H+-induced vasoconstriction ( −3 . 2 ± 0 . 4% , p=0 . 0002 , N = 7 vessels ) ( Figure 1C ) . Likewise , incubation in sodium metatungstate ( POM 1; 100 µM ) to inhibit ectonucleotidase activity also minimally affected the CO2/H+-vascular response of RTN arterioles ( −3 . 1 ± 0 . 6% , p=0 . 0195 , N = 5 vessels ) ( Figure 1C ) . These results suggest that nucleoside metabolites are not essential for CO2/H+-dependent regulation of vascular tone in the RTN . Previous evidence ( Gourine et al . , 2010 ) suggests CO2/H+-evoked ATP release from RTN astrocytes is mediated by intracellular Ca2+ . Therefore , in the absence of high CO2 , pharmacological activation of RTN astrocytes should trigger arteriole constriction by a purinergic-dependent mechanism . We test this by bath application of t-ACPD ( 50 µM ) , an mGluR agonist used to elicit Ca2+ elevations in cortical astrocytes ( Filosa et al . , 2004; Zonta et al . , 2003 ) . Exposure to t-ACPD caused vasoconstriction of RTN arterioles under baseline conditions ( −3 . 5 ± 0 . 5% , p=0 . 0007 , N = 7 ) but not in PPADS ( −0 . 6 ± 0 . 4% , p=0 . 2163 , N = 7 vessels ) ( Figure 2A–B , E ) . These results are consistent with our hypothesis that purinergic signaling , possibly from CO2/H+-sensitive RTN astrocytes ( Gourine et al . , 2010 ) , serves to maintain tone of arteriole in the RTN during hypercapnia . 10 . 7554/eLife . 25232 . 004Figure 2 . t-ACPD-mediated astrocyte activation has opposite effects on arteriole diameter in the RTN and cortex . ( A ) diameter trace of an RTN arteriole show the response of an RTN arteriole to t-ACPD ( 50 µM ) under baseline conditions and during P2-receptor blockade with PPADS ( 5 µM ) . ( B ) example RTN vessel image under baseline conditions and corresponding fluorescent intensity profile plots also show that exposure to tACPD decreased vessel diameter . ( C ) diameter trace of an cortical arteriole and corresponding vessel image with example profile plots ( D ) show that exposure to tACPD ( 50 µM ) increase cortical arteriole diameter . Profile plot scale bars: 2000 a . u . , 10 µm . ( E ) summary from the RTN ( N = 7 vessels ) and cortex ( N = 5 vessels ) data show that t-ACPD caused vasoconstriction of RTN arterioles under control conditions but not in the presence of PPADS , suggesting purinergic signaling most likely from astrocytes mediate constriction of arterioles in the RTN . Conversely , in the cortex t-ACPD caused vasodilation . ## , difference in µm from baseline ( paired t-test , p<0 . 01 ) . ### , difference in µm from baseline ( RM-one-way ANOVA and Fishers LSD test , p<0 . 001 ) . ** , difference in t-ACPD-induced % change under baseline conditions vs in PPADS ( paired t-test , p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25232 . 004 In marked contrast to the RTN , we found that cortical arterioles dilated in response to astrocyte activation by t-ACPD . For example , bath application of t-ACPD ( 50 µM ) dilated cortical arterioles by 3 . 2 ± 0 . 6% ( p=0 . 0030 , N = 5 vessels ) ( Figure 2C–E ) . This response is consistent with previous cortical studies ( Filosa et al . , 2004; Zonta et al . , 2003 ) , and suggests that astrocytes in the RTN and cortex have fundamentally different roles in regulation of vascular tone . Also consistent with previous work ( Ainslie and Duffin , 2009 ) , we confirmed that cortical arterioles dilate in response to CO2/H+ ( 5 . 7 ± 1 . 1% , p=0 . 0057 , N = 11 vessels ) ( Figure 3A–C ) . Interestingly , we also found that the CO2/H+-vascular response of cortical arterioles was reduced to 0 . 5 ± 0 . 5% in PPADS ( p=0 . 004 , N = 6 vessels ) ( Figure 3A–C ) , suggesting involvement of endogenous ATP in cortical arteriole CO2/H+-dilation . As in the RTN , we also found that endothelial cells , smooth muscle and astrocytes associated with cortical arterioles were immunoreactive for P2Y2 and P2Y4 ( Figure 3D–E ) , suggesting the differential roles of purinergic signaling in these regions is not due to the presence or absence of P2Y2 and P2Y4 . However , since the vascular responses to activation of P2Y2/4 can vary depending on which cells express the receptor ( Burnstock and Ralevic , 2014 ) , it remains possible that differential expression of P2Y2/4 by endothelial and smooth muscle may mediate vasodilation in the cortex and constriction in the RTN , respectively . It is also possible that other purinergic receptors contribute to regulation of arteriole tone in these regions . For example , endothelial P2Y1 receptors are known to mediate vasodilation in the cortex ( Burnstock and Ralevic , 2014 ) . However , we found that in vivo application of a selective P2Y1 receptor blocker ( MRS2179 , 100 µM ) had no measurable effect on the CO2/H+ response of pial vessels in the RTN ( −3 . 7 ± 0 . 8% , vs . saline plus CO2: −4 . 3 ± 0 . 7%; p=0 . 068; N = 5 vessels ) or cortex ( 4 . 8 ± 0 . 5% , vs . saline plus CO2: 4 . 7 ± 0 . 6%; p=0 . 24; N = 5 vessels ) ( data not shown ) . Alternatively , arachidonic acid metabolites are also potent regulators of vascular tone ( MacVicar and Newman , 2015 ) and recent evidence showed that CO2/H+-mediated vasodilation in the cortex and hippocampus involved activation of cyclooxygenase-1 and prostaglandin E2 release by astrocytes ( Howarth et al . , 2017 ) . Considering that purinergic signaling can elicit Ca2+ responses in astrocytes to facilitate prostaglandin E2 synthesis ( Xu et al . , 2003 ) , it remains possible that purinergic signaling contributes to cortical CO2/H+ dilation by influencing synthesis and release of prostaglandin E2 by astrocytes . However , currently the cellular and molecular basis of purinergic dilation in the cortex remains unknown . 10 . 7554/eLife . 25232 . 005Figure 3 . Cortical arterioles dilate in response to CO2/H+ . ( A ) diameter trace of a cortical arteriole with an example vessel image and fluorescence profile plots ( B ) show that exposure to CO2/H+ caused vasodilation under baseline conditions and this response was blunted by PPADS ( 5 µM ) . Profile plot scale bars: 2000 a . u . , 10 µm . ( C ) summary data show CO2/H+-induced vasodilation of cortical arterioles under bassline conditions ( N = 11 vessels ) but not in PPADS ( N = 6 vessels ) . ( D–E ) , immunoreactivity for P2Y2 ( D ) and P2Y4 ( E ) receptors was detected as brightly label puncta near endothelial cells ( DyLight 594 Isolectin B4 conjugate; IB4 ) , arteriole smooth muscle ( α-smooth muscle actin; αSMA ) , and astrocytes ( glial fibrillary acidic protein; GFAP ) associated with cortical arterioles ( N = 3 animals ) . Arrows identify receptor labeling close to endothelial or smooth muscle cells and arrowheads identifies receptor labeling of astrocyte processes . Scale bar 10 µM . ## , difference in µm from baseline ( paired t test , p<0 . 01 ) . ** , difference in CO2/H+-induced % change under control conditions vs in PPADS ( paired t-test , p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25232 . 005 To determine whether regulation of vascular tone in the RTN impacts respiratory behavior , we pharmacologically manipulated RTN vessels in anesthetized rats while simultaneously measuring systemic blood pressure and diaphragm EMG activity ( as a measure of respiratory activity ) during exposure to high CO2 . We found that localized application of the vasoconstrictors phenylephrine ( Phe; 1 µM ) or U46619 ( 1 µM ) to the ventral medullary surface ( VMS ) enhanced the ventilatory response to CO2 by increasing diaphragm electromyogram ( EMG ) amplitude 15 ± 2% and 18 ± 1 . 8% , respectively ( Figure 4A–C ) ( p=0 . 02; N = 6 animals ) but with no change in frequency ( Figure 4D ) ( p>0 . 05; N = 6 animals ) . Also consistent with the possibility that increased blood flow will facilitate removal of tissue CO2/H+ , and thus decrease the stimulus to chemosensitive neurons , we found that VMS application of the vasodilator sodium nitroprusside ( SNP; 1 µM ) decreased ventilatory response to CO2 by decreasing diaphragm amplitude by 24 ± 2 . 6% ( Figure 4A–C ) ( p=0 . 02; N = 6 animals ) but with no change in frequency ( Figure 4D ) ( p>0 . 05; N = 6 animals ) . These treatments had negligible effect on systemic mean arterial pressure ( MAP ) ( Phe: 110 ± 2; SNP: 110 ± 2; U4619: 108 ± 3 vs . saline: 109 ± 2 mmHg; p>0 . 05 ) ( Figure 4E ) . These results are consistent with the possibility that regulation of vascular tone in the RTN can influence respiratory output . However , we cannot exclude potential direct effects of these drugs on activity of neurons or astrocytes in the region . For example , phenylephrine can directly stimulate activity of chemosensitive RTN neurons ( Kuo et al . , 2016 ) . Therefore , effects of phenylephrine on chemoreception likely involves both vasoconstriction and direct neural activation . It remains to be determined whether U46619 or SNP also affect activity of neurons or astrocytes in the RTN . 10 . 7554/eLife . 25232 . 006Figure 4 . Local constriction and dilation of RTN vessels reciprocally modulates the ventilatory response to CO2in vivo . ( A ) end expiratory CO2 ( EtCO2 ) , arterial pressure ( AP ) and diaphragm EMG ( DiaEMG ) traces show that application of vasoconstrictors ( phenylephrine , Phe , 1 µM or U46619 , 1 µM ) or a vasodilator ( sodium nitroprusside , SNP , 1 µM ) to the RTN increased and decreased the ventilatory response to 7–8% CO2 , respectively . ( B ) diaphragm EMG ( DiaEMG ) traces expanded in time show that application of Phe , U46619 or SNP , to the VMS in the region of the RTN increased and decreased the DiaEMG amplitude response to 7–8% CO2 . ( C–E ) summary data show effects of saline , SNP , Phe and U46619 applications to the VMS near the RTN on DiaEMG amplitude ( N = 6 animals per group ) ( C ) , DiaEMG frequency ( D ) and mean arterial pressure ( MAP ) ( E ) . * , difference in CO2/H+-induced % change under control conditions ( saline ) vs during vasodilation or vasoconstriction ( RM-ANOVA followed by Bonferroni multiple-comparison test , p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25232 . 006 To determine whether purinergic signaling regulates CO2/H+-mediated constriction in vivo , we first measured the diameter of pial vessels on the VMS in the region of the RTN or on the surface of the cortex during exposure to high CO2 under control conditions and when P2 receptor are blocked with PPADS ( 10 µM ) . Consistent with our in vitro data , we found that increasing end-expiratory CO2 to 9 . 5–10% , which corresponds with an arterial pH of 7 . 2 pH units ( Guyenet et al . , 2005 ) , constricted VMS vessels by −4 . 5 ± 0 . 5% ( p=0 . 014 , N = 5 animals ) ( Figure 5A–B ) . However , when P2-receptors are blocked with PPADS ( 10 µM ) , increasing inspired CO2 resulted in a vasodilation of 4 . 3 ± 0 . 4% ( Figure 5A–B ) ( p=0 . 036; N = 5 animals ) . This suggests that in the RTN , purinergic-mediated vasoconstriction is working against a background CO2/H+ dilation , possibly mediated by a cyclooxygenenase/prostroglandine E2-dependent mechanism as described elsewhere in the brain ( Howarth et al . , 2017 ) . Therefore , disruption of CO2/H+ dilation would be expected to enhance purinergic-dependent vasoconstriction of RTN arterioles , and thus increase baseline breathing and the ventilatory response to CO2 . Consistent with this , administration of a cyclooxygenase inhibitor ( indomethacin ) has been shown to increase baseline breathing and the ventilatory response to CO2 in humans ( Xie et al . , 2006 ) . However , it is also possible that decreasing cerebral blood flow globally by indomethacin treatment or cerebral ischemia ( Chapman et al . , 1979 ) will cause widespread acidification leading to enhanced activation of multiple chemoreceptor regions including those outside the RTN ( Nattie and Li , 2012 ) , thus further increasing chemoreceptor drive . It should also be noted that global disruption of cerebrovascular CO2/H+ reactivity as associated with certain pathological states including heart failure and stroke ( Yonas et al . , 1993; Howard et al . , 2001 ) or by systemic administration of indomethacin can increase chemoreceptor gain to the extent that breathing becomes unstable ( Fan et al . , 2010; Xie et al . , 2009 ) . These results underscore the need to understand how CO2/H+ regulates vascular tone at other levels of the respiratory circuit . 10 . 7554/eLife . 25232 . 007Figure 5 . Purinergic signaling opposes CO2/H+-dilation of VMS pial vessels in vivo and contributes to the ventilatory response to CO2 . ( A ) summary data plotted as % change in RTN pial vessel diameter in response to an increase in end expiratory CO2 after VMS application of saline ( 100 nL; N = 5 animals ) or PPADS ( 10 µM , 100 nL; N = 5 animals ) . Also shown are the vascular responses to VMS application of ATP ( 1 mM , 100 nL; N = 5 ) and UTPγS ( 1 mM , 100 nL; N = 5 animals ) . ( B ) Photomicrographs ( 40X ) show pial vessel distribution on the VMS , arrow; representative vessel analyzed . ( C ) Summary data shows the response of cortical pial vessels to CO2 after local application of saline or PPADS ( 10 µM , 100 nL; N = 5 animals ) . Also shown are vascular responses to exogenous application of ATP ( 1 mM - 100 nL; N = 5 animals ) or UTPγS ( 1 mM , 100 nL; N = 5 ) . ( D ) End expiratory CO2 ( EtCO2 ) , arterial pressure ( AP ) and diaphragm EMG ( DiaEMG ) traces show that bilateral VMS application of PPADS ( 10 µM , 100 nL ) attenuated the ventilatory response to CO2 . ( E ) summary data show CO2-induced changes in DiaEMG frequency and amplitude after bilateral VMS application of saline and PPADS ( 10 µM; N = 5 animals ) . ( F ) summary data show that PPADS ( 10 µM ) application to the cortex surface had no measurable effect on CO2-induced changes in DiaEMG frequency and amplitude ( N = 5 animals ) . Hash marks designate a difference in µm from baseline ( RM-ANOVA followed by Bonferroni multiple-comparison test , # , p<0 . 05 ) . Asterisks identify a difference in CO2/H+--induced % change under control conditions ( saline ) vs in the presence of PPADS ( RM-ANOVA followed by Bonferroni multiple-comparison test , * , p<0 . 05; ** , p<0 . 01 ) ( panels A and C ) or paired t-test ( panel E , * , p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25232 . 007 Also consistent with our slice data , we found that exogenous application of ATP ( 1 mM ) or UTPγS ( 1 mM ) constricted VMS vessels by −5 . 1 ± 0 . 6% and −5 . 0 ± 0 . 5% , respectively ( p=0 . 001; N = 5 animals ) ( Figure 5A ) . Conversely , cortical pial vessels dilated in response to an increase in inspired CO2 under control conditions ( 5 ± 0 . 5% ) and after application of 10 µM PPADS ( 3 . 2 ± 0 . 3% ) ( Figure 5C ) ( p=0 . 03; N = 5 animals ) ; however , the CO2/H+-induced vasodilation of cortical vessels was blunted in the presence of PPADS ( p=0 . 03; N = 5 animals ) , suggesting endogenous purinergic signaling may facilitate dilation of cortical vessels in response to CO2/H+ . However , we failed to mimic this response by exogenous application of purinergic agonists; exposure to ATP ( 1 mM ) or UTPγS ( 1 mM ) constricted cortical vessels by −4 . 9 ± 0 . 7% and −4 . 1 ± 0 . 7% , respectively ( p=0 . 024; N = 5 animals ) . These divergent results are not entirely unexpected since we detected P2Y2 and P2Y4 immunoreactivity on endothelial cells and smooth muscle of cortical arterioles ( Figure 3D–E ) , and exogenous application of P2 agonists may activate P2 receptors not necessarily targeted by endogenous purinergic signaling . Future work is required to identify the source of purinergic drive and effector P2 receptors contributing to vascular CO2/H+-reactivity in the cortex . In addition , as previously described ( Gourine et al . , 2005 ) , we found that application of PPADS ( 10 µM ) to the RTN blunted the ventilatory response to CO2 both in terms of DiaEMG frequency ( 82 ± 3 , vs . saline: 97 ± 4% ) ( p=0 . 042 , N = 5 ) and amplitude ( 83 ± 6 , vs . saline: 100 ± 5% ) ( p=0 . 035 , N = 5 ) ( Figure 5D–E ) . Considering that RTN manipulations of vascular tone preferentially affect respiratory amplitude ( Figure 4A–D ) , whereas application of PPADS to the same region , which likely disrupts both direct excitatory effects of ATP on RTN chemoreceptors ( Wenker et al . , 2012 ) and indirect effects of ATP on vascular tone ( Figure 5A ) , blunts respiratory frequency and amplitude , suggests that purinergic signaling in the RTN might regulate discrete aspects of respiratory output . In the cortex , application of PPADS had no measurable effect on CO2-induced changes in DiaEMG frequency ( p=0 . 33 , N = 5 animals ) and amplitude ( p=0 . 42 , N = 5 ) ( Figure 5F ) . Together with previous evidence , these findings suggest that purinergic signaling contributes to RTN chemoreception by directly activating RTN neurons ( Gourine et al . , 2010 ) and indirectly by opposing CO2/H+-vasodilation . It should be acknowledged that our study is limited to the use of pharmacological tools that potentially have off-target effects . We have minimized this concern by ( i ) using low concentrations of PPADS that are reported to be specific for P2 receptor’s ( Lorier et al . , 2004 ) ; ( ii ) mimicking CO2/H+-induced vasoconstriction in vitro and in vivo by exogenous application of ATP and a specific P2Y2/4 receptor agonist ( UTPγS ) , but not by a non-specific P2X receptor agonist ( α , β-mATP ) ; ( iii ) confirming that candidate P2Y2 and P2Y4 receptors are expressed in the RTN at the astrocyte-arteriole interface; and ( iv ) for in vitro experiments by confirming that CO2/H+-induced vasoconstriction was retained when neuronal action potentials were blocked with TTX . Therefore , our results suggest that purinergic signaling possibly through P2Y2/4 receptors in the RTN provides specialized control of vascular tone by preventing CO2/H+-induced dilation . Our results also suggest that regulation of vascular tone in the RTN contributes functionally to the ventilatory response to CO2 . This is the first evidence to suggest that regulation of vascular tone in a chemoreceptor region contributes to the drive to breathe . This discovery may be of fundamental importance to understanding how regulation of vascular tone impacts neural network function and ultimately behavior .
All procedures were performed in accordance with National Institutes of Health and University of Connecticut Animal Care and Use Guidelines . A total of 93 adult Sprague Dawley rats ( 60–100 days of age ) were used for in vitro experiments . Animals were decapitated under isoflurane anesthesia , and transverse brainstem slices ( 200 μm ) were prepared using a vibratome in ice-cold substituted artificial cerebrospinal fluid ( aCSF ) solution containing ( in mm ) : 130 NaCl , 3 KCl , 2 MgCl2 , 2 CaCl2 , 1 . 25 NaH2PO4 , 26 NaHCO3 , and 10 glucose , with 0 . 4 mM L-ascorbic acid added ( all Sigma-Aldrich ) . Slices were incubated for ∼30 min at 37°C and subsequently at room temperature in aCSF . Prior to imaging slices were incubated for 1 hr with 10 µg/ml TRITC-lectin conjugate ( Sigma-Aldrich , St . Louis , MO ) or 6 µg/ml DyLight 594 Isolectin B4 conjugate ( Vector Labortories ) to label endothelial cells as previously described ( Mishra et al . , 2014 ) . Slicing solutions were equilibrated with a high oxygen carbogen gas ( 95% O2-5% CO2 ) ( Mulkey et al . , 2001 ) . An individual slice containing the RTN was transferred to a recording chamber mounted on a fixed-stage microscope ( Zeiss Axioskop FS ) and perfused continuously ( ∼2 ml min−1 ) with aCSF bubbled with 5% CO2 , 21% O2 ( balance N2; pHo ~7 . 35; 37⁰ C ) ( Gordon et al . , 2008 ) . Hypercapnic solution was made by equilibrating aCSF with 15% CO2 , 21% O2 ( balance N2; pHo ~6 . 90; 37⁰ C ) . Arterioles were identified as previously described ( Mishra et al . , 2014; Filosa et al . , 2004 ) by the following criteria: clear evidence of vasomotion under IR-DIC , bulky fluorescent labeling and a thick layer of smooth muscle surrounding the vessel lumen . Vessels that appeared collapsed and unhealthy were excluded , as were those with little fluorescence staining and thin walls , indicative of a lack of smooth muscle ( Mishra et al . , 2014 ) . All arterioles selected for analysis had a resting luminal diameter of between 8–45 µm; RTN vessels were located within 200 µm of the ventral surface and below the caudal end of the facial motor nucleus and cortical vessels were located in layers 1–3 . For an experiment , fluorescent images were acquired at a rate of 1 frame/20 s using a x40 water objective , a Clara CCD Andor camera and NIS Elements software . To induce a partially constricted state we bath applied a thromboxane A2 receptor agonist ( U46619; 125 nM; Sigma-Aldrich , St . Louis , MO ) . At this concentration , U46619 has been shown to decrease vessel diameter by 20–30% under similar experimental conditions , thus allowing assessment of both vasodilation and vasoconstriction ( Filosa et al . , 2004; Girouard et al . , 2010 ) . In the continued presence of U-46619 , we then characterized the effects of hypercapnia , ATP ( 100 µM; Sigma-Aldrich , St . Louis , MO ) , α , β-methyleneATP ( 100 µM ) , UTPγS ( 0 . 5 µM ) , and adenosine ( 1 µM ) , or the mGluR agonist t-ACPD ( ( ± ) −1-aminocyclopentane-trans-1 , 3-dicarboxylic acid; 50 µM ) alone or in the presence of P2-recepetor blocker PPADS ( 5 µM; Tocris Bioscience , Minneapolis , MN ) , the P1 receptor antagonist 8-Phenyltheophylline ( 8-PT; 10 µM; Sigma ) or the Ecto-NTPDase antagonist sodium metatungstate ( POM 1; 100 µM; Tocris ) . In a subset of experiments we also tested CO2 , ATP and PPADS in the presence of TTX ( 0 . 5 µM; Alomone Laboratories ) . As previously described ( Girouard et al . , 2010 ) , at the end of each experiment we assessed arteriole viability by inducing a constriction with a high K+ solution ( 60 mM ) and then maximal dilation with a Ca2+ free solution containing EGTA ( 5 mM ) , papaverine ( 200 μM , a phosphodiesterase inhibitor ) and diltiazem ( 50 µM , to block L-type Ca channels ) . Vessels from both regions of interest ( RTN and cortex ) show similar responses to these conditions , and vessels that did not respond were excluded from analysis . Rat brain slices were prepared from three rats and labelled with DyLight 594 Isolectin B4 conjugate as above followed by immersion fixation overnight in 1% paraformaldehyde in pH 7 . 4 PBS at 4°C . Excess fixative was removed by three washes in PBS , and prior to antibody incubations , tissue sections were treated to unmask epitopes with 0 . 2 mg/ml pepsin ( Sigma-Aldrich , St . Louis , MO ) in 0 . 2 M HCl for 10 mins at 37°C ( Corteen et al . , 2011 ) followed by three washes in PBS for P2Y4 labelling only . A blocking stage was then performed by incubating tissue in 10% normal horse serum in PBS with 10% Triton X-100 ( Sigma-Aldrich , St . Louis , MO ) for 1 hr at room temperature ( RT ) . Sections were then incubated overnight at RT with primary antibodies diluted in blocking solution as follows: 1:200 rabbit anti-P2Y2 ( RRID: AB_2040078 ) or P2Y4 receptor ( RRID: AB_2040080 ) ( Alomone Labs , Alomone Labs , Jerusalem Israel ) , 1:200 chicken anti-glial fibrillary acidic protein ( RRID: AB_177521 ) ( Chemicon ) and 1:500 mouse anti-α-smooth muscle actin ( RRID: AB_262054 ) ( Sigma-Aldrich , St . Louis , MO ) . After washes in PBS , tissues were incubated for 1 hr at RT with the appropriate secondary antibodies raised in donkey , conjugated with 488DyLight 1:800 ( RRID: AB_2492289 ) , 405DyLight 1:200 ( RRID: AB_2340373 ) or Cy5 1:500 ( RRID: AB_2340820 ) ( Jackson Immunoresearch Laboratories ) . Sections were washed in PBS again before being mounted with Vectasheild ( VectorLabs ) . Images were acquired using a Nikon A1R confocal microscope ( Nikon Instruments ) , with minimal background staining observed in the control reactions where primary antibodies were omitted or P2 receptor antibodies were pre-absorbed with control antigen prior to exposure to tissues . For confocal photomicrographs , two-dimensional flattened images of the projected z-stacks are presented . Animal use was in accordance with guidelines approved by the University of São Paulo Animal Care and Use Committee . A total of 21 adult male Wistar rats ( 60–90 days of age; 270–310 g ) were used for in vivo experiments . General anesthesia was induced with 5% halothane in 100% O2 . A tracheostomy was made and the halothane concentration was reduced to 1 . 4–1 . 5% until the end of surgery . The femoral artery was cannulated ( polyethylene tubing , 0 . 6 mm o . d . , 0 . 3 mm i . d . , Scientific Commodities ) for measurement of arterial pressure ( AP ) . The femoral vein was cannulated for administration of fluids and drugs . Rats were placed supine onto a stereotaxic apparatus ( Type 1760; Harvard Apparatus ) on a heating pad and core body temperature was maintained at a minimum of 36 . 5°C via a thermocouple . The trachea was cannulated . Respiratory flow was monitored via a flow head connected to a transducer ( GM Instruments ) and CO2 via a capnograph ( CWE , Inc , ) connected to the tracheal tube . Paired EMG wire electrodes ( AM-System ) were inserted into the diaphragm muscle to record respiratory-related activity . After the anterior neck muscles were removed , a basio-occipital craniotomy exposed the ventral medullary surface , and the dura was resected . After bilateral vagotomy , the exposed tissue around the neck and the mylohyoid muscle was covered with mineral oil to prevent drying . Baseline parameters were allowed to stabilize for 30 min prior to recording . Mean arterial pressure ( MAP ) , diaphragm muscle activity ( DiaEMG ) and end-expiratory CO2 ( etCO2 ) were digitized with a micro1401 ( Cambridge Electronic Design ) , stored on a computer , and processed off-line with version 6 of Spike 2 software ( Cambridge Electronic Design ) . Integrated diaphragm activity ( ∫DiaEMG ) was collected after rectifying and smoothing ( τ = 0 . 03 ) the original signal , which was acquired with a 300–3000 Hz bandpass filter . Noise was subtracted from the recordings prior to performing any calculations of evoked changes in DiaEMG . A direct physiological comparison of the absolute level of DiaEMG activity across nerves is not possible because of non-physiological factors ( e . g . , muscle electrode contact , size of muscle bundle ) and the ambiguity in interpreting how a given increase in voltage in one EMG relates to an increase in voltage in another EMG . Thus , muscle activity was defined according to its baseline physiological state , just prior to their activation . The baseline activity was normalized to 100% , and the percent change was used to compare the magnitude of increases or decreases across muscle from those physiological baselines . Each in vivo experiment began by testing responses to hypercapnia by adding pure CO2 to the breathing air supplied by artificial ventilation . In each rat , the addition of CO2 was monitored to reach a maximum end-expiratory CO2 between 9 . 5% and 10% , which corresponds with an estimated arterial pH of 7 . 2 based on the following equation: pHa = 7 . 955–0 . 7215 × log10 ( ETCO2 ) ( Guyenet et al . , 2005 ) . This maximum end-expiratory CO2 was maintained for 5 min and then replaced by 100% O2 . To determine whether local regulation of vascular tone in the region of the RTN contributes to the CO2/H+-dependent drive to breathe , we made bilateral injections of saline , phenylephrine ( Phe , 1 µM ) , U46619 ( 1 µM ) or sodium nitroprusside ( SNP , 1 µM ) while monitoring DiaEMG amplitude and frequency . These drugs were diluted to 1 µM with sterile saline ( pH 7 . 4 ) and applied using single-barrel glass pipettes ( tip diameter of 25 µm ) connected to a pressure injector ( Picospritzer III , Parker Hannifin Corp , Cleveland , OH ) . For each injection , we delivered a volume of 100 nl over a period of 5s . Injections in the VMS region were placed 1 . 9 mm lateral from the basilar artery , 0 . 9 mm rostral from the most rostral hypoglossal nerve rootlet , and at the VMS . The second injection was made 1–2 min later at the same level on the contralateral side . In separate series of experiments saline , ATP ( 1 mM ) , UTPγS ( 1 mM ) or PPADS ( 10 µM ) were applied similarly to the VMS to test the effect of P2-blockade on vascular CO2/H+ reactivity and the ventilatory response to CO2 . A cranial optical window was prepared using standard protocols previously described ( Kim et al . , 2015 ) . Briefly , a dental drill ( Midwest Stylus Mini 540S , Dentsply International ) was used to thin a circumference of a 4–5 mm-diameter circular region of the skull over somatosensory cortex ( stereotaxic coordinates: AP: −1 . 8 mm from bregma and ML: 2 . 8 mm lateral to the midline ) . For the VMS , the anterior neck muscles were removed , a basio-occipital craniotomy exposed the ventral medullary surface , and the dura was resected . Pial vessels in the VMS had an average and were located 1 . 9 mm lateral from the basilar artery and 0 . 9 mm rostral to the most rostral portion of the hypoglossal nerve rootlet . Both thinned bone were lifted with forceps . The surface of the cortex or the VMS were cleaned with buffer containing ( in mmol/L ) the following: 135 NaCl , 5 . 4 KCl , 1 MgCl2 , 1 . 8 CaCl2 , and 5 HEPES , pH 7 . 3 . , and a chamber ( home-made 1 . 1-cm-diameter plastic ring was glued with dental acrylic cement attached to a baseplate ) . The chamber was sealed with a circular glass coverslip ( #1943–00005 , Bellco ) . The baseplate was affixed to the Digital Camera ( Sony , DCR-DVD3-5 ) and a light microscope was used for vessel imaging ( x40 magnification ) . Vessel diameter was determined offline using ImageJ . For in vitro experiments , stack registration and selection of linear regions of interest ( ROIs ) ( three regions per arteriole ) was carried out . Linear ROI’s were used to create a fluorescent intensity profile plot as described previously ( Mishra et al . , 2014 ) and a macro ( available at https://github . com/omsai/blood-vessel-diameter [Nanda , 2017]; copy archived at https://github . com/elifesciences-publications/blood-vessel-diameter ) . was used to determine the peak-peak distance as a measure of vessel diameter for each frame . In most cases , we also confirmed vessel diameter by manually measuring at least one point per frame . In vivo data was also analyzed using three linear ROI’s drawn perpendicular to the vessel in each image and the Diameter plug-in function in ImageJ was used to calculate changes in diameter ( Kim et al . , 2015; Fischer et al . , 2010 ) . All in vitro images were calibrated and pixel distances converted to diameter ( µM ) and these values were used for analysis of CO2/H+ or agonist-induced changes in vessel diameter from baseline by RM-one-way ANOVA and Fishers LSD test or paired t-test . Hash marks were used to identity differences from baseline ( vasoconstriction or vasodilation ) . Mean percent changes in vessel diameter was used to compare between agonist responses or CO2/H+ responses under control conditions vs during purinergic receptor blockade or in the presence of an ectonucleotidase inhibitor by one-way ANOVA and Fishers LSD test or t-test . Asterisks were used to identify differences in % change in vessel diameter . For in vivo experiments , respiratory muscle activity was calculated as the mean amplitude of the integrated DiaEMG over 20 respiratory cycles . To obtain control values , the 20 cycles preceding each experimental manipulation for all parameters were averaged . Under hypercapnic conditions , measurements from the 20 cycles preceding stimulus cessation were averaged . Respiratory frequency ( fR ) was ( 1/ ( inspiratory time + expiratory time ) . Differences in the ventilatory response to CO2 were determined using either paired t-test or one-way analysis of variance ( ANOVA ) followed by the Bonferroni multiple-comparisons as appropriate . Power analysis was used to determine sample size , all data sets were tested for normality using Shapiro-Wilk test . All data values are expressed as means ± SEM and specific statistical test and relevant p values are reported in the text and figure legends . | We breathe to help us take oxygen into the body and remove carbon dioxide . Our cells use the oxygen to break down food to release energy , and as they do so they produce carbon dioxide as a waste product . Cells release this carbon dioxide back into the bloodstream so that it can be transported to the lungs to be breathed out . Carbon dioxide also makes the blood more acidic; if the blood becomes too acidic , tissues and organs may not work properly . The brain uses roughly 25% of the oxygen consumed by the body and is particularly sensitive to the levels of gases and acidity in the blood . It has been known for more than a century that increased carbon dioxide causes blood vessels in the brain to widen , allowing the excess carbon dioxide to be carried away quickly . More recent work has shown that increased carbon dioxide also activates neurons called respiratory chemoreceptors . These in turn activate the brain centers that drive breathing , causing us to breathe more rapidly to help us remove surplus carbon dioxide . But this scenario contains a paradox . If high levels of carbon dioxide cause widening of the blood vessels in the brain regions that contain respiratory chemoreceptors , this should , in theory , wash out that important stimulus , reducing the drive to breathe . So how does the brain prevent this unhelpful response ? By studying the brains of adult rats , Hawkins et al . show that different rules apply to the brain centers that control breathing compared to other areas of the brain . In one such region , if the blood becomes too acidic , support cells called astrocytes release chemical signals called purines . This counteracts the tendency of high carbon dioxide levels to widen blood vessels in this region , and instead causes these vessels to become narrower . This mechanism ensures that local levels of carbon dioxide in respiratory brain centers remain in tune with the demands of local networks , thereby maintaining the drive to breathe . The next challenges are to identify the molecular mechanisms that control the diameter of blood vessels in brain regions containing respiratory chemoreceptors , and to find out whether drugs that modulate these mechanisms have the potential to treat some respiratory conditions . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2017 | Purinergic regulation of vascular tone in the retrotrapezoid nucleus is specialized to support the drive to breathe |
The centrosome acts as a microtubule organizing center ( MTOC ) , orchestrating microtubules into the mitotic spindle through its pericentriolar material ( PCM ) . This activity is biphasic , cycling through assembly and disassembly during the cell cycle . Although hyperactive centrosomal MTOC activity is a hallmark of some cancers , little is known about how the centrosome is inactivated as an MTOC . Analysis of endogenous PCM proteins in C . elegans revealed that the PCM is composed of partially overlapping territories organized into an inner and outer sphere that are removed from the centrosome at different rates and using different behaviors . We found that phosphatases oppose the addition of PCM by mitotic kinases , ultimately catalyzing the dissolution of inner sphere PCM proteins at the end of mitosis . The nature of the PCM appears to change such that the remaining aging PCM outer sphere is mechanically ruptured by cortical pulling forces , ultimately inactivating MTOC function at the centrosome .
Numerous cell functions such as transport , migration , and division are achieved through the specific spatial organization of microtubules imparted by microtubule organizing centers ( MTOCs ) . The best-studied MTOC is the centrosome , a membrane-less organelle composed of two barrel-shaped microtubule-based centrioles surrounded by a cloud of pericentriolar material ( PCM ) . Microtubules at the centrosome are mainly nucleated and localized by complexes within the PCM , which generate a radial array of microtubules in dividing animal cells and some specialized cell types such as fibroblasts . The PCM is a central hub for the regulation of a number of cellular processes including centriole duplication , ciliogenesis , cell cycle regulation , cell fate determination , and microtubule organization ( Chichinadze et al . , 2013; Fry et al . , 2017; Stubenvoll et al . , 2016 ) . In Drosophila and human cell lines , PCM proteins including a subset of scaffolding proteins are organized in cumulative layers ultimately recruiting microtubule nucleation and organization factors , such as the conserved microtubule nucleating γ-tubulin ring complex ( γ-TuRC ) ( Fu and Glover , 2012; Lawo et al . , 2012; Mennella et al . , 2012 ) . In C . elegans , the PCM is much simpler in composition , built from the interdependent recruitment of two scaffolding proteins , SPD-2/CEP192 and SPD-5 , the functional homologue of CDK5RAP2/Cnn ( Hamill et al . , 2002; Kemp et al . , 2004 ) . Together with the highly conserved kinase AIR-1/Aurora-A , SPD-2 and SPD-5 are required to localize γ-TuRC , which in C . elegans is composed of TBG-1/γ-tubulin , GIP-1/GCP3 , GIP-2/GCP2 and MZT-1/MZT1 ( Bobinnec et al . , 2000; Hamill et al . , 2002; Hannak et al . , 2002; Kemp et al . , 2004; Lin et al . , 2015; Oakley et al . , 2015; Sallee et al . , 2018 ) . γ-TuRC and AIR-1 have been shown to together be required to build microtubules at the centrosome in the C . elegans zygote ( Motegi et al . , 2006 ) . Additionally , the major role of the γ-TuRC at the PCM in C . elegans might be to anchor microtubules at the periphery as loss of γ-TuRC results in microtubules distributed throughout the PCM ( O'Toole et al . , 2012 ) . In addition to the γ-TuRC , several other microtubule regulating proteins are recruited to the PCM to promote its microtubule organizing center function through the stabilization and growth of microtubules . The conserved complex of ZYG-9/XMAP-215/Alp14 , a processive microtubule polymerase ( Matthews et al . , 1998; Thawani et al . , 2018 ) , and TAC-1/TACC/Alp7 promotes microtubule polymerization ( Bellanger and Gönczy , 2003; Bellanger et al . , 2007 ) . This complex is also involved in microtubule nucleation as has been recently shown in yeast and Xenopus egg extract ( Flor-Parra et al . , 2018; Thawani et al . , 2018 ) . In addition . the microtubule stabilizing and nucleation-promoting factor TPXL-1/TPX2 also localizes to the PCM where it interacts with and activates AIR-1 ( Bayliss et al . , 2003; Zhang et al . , 2017 ) . Although the pathways required to build the PCM are largely known in C . elegans , the organization of proteins within the PCM has been generally unexplored . One notable exception is that microtubules have been shown to localize to the periphery of the PCM by electron microscopy ( O'Toole et al . , 2012 ) . As TBG-1 is found throughout the PCM , these studies suggest the existence of different pools of TBG-1 at the PCM , with an active population at the periphery that organizes microtubules . The centrosome is not a static organelle; during each cell cycle , MTOC activity at the centrosome is massively increased to ultimately build the mitotic spindle ( Dictenberg et al . , 1998; Woodruff et al . , 2014 ) . This increase in centrosomal MTOC activity relies on the recruitment of PCM proteins to the centrosome , a process that is controlled by the concentration and availability of PCM proteins and their phosphorylation by mitotic kinases ( Decker et al . , 2011; Wueseke et al . , 2014; Wueseke et al . , 2016; Yang and Feldman , 2015 ) . Three main kinases are involved in the regulation of PCM activity: CDK-1/CDK1 , PLK-1/PLK1 and AIR-1/Aurora A ( Pintard and Archambault , 2018 ) . CDK-1 acts as the main driver of PLK-1 and AIR-1 activity , which likely directly phosphorylate PCM proteins to promote PCM assembly ( Woodruff et al . , 2014 ) . During mitotic exit , MTOC activity of the centrosome rapidly decreases , marked by the reduction of the PCM and microtubule association . This cycle of centrosomal MTOC activity continues every cell cycle , but can also be naturally discontinued during cell differentiation when MTOC function is often reassigned to non-centrosomal sites ( Sanchez and Feldman , 2017 ) . Although the mechanisms controlling PCM disassembly have been relatively unexplored , inhibition of CDK activity can drive precocious PCM disassembly and inhibition of the PP2A phosphatase LET-92 perturbs SPD-5 removal from the centrosome , suggesting that phosphatase activity could be more generally required for the inactivation of MTOC function at the centrosome ( Enos et al . , 2018; Yang and Feldman , 2015 ) . Additionally , stabilization of CDK1 activity has been shown to inhibit PCM disassembly and promote PCM maintenance ( Rusan and Wadsworth , 2005 ) . Although kinase and phosphatase activity are implicated in this MTOC cycle , an understanding of how and when these factors act to inactivate MTOC function at the centrosome and whether all PCM proteins behave in the same manner during disassembly in vivo is currently lacking . The inactivation of MTOC activity of the centrosome is likely critical in a number of cellular and developmental contexts . For example , asymmetric cell division is often associated with unequal PCM association at the mother vs . daughter centrosome and terminal differentiation of murine cardiomyocytes and keratinocytes has been linked to centrosome inactivation ( Cheng et al . , 2011; Conduit and Raff , 2010; Muroyama et al . , 2016; Zebrowski et al . , 2015 ) . In an extreme example , female gametes in a range of organisms completely eliminate centrosomes and this elimination can be a critical step in gametogenesis ( Borrego-Pinto et al . , 2016; Lu and Roy , 2014; Luksza et al . , 2013; Mikeladze-Dvali et al . , 2012; Pimenta-Marques et al . , 2016 ) . Moreover , hyperactive MTOC function at the centrosome has been linked to several types of epithelial cancers and invasive cell behavior , and is a hallmark of tumors ( Godinho and Pellman , 2014; Lingle et al . , 1998; Pihan , 2013; Pihan et al . , 2001; Salisbury et al . , 1999 ) . Despite the clear importance of properly regulating MTOC activity , little is known about the mechanisms that inactivate MTOC function at the centrosome , either what initiates the removal of PCM and microtubules during the cell cycle or what keeps them off the centrosome in differentiated cells . To better understand how MTOC activity is regulated at the centrosome , here we investigate the localization and dynamics of endogenously tagged PCM proteins in the C . elegans embryo . We find that C . elegans PCM is composed of overlapping spheres of proteins similar to what has been observed in other systems , with SPD-5 and γ-TuRC occupying distinct regions from known binding partners SPD-2 and MZT-1 , respectively . Live imaging of PCM components at the end of mitosis revealed two phases of disassembly , beginning with the gradual dissolution of PCM proteins such as PLK-1 , SPD-2 , TAC-1 , and MZT-1 , followed by the rupture of the remaining PCM proteins ZYG-9 , SPD-5 , γ-TuRC , TPXL-1 , and AIR-1 into microtubule associated packets . Using pharmacological and genetic perturbations , we found a role for phosphatases in PCM disassembly throughout mitosis , opposing CDK activity during PCM assembly and catalyzing PCM dissolution once CDK activity naturally dissipated . Cell fusion and RNAi experiments indicated that the nature of the remaining PCM was transformed and mechanically cleared from the centrosome by cortical pulling forces . Delay in PCM removal impacted subsequent centriole separation and PCM maturation in the next cell cycle . These data indicate that the inactivation of MTOC function at the centrosome involves a regulated two-step process of PCM disassembly , the timing of which is critical to the developing embryo .
In order to better understand how PCM proteins behave during disassembly , we first characterized the spatial organization of the PCM during mitosis in the ABp cell of the 4-cell C . elegans embryo . ABp has relatively large centrosomes oriented during mitosis along the left-right axis of the embryo , with one of the centrosomes positioned very close to the coverslip in an end-on orientation ( Figure 1A ) . We analyzed the localization of endogenously-tagged PCM proteins immediately after nuclear envelope breakdown ( NEBD ) in the ABp cell ( Figure 1A ) . At this time , the centrosome still functions as an MTOC , actively growing and organizing microtubules ( Figure 1A ) . We assessed the localization of the centriole component SAS-4 , the PCM scaffolding proteins SPD-2 and SPD-5 , the γ-TuRC components GIP-1 and MZT-1 , the mitotic kinases AIR-1 and PLK-1 , and the microtubule associated proteins ZYG-9 , TAC-1 and TPXL-1 ( Figure 1B , Figure 1—figure supplement 1 ) . As expected , the centrioles sit at the center of the centrosome ( Figure 1B–C , Figure 1—figure supplement 1B and D ) surrounded by PCM proteins which formed ordered layers of protein localization . SPD-2 and SPD-5 localization at the PCM is co-dependent ( Hamill et al . , 2002; Kemp et al . , 2004; Pelletier et al . , 2004 ) , however these proteins displayed distinct outer localization boundaries within the PCM; both SPD-2 and SPD-5 localized to a more proximal region surrounding the centrioles ( distance from center at half maximum intensity for SPD-2: −0 . 575–0 . 575 ± 0 . 02 µm; 77 . 8 ± 0 . 8% of total SPD-5 overlapping with SPD-2 in this region ) , and SPD-5 extended to a more distal region lacking SPD-2 ( distance from center at half maximum intensity for SPD-5: −0 . 83–0 . 83 ± 0 . 03 µm; Figure 1B–C ) . Based on the outer edge of these two matrix proteins , we divide the PCM into an ‘inner’ and ‘outer’ sphere , with the smaller inner sphere defined by the outer edge of SPD-2 localization and the larger outer sphere defined by the outer edge of SPD-5 localization ( Figure 1D ) . PLK-1 showed the most restricted localization , occupying a relatively proximal localization in the inner sphere ( Figure 1B , C , E ) . GIP-1 localization was indistinguishable from SPD-5 , extending into the outer sphere ( Figure 1B–C , Figure 1—figure supplement 1B and D ) , however another γ-TuRC component MZT-1 showed an intermediary localization , extending only partially into the outer sphere ( Figure 1B–C and Figure 1—figure supplement 1B , D ) . TAC-1 and ZYG-9 also shared this intermediate localization pattern ( Figure 1C , Figure 1—figure supplement 1B–D ) . Finally , the localization of AIR-1 was mainly restricted to the outer sphere , forming a toroid as previously reported ( Hannak et al . , 2001 ) and a complimentary localization pattern to PLK-1 . As expected , TPXL-1 and AIR-1 co-localized , consistent with the fact that TPXL-1 is important for AIR-1 recruitment to the centrosome and microtubules ( Toya et al . , 2011 ) . Both TPXL-1 and AIR-1 localization extended further than the boundary of SPD-5 and GIP-1 ( Figure 1—figure supplement 1B–D ) . Based on these observations , we conclude that the PCM has a layered structure with an inner sphere delimited by SPD-2 localization ( Figure 1D ) that also localizes SPD-5 , PLK-1 , ZYG-9 , TAC-1 , γ-TuRC components , AIR-1 , and TPXL-1 , and an outer sphere delimited by SPD-5 localization that also contains ZYG-9 , TAC-1 , GIP-1 , MZT-1 , AIR-`1 , and TPXL-1 ( Figure 1D ) . Although our imaging approach did not allow us to resolve toroidal localization patterns of the majority of the PCM proteins we analyzed , the boundaries of PCM protein localization follows the general pattern of the predicted orthologs in Drosophila and human cells ( Fu and Glover , 2012; Lawo et al . , 2012; Mennella et al . , 2012; Mennella et al . , 2014 ) . This localization pattern is also noteworthy as SPD-5 and GIP-1 are found in a region lacking known binding partners SPD-2 and MZT-1 , respectively . Based on their distinct localization within the PCM , we hypothesized that different PCM proteins would disassemble with different kinetics and behaviors . To test this hypothesis , we examined the dynamics of disassembly of each of the endogenously-tagged PCM proteins described above by live-imaging in the ABp cell beginning at NEBD ( Figure 1E and Figure 1—figure supplement 2A ) . SPD-2 ( Video 1 ) , MZT-1 ( Video 2 ) , TAC-1 , and PLK-1 displayed similar disassembly behaviors , leaving the centrosome by gradual ‘dissolution’ over time and eventually only remaining at the two centrioles that will duplicate and mature into new centrosomes ( Figure 1E and Figure 1—figure supplement 2A ) . In contrast , SPD-5 ( Video 3 ) , GIP-1 ( Video 4 ) , ZYG-9 , AIR-1 , and TPXL-1 initially showed some gradual disassembly , however the structure containing these proteins then appeared to ‘rupture’ and fragment into ‘packets’ that were distinct from the centrioles ( Figure 2A–C ) . These sub-PCM packets localized SPD-5 , GIP-1 ( Figure 2A , early packets ) , microtubules ( Figure 2B , early packets ) , AIR-1 ( Figure 2C , early packets ) , and TPXL-1 , but neither SPD-2 nor MZT-1 ( Figure 2E , see below ) . Intriguingly , packets appeared to retain MTOC potential as EBP-2/EB1 comets , a marker of growing microtubule plus ends , dynamically moved from the SPD-5/GIP-1 foci ( Figure 2D ) . The packets appeared to be further disassembled in the cytoplasm following their removal from the PCM , with GIP-1 and microtubules first losing their association , followed by SPD-5 ( Figure 2A–C , late packets , Figure 2F ) . To gain a better sense of the timing of the disassembly of the different PCM proteins , we imaged each protein in combination with SPD-5 . SPD-2 ( Figure 3A ) and MZT-1 ( Figure 3B ) showed a gradual decrease in intensity , beginning at 2 ( 2 . 20 ± 0 . 13 min , n = 10 ) or 3 min ( 3 . 00 ± 0 . 27 min , n = 8 ) post-NEBD , respectively , and a decrease in PCM volume beginning 3 min post-NEBD ( SPD-2: 3 . 00 ± 0 . 21 min , n = 10; MZT-1: 2 . 88 ± 0 . 23 min , n = 8 ) . These changes occurred several minutes before the decrease in either SPD-5 or GIP-1 ( Figure 3D–E ) . PLK-1 and TAC-1 showed a similar disassembly behavior as SPD-2 and MZT-1 , with a gradual decrease in intensity beginning at 2 min post-NEBD ( n = 7 , Figure 3—figure supplement 1A ) . As expected from our observation of their individual localization behaviors , both SPD-5 and GIP-1 colocalized during the process of disassembly ( Figure 3C , Video 5 ) . Both proteins began a rapid decrease in intensity following their peak at 3 min post-NEBD ( SPD-5: 3 . 00 ± 0 . 14 min , n = 11; GIP-1: 3 . 18 ± 0 . 12 min , n = 11; Figure 3D–E ) . Their volume , however , remained unchanged until 6 min post-NEBD ( SPD-5: 5 . 91 ± 0 . 17 min , n = 11; GIP-1: 6 . 00 ± 0 . 19 min , n = 11 ) , at which time we began to see changes in the structural integrity of the PCM as holes appeared . A qualitative assessment of when these holes began to appear tracked perfectly with the quantitative changes we observed in SPD-5 and GIP-1 PCM volume . We refer to the appearance of these holes and the concomitant change in PCM volume as ‘rupture’ . Following rupture , SPD-5 and GIP-1 deformation continued until distinct sub-PCM ‘packets’ could be observed individualized from the two future centrosomes ( Figure 3A ) . Intriguingly , both AIR-1 and TPXL-1 appeared to spread onto the microtubules starting 3 min post-NEBD ( Figure 2C , Figure 3—figure supplement 1B ) , 3 min ahead of SPD-5 and GIP-1 rupture , before also localizing in packets . Together , these data indicate that the PCM disassembles in two distinct steps: a dissolution step that is characterized by the decrease in intensity of PCM proteins that starts with the removal of the most internal proteins , PLK-1 , SPD-2 , TAC-1 and MZT-1; and a rupture step where the deformation and subsequent separation of the PCM leads to further disassembly into individual packets . The formation of packets that appear to be pulled away from the centrioles suggests that mechanical forces underlie this aspect of PCM disassembly . Forces can be exerted on the PCM by a conserved cortically anchored complex of LIN-5/NuMA , ( GPR-1/2 ) /LGN , and ( GOA-1/GPA-16 ) /Gαi , which localizes dynein-dynactin and can pull on the astral microtubules extending from the PCM ( Kotak and Gönczy , 2013 ) . Given that greater cortical forces exist in the posterior of the one-cell C . elegans embryo , it has been hypothesized that these forces could be responsible for the asynchrony observed in the disassembly of the anterior vs . the posterior centrosome ( Grill et al . , 2001 ) . Moreover , a recent study implicated the ( GPR-1/2 ) /LIN-5/DHC-1 complex in SPD-5 disassembly from the PCM ( Enos et al . , 2018 ) . To assess the involvement of cortical forces in the general disassembly of the PCM and specifically in rupture and packet formation , we used RNAi to either decrease ( gpr-1/2 ( RNAi ) ) or increase ( csnk-1 ( RNAi ) ) cortical forces . In control embryos treated with lacZ RNAi , SPD-5 ruptured starting 6 min post-NEBD ( 5 . 91 ± 0 . 16 min , n = 11; Figure 4A ) . In contrast , we did not observe SPD-5 rupture or packet formation in gpr-1/2 ( RNAi ) treated embryos ( Figure 4A ) . Instead , SPD-5 , like SPD-2 , disassembled by gradual dissolution as indicated by the steady decrease in SPD-5 centrosomal volume which was in sharp contrast to the precipitous drop off seen in control embryos ( Figure 4B ) . In csnk-1 ( RNAi ) treated embryos , we observed slightly earlier SPD-5 rupture ( 5 . 4 + 0 . 2 min , n = 7; Figure 4A ) . In contrast to SPD-5 , SPD-2 disassembly was unaffected following depletion of either gpr-1/2 or csnk-1 by RNAi ( Figure 4C ) . Interestingly , SPD-5 levels at the PCM were increased by gpr-1/2 and decreased by csnk-1 depletion ( Figure 4A , Figure 4—figure supplement 1A ) . Together , these results suggest that cortical forces generate the mechanical forces necessary for rupture and packet formation , allowing for the efficient removal of the outer sphere protein SPD-5 but not the exclusively inner sphere protein SPD-2 . Cortical forces could be present and constant throughout mitosis or instead intensify at the time of disassembly as is the case in the zygote , providing forces only when necessary ( Gönczy , 2005; Rose and Gonczy , 2014 ) . To distinguish between these possibilities , we tracked the localization of microtubules , LIN-5 , DNC-1/dynactin and DHC-1/dynein heavy chain , during different stages of mitosis . Astral microtubules showed a striking network reorganization post-NEBD , growing progressively longer and contacting the cell cortex , sometimes wrapping around the membrane prior to rupture and packet formation ( Figure 4—figure supplement 2A ) . We saw a similar reorganization of AIR-1 and TPXL-1 , which coat these astral microtubules ( Figure 4—figure supplement 1B ) . In contrast , we saw no change in the gross cortical distribution or intensity of LIN-5 ( Figure 4—figure supplement 2B ) , DNC-1 ( Figure 4—figure supplement 2C ) , or DHC-1 ( Figure 4—figure supplement 2D ) post-NEBD . Interestingly , we observed an ephemeral redistribution of DHC-1 coincident with rupture ( Figure 4—figure supplements 2E and 4 min . ) . This pattern of localization suggests that although cortical complexes are present throughout the cell cycle , they may only make productive contact with astral microtubules at a particular time period to allow for outer sphere disassembly . The rapid rounds of PCM assembly and disassembly during the early embryonic divisions suggest that efficient and robust PCM disassembly might be critical for subsequent carefully timed events such as centriole separation and the assembly of new PCM in the next cell cycle ( Cabral et al . , 2013 ) . We tested whether force dependent PCM removal corresponds to centriolar separation by tracking SAS-4::GFP during disassembly ( Figure 4D ) . In control embryos , the centriolar pair appeared as a single SAS-4 focus up to 5 min post-NEBD ( Figure 4D ) . Two closely apposed SAS-4 foci became apparent beginning at 5 min post-NEBD ( Stage 1 , Figure 4D ) , which quickly separated by greater than 1 µm beginning about 1 min later ( Stage 2 , Figure 4D ) . We saw a significant delay in the onsets of both Stage one and Stage two in gpr-1/2 ( RNAi ) treated embryos , but no significant change in csnk-1 ( RNAi ) treated embryos ( Figure 4D ) . These results suggest that cortical forces facilitate centriole separation either through direct force transmission or indirectly through their role in PCM removal . That csnk-1 RNAi had no effect on the timing of centriole separation suggests that a force-independent licensing event is necessary to initiate separation ( Cabral et al . , 2013; Tsou and Stearns , 2006 ) , but that centrioles are subsequently held together by PCM . In addition to defects in centriole separation , we observed that gpr-1/2 ( RNAi ) treated embryos had defects in effectively clearing SPD-5 , but not SPD-2 , from the PCM prior to the subsequent round of PCM accumulation in the next cell cycle ( Figure 4B and E ) . Consistent with these defects , the timing of subsequent SPD-5 accumulation was significantly delayed as compared to control embryos ( Figure 3—figure supplement 1C ) . Together , these results underscore the importance of the timely removal of PCM to the developing embryo . As the growth of the PCM is highly dependent on phosphorylation and CDK inhibition causes precocious removal of PCM proteins ( Woodruff et al . , 2014; Yang and Feldman , 2015 ) , we hypothesized that the dissolution of the PCM that precedes rupture and packet formation requires phosphatase activity . To test this hypothesis , we treated cycling embryonic cells at anaphase with either a broad-spectrum serine/threonine phosphatase inhibitor ( okadaic acid , OA ) or a PP2A inhibitor ( rubratoxin A , Figure 5A ) . We observed a stabilization of the PCM in both OA and rubratoxin A treated embryos compared to control embryos treated with DMSO . Notably , treatment with either drug led to depolymerization of the microtubules , perhaps due to the hyperactivation of the depolymerizing kinesin KLP-7 during PP2A inactivation ( Schlaitz et al . , 2007 ) . Consistent with these pharmacological inhibition results , a recent study implicated the PP2A subunit LET-92 in SPD-5 disassembly ( Enos et al . , 2018 ) . To assess the function of LET-92 on PCM disassembly in general and more specifically on dissolution and packet formation , we treated SPD-2::GFP; tagRFP::SPD-5 expressing embryos with let-92 ( RNAi ) . As previously reported , let-92 inhibition caused severe defects in cell division , necessitating analysis in the one-cell zygote rather than 4-cell embryo ( Song et al . , 2011 ) . We monitored PCM disassembly in the one-cell zygote beginning when the membrane invagination that occurs during cytokinetic furrow formation was visible . At this stage in control embryos , PCM disassembly occurs in a similar manner to ABp cells , with SPD-2 dissolution preceding SPD-5 rupture and packet formation ( Figure 5B ) . let-92 depletion impaired the disassembly of SPD-2 and SPD-5 from the centrosome in three distinct ways ( Figure 5C ) . First , we never saw holes appearing in centrosomal SPD-5 , indicating a defect in rupture . Moreover , SPD-5 was only partially cleared into packets , however these packets were more fluid and persisted significantly longer in the cytoplasm than control . Unlike in the 4-cell embryo , we occasionally saw a small fraction of SPD-2 being cleared from the centrosome by rupture in this first cell division , a short-lived phenomenon that was exacerbated when both SPD-2 and SPD-5 were endogenously tagged . In contrast , following let-92 depletion , SPD-2 consistently ruptured and appeared in packets that persisted in the cytoplasm long after those of control embryos . Second , the rate and time of SPD-2 and SPD-5 disassembly were significantly slower in let-92 depleted embryos than in control , as indicated by tracking the total centrosomal SPD-2 and SPD-5 over time ( Figure 5E , F ) . Centriole duplication fails following let-92 depletion such that each centrosome at this stage contains only one rather than two centrioles ( Song et al . , 2011 ) . Thus , total centrosome intensity measurements underestimate differences between control and let-92 depletion conditions because centriole number defects alter the underlying amounts of centriole-localized SPD-2 or SPD-5 . Finally , we found that although much of the SPD-2 and SPD-5 appeared to be cleared from the PCM into packets , let-92 depletion inhibited the complete removal of either protein from the centrosome ( Figure 5C , G ) . The partial removal of SPD-2 and SPD-5 in packets suggested that let-92 depletion affected mainly dissolution , and that much , but not all , of the remaining PCM was cleared by cortical forces . To test this model , we inhibited let-92 together with gpr-1/2 and observed a strong stabilization of both SPD-2 ( Figure 5D , E ) and SPD-5 ( Figure 5D , F ) at the PCM without rupture or packet formation . Together , these results indicate that PP2A phosphatases control the dissolution of SPD-2 and SPD-5 , and that both PP2A and cortical forces are required for the efficient and timely removal of the PCM from the centrosome . The timing of centrosome disassembly is critical . For example , precocious removal of PCM would impair the ability of the centrosome to build the mitotic spindle , and delayed disassembly affects the subsequent centrosome duplication cycle ( see above , Figure 4 ) . While cortical pulling forces appear to act on the centrosome post-NEBD ( Figure 4—figure supplement 2A ) , it is unclear when phosphatases such as LET-92 are active to help drive disassembly . Phosphatases could be active at the centrosome throughout the cell cycle or could instead be activated only at the time of disassembly . To distinguish between these possibilities , we first assessed the localization of endogenously-tagged LET-92 throughout mitosis . LET-92 localized to the centrosome through the entire process of assembly and disassembly ( Figure 6A ) , extending into the outer sphere in a similar localization pattern to SPD-5 and GIP-1 ( Figure 1—figure supplement 1D ) . Similarly , LET-92 displayed disassembly behavior and kinetics similar to that of SPD-5 and GIP-1 ( Figure 6A ) , however the low expression of LET-92 made it difficult to reliably determine whether it localized to packets . These localization data raised the possibility that phosphatases are active at the centrosome throughout mitosis rather than just during the disassembly phase . To test this possibility further , we treated cycling SPD-2::GFP; tagRFP::SPD-5 embryos with OA and/or the CDK inhibitor flavopiridol ( FP ) at three defined timepoints during mitosis ( Figure 6B ) : ( 1 ) just prior to PCM growth ( ‘pre-growth’ ) ; ( 2 ) during PCM growth ( ‘growth’ ) ; and ( 3 ) at metaphase immediately before NEBD when PCM levels at the centrosome are near their peak and just prior to the initiation of disassembly ( ‘metaphase’ ) . FP treatment in pre-growth cells inhibited the accumulation of SPD-2 and SPD-5 at the centrosome and growth-stage FP treatment induced their precocious disassembly . Metaphase stage FP treatment had relatively little effect on SPD-2 and SPD-5 centrosomal localization , consistent with the fact that CDK is normally inactivated shortly after metaphase ( Kipreos and van den Heuvel , 2019 ) . SPD-2 and SPD-5 both accumulated at centrosomes following pre-growth OA treatment , albeit to a lesser extent than in control cells . As predicted by the let-92 RNAi phenotype ( Figure 5B–D ) , growth and metaphase stage PCM was stabilized by OA treatment as indicated by the continued presence of SPD-5 ( Figure 6B ) or GIP-1 ( Figure 5A ) at the centrosome during the disassembly period . Surprisingly and in contrast to SPD-5 , SPD-2 was precociously disassembled in the presence of OA in growth and metaphase stage embryos . The precocious disassembly of PCM following CDK inhibition suggested that the association of PCM proteins with the centrosome is normally actively opposed such that turning off assembly immediately triggers disassembly . To test if this opposition is phosphatase dependent , we treated embryos with both FP and OA . Pre-growth stage treated embryos showed no addition of SPD-2 or SPD-5 , consistent with a requirement for CDK activity in centrosome maturation . Treatment with both inhibitors at growth or metaphase stage led to a stabilization of SPD-5 , consistent with the hypothesis that SPD-5 assembly driven by CDK activity is normally opposed by phosphatase activity . In contrast , SPD-2 was precociously disassembled in the presence of OA and FP in growth and metaphase stage embryos , identically to what was observed in embryos treated with OA alone . These results suggest that the maintenance of SPD-2 at the centrosome is controlled by an OA sensitive phosphatase and indicate that regulation of SPD-2 and SPD-5 can be uncoupled . The differential behavior of PCM proteins in response to the timing of kinase and phosphatase inhibition suggests that assembling and disassembling PCM are inherently different structures . We therefore wanted to test whether the factors that contribute to PCM assembly had any impact on disassembling PCM or vice versa . Using in vivo cell fusion experiments , we previously found that cytoplasm from pre-anaphase cells could rapidly induce the assembly of PCM and microtubules onto inactive centrosomes in both cycling and differentiated cells , indicating that mitotic cytoplasm dominantly selects for PCM assembly ( Yang and Feldman , 2015 ) . Using a similar cell fusion approach , we fused a pre-metaphase cell in which the PCM was assembling ( P2 , Figure 6C ) and a post-anaphase cell in which the PCM was disassembling ( ABp , Figure 6C ) to examine the relationship between assembling and disassembling PCM and the cytoplasmic environments that maintain them . If PCM assembly is dominant , we would expect the PCM in ABp to be stabilized following cell fusion . Conversely , if PCM disassembly is dominant , we would expect cell fusion to induce disassembly of the P2 centrosome . Finally , the process of assembly and disassembly could be mutually resistant to the factors that induce the converse process , that is cell fusion would have no impact on the assembling P2 centrosome or the disassembling ABp centrosome . We fused ABp ( Figure 6D , blue , n = 13 ) with P2 ( Figure 6D , magenta ) 2 min after NEBD in the ABp cell in embryos expressing a membrane localized mCherry and endogenously tagged tagRFP::SPD-5 and GFP::GIP-1 . Microtubules associated with the disassembling ABp centrosome invaded P2 following fusion , confirming an exchange between the cytoplasm of the two cells ( Figure 6—figure supplement 1A ) . Following fusion , the ABp centrosome ( Figure 6D , blue arrows ) exhibited normal disassembly , showing packet formation as in the control unfused ABa cell ( Figures 6D and 5 min , white arrowheads ) . Similarly , the P2 centrosome showed normal assembly following cell fusion ( Figure 6D , magenta arrows ) . Interestingly , as soon as the existing PCM was stripped from the ABp centrosome into packets , new PCM rapidly assembled at the ABp centrosome ( Figures 6D and 7 min , blue double arrow ) in a similar manner to that of P2 ( Figures 6D and 7 min , pink double arrow ) . The addition of PCM in ABp was precocious as the control ABa cell had not yet started adding PCM to its centrosome . This precocious assembly also induced the assembly of microtubules but did not lead to the clustering of the ABp and P2 centrosome ( Figure 6—figure supplement 1A ) . Together , these results indicate that PCM assembly and disassembly are mutually resistant with each state being locked in place; a disassembling centrosome and PCM packets are unaffected by cytoplasm that normally promotes assembly and an assembling centrosome is unaffected by cytoplasm that promotes disassembly . Moreover , the addition of new ‘assembly state’ PCM occurs once the old ‘disassembly state’ PCM is removed . Previous experiments indicated that fusion induced PCM assembly requires CDK activity ( Yang and Feldman , 2015 ) , lending further evidence to the idea that the nature of the PCM changes throughout mitosis and becomes resistant to phosphoregulation .
Here we present evidence that MTOC function at the centrosome is inactivated through a two-step PCM disassembly process involving the gradual dissolution of proteins localized close to the centrioles followed by the forceful rupture and ejection of proteins that extend more distally . Our data suggest that PCM dissolution is controlled by phosphatase activity , including that of PP2A , and that cortical forces drive the rupture of remaining PCM , pulling it into packets ( Figure 7 ) . While previous studies indicated a role for both LET-92 and cortical forces in the disassembly of SPD-5 ( Enos et al . , 2018 ) , here we have presented a more complete picture of centrosome disassembly in two steps as discussed below . Furthermore , our data indicate that PCM is not a mass of proteins that is assembled and disassembled as a batch by common regulation , but rather a complicated meshwork of proteins with distinct mechanisms of intricate regulation . Our two-step model for PCM disassembly is predicated on the duality of localization patterns and disassembly behaviors we discovered for different PCM proteins . In particular , we found that the C . elegans centrosome is organized into discrete layers which we propose to be part of two spheres based on the localization boundaries of the matrix proteins SPD-2 and SPD-5 . While our analysis of PCM composition is limited by our choice of diffraction-limited imaging platforms , this layered organization appears to be generally conserved between direct and functional orthologs in C . elegans , Drosophila , and human PCM , suggesting evolutionary pressure to create specific functional PCM domains and that the mechanisms of disassembly described here might be generally conserved . We found that known binding partners separate between these two distinct PCM regions . For example , SPD-5 and GIP-1 localize to the outer sphere region which lacks binding partner SPD-2 and MZT-1 , respectively . Similarly , we found that both SPD-2 and MZT-1 normally disassemble from the centrosome before either SPD-5 or GIP-1 . Furthermore , the precocious removal of SPD-2 by OA treatment did not affect SPD-5 or GIP-1 localization , suggesting that these proteins have the ability to form a matrix in the absence of SPD-2 and MZT-1 . SPD-5 can form a matrix in vitro and perhaps its self-association drives outer sphere assembly and maintains PCM structure in the absence of SPD-2 ( Woodruff et al . , 2015 ) . Finally , the differential localization patterns of PCM proteins correlate with the two different disassembly modes we observed ( dissolution vs . rupture ) : Proximal proteins ( PLK-1 , SPD-2 , MZT-1 , TAC-1 ) disassembled by gradual dissolution while more distally extending proteins ( ZYG-9 , GIP-1 , SPD-5 , AIR-1 ) ruptured and formed packets . These differences in disassembly behaviors might reflect differences in diffusion of individual components as SPD-2 and PLK-1 are known to have increased mobility within the PCM as compared to SPD-5 and GIP-1 ( Laos et al . , 2015; Woodruff et al . , 2017 ) . Thus , removal of more fluid inner sphere proteins could rely on active turnover while disassembly of more stable outer sphere proteins might require physical disruption . Our data suggest that PCM disassembly is initiated through the active turnover of inner sphere proteins by dephosphorylation , either through the direct action of phosphatases on these proteins or through their inactivation of mitotic kinases . Indeed , the removal of both SPD-2 and MZT-1 appears to exclusively depend on phosphatase activity as they do not localize in packets and their disassembly was not affected by the inhibition of cortical forces . Furthermore , a pool of both SPD-2 and SPD-5 remained at the centrosome following LET-92 depletion , suggesting that cortical forces alone are not sufficient for their effective clearance . Thus , PCM disassembly appears to be initiated by dephosphorylation by the PP2A subunit LET-92 . As LET-92 plays a number of roles at the centrosome and phosphatase activity can directly regulate mitotic kinases ( Enos et al . , 2018; Kitagawa et al . , 2011; Song et al . , 2011 ) , further studies will be necessary to determine if its role in PCM dissolution is direct or indirect . Our inhibitor experiments also uncovered key roles for phosphatases in both centrosome assembly and disassembly . Centrosome assembly appears to be the result of a balance of kinase and phosphatase activity acting on PCM proteins . Both TBG-1 and SPD-5 could be prematurely forced from the centrosome by dissolution in the presence of the CDK inhibitor FP ( this study and Yang and Feldman , 2015 ) . This precocious dissolution was inhibited by additional treatment with OA , suggesting that the ability of CDK to drive the addition of PCM proteins is actively opposed by serine/threonine phosphatase activity . This phosphatase-based opposition is independent of inhibition of CDK activity and might instead act on other kinases such as PLK-1/PLK1 or AIR-1/Aurora A . Consistently , PP2A can remove activating phosphates from both PLK1 and Aurora A ( Horn et al . , 2007; Wang et al . , 2015 ) . Alternatively , our data are also consistent with phosphatases being directly inhibited by CDK , thus forced CDK inactivation would relieve phosphatase inhibition and result in dissolution . However , we favor a model in which phosphatases actively oppose PCM assembly as this type of model can account for the observed mobility of exclusively inner sphere proteins such as SPD-2 ( Laos et al . , 2015 ) . Indeed , phosphatases might directly remove PCM protein phosphorylation which in turn could lead to their dissociation from the centrosome . This model seems plausible as SPD-5 can be dephosphorylated in vitro by LET-92 and has been shown to interact with the PP2A targeting subunits RSA-1 and RSA-2 ( Enos et al . , 2018; Schlaitz et al . , 2007 ) . As LET-92 is localized to the centrosome throughout mitosis , PCM dissolution might simply be the result of the normal inhibition of CDK activity in the cell cycle coupled with the continued presence of LET-92 at the centrosome . These experiments have also revealed differential regulation for SPD-2 and SPD-5 . In contrast to LET-92 inhibition which stabilized both SPD-2 and SPD-5 at the centrosome , we found that OA treatment inhibited the removal of SPD-5 from the centrosome but expedited SPD-2 disassembly . These experiments suggest that SPD-2 association with the centrosome is positively regulated by another OA sensitive phosphatase , further separating the localization and regulation of SPD-2 from that of SPD-5 . OA induced SPD-2 removal only occurred after NEBD , suggesting that the phosphatase that maintains it at the centrosome is regulated in time and/or space . PP1 and/or PP4 could play this role as both have been shown to positively regulate PCM association with the centrosome , and PP1 can interact directly with the SPD-2 homologue CEP192 ( Martin-Granados et al . , 2008; Nasa et al . , 2017 ) . Thus , PCM disassembly and assembly are regulated by phosphatases and SPD-2 appears to have additional levels of dephosphorylation dependent mechanisms to maintain it the centrosome . In the future , it will be interesting to determine if other inner sphere proteins have similar regulation . The sensitivity of PCM to kinase and phosphatase activity appears to change during mitosis . Indeed , disassembling PCM appears to be resistant to assembly-competent cytoplasm; PCM continued to disassemble into packets despite exposure to active mitotic kinases following cell fusion . Moreover , the aging PCM matrix appeared to protect centrosomes from the addition of new PCM , which was only added onto centrioles once existing PCM had been stripped away . Likewise , assembling PCM was unaffected by the presence of disassembly-competent cytoplasm following fusion , further underscoring that disassembly by phosphatases is a normal aspect of assembly that is dominated by kinase activity . An alternative explanation for these observed phenomena is that PCM assembly is slower than the off rate of PCM proteins or that diffusion between ABp and P2 is too slow to induce assembly prior to disassembly . However , we previously found that new PCM can add onto inactive centrosomes in interphase cells in under three minutes following fusion to a mitotic cell ( Yang and Feldman , 2015 ) . Given that in the present study we observe disassembly and subsequent reassembly in ABp well after this three-minute window , our results instead favor the model that centrosomes become locked in mutually resistant assembly or disassembly states perhaps due to a change in the biophysical nature of the PCM . Recent studies of in vitro assembled PCM point to different physical properties between ‘young’ and ‘old’ condensates of SPD-5 , with young condensates behaving more like a liquid and old condensates acting more like a gel ( Woodruff et al . , 2017 ) . This change in the nature of the PCM could also be regulated by phosphorylation . For example , Cnn is proposed to live in different states in the PCM in Drosophila , assembling first near the centrioles in a phosphorylated state and transiting towards the PCM periphery as a higher order multimerized scaffold where Cnn molecules are likely eventually dephosphorylated and lose PCM association ( Conduit et al . , 2014 ) . Similarly , the inner sphere of SPD-5 may represent a specific pool that can be readily dissociated by dephosphorylation , while the outer sphere may represent a macromolecular scaffold that relies on physical disruption for disassembly . More mobile inner sphere proteins such as SPD-2 could be more likely to escape an aging outer sphere matrix of SPD-5 and other proteins which would then be torn apart by cortical forces as it matured into a gel . Indeed , let-92 depletion inhibited rupture and led to the appearance of more fluid packets , consistent with a role for phosphorylation in regulating the nature of the PCM . Interestingly , a recent study reported similar defects upon depletion of PCMD-1 ( Erpf et al . , 2019 ) . Thus , PCMD-1 could be involved in this phosphorylation dependent regulation of PCM structural integrity . Different landscapes of phosphorylation could be provided by the complementary localization patterns of the two mitotic kinases PLK-1 and AIR-1 . Although previous studies had suggested that only more proximally localized AIR-1 is activated by auto-phosphorylation ( Hannak et al . , 2001; Toya et al . , 2011 ) , more recent studies have found that human Aurora A can also be activated by interaction with TPX2 ( Zorba et al . , 2014 ) . These results suggest that AIR-1 might be similarly active throughout the outer sphere where it colocalizes with TPXL-1 and therefore could phosphorylate a complementary set of PCM proteins to that of PLK-1 . Following dissolution , we found that the PCM fragments into small packets that retain MTOC potential . These packets are reminiscent of PCM flares described in Drosophila ( Megraw et al . , 2002 ) and to the fragments that are released from the centrosome in anaphase in the LLC-PK1 kidney cell line ( Rusan and Wadsworth , 2005 ) . PCM flares are reported to be present to some extent throughout the cell cycle rather than exclusively during centrosome disassembly like the packets we describe ( Lerit et al . , 2015; Megraw et al . , 2002 ) . However , like packets , flare activity dramatically increases in telophase and centrosome fragments in LLC-PK1 cells appear in anaphase . Flares were first defined by their association with Cnn , the proposed functional ortholog of SPD-5 ( Megraw et al . , 2002 ) . However , flares also localize D-TACC while the C . elegans orthologue TAC-1 does not localize to packets . Furthermore , γ-TuRC does not localize to flares but does localize to both packets and centrosome fragments . Finally , packets , flares , and centrosome fragments all appear to be dependent on microtubules for their formation . Thus , these remnants of PCM fragmentation appear to be conserved , although the molecular composition and timing of appearance of the resulting structures can vary . Because packets still disassemble following let-92 inhibition and exposure to assembly competent cytoplasm , other kinase- and phosphatase-independent mechanisms must be required for packet disassembly . These mechanisms could include rapid diffusion or proteasome-based degradation and further studies will be required to uncover their mechanism of disassembly . Finally , our results indicate that cortical forces can shape the PCM mainly through an effect on outer sphere proteins . The balance of cortical forces appears to tune the levels of SPD-5 incorporation into the PCM , independently of SPD-2; decreasing or increasing cortical forces caused more or less SPD-5 incorporation , respectively , but had no effect on the levels of SPD-2 . Thus , cortical forces negatively regulate the growth of the PCM , hypothetically by physically changing the nature of PCM in the outer sphere . We found that the effect of cortical forces on the PCM was temporally restricted , with the PCM only becoming sensitive to these forces in late anaphase . This claim is supported by multiple observations . First , when PCM is precociously removed from the centrosome by FP treatment , SPD-5 and γ-TuRC disassemble by dissolution , suggesting that cortical forces are not capable of rupturing the PCM and forming packets at the time of normal PCM growth ( this study and Yang and Feldman , 2015 ) . Second , we see astral microtubule rearrangements starting in anaphase that result in a large increase in the number of microtubules that reach the plasma membrane , consistent with what has been seen in other cell types ( Rusan and Wadsworth , 2005 ) . Thus , productive force can only act on the PCM beginning in anaphase . Consistently , increasing cortical forces by CSNK-1 inhibition only slightly expedited PCM disassembly . Finally , we see an apparent movement of AIR-1 and TPXL-1 from the PCM along the microtubules also beginning at about the end of anaphase . This redistribution could simply be a biproduct of the microtubule network reorganization . Alternatively , AIR-1 and TPXL-1 relocalization could contribute to microtubule reorganization by stabilizing the microtubules directly or promoting their efficient outgrowth ( Bayliss et al . , 2003; Zhang et al . , 2017 ) . In total , these results suggest that PCM is disassembled through the removal of the inner sphere of PCM by phosphatase activity , including that of PP2A . This dissolution is followed by the clearance of an aging outer sphere matrix by cortical pulling forces , which liberate dynamic microtubules and inactivate MTOC function at the centrosome ( Figure 7 ) . With an understanding of the mechanisms underlying this process , future studies will reveal whether hyperactive MTOC function at the centrosome has a direct effect on the cell cycle or cell differentiation in a developing organism , as has been previously postulated .
C . elegans strains were maintained at 20°C unless otherwise specified and cultured as previously described ( Brenner , 1974 ) . Experiments were performed using embryos from one-day adults . Unless otherwise indicated , at least five embryos were scored in each experimental condition . Strains used in this study are as follows . Strain nameGenotypeSourceN2Bristol N2CGCJLF14gip-1 ( wow3[gfp::gip-1] ) III ( Sallee et al . , 2018 ) JLF432spd-2 ( wow60[spd-2::gfp^3xflag] ) IThis studyJLF359spd-5 ( wow36[tagrfp-t^3xmyc::spd-5] ) IThis studyJLF361spd-5 ( wow52[gfp^3xflag::spd-5] ) IThis studyJLF342zif-1 ( gk117 ) ; mzt-1 ( wow51[gfp^3xflag::mzt-1] ) I ( Sallee et al . , 2018 ) JLF198Zif-1 ( gk117 ) ; sas-4 ( wow32[zf^gfp^3xflag::sas-4] ) IIIThis studyJLF50zif-1 ( gk117 ) , outcrossed 6x ( Sallee et al . , 2018 ) JLF427spd-5 ( wow36[tagrfp-t^3xmyc::spd-5] ) I; unc-119 ( ed3 ) ; ruIs57[pie-1p::GFP::tbb/β-tubulin; unc-119 ( + ) ]This study/CGCJLF428spd-5 ( wow36[tagrfp-t^3xmyc::spd-5] ) I; ebp-2 ( wow47[ebp-2:: gfp^3xflag] ) IIThis study/ ( Sallee et al . , 2018 ) JLF430spd-5 ( wow36[tagrfp-t^3xmyc::spd-5] ) I; gip-1 ( wow3[gfp^3xflag::gip-1] ) IIIThis study/ ( Sallee et al . , 2018 ) JLF426spd-5 ( wow36[tagrfp-t^3xmyc::spd-5] ) I; mzt-1 ( wow51[gfp^3xflag::mzt-1] ) IThis studyJLF425spd-5 ( wow36[tagrfp-t^3xmyc::spd-5] ) I; spd-2 ( wow60[spd-2:: gfp^3xflag] ) IThis studyJLF429zif-1 ( gk117 ) ; spd-5 ( wow36[tagrfp-t^3xmyc::spd-5] ) I; sas-4 ( wow32[zf^gfp^3xflag::sas-4] ) IIIThis studyLP585lin-5 ( cp288[lin-5::mNG-C1^3xFlag] ) IICGCLP560dhc-1 ( cp268[dhc-1::mNG-C1^3xFlag] ) ICGCLP563dnc-1 ( cp271[dnc-1::mNG-C1^3xFlag] ) ICGCOD2425plk-1 ( it17[plk-1::sgfp]loxp ) III ( Martino et al . , 2017 ) JLF158tac-1 ( wow19[tac::zf^gfp^3xflag] ) ; zif-1 ( gk117 ) This studyJLF105zyg-9 ( wow12[zf::gfp::zyg-9] ) II; zif-1 ( gk117 ) ( Sallee et al . , 2018 ) JLF518let-92 ( wow88[let-92::gfp^aid^3xFlag] ) IV/nT1This studyJLF216tpxl-1 ( wow34[zf^gfp^3xflag::tpxl-1 ) I; zif-1 ( gk117 ) ( Sallee et al . , 2018 ) JLF166itSi569 ( tbg-1::mcherry ) ; air-1 ( wow14[air-1::zf^gfp^3xflag] ) V; zif-1 ( gk117 ) ( Sallee et al . , 2018 ) JLF517gip-1 ( wow3[gfp^3xflag::gip-1] ) IIIGIP-1 , ; spd-5 ( wow36[tagrfp-t^3xmyc::spd-5] ) I; ltIs44 [pie-1p::mCherry:: PH ( PLC1delta1 ) +unc-119 ( + ) ] VThis study/ ( Sallee et al . , 2018 ) JLF8ruIs75 ( tubulin::gfp ) ; itIs37 [pie-1p::mCherry::H2B::pie-1 3'UTR + unc-119 ( + ) ] IV; ltIs44 [pie-1p::mCherry::PH ( PLC1delta1 ) +unc-119 ( + ) ] VThis study Endogenously tagged proteins used in this study were generated using the CRISPR Self Excising Cassette ( SEC ) method that has been previously described ( Dickinson et al . , 2015 ) . DNA mixtures ( sgRNA and Cas9 containing plasmid and repair template ) were injected into young adults , and CRISPR edited worms were selected by treatment with hygromycin followed by visual inspection for appropriate expression and localization ( Dickinson et al . , 2015 ) . sgRNA and homology arm sequences used to generate lines are as follows: AllelesgRNA sequenceHomology armSEC usedspd-2 ( wow60[spd-2::gfp^3xflag] ) cagagaatatttggaaagttagg ( pJM31 ) HA1 Fwd: ttgtaaaacgacggccagtcgccggcaGTGTTGACATTCGCATCGACpDD282HA1 Rev: CATCGATGCTCCTGAGGCTCCCGATGCTCCCTTTCTATTCGAAAATCTTGTATTGGHA2 Fwd: CGTGATTACAAGGATGACGATGACAAGAGATAAaatcttaagataactttccaaatattcHA2 Rev: ggaaacagctatgaccatgttatcgatttcatcctcaatatgccagatgcspd-5 ( wow36[tagrfp-t^3xmyc::spd-5] ) gaaaacttcgcgttaaATGGAGG ( pJM13 ) HA1 Fwd: cacgacgttgtaaaacgacggccagtcgacgcaaggaaatcgtcacttpDD286HA1 Rev: CTTGATGAGCTCCTCTCCCTTGGAGACCATttaacgcgaagttttctgHA2 Fwd: GAGCAGAAGTTGATCAGCGAGGAAGACTTGGAGGATAATTCTGTGCTCAACGHA2 Rev: tcacacaggaaacagctatgaccatgttatCTTTCCTCCATTGCATGCTTspd-5 ( wow52[gfp^3xflag::spd-5] ) HA1 Fwd: acgttgtaaaacgacggccagtcgccggcaacgcaaggaaatcgtcacttpDD282HA1 Rev: TCCAGTGAACAATTCTTCTCCTTTACTCATttaacgcgaagttttctgHA2 Fwd: CGTGATTACAAGGATGACGATGACAAGAGAGAGGATAATTCTGTGCTCAACGHA2 Rev: tcacacaggaaacagctatgaccatgttatCTTTCCTCCATTGCATGCTTtac-1 ( wow19[tac-1::zf^gfp^3xflag] ) cagagaatatttggaaagttagg ( pJF283 ) HA1 Fwd: ttgtaaaacgacggccagtcgccggcagctttctaggccaactgcacpJF250HA1 Rev: ACAAAGTCGCGTTTTGTATTCTGTCGGCATctgaaaatcggatgaatttaatagHA2 Fwd: CGTGATTACAAGGATGACGATGACAAGAGATCGCTCAACACAACCTTCACHA2 Rev: tcacacaggaaacagctatgaccatgttatACTCCACGGATGCTctgaatlet-92 ( wow88[let-92::gfp^aid^3xflag] ) GAAAACGGCGATTTGAACGGAGG ( pJM51 ) HA1 Fwd: ttgtaaaacgacggccagtcgccggcaCCTTCACGGAGGTCTTTCACpJW1583HA1 Rev: CATCGATGCTCCTGAGGCTCCCGATGCTCCCAGGAAGTAGTCAGGCGTTCTHA2 Fwd: CGTGATTACAAGGATGACGATGACAAGAGATAGatagatacctccgttcaaatcgHA2 Rev: ggaaacagctatgaccatgttatcgatttcgggaagtggtgaaaaggatgsas-4 ( wow32[zf::gfp^3xflag::sas-4] ) GGAAAACAACTTTGTTCCAG ( pJF296 ) HA1 Fwd: ttgtaaaacgacggccagtcgccggcaaattgtaaaatttggcgccttcaapJF250HA1 Rev: CATCGATGCTCCTGAGGCTCCCGATGCTCCTTTTTTCCATTGAAACAATGTAGTCTHA2 Fwd: CGTGATTACAAGGATGACGATGACAAGAGATGAgaaattccaacccctttHA2 Rev: ggaaacagctatgaccatgttatcgatttcaagatgctgctcctggatgt Embryos dissected from one-day old adults were mounted on a pad ( 3% agarose dissolved in M9 ) sandwiched between a microscope slide and no . 1 . 5 coverslip . Time-lapse images were acquired on a Nikon Ti-E inverted microscope ( Nikon Instruments ) equipped with a 1 . 5x magnifying lens , a Yokogawa X1 confocal spinning disk head , and an Andor Ixon Ultra back thinned EM-CCD camera ( Andor ) , all controlled by NIS Elements software ( Nikon ) . Images were obtained using a 60x Oil Plan Apochromat ( NA = 1 . 4 ) or 100x Oil Plan Apochromat ( NA = 1 . 45 ) objective . Z-stacks were acquired using a 0 . 5 µm step every minute . Images were adjusted for brightness and contrast using ImageJ software . Drugs treatments were performed as previously described ( Yang and Feldman , 2015 ) . Briefly , embryos were mounted between a slide and coverslip , supported with 22 . 5 uM beads ( Whitehouse Scientific ) , and bathed in an osmotically balanced control buffer ( embryonic growth medium – EGM [Shelton and Bowerman , 1996] ) supplemented with either 10% DMSO , 30 µM okadaic acid , or 60 µM rubratoxin A , or 200 µM flavopiridol . Embryos were laser permeabilized at appropriate times using a Micropoint dye laser ( coumarin 435 nm ) mounted on the spinning-disk confocal described above . Embryos were prepared and mounted the same way as described above for image acquisition . ABp and P2 cells were fused using the Micropoint dye laser ( coumarin 435 nm ) and confocal described above . RNAi treatment was performed by feeding as previously described using csnk-1 ( RNAi ) , gpr-1/2 ( RNAi ) , and let-92 ( RNAi ) expressing HT115 bacteria from the Ahringer RNAi library ( Ahringer , 2006; Fraser et al . , 2000; Kamath et al . , 2003 ) . L4 stage worms were grown on RNAi plates ( NGM supplemented with IPTG and Ampicillin ) at 25°C for 24 h-48h . RNAi plates were seeded with a bacterial culture grown overnight and subsequently grown 48 hr at room temperature protected from light . | New cells are created when existing cells divide , a process that is critical for life . A structure called the spindle is an important part of cell division , helping to orient the division and separate parts of the old cell into the newly generated ones . The spindle is built using filamentous protein structures called microtubules which are arranged by microtubule organizing centers ( or MTOCs for short ) . In animals , an MTOC forms at each end of the spindle around two structures called centrosomes . A network of proteins called the pericentriolar material ( PCM ) form around centrosomes , converting them into MTOCs . The PCM grows around centrosomes as a cell prepares to divide and is removed again afterward . Enzymes called kinases are important in controlling cell division and PCM assembly; they are opposed by other enzymes known as phosphatases . The processes involved in organization and removal of the PCM are not well understood . The microscopic worm Caenorhabditis elegans provides an opportunity to study details of cell division in a living animal . Magescas et al . used fluorescent labels to view proteins from the PCM under a microscope . The images showed two partially overlapping spherical parts to the PCM – inner and outer . Further examination revealed that the inner PCM is maintained by a careful balance of kinase and phosphatase activity . When kinases shut down at the end of cell division , the phosphatases break down the inner PCM . By contrast , the outer PCM is physically torn apart by forces acting through the attached microtubules . Future work will seek to examine which proteins are specifically affected by phosphatases to identify the key regulators of PCM persistence in the cell and to reveal the proteins needed for MTOC activity at the centrosome . Since poor MTOC regulation can play a part in the growth and spread of cancer , this could lead to targets for new treatments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2019 | A two-step mechanism for the inactivation of microtubule organizing center function at the centrosome |
Negative-strand RNA viruses condense their genome into helical nucleocapsids that constitute essential templates for viral replication and transcription . The intrinsic flexibility of nucleocapsids usually prevents their full-length structural characterisation at high resolution . Here , we describe purification of full-length recombinant metastable helical nucleocapsid of Hantaan virus ( Hantaviridae family , Bunyavirales order ) and determine its structure at 3 . 3 Å resolution by cryo-electron microscopy . The structure reveals the mechanisms of helical multimerisation via sub-domain exchanges between protomers and highlights nucleotide positions in a continuous positively charged groove compatible with viral genome binding . It uncovers key sites for future structure-based design of antivirals that are currently lacking to counteract life-threatening hantavirus infections . The structure also suggests a model of nucleoprotein-polymerase interaction that would enable replication and transcription solely upon local disruption of the nucleocapsid .
The Bunyavirales order is one of the largest groups of segmented negative-strand RNA viruses ( sNSV ) that include many pathogenic strains ( Sun et al . , 2018 ) . In particular , the Hantaviridae family comprises the virus Hantaan ( HTNV ) that gives rise to haemorrhagic fevers with renal syndrome and the virus Sin Nombre that is linked to severe pulmonary illnesses with fatality rates up to 40% . Neither treatment nor vaccine is currently available to counteract them . The Bunyavirales genome is usually divided into three RNA segments enwrapped by the viral nucleoproteins ( NP ) . The resulting nucleocapsids ( NCs ) protect the genome and serve as a replication/transcription template for the viral polymerase ( Reuter and Krüger , 2018 ) . They coat the genomic and anti-genomic RNA during replication but not the mRNA produced by transcription . As they are specific and essential to the viral cycle , NCs constitute an attractive potential target for antiviral drugs . Nucleoproteins of segmented NSV ( sNSV ) present a large diversity of folds ( Sun et al . , 2018 ) . Most of the available structures have been determined as rings and monomers that present the advantage of being rigid enough for crystallisation . However , the relevant conformations of assembled NPs in the viral context correspond to flexible helices or pearl-necklaces that encapsulate long RNA segments . Helical crystal structures of La Crosse virus NP ( Peribunyaviridae , LACV ) ( Reguera et al . , 2013 ) and Crimean Congo Fever Virus NP ( Nairoviridae , CCFHV ) ( Wang et al . , 2012 ) have been determined but they correspond to local organisations of pearl-necklace-like native NCs . Influenza double-helical NCs 3D structures have been described at low resolution but display different helical parameters , also reflecting their malleability ( Arranz et al . , 2012; Moeller et al . , 2012 ) . NC flexibility appears as a hallmark in NSVs as NCs from non-segmented NSV such as Ebola or measles virus require C-terminal NP truncations in order to obtain rigid NCs and high-resolution 3D structures ( Gutsche et al . , 2015; Sugita et al . , 2018; Wan et al . , 2017 ) . In this context , HTNV-NCs are intriguing as they were shown to be able to form rather rigid helices of 10 nm diameter within viruses ( Battisti et al . , 2011; Huiskonen et al . , 2010 ) and during cell infection ( Goldsmith et al . , 1995 ) . We therefore aimed at obtaining their high-resolution 3D structure , identifying the determinants of NP polymerisation and visualising RNA organisation .
Expression of recombinant full-length HTNV-NP in insect cells led to formation of recombinant NCs that have a diameter consistent with native NCs ( Figure 1—figure supplement 1 ) . Cryo-EM images collected on a Titan Krios enabled structural determination of HTNV-NC at 3 . 3 Å resolution ( Figure 1A , Video 1 , Figure 1—figure supplements 1 and 2 , Figure 1—source data 1 ) . HTNV-NC is a left-handed helix with a pitch of 68 . 03 Å and 3 . 6 subunits per turn ( Figure 1A and Figure 1—figure supplement 1 ) . To derive an atomic model of HTNV-NP , the monomeric crystal structure of HTNV-NP comprising residues 113 to 429 ( Olal and Daumke , 2016 ) was fitted into the EM map , and the N-terminal residues and loops not present in the crystal structures were unambiguously built ( Figure 1D ) . The HTNV-NC model was then iteratively rebuilt and the all-atom model refined using stereochemical restraints . Although we expressed the full-length NP , residues 1 to 79 are missing in the final map , indicating that they do not follow the helical symmetry . Interestingly , a NP74-429 construct obtained by trypsin limited proteolysis still forms a rigid helix , which means that the N-ter1-73 is not necessary for helix stabilisation ( Figure 2—figure supplement 1A , B ) . The N-ter1-73 might correspond to a flexibly-linked coiled-coil as previously visualised in the structure of an N-ter construct ( Boudko et al . , 2007 ) . The structure of NP80-429 is composed of a core comprising residues 117 to 398 ( NPcore ) that defines two lobes surrounding a positively charged groove ( Figure 1B , C , Video 1 ) . The NPcore structure is relatively conserved compared to the apo monomeric truncated crystal structure , with a Cα RMSD of 0 . 808 Å over 205 atoms ( Figure 1D ) . N-terminal and C-terminal arms ( Ntarm , Ctarm ) are connected to each extremity of NPcore by flexible hinges ( Figure 1B , Video 1 ) . Superimposition of HTNV-NC with the truncated monomeric structure ( Olal and Daumke , 2016 ) shows that the Ctarm undergoes a 130 . 4° rotation upon multimerisation ( Figure 1D ) . As a result , Nt- and Ctarm of the same subunit embrace each other: residues 97–98 of Ntarm contact residues 423–424 of Ctarm , revealing a unique intra-arms interaction specific to HTNV-NC ( Figure 1D , Video 1 ) . Recombinant HTNV-NC are rigid and remain stable in a large range of salt , pH and temperature conditions ( Figure 2—figure supplement 1C ) . This can be explained by the multiplicity of interactions between protomers , each NP interacting with six other subunits ( Figure 2A , Video 2 ) . Successive subunit interactions rely on exchange of their Ntarm , and Ctarm that make intimate contacts with the core domain of neighbouring protomers ( Figure 2A , B , C ) , resulting in a buried area of 2704 Å2 at each NP-NP interface . The NPi Ntarm forms an elongated structure that binds to residues 155–160 , 177–181 , 189–192 and 136–140 of the NPi-1 subunit ( Figure 2B , Video 2 ) . Since constructs lacking the Ntarm remain monomeric ( Olal and Daumke , 2016 ) , the identified contacts appear to be essential for multimerisation . The NPi amphipatic Ctarm binds in a hydrophobic pocket of the protomer NPi+1 comprising residues 334–346 and 378–394 ( Figure 2C , Video 2 ) . This agrees with the results of double hybrid experiments ( Kaukinen et al . , 2004; Yoshimatsu et al . , 2003 ) which suggested that interaction of C-terminal helices of neighbouring protomers are critical to oligomerisation . Another key actor of multimerisation is the β-hairpin152-181 that protrudes towards the exterior of the NP-core ( Figure 1B ) . The two β-strands of the hairpin ( residues 155–160 and 177–181 ) interact with the Ntarm residues 101–103 from the following subunit to form a 3-stranded β-sheet ( Figure 2A , B , Video 2 ) . As shown by pull-down assays of mutants L102A , V104A , this structure has a main contribution in NC stabilisation ( Guo et al . , 2016 ) . In addition , the β-hairpin152-181 tip ( residues 162–175 ) acts as a clamp and seals the C-terminus-mediated contacts between NPi+2 ( residues 409 to 419 ) and NPi+3 ( residues 385–390 ) , thereby buttressing and rigidifying the metastable helical form of the HTNV-NC ( Figure 2A , D , Video 2 ) . HTNV recombinant NP was expressed in the absence of viral RNA , but after purification the optical density ( OD ) at 260–280 nm was measured to be around 1 . 0 , which strongly suggests the binding of insect cell RNAs during expression . Consistently , three nucleotides with a partial occupancy can be visualised for each NP ( Figure 3A ) . Conserved residues R197 , R314 , R368 , R146 and K153 interact via salt bridges with RNA phosphates , in line with the versatility of RNA sequence to be incorporated by NP . In addition , F361 stabilises a stacking interaction network between the nucleotide bases . The nucleotides bind in a continuous positively charged groove oriented towards the interior of the NC . Density corresponding to additional nucleotides is visible in this groove , but present at low occupancy , suggesting the lower affinity of NC-RNA binding in these regions . Interestingly , the observed nucleotides occupancy is in accordance with electromobility shift assays of Sin Nombre NP mutants ( Guo et al . , 2016 ) : indeed , in this work , residues directly interacting with the three visible RNA nucleotides were identified as the ones displaying the highest affinity with RNA . Altogether , these results reveal that HTNV-NC structure is compatible with binding of long viral RNA , thereby reinforcing the biological relevance of this structure ( Figure 3B , C ) .
Structural determination of a full-length RNA-bound helical NSV NC at 3 . 3 Å resolution is particularly valuable because the usual intrinsic flexibility of NSV full-length NCs prevents their high-resolution analysis . It fits together the pieces of the jigsaw accumulated over several years of biochemistry and structural analysis on Hantavirus NP . HTNV recombinant NC structure indeed displays 3 . 6 subunits per turn and is thus consistent with observations of HTNV-NP trimers made by several groups ( Alfadhli et al . , 2001; Kaukinen et al . , 2004 ) . The present structure is also compatible with the proposed model ( Kaukinen et al . , 2001 ) which suggested that NPs first trimerise around the viral RNA and then gradually assemble to form longer multimers . The role of Ntarm and Ctarm exchange between successive subunits , identified by double-hybrid and pull-down experiments ( Guo et al . , 2016; Kaukinen et al . , 2004; Yoshimatsu et al . , 2003 ) , is in line with HTNV recombinant NC structure . These observations , together with the visualisation of a continuous positively charged groove , strongly suggest that the present structure is biologically pertinent . HTNV recombinant NCs are thus likely to be similar to helical NCs observed within viral particles , although the fact that the latter are less straight implies that at least the β-hairpin152-181 might change conformation in the viral context , thereby enabling more flexibility ( Battisti et al . , 2011; Huiskonen et al . , 2010 ) . Other conformations of NCs are in addition likely to exist because the 5’ and 3’ end of HTNV viral RNA are known to bind to the viral polymerase , implying that the NC must somehow be circularised ( Garcin et al . , 1995 ) . Accordingly , flexible pearl-necklaces are also visible within virions , in infected cells and in NC extracted from viral particles ( Battisti et al . , 2011; Goldsmith et al . , 1995; Guo et al . , 2016; Huiskonen et al . , 2010 ) . Combination of helical NCs observed here and flexible pearl-necklaces thus represent relevant genome-encapsidating conformations . HTNV and LACV NPcore share a common fold ( Guo et al . , 2016; Olal and Daumke , 2016 ) enabling their structural superimposition ( Figure 3—figure supplement 1A ) and comparison of their RNA binding mode . This reveals that the three nucleotides present in HTNV-NC adopt similar conformations as the nucleotides 5 , 6 and 7 of LACV-NP ( Figure 3—figure supplement 1B ) ( Reguera et al . , 2013 ) . Key residues involved in the binding of HTNV-NC nucleotides , namely R197 , R367 and F361 are conserved in LACV-NP ( R94 , R183 and Y177 ) . Superposition of RNA-bound LACV-NP monomers on HTNV-NC shows that LACV RNA fits reasonably well into the HTNV-NC positively charged groove ( Figure 3—figure supplement 1C ) . However , the proposed HTNV-NC RNA path is slightly shorter , suggesting that each HTNV-NP can contain between 8 and 10 nucleotides ( Figure 3—figure supplement 1C ) . Modelling of RNA binding in HTNV-NC based on RNA position in LACV-NP ( see Materials and methods ) indicates that RNA phosphates and riboses interact with residues R146 , K150 , K153 , R197 , R314 , R339 , R367 and R368 , while bases are surrounded by residues 113–116 , 143–151 , 182–188 and 217–222 ( Figure 3—figure supplement 1D ) . While the structures of HTNV and LACV NPs share a common fold for their cores , they however differ in their N-terminal organisation , HTNV Ntarm being significantly longer and linked to the unresolved N-terminal1-73 region ( Figure 3—figure supplement 1A ) . The N-terminal1-73 region may bring the polymerase in close contact to NP as suggested ( Cheng et al . , 2014 ) , and thus play a role similar to the intrinsically disordered phosphoproteins and NP C-terminal regions in nsNSV ( Ivanov et al . , 2011 ) . One may therefore hypothesise that during genome reading by the polymerase , the Ntarm of one HTNV-NP could detach from its adjacent subunit , thereby affecting the conformation of the next NP β-hairpin and removing the seal over the two following NPs . This would provoke a local disruption of the metastable NC and provide RNA access to the polymerase , enabling replication and transcription ( Figure 4 ) . This opens new routes for future experiments , such as cryo-EM analysis of in vivo reconstituted mini-ribonucleoprotein particles ( RNP ) or cryo-electron tomography of viral RNPs , that would decipher the exact mode of interaction between NPs , polymerase and RNA during RNA replication/transcription . The rigidity of HTNV-NC and the absence of disordered phosphoprotein in HTNV should facilitate this study compared to the more complex NP/phosphoprotein/polymerase complex of nsNSV , and thus represents a unique opportunity to unravel significant properties of NSV replication . In addition to these fundamental aspects , the HTNV-NC structure reveals the position of specific antigenic sites on variable regions of the NC surface ( Tischler et al . , 2008 ) thus providing a structure-based rationale for diagnosis ( Figure 1—figure supplement 3 ) . It may also stimulate the design of antivirals , because it defines key regions involved in NP oligomerisation and RNA encapsidation .
Sequence-optimised synthetic DNA encoding a N-terminal his-tag , a TEV protease recognition site and the HTNV-NP ( NCBI accession code NC_005218 ) was synthetised ( Geneart ) and cloned into a pFastBac1 vector between NdeI and NotI restriction sites . The HTNV-NP-expressing baculovirus was generated via the standard Bac-to-Bac method ( Invitrogen ) . For large scale expression , Sf21 cells at 0 . 5 × 106 cells/mL concentration were infected by adding 0 . 1% of virus . Expression was stopped 72 hr after the day of proliferation arrest . The cells were disrupted by sonication 3 min ( 10 s ON , 20 s OFF , 40% amplitude ) on ice in lysis buffer ( 20 mM Tris-HCl pH 8 , 300 mM NaCl , 10 mM MgCl2 , 2 mM β-mercaptoethanol and 20 mM Imidazole ) with EDTA free protease inhibitor complex ( Roche ) . After lysate centrifugation at 20 , 000 rpm during 45 min at 4°C , protein from the soluble fraction was loaded on a nickel column , washed with 10 volumes of lysis buffer , 10 volumes lysis buffer supplemented with 50 mM Imidazole and eluted with 5 volumes of lysis buffer supplemented with 500 mM Imidazole . The eluted protein was cleaved with TEV protease in a 20:1 w/w ratio overnight at 4°C in dialysis against lysis buffer resulting in an almost complete cleavage . A second nickel column step was performed to remove unwanted material . The resulting protein was concentrated by ultracentrifugation using Optima XE SW55Ti rotor ( Beckman Coulter ) and 0 . 8 mL Ultra Clear tube during 2 hr at 45 , 000 rpm , 4°C . Concentrated HTNV-NC present in the 20 µl bottom fraction of each tube was gently resuspended . The 260/280 nm absorbance ratio was measured to be around one at the end of the purification indicating presence of nucleic acid . Three biologically independent batches of large-scale expression and fifteen biologically independent batches of purifications were performed and gave reproducible results . Limited proteolysis of HTNV-NC was performed at 20°C in lysis buffer with a 3:2 w/w ratio of HTNV-NP/trypsin . Proteolysis was stopped by the addition of denaturing SDS-PAGE loading dye and incubation at 95°C for 5 min . Digestion was visualised in 10% acrylamide SDS-Page gels . For N-terminal sequencing , proteins were transferred on PVDF membrane previously incubated in 10 mM CAPS pH 11 , 10% methanol buffer . PVDF membrane was stained using 0 . 1% Coomassie Brilliant Blue R-250 , 40% methanol , 1% acetic acid buffer until bands were visible , washed using 50% methanol and dried . Visible bands were cut and subjected to N-terminal sequencing . Amino acid sequence determination based on Edman degradation was performed using an Applied Biosystems gas-phase sequencer model 492 ( s/n: 9510287J ) . Phenylthiohydantoin amino acid derivatives generated at each sequence cycle were identified and quantitated on-line with an Applied Biosystems Model 140C HPLC system using the data analysis system for protein sequencing from Applied Biosystems ( software Procise PC v2 . 1 ) . For negative stain EM grid preparation , 4 µl of sample was applied between mica and carbon layer and stained using sodium silicotungstate ( SST ) 2% . After removing the mica part , the grid was deposited on the carbon layer and dried at room temperature . Micrographs were collected at 2 . 5 µm defocus using a FEI Tecnai F20 operated at 200 kV on 4 k*4 k CETA FEI CCD camera . For cryo-EM grid preparation , Quantifoil grids 400 mesh 2/1 were glow-discharged at 30 mA for 1 min . 3 µl of sample were applied on the resulting glow-discharged grids and excess solution was blotted during 2 . 5 s force seven with a Vitrobot Mark IV ( FEI ) and the grid frozen in liquid ethane . Two biologically independent datasets , arising from different expression and purification batches , were collected on two high-end cryo-electron microscopes and gave consistent structures . The first cryo-EM dataset was collected on a FEI Polara F30 microscope operated at 300 kV equipped with a K2 summit GATAN direct electron detector and resulted in a 3 . 8 Å resolution structure . The second dataset , that gave rise to the 3 . 3 Å resolution structure presented in the present article , was collected on a FEI Titan Krios operated at 300 kV equipped with a Gatan Bioquantum LS/967 energy filter coupled to a Gatan K2 direct electron detector camera . For this second dataset , automated data collection was performed using EPU FEI software . Zero-loss micrographs were recorded at a 46 , 860x magnification giving pixel size of 1 . 067 Å with defocus ranging from 0 . 8 to 3 . 5 µm . In total , 4328 movies with 28 frames per movie were collected with a total dose of 40 e-/Å2 . Micrographs were initially selected based on visual quality inspection . Movie drift correction was performed in Motioncor2 ( Li et al . , 2013 ) excluding the two first frames . CTF parameters were determined in Gctf ( Zhang , 2016 ) ( RRID:SCR_016500 ) . All subsequent processing steps were performed in Relion2 . 1 and Relion3 software ( Scheres , 2012; Zivanov et al . , 2018 ) ( RRID:SCR_016274 ) . Straight HTNV-NC were manually picked and computationally cut with an inter-box distance of 38 Å along the helical axis into overlapping boxes of 400*400 pixels , resulting in 168 , 709 extracted segments . 2D classification was used to eliminate bad quality filaments . The best 2D classes were aligned and padded in a square of 1200*1200 pixels2 . The individual 2D Power Spectra ( PS ) of each best class were averaged and used for a first estimation of the helical parameters using Fourier Bessel indexing ( Cochran et al . , 1952 ) . An estimation of the repeat c was first inferred from the layer lines regularly spaced at multiples of ~335 . 6 Å−1 ( Figure 1—figure supplement 1E ) . Several strong meridional layer lines regularly spaced at multiples of 18 . 64 Å−1 indicated the axial rise p between subunits . According to both c estimation and p , pitch P was inferred to be around 67 . 12 Å−1 as a strong layer line with a first intensity near the meridian can be seen at l = 5 . Therefore , the structure repeats after u = 18 subunits ( 18 ( u ) = 335 . 6 ( c ) /18 . 64 ( p ) ) , in t = 5 turns , resulting in a number of units per turn u/t of ~3 . 6 . 3D helical reconstruction was performed using the 111 , 248 selected segments from the best 2D classes . A 130 Å diameter featureless cylinder was used as an initial model . Helical symmetry search was performed between −100 ± 4 ° for the helical twist ( 360°/u/t ) and 18 . 64 ± 0 . 5 Å for the helical rise . Helical twist and rise after refinement were respectively −99 . 97° and 18 . 96 Å . For the final 3D helical reconstruction , poorly aligned segments with an angular tilt >±20 ° were discarded resulting in 105 , 665 segments . Local CTF-determination was calculated for each segment using Gctf ( Zhang , 2016 ) and finally CTF-refinement was performed using Relion 3 . 018 . The 10 last frames of each motion corrected micrographs were removed resulting in a total dose of ~22 e-/Å2 . The latest 3D map filtered at 15 Å resolution was used as initial model and helical symmetry search further refined during a last refinement . The resulting final reconstruction displays refined helical parameters of −99 . 95° for the twist and 18 . 87 Å for the rise . Post-processing , done using a B-factor of −103 Å2 , resulted in a 3 . 3 Å resolution reconstruction using the FSC 0 . 143 cutoff criteria ( Figure 1—figure supplement 2A ) . Local resolution variations were estimated in Relion ( Figure 1—figure supplement 2B ) . Observation of nucleotides with partial occupancy suggests that only a fraction of HTNV-NC segments analysed contains RNA . We thus attempted to perform both 2D and 3D classifications in order to separate RNA-bound and apo HTNV-NC segments using several strategies including usage of different masks , subtraction/absence of subtraction of protein density , symmetry release . However , due to the low molecular weight of RNA compared the protein ( 9 RNA nucleotides bound per HTNV-NP would correspond to 3 kDa of RNA versus 50 kDa of protein ) , these classifications did not succeed in separating the apo and RNA-bound states . The monomeric crystal structure of HTNV-NP comprising residues 113 to 429 ( PDB code: 5FSG ) was initially fitted into the EM density as two separate rigid bodies containing residues 113–398 and 399–429 . Loops comprising residues 146–155 and 349–360 , β-hairpin 160–172 and the N-terminal arm residues 80 to 112 - previously not present in the crystal structure - were manually built in Coot ( Emsley et al . , 2010 ) . The three RNA nucleotides visible in the density map were manually built and adjusted using Coot ( Emsley et al . , 2010 ) ( RRID:SCR_014222 ) and RCrane ( Keating and Pyle , 2012 ) . NP monomeric model was symmetrised according to the helical parameters to form a filament model of 7 subunits . Fiber model was then iteratively rebuilt and all-atom refined using stereochemical and NCS restraints within PHENIX ( Adams et al . , 2010 ) ( RRID:SCR_014224 ) . In order to model HTNV RNA , two HTNV-NP were extracted from HTNV-NC , together with their respective three nucleotides ( called ‘HTNV-nts’ ) . Five , six or seven LACV-nts were used to link the three HTNV-nts , giving rise to hybrid models containing between eight and ten nucleotides per NP . They were regularised in Coot , minimised in Chimera ( Pettersen et al . , 2004 ) ( RRID:SCR_004097 ) and were then subjected to modelling in Haddock ( de Vries et al . , 2010; van Zundert et al . , 2016 ) . During the modelling procedure , the interaction between HTNV-NP and HTNV-nucleotides was maintained in the same conformation as in the EM map using unambiguous distance restraints . In addition , Haddock ambiguous interaction restraints were defined: residues involved in interactions ( Guo et al . , 2016 ) were considered as ‘active’ , while all solvent accessible surface neighbours of active residues were defined as ‘passive’ . The initial starting orientations were not randomised . Rigid body minimisation , semi-flexible simulated annealing and flexible explicit solvent refinement was performed . For each starting RNA , the resulting models were very similar and gathered in one single cluster . The model with the lowest energy and the lowest Haddock score was thus considered as being the modelling result . The atomic coordinates and the cryo-EM map have respectively been deposited in the PDB and EMDB under the accession codes 6I2N and EMD-0333 . | Rats and mice sometimes transmit hantaviruses , a family of microbes that can cause deadly human diseases . For example , the Hantaan virus leads to haemorrhagic fevers that are potentially fatal . There are no vaccine or even drugs against these infections . To multiply , viruses must insert their genetic material inside a cell . While the body often detects and destroys foreign genetic information , hantaviruses can still evade our defences . Molecules called nucleoproteins bind to the viral genome , hiding it away in long helices called nucleocapsids . When the virus needs to replicate , an enzyme opens up the nucleocapsid , reads and copies the genetic code , and then closes the helix . Yet , researchers know little about the details of this process , or even the structure of the nucleocapsid . Here , Arragain et al . use a method called cryo-electron microscopy to examine and piece together the exact 3D structure of the Hantaan virus nucleocapsid . This was possible because the new technique allows scientists to observe biological molecules at an unprecedented , near atomic resolution . The resulting model reveals that the viral genome nests into a groove inside the nucleocapsid . It also shows that specific interactions between nucleoproteins stabilise the helix . Finally , the model helps to provide hypotheses on how the enzyme could read the genome without breaking the capsid . Mapping out the structure and the interactions of the nucleocapsid is the first step towards finding molecules that could destabilise the helix and neutralise the virus: this could help fight both the Hantaan virus and other members of its deadly family . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2019 | High resolution cryo-EM structure of the helical RNA-bound Hantaan virus nucleocapsid reveals its assembly mechanisms |
Lymphomagenesis in the presence of deregulated MYC requires suppression of MYC-driven apoptosis , often through downregulation of the pro-apoptotic BCL2L11 gene ( Bim ) . Transcription factors ( EBNAs ) encoded by the lymphoma-associated Epstein-Barr virus ( EBV ) activate MYC and silence BCL2L11 . We show that the EBNA2 transactivator activates multiple MYC enhancers and reconfigures the MYC locus to increase upstream and decrease downstream enhancer-promoter interactions . EBNA2 recruits the BRG1 ATPase of the SWI/SNF remodeller to MYC enhancers and BRG1 is required for enhancer-promoter interactions in EBV-infected cells . At BCL2L11 , we identify a haematopoietic enhancer hub that is inactivated by the EBV repressors EBNA3A and EBNA3C through recruitment of the H3K27 methyltransferase EZH2 . Reversal of enhancer inactivation using an EZH2 inhibitor upregulates BCL2L11 and induces apoptosis . EBV therefore drives lymphomagenesis by hijacking long-range enhancer hubs and specific cellular co-factors . EBV-driven MYC enhancer activation may contribute to the genesis and localisation of MYC-Immunoglobulin translocation breakpoints in Burkitt's lymphoma .
Epstein-Barr virus ( EBV ) is associated with the development of numerous lymphomas including Burkitt's ( BL ) , post-transplant , Hodgkin and certain NK and T-cell lymphomas . EBV was discovered in BL biopsies from sub-Saharan Africa ( Epstein et al . , 1964 ) , where BL is endemic ( eBL ) and almost always EBV associated . BL also occurs world-wide as sporadic BL ( sBL ) and immunodeficiency-associated BL , where EBV positivity is approximately 20% and 60% , respectively ( Mbulaiteye et al . , 2014 ) . Irrespective of origin or EBV status , the defining feature of BL is a chromosomal translocation involving MYC on chromosome 8 and an immunoglobulin ( IG ) gene . MYC translocations detected in BL involve either the IG heavy , or lambda or kappa light chain loci on chromosomes 14 , 2 or 22 respectively . t ( 8:14 ) translocations occur in 85% of BL cases ( Boerma et al . , 2009 ) . The position of the MYC/IG translocation breakpoint is usually far 5’ of MYC in endemic ( EBV positive ) BL . In sporadic BL , breakpoints are in the first exon or intron , implicating different , but unknown , mechanisms in their generation ( Neri et al . , 1988; Shiramizu et al . , 1991 ) . The placement of MYC adjacent to highly active regulatory regions at these IG loci leads to constitutive high-level MYC expression and the uncontrolled proliferation of BL cells . Despite intensive study , the role of EBV in the development of BL is still unclear . The oncogenic potential of EBV is evident from its potent transforming activity in vitro . On infection , resting B lymphocytes are growth-transformed into permanently proliferating lymphoblastoid cell-lines ( LCLs ) . In common with other herpesviruses , EBV establishes a latent infection in infected cells . Nine viral latent proteins are expressed in EBV-immortalised LCLs; six Epstein-Barr nuclear antigens ( EBNAs 1 , 2 , 3A , 3B , 3C and LP ) and three latent membrane proteins ( LMP1 , 2A and 2B ) . EBNA2 and the EBNA3 family of distantly-related transcription factors ( TF ) ( EBNA3A , EBNA3B and EBNA3C ) play important roles in the transcriptional reprogramming of host B cells . The actions of these four EBV TFs results in the deregulation of numerous cellular genes involved in the control of B-cell growth and survival ( Zhao et al . , 2011a , 2006; Spender et al . , 2002; Maier et al . , 2006; McClellan et al . , 2012; Hertle et al . , 2009; White et al . , 2010 ) . EBNA2 , EBNA3A and EBNA3C are required for B-cell immortalisation and the continuous proliferation of infected cells ( Cohen et al . , 1989; Tomkinson et al . , 1993; Maruo et al . , 2003 , 2006; Kempkes et al . , 1995 ) . These TFs cannot however bind DNA directly; they control gene transcription through interactions with cellular DNA-binding proteins ( e . g . RBP-Jκ and PU . 1 ) ( Johannsen et al . , 1995; Ling et al . , 1994; Waltzer et al . , 1994 , 1996; Robertson et al . , 1995; Le Roux et al . , 1994; Zhao et al . , 1996; Robertson et al . , 1996 ) . Following initial B-cell transformation in vivo , EBV-infected cells sequentially reduce the number of latent genes they express to enable progression through the B-cell differentiation pathway ( Thorley-Lawson and Babcock , 1999 ) . This allows entry into the memory B-cell compartment , where the virus persists . Many EBV-associated tumour cells display restricted patterns of viral latent gene expression that may reflect the differentiation state of the neoplastic precursor cell . During B-cell transformation by EBV , EBNA2 plays a key role in upregulating numerous genes involved in driving cell proliferation , including the proto-oncogene MYC ( Kaiser et al . , 1999 ) . Whether EBNA2 activation of MYC contributes to the genesis of the MYC translocation in BL cells however , is not known . EBNA2 contains an acidic activation domain and mediates gene activation by binding histone acetyl transferases and chromatin remodellers ( reviewed in ( Kempkes and Ling , 2015 ) . EBNA3A , EBNA3B and EBNA3C individually and co-operatively activate and repress cellular gene expression ( White et al . , 2010; McClellan et al . , 2012; Hertle et al . , 2009 ) . EBNA3B is dispensable for B-cell immortalisation by EBV , but appears to play a role in suppressing tumour formation in vivo , since its loss accelerates lymphoma development ( Tomkinson and Kieff , 1992; White et al . , 2012 ) . The role of EBNA3A and EBNA3C as cellular gene repressors has been most extensively studied ( reviewed in ( Allday et al . , 2015 ) . These two viral TFs work cooperatively to silence key tumour suppressor gene . These include the cyclin-dependent kinase inhibitor p16INK4a and the pro-apoptotic Bcl-2 family binding protein Bim ( BCL2L11 ) ( Skalska et al . , 2010; Anderton et al . , 2008 ) . EBNA3A/EBNA3C directed silencing is associated with the recruitment of polycomb repressor complex 1 and 2 ( PRC1 , 2 ) and the deposition of the histone H3 lysine 27 trimethyl mark ( H3K27me3 ) ( McClellan et al . , 2012 , 2013; Paschos et al . , 2012 , 2009; Skalska et al . , 2010; Kalchschmidt et al . , 2016b ) At BCL2L11 , PRC-mediated repression leads to longer-term silencing through the accumulation of CpG promoter methylation ( Paschos et al . , 2009 ) . By activating MYC to drive cell proliferation and counteracting MYC-triggered apoptosis by silencing BCL2L11 , EBV TFs are manipulating the same pathways that are deregulated in non-viral lymphomas . In fact , lymphomagenesis only occurs in the presence of deregulated MYC expression when the p53-MDM2-p14ARF or BCL2L11 apoptotic axes are disabled ( reviewed in Thorley-Lawson and Allday , 2008 ) . The mechanisms through which MYC and BCL2L11 are deregulated by EBV TFs however , are not fully defined . Genome-wide analyses indicate that binding of long-range regulatory elements by EBNA2 and EBNA3 proteins plays a key role in cellular gene reprogramming ( McClellan et al . , 2013; Zhao et al . , 2011b; Schmidt et al . , 2015; Jiang et al . , 2014; Zhou et al . , 2015; McClellan et al . , 2012 ) . Indeed , upstream MYC enhancer regions bound by EBNA2 have been identified ( Zhao et al . , 2011b ) . MYC is one of the most commonly deregulated oncogenes in human cancers and the mapping of multiple cancer risk loci and regions of focal amplification to MYC enhancer regions provides strong evidence that inappropriate MYC expression can result from the perturbation of long-range control ( Yochum , 2011; Ahmadiyeh et al . , 2010; Tuupanen et al . , 2009; Pomerantz et al . , 2009; Shi et al . , 2013; Zhang et al . , 2016; Herranz et al . , 2014 ) . BCL2L11 silencing by EBV has only been studied in the context of EBNA3A and EBNA3C binding to the gene promoter ( McClellan et al . , 2013; Paschos et al . , 2012 ) , but interestingly , inactivation of a murine-specific BCL2L11 enhancer has recently been reported in B lymphoblastic leukaemia ( Wang et al . , 2015 ) . This indicates that enhancer control of BCL2L11 may be important in other contexts and may be a target for disruption in tumour cells . Here we demonstrate that EBV EBNA2 upregulates MYC by activating specific upstream enhancers . This promotes upstream and reduces downstream enhancer-promoter interactions . These MYC enhancer interactions in EBV-infected cells are dependent on the chromatin remodelling function of SWI/SNF . At BCL2L11 , we show that the EBV repressors EBNA3A and EBNA3C inactivate a newly-described enhancer-promoter hub in a manner dependent on the activity of PRC2 . Lymphomagenesis by EBV therefore involves the hijack and reorganisation of large-scale enhancer hubs through the recruitment of specific cellular co-factors .
To study the mechanism of MYC activation by the EBV TF EBNA2 , we examined EBNA2 binding at the MYC locus using ChIP-sequencing data we obtained previously from two EBV-infected cell lines ( Gunnell et al . , 2016; McClellan et al . , 2013 ) . These included the EBV-transformed lymphoblastoid cell-line GM12878 ( ENCODE Tier 1 ) generated by in vitro infection of resting B cells and the EBV-positive BL cell line Mutu III ( Gregory et al . , 1990 ) . The Mutu III line is derived from a BL tumour cell line ( Mutu I ) which underwent broadening of its virus latent gene expression profile in culture . Both cell-lines therefore express the full panel of EBV latent genes , including EBNA2 . We found that EBNA2 bound multiple elements both upstream ( −556 , −428 , and −186/168 kb ) and downstream ( +450 , +570 , +900 kb and +1 . 8 Mb ) from MYC ( Figure 1A and B ) . Upstream regions bound by EBNA2 in particular , have H3K27 acetylation ( H3K27ac ) signals characteristic of active enhancers ( Figure 1C ) . In fact all regions bound by EBNA2 , apart from the +900 kb enhancer , are classified as super-enhancers in numerous cancer cell lines and/or primary cells ( dbSUPER , http://bioinfo . au . tsinghua . edu . cn/dbsuper/ [Khan and Zhang , 2016] ) ( Supplementary file 1 ) . Super-enhancers ( SE ) are large lineage-specific regulatory elements typically comprising multiple TF binding sites ( Whyte et al . , 2013 ) . The −556 and −428 regions also contain susceptibility loci for chronic lymphocytic leukaemia ( CLL ) and prostate cancer ( rs2466024 , rs2466035 , rs18814048 , rs16902094 and rs445114 ) ( Speedy et al . , 2014; Gudmundsson et al . , 2009 , 2012 ) . EBV therefore targets MYC enhancers that function in multiple cell-type specific and tumourigenic contexts . 10 . 7554/eLife . 18270 . 003Figure 1 . EBNA2 binding induces directional reorganisation of MYC promoter-enhancer interactions . ( A ) EBNA2 ChIP-sequencing reads in EBV-infected GM12878 cells ( B ) in EBV positive Mutu III BL cells ( that express all latent EBV proteins ) . ( C ) H3K27ac signals in GM12878 from ENCODE . Numbering indicates the location of the major enhancer clusters relative to the MYC transcription start site . ( D ) Sequencing reads from circularised chromosome conformation capture-sequencing ( 4C-seq ) using the MYC promoter as bait in ER-EB 2 . 5 cells expressing an EBNA2-ER fusion protein cultured in the absence of β-estradiol ( -EBNA2 ) . Reads shown are from one of two replicates . The scale bar shows reads per 10 kb window per million reads of sequencing library . ( E ) 4C-seq data from cells incubated in the presence of β-estradiol ( +EBNA2 ) . ( F ) Subtraction of -EBNA2 4C-sequence reads from +EBNA2 4C-sequence reads . The scale bar shows the normalised interaction read count difference ( see Materials and methods for more details ) . Asterisks indicate the positions of CTCF sites . ( G ) Capture Hi-C sequencing reads using a MYC promoter bait and a CD34+ haemopoietic progenitor cell Hi-C library . Arrows denote positions where statistically significant MYC interactions correspond to EBNA2 binding sites . The scale bar shows reads for five merged consecutive genome fragments per million reads of sequencing library . ( H ) MYC promoter Capture Hi-C reads obtained from a GM12878 CHi-C libary . DOI: http://dx . doi . org/10 . 7554/eLife . 18270 . 00310 . 7554/eLife . 18270 . 004Figure 1—figure supplement 1 . MYC mRNA induction in ER-EB 2 . 5 cells . RT-QPCR analysis of MYC mRNA expression in ER-EB 2 . 5 cell samples used for 4C-seq analysis . Cells were cultured in the absence of β-estradiol for 4 days and then for a further 17 hr with ( +EBNA2 ) or without β-estradiol ( -EBNA2 ) . Signals were normalised to GAPDH mRNA levels . Results show the mean ± standard deviation of QPCR duplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 18270 . 004 Although EBNA2 enhances the association of the −428 MYC SE with the MYC promoter ( Zhao et al . , 2011b ) , there is no information on how the targeting of multiple upstream and downstream long-range MYC enhancers by EBNA2 affects enhancer-promoter interactions across the entire MYC locus . We examined this using a 4C-sequencing approach . We used a MYC promoter fragment as bait to capture interacting regions in an EBV-transformed LCL expressing a conditionally-active oestrogen receptor-EBNA2 fusion protein ( ER-EB 2 . 5 ) ( Kempkes et al . , 1995 ) . In these cells , EBNA2 can be reversibly inactivated through the withdrawal and re-addition of oestrogen to the culture medium , providing a useful system in which to study the effects of EBNA2 on gene transcription . Re-addition of β-estradiol to ER-EB 2 . 5 cells cultured for 4 days in its absence , upregulated MYC as previously described ( Kaiser et al . , 1999 ) ( Figure 1—figure supplement 1 ) and resulted in the substantial directional reorganisation of interactions between EBNA2-bound regions and the MYC promoter ( Figure 1D–F ) . In the presence of functional EBNA2 , MYC promoter interactions with upstream elements , including the −556 , −428 , and −186/168 kb regions were increased . In contrast , interactions with downstream elements , including the +450 kb , +570 kb and +1 . 8/1 . 9 Mb regions were decreased ( Figure 1D–F ) . Subtraction of 4C-sequencing reads obtained in the absence of functional EBNA2 from those obtained in its presence demonstrated the clear directionality of EBNA2-directed MYC reorganisation ( Figure 1F ) . The high frequency of upstream enhancer interactions with the MYC promoter in EBV-infected cells was confirmed in genome-wide capture Hi-C data from GM12878 cells ( Figure 1H ) . The GM12878 interaction profile contrasts with that obtained from CD34+ haemopoietic progenitor cells ( Figure 1G ) ( Mifsud et al . , 2015 ) and leukaemic cells ( Shi et al . , 2013 ) , where interactions with the +1 . 8/1 . 9 Mb enhancers dominate . We used chromosome conformation capture ( 3C ) to verify the effects of EBNA2 on upstream MYC promoter-enhancer interactions ( Figure 2 ) . We detected 2–3-fold increases in promoter interaction frequency in the presence of EBNA2 for the largest EBNA2 binding peak in the −556 SE and for the four EBNA2 peaks in the −186/168 region ( Figure 2C and D ) . EBNA2 effects on −428 SE interactions have been documented previously ( Zhao et al . , 2011b ) . 4C-sequencing analysis also identified a distinct region within the −556 SE that displayed reduced MYC promoter interactions in the presence of EBNA2 ( Figure 1F ) . A 2 . 6-fold decrease in the frequency of interactions between this region and the MYC promoter was confirmed using 3C ( Figure 2C ) . This region contains three binding sites for the chromatin boundary and looping factor CTCF ( Figure 2 and Figure 2—figure supplement 1 ) . Two of these CTCF sites are located adjacent to one another and immediately upstream of the main −556 SE EBNA2 binding peak that shows increased promoter interactions in the presence of EBNA2 . CTCF binding may therefore delineate a chromatin interaction boundary . All three CTCF sites are bound by CTCF and components of the chromatin looping complex cohesin ( SMC3 and RAD21 ) in GM12878 cells ( Figure 2—figure supplement 1 ) . Using 3C , we investigated whether EBNA2 influenced interactions between these CTCF sites to change the three-dimensional arrangement of promoter interactions in this region . We found a four-fold increase in interactions between CTCF site-containing fragments in the presence of EBNA2 ( Figure 2E ) . These data indicate that the process of MYC activation by EBNA2 through the directed reorganisation of upstream MYC chromatin promotes CTCF site interactions . This results in the looping out of this specific region from the enhancer-promoter chromatin hub and presumably facilitates other upstream enhancer-promoter interactions . 10 . 7554/eLife . 18270 . 005Figure 2 . Chromosome conformation capture ( 3C ) confirms EBNA2-induced changes at MYC and detects altered CTCF site interactions . ( A ) EBNA2 binding at the −556 super-enhancer region . The positions of the HindIII restriction enzyme sites and the primers used for 3C are indicated . Red arrows indicate the position of the MYC enhancer primers used for promoter interaction analysis . Blue arrows indicate the position of the CTCF site primers used to analyse CTCF site interactions . Asterisks indicate the position of CTCF sites . There are two adjacent sites at the 3’ end of the region ( see Figure 2—figure supplement 1 ) . Primer design is unidirectional ( Naumova et al . , 2012 ) . Purple and green lines indicate the regions that show reduced or increased interactions with the MYC promoter in 4C , with the transition area displaying a mix of increased and decreased interactions indicated by the checked line ( see Figure 1 ) . ( B ) EBNA2 binding and MYC enhancer primer positions in the −186/168 enhancer region ( C ) 3C analysis of interactions between the indicated −556 super-enhancer regions and the MYC promoter in the absence or presence of EBNA2 in ER-EB 2 . 5 cells . Promoter interactions with a region upstream of the −556 super-enhancer not bound by EBNA2 were analysed as a control ( Con ) . Results show the mean ± standard deviation of signals from duplicate PCRs . ( D ) 3C analysis of interactions between the indicated EBNA2-bound −186/168 enhancer regions and the MYC promoter in the absence or presence of EBNA2 . Control interaction analysis ( Con ) as in C . ( E ) 3C analysis of interactions between the CTCF sites in the −556 super-enhancer region in the absence and presence of EBNA2 . CTCF site 2 interactions with the upstream control region were also analysed ( Con ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18270 . 00510 . 7554/eLife . 18270 . 006Figure 2—figure supplement 1 . CTCF and Cohesin binding in the MYC -556 super-enhancer region . ( A ) EBNA2 ChIP-sequencing reads in the GM12878 LCL and ( B ) Mutu III BL cells in the −556 region ( as in Figure 2 ) . ENCODE GM12878 ChIP-sequencing data for CTCF ( C ) and the cohesin subunits SMC3 ( D ) and RAD21 ( E ) . ( F ) HindIII restriction enzyme sites and the location of the CTCF site primers used for 3C analysis ( Figure 2 ) . Asterisks indicate the position of three CTCF consensus binding motifs . DOI: http://dx . doi . org/10 . 7554/eLife . 18270 . 006 We conclude that EBNA2 activation of MYC is associated with the large-scale directional reorganisation of the MYC enhancer hub . EBNA2 increases upstream enhancer interactions , reduces downstream enhancer interactions and alters interactions between CTCF sites . EBV infection of naïve B cells results in B-cell activation in a manner that resembles physiological B-cell activation by CD40 ligand ( CD40L ) and IL-4 . B-cell activation by CD40L/IL-4 however results in short-term proliferation whereas EBV-infected B cells grow out into immortal cell-lines . We used 4C-sequencing to determine whether infection of resting B cells by EBV induced the same changes in MYC enhancer interactions observed in EBV-infected cell lines . We also examined whether the effects of EBV on MYC were distinct from any changes induced by B-cell activation by CD40L/IL-4 . We found that in naïve CD19+ B cells , the MYC promoter interacted with the −556 , −428 , +450 and +570 kb enhancer regions ( Figure 3B ) . This is consistent with the existence of some baseline ‘static’ enhancer-promoter interactions in resting cells , and the classification of the −556 region as a SE in CD19+ B cells ( Supplementary file 1 ) . MYC mRNA levels increased to maximum levels 48 hr post-EBV infection ( Figure 3—figure supplement 1 ) . 4C-sequencing carried out at this time point detected increases in interactions between all upstream enhancers and intervening regions and the MYC promoter ( Figure 3B , C and E ) . This is consistent with the effects we observed on EBNA2 activation in the LCL ( Figure 1 ) . We also observed reduced interactions between the +450 , +570 kb and +900 kb enhancers and the MYC promoter on EBV infection consistent with our LCL data ( Figure 3B , C and E ) . In contrast to our previous observations however , we detected some small localised increases in interactions with regions downstream from the MYC promoter ( Figure 3B ) , which may reflect the low level of interactions present in this region in resting B cells ( Figure 3E ) . In contrast , 48 hr after B-cell activation by CD40L/IL-4 we observed a reverse effect on MYC promoter-enhancer interactions ( Figure 3B , D and F ) . Upstream interactions were reduced and downstream interactions were increased . In particular , the region +235 to +432 kb interacted with the MYC promoter at high level . This downstream region is not bound by EBNA2 and does not interact with the MYC promoter significantly in EBV-infected cells . This region does however interact at high frequency in CD34+ cells ( Figure 1G ) . These data therefore highlight specificity in the remodelling of MYC promoter-enhancer interactions by different stimuli in B cells . This is particularly evident when the effects of EBV and CD40L/IL-4 on MYC promoter interactions are compared ( Figure 3G ) . 10 . 7554/eLife . 18270 . 007Figure 3 . EBV infection of naïve B cells induces directional reorganisation of MYC promoter-enhancer interactions . ( A ) EBNA 2 ChIP-sequencing reads in Mutu III BL cells ( as in Figure 1 ) . Interactions captured by 4C-seq using the MYC promoter as bait ( as in Figure 1 ) in uninfected naïve B cells ( B ) , B cells 48 hr post-EBV infection ( C ) and B cells 48 hr post-stimulation with CD40L/IL-4 ( D ) . Subtraction of 4C-seq reads from uninfected B cells from those obtained from EBV-infected cells ( E ) or CD40L/IL-4 treated cells ( F ) . Reads shown are from both replicates combined . The scale bar shows reads per 10 kb window per million reads of sequencing library . ( G ) Subtraction of 4C-seq reads from CD40L/IL4-treated cells from those obtained from EBV-infected cells . The scale bar shows the normalised interaction read count difference . DOI: http://dx . doi . org/10 . 7554/eLife . 18270 . 00710 . 7554/eLife . 18270 . 008Figure 3—figure supplement 1 . MYC mRNA induction on EBV infection . ( A ) RT-QPCR analysis of MYC mRNA expression over a B-cell infection time course to determine the optimum time for MYC induction . ( B ) RT-QPCR analysis of MYC mRNA expression in the naïve , EBV-infected and CD40L/IL-4 B cell samples used for 4C-seq . Signals were normalised to β2-microglobulin mRNA levels as GAPDH is induced on B-cell infection . Results show the mean ± standard deviation of QPCR duplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 18270 . 008 In summary , EBV infection of naïve B cells reconfigures MYC chromatin architecture in a manner distinct from physiological B-cell stimulation . EBV infection results in the same selective enhancement of upstream enhancer-promoter interactions observed on EBNA2 activation in EBV-infected cell-lines . We next investigated the mechanism of MYC activation by EBNA2 . In the presence of EBNA2 , we found increased levels of H3K27ac across EBNA2-bound enhancer regions , consistent with enhancer activation . EBNA2 increased H3K27ac levels seven-fold at the main MYC P2 promoter and 16-fold at the −556 SE ( Figure 4A ) . BRG1 , the ATPase subunit of the chromatin remodeller SWI/SNF , is required for the interaction of the +1 . 8/1 . 9 Mb leukaemic-cell MYC enhancer with the MYC promoter ( Shi et al . , 2013 ) , so we examined the involvement of BRG1 in EBNA2 activation of MYC enhancers . We found that BRG1 associated with specific MYC enhancers , with highest levels at the −186/168 enhancer and the −556 SE main peak ( Figure 4B ) . BRG1 also bound the +1 . 8 Mb enhancer but at lower levels . EBNA2 increased BRG1 binding at these sites , consistent with the ability of EBNA2 to interact with BRG1 via the Snf5 subunit of SWI/SNF ( Wu et al . , 1996 ) . Interestingly , EBNA2 reduced BRG1 binding to the region of the −556 SE that is looped out through CTCF site association ( Figure 4B ) . The effect of EBNA2 on BRG1 binding at upstream regions therefore correlates with the effect of EBNA2 on promoter interaction frequency . We next investigated whether BRG1 was required for the interaction of EBNA2-bound enhancers with the MYC promoter in EBV-infected cells . We found that siRNA-mediated BRG1 knockdown in GM12878 cells led to a loss of MYC promoter interactions with the −556 , −428 and −186/168 enhancers ( Figure 4C and D ) . 3C did not detect any interactions between the +1 . 8 Mb region and the MYC promoter in the presence or absence of BRG1 ( data not shown ) , consistent with its low-level interaction frequency in EBV-infected cells ( Figures 1 and 3 ) . We conclude that BRG1 is required to maintain the active upstream MYC enhancer-promoter hub in EBV-infected B cells . 10 . 7554/eLife . 18270 . 009Figure 4 . BRG1 is required for upstream MYC enhancer-promoter interactions in EBV-infected cells . ( A ) ChIP-QPCR analysis of H3 acetylation at MYC in ER-EB 2 . 5 cells minus or plus β-estradiol ( ± EBNA2 ) . Precipitated DNA was analysed using primer sets located at the main EBNA2-bound enhancers . For the −556 SE analysis included a region where decreased interactions were observed ( 1 ) and the −556 main peak where increased interactions were observed ( 3 ) ( see Figure 2 ) . The signal at a control region not bound by EBNA2 ( used for 3C analysis in Figure 2 ) was used as a negative control for binding ( Con ) . Mean percentage input signals , after subtraction of no antibody controls , are shown ± standard deviation for two independent ChIP experiments . ( B ) ChIP-QPCR analysis of BRG1 binding at MYC in ER-EB 2 . 5 cells minus or plus β-estradiol ( ± EBNA2 ) . ( C ) Western blot analysis of BRG1 expression in GM12878 transiently transfected with control or BRG1-specific siRNAs . Actin was used as a loading control . ( D ) Chromosome conformation capture analysis of the interaction of EBNA2-bound upstream enhancerswith the MYC promoter in control and BRG1 siRNA transfected GM12878 cells . Results show the mean ± standard deviation of signals from duplicate PCRs . DOI: http://dx . doi . org/10 . 7554/eLife . 18270 . 009 We previously demonstrated that repression of BCL2L11 was associated with EBNA3A and 3C-specific binding to the BCL2L11 promoter ( McClellan et al . , 2013 ) . However , ChIP-sequencing using an antibody that precipitates all EBNA3 proteins also revealed the presence long-range EBNA3 binding sites at the BCL2L11 locus ( Figure 5A ) ( McClellan et al . , 2012 , 2013 ) . These include three major sites upstream of BCL2L11 ( up to −374 kb ) that lie within the neighbouring acyl-CoA oxidase-like gene ACOXL and three sites 310 to 587 kb downstream of BCL2L11 ( Figure 5A ) . These elements represent new putative BCL2L11 enhancers that we designated enhancers 1–6 . Enhancers 1 , 4 , and 6 are predicted as SEs in blood-derived primary cells or cell-lines ( http://bioinfo . au . tsinghua . edu . cn/dbsuper/ [Khan and Zhang , 2016] ) indicative of a functional role . In fact , enhancer 4 is predicted to have SE function in 30 different cell types , 22 of which are blood-derived primary or cancer cells ( Supplementary file 2 ) . Enhancer 4 may therefore play a key role in BCL2L11 control in blood cells . EBNA3 binding sites are also present in the ACOXL promoter and the adjacent BUB1 promoter . 10 . 7554/eLife . 18270 . 010Figure 5 . EBNA3A and 3C repress BCL2L11 by inactivating a long-range enhancer hub . ( A ) ChIP-sequencing reads for EBNA3A/3B/3C at the BCL2L11 locus in Mutu III BL cells . The major EBNA3-bound sites are numbered 1–6 and their location relative to the transcription start site of the BCL2L11 promoter is indicated . Binding peaks at the ACOXL ( Ap ) and BCL2L11 promoters ( Bp ) are also indicated . ChIP-QPCR analysis of EBNA3A binding ( B ) EBNA3B binding ( C ) and EBNA3C binding ( D ) in the EBV-negative BL31 BL cell-line infected with wild-type recombinant EBV ( BL31 wtBAC2 ) . Precipitated DNA was analysed using primer sets located at EBNA3A/3B/3C binding sites ( binding at the BCL2L11 promoter has been previously characterised [McClellan et al . , 2013] ) . The signal at the transcription start site of PPIA was used as a negative control for binding ( − ) . The previously characterised CTBP2 binding site was used as a positive control for EBNA3A and EBNA3C binding ( + ) . The RUNX3 superenhancer was used as a positive control for EBNA3B binding ( Gunnell et al . , 2016 ) . Mean percentage input signals , after subtraction of no antibody controls , are shown ± standard deviation for two independent ChIP experiments . ( E ) Chromosome conformation capture ( 3C ) analysis of BCL2L11 promoter interactions between enhancers 1–6 and the ACOXL promoter in the EBV-negative BL cell-line BL31 and in BL31 cells infected with wild-type recombinant EBV ( wt BAC2 ) , EBNA3A knock-out EBV ( EBNA3AKO ) , EBNA3B knock-out EBV ( EBNA3BKO ) and EBNA3C knock-out EBV ( EBNA3CKO ) . A control region ( C ) not bound by the EBNAs was also included in the analysis . Positive controls show amplification of a digested and ligated genomic PCR fragment library containing all ligation junctions . The expression status of BCL2L11 in each line is indicated on the right . The red asterisks indicates non-specific amplification products of incorrect size . DOI: http://dx . doi . org/10 . 7554/eLife . 18270 . 01010 . 7554/eLife . 18270 . 011Figure 5—figure supplement 1 . ACOXL is repressed by EBNA3C . LCLs expressing HT-EBNA3C were grown in the presence of HT for 25 days ( EBNA3C on ) and then HT was washed off and cells cultured in its absence for 21 days ( EBNA3C off ) . Cells then either had HT re-added and were cultured for up to another 14 days to re-instate EBNA3C function or were maintained in HT for a further 10 days to maintain EBNA3C inactivity . ( A ) ACOXL mRNA levels were determined by RT-PCR and signals were normalised to GAPDH and expressed relative to the expression level at the start of the experiment . Results show the mean ± standard deviation for duplicate QPCR samples from a representative experiment . ( B ) ADAMDEC1 mRNA analysis . ADAMDEC1 ( McClellan et al . , 2012 ) is a known EBNA3C repressed gene and serves as a control for the EBNA3C-HT withdrawal and add back . DOI: http://dx . doi . org/10 . 7554/eLife . 18270 . 01110 . 7554/eLife . 18270 . 012Figure 5—figure supplement 2 . Additional BCL2L11 chromosome conformation capture controls . ( A ) EBNA3A/3B/3C binding at the BCL2L11 locus as in Figure 5 showing the locations of the primers used for 3C analysis . Primer design is unidirectional . ( B ) 3C analysis in BL31 cells to examine BCL2L11 promoter interactions with additional intervening control regions where there is no EBNA3A or EBNA3C binding ( C2 , C3 , C4 ) . ( C ) Control 3C analysis using BL31 cell chromatin that was digested but incubated in the absence of ligase ( -ligase ) . Analysis was performed using primers to detect enhancer-promoter and promoter-promoter interactions as in Figure 5 and using the additional control primers . Red asterisks indicate the position of non-specific PCR products of the incorrect sizes ( verified by sequencing ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18270 . 012 ChIP-QPCR using specific antibodies demonstrated that only EBNA3A and EBNA3C bind to BCL2L11 long-range elements ( Figure 5B–D ) , consistent with BCL2L11 silencing through the combined actions of EBNA3A and EBNA3C , but not EBNA3B ( Anderton et al . , 2008 ) . Only EBNA3C bound at significant levels to ACOXL sites , with most binding at enhancers 2 and 3 ( Figure 5B–D ) . Although ACOXL was not expressed in most B cell-lines we examined , we found that inactivation of EBNA3C in an LCL expressing a conditionally-active EBNA3C-hydroxytamoxifen fusion protein led to increased ACOXL expression ( Figure 5—figure supplement 1 ) . This points to a role for EBNA3C as a repressor of ACOXL . To determine whether any of the long-range EBNA3A/3C binding sites interact with the BCL2L11 promoter , we performed 3C analysis in a previously described BL cell-line series ( BL31 ) ( Anderton et al . , 2008 ) . BL31 cells derive from an EBV-negative BL and have been used to study the effects of EBV and EBNA3A , 3B and 3C on cellular gene expression by infecting them with recombinant wild-type or knock-out EBVs ( Anderton et al . , 2008; White et al . , 2010; McClellan et al . , 2013 ) . The co-operative repression of BCL2L11 by EBNA3A and EBNA3C was first described and characterised in this cell line series ( Anderton et al . , 2008; Paschos et al . , 2012 ) , providing an excellent background in which to study the effects of these EBV repressors on BCL2L11 enhancer-promoter interactions . Using the BCL2L11 promoter as bait , we found that in uninfected BL31 cells ( where BCL2L11 is expressed ) , the ACOXL promoter and all long-range elements ( apart from enhancer 1 within ACOXL ) interacted with the BCL2L11 promoter ( Figure 5E ) . An active enhancer-promoter hub encompassing ACOXL therefore directs BCL2L11 expression in these cells . 3C did not detect any interactions between the BUB1 promoter and the BCL2L11 promoter ( data not shown ) . We detected no interactions between the BCL2L11 promoter and four intervening control regions not bound by EBNA3A or EBNA3C , indicating that the enhancer-promoter interactions we detected were specific . ( Figure 5E and Figure 5—figure supplement 2 ) . In contrast to EBV-negative BL31 cells , in BL31 cells infected with either wild-type EBV or EBNA3B knock-out EBV ( where BCL2L11 is repressed ) , we observed a loss of all promoter-enhancer interactions ( Figure 5E ) . Accordingly , in cells infected with EBNA3A or EBNA3C knock-out viruses , BCL2L11 was expressed and all enhancer-promoter interactions were preserved . BCL2L11 silencing by EBNA3A and EBNA3C is therefore associated with the inactivation of an active enhancer-promoter hub encompassing ACOXL . We next investigated whether the disruption of enhancer-promoter interactions by EBNA3A and EBNA3C was also associated with enhancer chromatin inactivation . Using ChIP-QPCR we examined the binding of the PRC2 H3K27 methyltransferase EZH2 across EBNA3-bound sites . Consistent with previous observations ( Paschos et al . , 2012 ) , we found that EZH2 was associated with the BCL2L11 promoter in cells infected with viruses expressing EBNA3A and EBNA3C ( wt BAC2 and EBNA3B KO , Figure 6A ) . We also detected EZH2 binding to all enhancers targeted by EBNA3A and EBNA3C and to the ACOXL promoter ( Figure 6A ) . These results are consistent with enhancer inactivation either initiated by , or resulting in , PRC-associated chromatin silencing . EBNA3A and EBNA3C have been shown to induce the deposition of the PRC silencing mark H3K27me3 across the BCL2L11 promoter ( Paschos et al . , 2009 ) . H3K27me3 ChIP-seq data from GM12878 cells ( that express EBNA3A and EBNA3C ) however , also demonstrates the presence of characteristically broad domains of H3K27me3 that coincide with the locations of the EBNA3-bound BCL2L11 enhancers ( Figure 6—figure supplement 1 ) . In fact a large H3K27me3 domain encompasses the entire ACOXL gene and the BCL2L11 promoter ( Figure 6—figure supplement 1 ) . The corresponding absence of active H3K27ac marks in these regions is consistent with silencing of both genes . 10 . 7554/eLife . 18270 . 013Figure 6 . EZH1/2 activity is required for the disruption of the BCL2L11 and ACOXL enhancer hub . ( A ) ChIP-QPCR analysis of EZH2 binding in the BL31 cell line series used in Figure 5 . The expression status of BCL2L11 in each cell line is shown . ( B ) RT-QPCR analysis of BCL2L11 mRNA expression in EBV negative BL31 cells or BL31 cells infected with wild-type recombinant EBV ( BL31 wtBAC2 ) treated with the EZH2 inhibitor UNC1999 for 8 hr . Signals were normalised to GAPDH mRNA levels and expressed as fold increase compared to untreated cells . ( C ) BCL2L11 mRNA expression in BL31 and BL31 wtBAC2 cells treated with UNC1999 for 18 hr . ( D ) Caspase 3/7 activity in BL31 or BL31 wtBAC2 cells treated with UNC1999 for 8 hr . Caspase signals shown are corrected for the number of live cells . ( E ) Caspase 3/7 activity in cells treated for 18 hr . ( F ) Chromosome conformation capture analysis of BCL2L11 promoter interactions between enhancers 1–6 and the ACOXL promoter in BL31 wtBAC2 cells treated with UNC1999 for 24 hr . Primers are as in Figure 5 and Figure 5—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 18270 . 01310 . 7554/eLife . 18270 . 014Figure 6—figure supplement 1 . EBNA3A and EBNA3C-bound enhancers at the BCL2L11/ACOXL locus are within H3K27me3 repressed domains . ( A ) EBNA3A/3B/3C ChIP-sequencing from Mutu III BL cells as in Figure 5 . ( B ) H3K27me3 ChIP-sequencing signals in GM12878 cells ( ENCODE ) . Boxes show the H3K27me3 domain encompassing the entire ACOXL gene and the BCL2L11 promoter ( and enhancers 1–3 ) and the domains that encompass enhancers 4 , 5 and 6 . ( C ) H3K27Ac ChIP-sequencing signals in GM12878 ( ENCODE ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18270 . 014 To determine whether BCL2L11 silencing by EBV could be reversed through the loss of EZH2 activity , we treated EBV-negative BL31 cells and EBV-infected BL31 wt BAC2 cells with the EZH1/2 inhibitor UNC1999 . In EBV-infected BL31 cells , BCL2L11 mRNA expression increased up to 3 . 6- and 5 . 4-fold after 8 and 18 hr treatment with UNC1999 , consistent with the inhibition of PRC2-mediated gene repression ( Figure 6B and C ) . In contrast , treatment of EBV-negative BL31 cells with UNC1999 resulted in only 1 . 8 to 2 . 7 fold increases in BCL2L11 expression after 8 and 18 hr , respectively ( Figure 6B and C ) . Since BCL2L11 is not repressed by EBV in BL31 cells , these small increases may reflect the fact that even though the gene is expressed , further de-repression can be achieved through EZH1/2 inhibition . Consistent with increased BCL2L11 expression and the pro-apoptotic function of BCL2L11 , EBV-infected BL31 cells treated with UNC1999 also displayed large increases in Caspase 3/7 activity ( Figure 6D and E ) . Caspase 3/7 activity was very low in EBV-negative BL31 cells but increased slightly in the presence of UNC1999 , consistent with the smaller increases in BCL2L11 expression . These data therefore indicate that PRC-mediated silencing of BCL2L11 by EBNA3A and EBNA3C can be reversed by EZH1/2 inhibition and results in the induction of apoptosis . To assess the effect of EZH2 inhibition on BCL2L11 promoter interactions , we performed 3C analysis of the BCL2L11 locus in EBV-infected BL31 cells following UNC1999 treatment . We found that treatment with 5 or 10 µM UNC1999 led to increased interactions between the BCL2L11 promoter and all enhancers , and between the BCL2L11 and ACOXL promoters ( Figure 6F ) . These data indicate that EZH1/2 activity is required for the inactivation of BCL2L11 enhancers and BCL2L11 silencing in EBV-infected cells . We conclude that the increased cell survival that results from EBV EBNA3A and EBNA3C silencing of BCL2L11 involves the recruitment of EZH2 and the inactivation of a long-range active enhancer hub encompassing the neighbouring ACOXL gene .
B-cell immortalisation by EBV plays a central role in the development of numerous B-cell lymphomas and is required for the persistence of EBV in infected hosts . EBNA2 is essential for B-cell immortalisation and the continuous growth of EBV-infected cells ( Cohen et al . , 1989; Kempkes et al . , 1995 ) . Upregulation of MYC by EBNA2 ( Kaiser et al . , 1999 ) plays a key role in stimulating B-cell proliferation early in infection , promoting immortalisation . Our data now show that EBV manipulates MYC enhancer function to drive tumourigenesis , inducing directional remodelling of enhancer-promoter interactions over 3 Mbs . EBV promotes upstream promoter-enhancer interactions and decreases downstream interactions . The MYC enhancer interaction landscape in EBV-infected cells is therefore distinct from leukaemia cells , where downstream enhancers are the major controllers of MYC transcription ( Herranz et al . , 2014; Shi et al . , 2013 ) . Whether MYC activation by EBNA2 also plays a role in predisposing immortalised infected cells to the MYC translocations that characterise EBV-positive BL has not been explored . In fact , the role of EBV in the pathogenesis of BL remains an enigma , since the defining feature of BL is a MYC/IG translocation rather than the presence of EBV . It has been proposed that EBV may contribute to BL simply by providing a pool of cells undergoing deregulated growth , in which a genetic accident becomes more likely . Most evidence however , now points to a role for EBV in providing a survival advantage to cells that express high-levels of MYC by repressing BCL2L11 ( see below ) . Pro-survival events presumably arise through genetic and epigenetic changes induced by non-viral mechanisms in EBV-negative BLs . Given that the BCL2L11 repressors EBNA3A and EBNA3C are expressed initially in growth-transformed B cells , but not in BL cells ( which express only EBNA1 ) , BCL2L11 repression is likely an early event that prevents apoptosis driven by the initial activation of MYC by EBNA2 . Since H3K27me3-mediated BCL2L11 repression leads to CpG methylation at the BCL2L11 promoter ( Paschos et al . , 2009 ) this ‘hit-and-run’ silencing event would provide a long-lived survival advantage to cells in which a MYC/IG translocation may subsequently arise . EBV increases the likelihood of a translocation event through the upregulation of AID by EBNA3C ( Kalchschmidt et al . , 2016a ) . The generation of double-strand DNA breaks as a result of aberrant AID activity is strongly implicated in the genesis of MYC/IG translocations ( Robbiani et al . , 2008; Dorsett et al . , 2007 ) . AID preferentially targets active enhancer regions , so our data implicate the activation of upstream MYC enhancers by EBNA2 in predisposing these regions to AID-induced breakpoints in EBV-infected cells . Interestingly , MYC-IG breakpoints in EBV-positive eBL carrying the common t ( 8;14 ) translocation are predominantly located far upstream of MYC whereas breakpoints in sBL are evenly distributed between the promoter region , the first exon and the first intron ( Neri et al . , 1988; Shiramizu et al . , 1991; Busch et al . , 2007; Joos et al . , 1992; Pelicci et al . , 1986 ) . When we examined the location of MYC-IG breakpoints in eBLs in light of the EBV-induced changes we detected in the chromatin region upstream of MYC , we found that previous studies mapped the majority of EBV-positive eBL breakpoints to upstream of the EcoRI site 7 kb upstream of MYC ( Neri et al . , 1988; Shiramizu et al . , 1991; Pelicci et al . , 1986 ) . Only eight EBV-positive eBLs have had their 5’ breakpoints further mapped or sequenced ( Haluska et al . , 1986; Neri et al . , 1988; Joos et al . , 1992 ) . These seven eBL cell-lines and one eBL biopsy sample have translocation junctions −215 to −46 kb upstream of MYC ( Haluska et al . , 1986; Neri et al . , 1988; Joos et al . , 1992 ) . These upstream eBL breakpoints therefore map to the vicinity of the EBNA2-bound −186/168 enhancer . This region has not previously been studied as a long-range MYC control region , but has SE properties in colorectal carcinoma and diffuse large B cell lymphoma cell-lines ( Supplementary file 1 ) . Our data therefore highlight a role for the −186/168 enhancer in the remodelling of upstream MYC chromatin that may create a ‘hotspot’ for eBL breakpoints . In fact all upstream breakpoints ( including those more than 7 kb upstream of MYC that have not been fully mapped ) likely fall within the upstream region that displays increased promoter looping in the presence of EBV . We therefore propose that the EBNA2-directed activation and remodelling of MYC upstream chromatin may increase the susceptibility of this region to a translocation event initiated by the off-target activity of activation-induced cytidine deaminase ( AID ) . In contrast to eBL , sporadic BL t ( 8:14 ) breakpoints cluster in two regions much further downstream that include the MYC promoter ( −400 bp to + 150 bp ) and a region immediately downstream ( +420 bp to +1 . 2 kb ) ( Busch et al . , 2007 ) . This suggests that specific chromatin changes or other factors contribute to the localisation of MYC breakpoints in different regions in the absence of EBV . Our data also demonstrate that the mechanism of MYC activation by EBNA2 through upstream enhancers involves the recruitment of the SWI/SNF ATPase BRG1 . Previous studies have demonstrated that AML cell growth is dependent on MYC activation by SWI/SNF ( Shi et al . , 2013 ) . This dependency appears to result from a requirement for BRG1 for the interaction of the +1 . 8/1 . 9 Mb ( +1 . 7 Mb in mouse ) MYC enhancer with the promoter ( Shi et al . , 2013 ) . In AML cells , BRG1 knockdown decreased downstream enhancer interactions and increased upstream enhancer interactions , suggesting some directionality in the effects of BRG1 on enhancer looping . In EBV-infected cells however , we find that BRG1 is required for the maintenance of upstream enhancer-promoter interactions . Our data are therefore consistent with a model where BRG1-dependent chromatin remodelling is required for MYC enhancer-promoter interactions . The specificity of BRG1 dependence however , is determined by which MYC enhancers are active in a particular cell-type or context . Repression of BCL2L11 is a key strategy employed by EBV to circumvent MYC-driven apoptosis and promote survival ( Thorley-Lawson and Allday , 2008 ) . This is consistent with observations that the loss of a single BCL2L11 allele accelerates lymphoma development in Eµ-MYC transgenic mice ( Egle et al . , 2004 ) . The tumour suppressor role of BCL2L11 is also supported by its deletion in 40% of mantle cell lymphomas ( Tagawa et al . , 2005 ) , silencing through CpG promoter methylation in natural killer cell lymphomas ( Küçük et al . , 2015 ) and its targeting by the oncogenic miR-32 and miR17-92 microRNAs ( Ventura et al . , 2008; Koralov et al . , 2008; Ambs et al . , 2008 ) . We now show that BCL2L11 repression in B cells results from the inactivation of multiple long-range enhancers . Our previous analysis focused on the low-level binding of the EBV repressors EBNA3A and EBNA3C to the BCL2L11 promoter ( McClellan et al . , 2013 ) . It is now clear that the BCL2L11 promoter is controlled through a long-range regulatory hub that is inactivated by EBNA3A and EBNA3C through a process that results in the recruitment of the PRC2 methyltransferase EZH2 and is dependent on EZH1/2 activity . No direct interaction between EBNA3A and EBNA3C and components of the PRC1 or PRC2 transcriptional repressors has been reported to date . PRC-dependent and H3K27me3-associated gene silencing by EBNA3A and EBNA3C has however been convincingly demonstrated at multiple gene loci ( Skalska et al . , 2010; Harth-Hertle et al . , 2013; Kalchschmidt et al . , 2016b; McClellan et al . , 2012; 2013 ) . The mechanism of PRC recruitment by these EBV repressors remains unclear . In fact , time-course studies have shown that the loss of active chromatin marks from gene promoters and enhancers repressed by EBNA3A and EBNA3C precedes the binding of PRCs and the deposition of H3K27me3 ( Harth-Hertle et al . , 2013; Kalchschmidt et al . , 2016b ) . PRC recruitment and H3K27me3 deposition may therefore be a secondary and perhaps default event . At the BCL2L11 locus , we have identified enhancers that control BCL2L11 expression in human B cells , that are manipulated by EBV to repress BCL2L11 in infected B cells . These enhancers have potential roles in BCL2L11 control in normal and malignant cells ( Supplementary file 2 ) . Interestingly , a murine BCL2L11 enhancer located 117 kb upstream and within ACOXL ( but distinct from the EBV-targeted enhancers described here ) was recently shown to be disrupted by binding of the TRIM33 transcription cofactor to prevent BCL2L11-induced apoptosis in B lymphoblastic leukaemia ( Wang et al . , 2015 ) . Importantly , we showed that EBV-induced repression of BCL2L11 through the disruption of enhancer-promoter interactions was reversed by EZH1/2 inhibition and resulted in increased apoptosis . This provides a therapeutic rationale for the use of EZH2 inhibitors in the treatment of EBV-positive lymphomas where BCL2L11 is repressed . Our work also revealed that the poorly characterised ACOXL gene is a target for repression by EBNA3C . Interestingly , ACOXL contains two risk loci for CLL ( Di Bernardo et al . , 2008; Berndt et al . , 2013 ) , one of which ( rs13401811 [Berndt et al . , 2013] ) maps to a smaller EBNA3A/B/C binding peak between enhancers 1 and 2 that we did not characterise in this study . This raises the possibility that this polymorphism deregulates the function of a BCL2L11 enhancer . ACOXL is also downregulated in prostate cancer tissues ( O'Hurley et al . , 2015 ) , but further studies are required to confirm any potential role in tumourigenesis . In summary , we show that EBV-directed lymphomagenesis involves the hijacking of long-range enhancer hubs at MYC and BCL2L11 ( Figure 7 ) . MYC enhancer activation by EBV may contribute to the genesis of MYC translocations in BL . Enhancer-mediated control of BCL2L11 may be exploited in other tumourigenic contexts to manipulate cell survival . 10 . 7554/eLife . 18270 . 015Figure 7 . Model showing the mechanism of MYC activation and BCL2L11 repression by EBV transcription factors . ( A ) In uninfected B cells , MYC promoter interactions with downstream enhancers dominate . MYC activation on EBV infection by the EBV TF EBNA2 occurs through the activation of three major clusters of upstream enhancers at −556 , −428 and −186/168 kb ( indicated by black boxes ) . This is associated with increased H3K27ac and BRG1 binding . EBNA2 promotes interactions between the MYC promoter and these upstream enhancers and reduces interactions with downstream enhancers . As part of this three-dimensional MYC enhancer reorganisation , EBNA2 also increases interactions between CTCF-bound regions ( asterisks ) in the −556 kb super-enhancer . ( B ) BCL2L11 is repressed on EBV-infection by the EBV repressors EBNA3A and EBNA3C through the inactivation of multiple enhancers in regulatory hub encompassing the ACOXL gene . Enhancer inactivation is associated with PRC2 ( EZH2 ) binding , increased H3K27me3 and the loss of enhancer-promoter interactions . Arrows indicate transcription start sites . Genes are indicated as expressed ( + ) or repressed ( − ) . ACOXL is repressed or expressed at low-level ( ± ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18270 . 015
All cell lines were routinely passaged twice-weekly in RPMI-1640 media ( Invitrogen , UK ) containing 10% Foetal Bovine Serum , Penicillin and Streptomycin ( Invitrogen ) . The EBV-positive latency III BL cell line Mutu III ( clone 48 ) derives from the Mutu I latency I BL cell-line ( Gregory et al . , 1990 ) . Mutu I cells display the ‘latency I’ restricted form of EBV gene expression characterised by the expression of only EBNA1 . Mutu III cell clones arose spontaneously during culture of Mutu I cells and display an expanded latent gene expression pattern ( latency III ) . The EBV-immortalised LCL GM12878 is an ENCODE Tier 1 cell line obtained from the Coriell Cell Repositories ( Camden , New Jersey ) ( RRID: CVCL_7526 ) . The EBV-negative BL31 cell line series infected with wild-type recombinant EBV bacmids or EBNA 3A , 3B and 3C knock-out and revertant bacmids ( kindly provided by Prof M . Allday ) were cultured with the appropriate selection and supplements , as previously described ( Anderton et al . , 2008 ) . The EBV-immortalised ER-EB 2 . 5 LCL , expressing a conditionally-active oestrogen receptor ( ER ) -EBNA2 fusion protein , was provided by Prof B . Kempkes , and was cultured in the presence of β-estradiol ( Kempkes et al . , 1995 ) . For β-estradiol withdrawal and add back experiments , ER-EB 2 . 5 cells were incubated in the absence of β-estradiol for 4 days , and 1 µM β-estradiol was re-added for 17 hr , prior to cell harvest . The LCL expressing conditionally-active EBNA3C is infected with recombinant EBV expressing EBNA3C fused at the C terminus to a 4-hydroxytamoxifen ( HT ) -sensitive murine oestrogen receptor ( LCL 3CHT ) and was provided by Prof M . Allday ( Skalska et al . , 2010 ) . For 4-hydroxytamoxifen withdrawal and add back experiments LCL 3CHT cells were initially cultured in the presence of 400 nM of 4-hydroxytamoxifen ( Sigma , UK ) for 25 days , HT was then washed off and cells cultured for 21 days . HT was then either re-added or cells grown for a further 10 days in HT . For EZH1/2 inhibition , UNC1999 ( Sigma ) was added to BL31 wtBAC2 cells seeded at a density of 5 × 105 cells/ml , and cells harvested after 24 hr for mRNA and chromosome conformation capture . All cell-lines were verified as mycoplasma free . Primary resting B cells were isolated from blood from fresh apheresis cones obtained from the NHSBT under the ethically approved study 14/WM/0001 . The Peripheral blood mononuclear cells were isolated by density centrifugation on lymphoprep ( Axis Shield , UK ) and the B cells were subsequently purified by positive isolation using pan-CD19 Dynabeads ( Dynal , ThermoFisher , UK ) . The Dynabeads were removed from the purified B cells by incubation with Detachabead ( Dynal , ThermoFisher ) . Purified B cells were incubated with 2089 EBV at an MOI of 100 for 1 hr at 37°C and the unbound virus was washed off . One million uninfected purified B cells were cultured in 4 ml of medium supplemented with 50 ng/ml soluble mega CD40L ( Enzo , UK ) and 50 ng/ml IL4 . The infected and CD40L-stimulated B cells were grown in RPMI , 10% FBS and supplemented with penicillin/streptomycin and glutamine . 200 nM ON-TARGETplus Human SMARCA4 ( BRG1 ) siRNA ( Dharmacon , GE Healthcare , UK; L-010431-00-0005 ) or ON-TARGETplus siRNA non-targeting siRNA #1 ( Dharmacon , GE Healthcare; D-001810-01-05 ) were transfected into 5 × 106 GM12878 cells resuspended in buffer T cells using the Neon transfection and 1 pulse of 1300 V for 30 msec . Following transfection the cells were further incubated for 72 hr in normal media without antibiotics . BL31 and BL31 wtBAC2 cells ( 100 µl ) were seeded into 96 well plates at a density of 20 , 000 cells/well and cultured for 8 or 18 hours in the presence or absence of UNC1999 . An equal volume of Caspase-Glo 3/7 Assay reagent ( Promega , UK ) was added , and cells were incubated for 30 min at room temperature . Luminescence was measured using a Glowmax multi detection system ( Promega ) . Previously published EBNA2 and EBNA3A/B/C Mutu III ChIP-sequencing data ( McClellan et al . , 2013 ) are available via GEO accession number GSE47629 and EBNA2 GM12878 data via accession number GSE76869 ( Gunnell et al . , 2016 ) . Note that Mutu III cells have an unmapped MYC-IG translocation , but sequence reads are mapped to the intact MYC locus . EBNA2 binding sites in Mutu III cells are also detected in GM1278 cells and/or in the IB4 EBV-infected LCL ( Zhao et al . , 2011b ) so the integrity of binding to these sites seems to be maintained despite their translocation . EBNA3A , EBNA3B and EBNA3C were precipitated as described previously using antibodies specific for each EBNA ( McClellan et al . , 2013 ) . ChIP for EZH2 was carried out using 4 µg mouse monoclonal antibody ( Millipore , UK; 17–662 ) , for BRG1 using 5 µg rabbit polyclonal antibody ( Santa Cruz sc-10768 ( H-88 X ) and for diacetylated Histone H3 using 5 µg rabbit polyclonal antibody ( Millipore 06–599 ) . Quantitative real-time PCR was carried out using specific and control primers ( Supplementary file 4 ) and the standard curve method as previously described ( McClellan et al . , 2013 ) . Immunoblotting was carried out as described previously ( Bark-Jones et al . , 2006; Gunnell et al . , 2016 ) using anti-BRG1 ( Santa Cruz biotechnology , Germany; sc-17796 ) and anti-actin antibodies ( Sigma; A-2066 ) . Chromosome conformation capture assays were carried out essentially as described previously ( McClellan et al . , 2013 ) using HindIII-HF ( New England Biolabs ) with baits consisting of either an 11 kb fragment encompassing the MYC promoter , an 18 . 2 kb fragment encompassing the −556 kb MYC SE 3’ CTCF site or a 10 . 8 kb fragment encompassing the BCL2L11 promoter . Samples were then analysed by semi-quantitative PCR using unidirectional ( rather than head to head ) primers designed to amplify across ligation junctions ( Naumova et al . , 2012 ) . Positive control PCRs across ligation junctions were carried out using libraries containing genomic DNA fragments representing expected ligation products . Positive control library fragmentswere either synthesised ( Genestrings , Life Technologies ) or generated from genomic PCR fragments covering the restriction sites of interest that were then digested and ligated . Titrations of positive control DNA were analysed by PCR using the same primers used for chromosome conformation capture to determine the linear range of the assay prior to analysis of the chromosome conformation capture library . Quantitation was carried out using a LiCOR imaging system following agarose gel electrophoresis and staining with GelRed ( Biotium , Fremont , California ) . Interaction frequencies were determined by dividing the chromosome conformation capture chromatin library signal for each ligation junction product by the signal obtained for a positive control sample from within the linear range . Positive control and 3C PCRs were carried out and analysed in duplicate . MYC promoter interacting fragments were captured using a 2 . 7 kb NlaIII fragment encompassing the MYC promoter , prior to further digestion by DpnII and 4C-seq was carried out using a previously described protocol ( Splinter et al . , 2012 ) . Cells were passed through a 70 µm filter to obtain a single cell preparation . 1 × 107 cells were then fixed in 2% formaldehyde in the presence of 10% FCS for 10 min at room temperature . The reaction was quenched with 0 . 125 M glycine , and cells collected by centrifugation at 400 g for 8 min at 4°C . The pellet was resuspended in 0 . 5 ml lysis buffer ( 50 mM Tris-HCl , pH 7 . 5; 150 mM NaCl , 0 . 5 mM EDTA , 0 . 5% NP-40 , 1% Triton X-100 ) with freshly added complete protease inhibitors ( Roche , UK ) , and lysed on ice for 10 min . The nuclei were collected by centrifugation at 750 g for 5 min at 4°C , then resuspended in 0 . 5 ml of 1 . 2X CutSmart Buffer ( New England Biolabs ) containing 0 . 3% SDS and incubated for 1 hr at 37°C , while shaking at 900 rpm . Triton X-100 was then added to the nuclei to give a final concentration of 3% and the samples incubated for 1 hr at 37°C , with shaking . 200 U NlaIII ( New England Biolabs , UK ) were added to the nuclei and the samples incubated for 4 hr at 37°C , while shaking at 900 rpm . The reaction was supplemented with a further 200 U NlaIII and incubated overnight at 37°C with shaking at 900 rpm . A further 200 U NlaIII was added , followed by an additional 4 hr incubation at 37°C while shaking at 900 rpm . The digestion reaction was stopped by incubation at 65°C for 20 min . The sample was then diluted to 7 mls with 1X ligation buffer ( Roche ) . 50 U DNA ligase ( Roche ) were added to the sample , and the reaction was incubated overnight at 16°C . 300 µg of Proteinase K ( Roche ) were added to the sample and the reaction incubated at 65°C overnight . RNA was removed by incubation with 300 µg of RNAse for 45 min at 37°C . Following two rounds of phenol-chloroform extraction , DNA was ethanol precipitated prior to resuspension in 150 µl of 10 mM Tris-HCL pH7 . 5 at 37°C . The samples were then diluted to 500 µl with 1X DpnII buffer ( New England Biolabs ) . 50 U DpnII ( New England Biolabs ) were added , followed by an overnight incubation at 37°C whilst shaking at 900 rpm . The digestion reaction was stopped by incubation at 65°C for 20 min . Samples were diluted in 14 mls 1X ligation buffer . 100 U DNA ligase were added and the reaction incubated overnight at 16°C . The samples were then ethanol precipitated in the presence of glycogen at −80°C , prior to purification over a QIAquick PCR purification Kit ( Qiagen ) , and eluted in 150 µl 10 mM Tris-HCL pH7 . 5 . Fragments captured by the bait region were then amplified by inverse PCR using primers designed to amplify outwards from the bait region ( Supplementary file 3 ) . Individual forward primers included a 5’ overhang of the Illumina sequence adapter P5 and a unique ‘barcode’ sequence and encompassed the primary NlaIII restriction site of the MYC promoter bait . Common reverse primers included a 5’ overhang of the Illumina sequence adapter P7 and were designed to bind less than 100 bps from the secondary restriction site ( DpnII ) in the MYC promoter bait . PCR was performed using Expand Long Template Polymerase ( Roche ) , with 3 . 2 µg of template and 1 . 12 nmol of the P5 and P7 primers for 2 min at 94°C , 10 s at 94°C , 1 min at 55°C , 3 min at 68°C for 29 cycles followed by 5 min at 68°C . 16 individual PCR reactions were carried out for each sample . The reactions were pooled and purified to separate the unused adapter primers from the PCR product using the High Pure PCR Product Purification Kit ( Roche ) . Replicate chromatin samples were generated from the same cell batch and processed separately through each stage . Samples were combined for multiplex 100bp single-end Illumina HiSeq sequencing . Four ER-EB 2 . 5 LCL samples were sequenced in a single lane using barcodes TSBC01 , TSBC02 , TSBC10 , and TSBC20 . Six primary B cell samples were combined in a single lane using barcodes TSBC02 , TSBC04 , TSBC05 , TSBC06 , TSBC07 , and TSBC12 . Initial data extraction was performed using a custom script ( available as a source code file ) to strip out and separate embedded barcodes , and to remove reads from the restriction fragment immediately adjacent to the bait , where no digestion had occurred . Reads were mapped to the Homo sapiens GRCh37 genome assembly using bowtie 2 v2 . 2 . 7 using default parameters , and were filtered to retain only those uniquely mapping reads with MAPQ >=42 . For absolute quantitation the genome was divided into 10 kb windows and quantitated with read counts normalised to the data set with the highest read coverage . For relative quantitation the genome was divided into windows each of which contained a total of 50 , 000 reads across all samples . Read counts for each region were then quantitated in each individual dataset , and the raw counts were corrected for the total read count to account for differing depths of sequencing . Interaction count differences were calculated by subtracting the normalised counts for one dataset from another . 4C-sequencing data are available via GEO accession GSE82150 . | The Epstein-Barr virus is a common virus that can cause mild illnesses as well as more severe diseases . The virus infects white blood cells called B cells and can drive the development of blood cancers , including Burkitt’s and Hodgkin’s lymphoma . In these cancers , the infected B cells multiply rapidly and continuously , free from the controls that exist in normal cells . This occurs because the Epstein-Barr virus can both switch on genes in the B cells that drive growth and turn off other genes that trigger cell death . To achieve this , the virus hijacks DNA regions called enhancers that are situated far away from the genes that they control . However , it was not clear how this hijacking process works . Wood et al . set out to determine how the Epstein-Barr virus uses enhancers to switch on MYC , a gene that is a key driver of lymphoma development , and switch off BCL2L11 , a gene that normally triggers cell death and prevents lymphoma . Using human B cells that had been infected with the Epstein-Barr virus , Wood et al . showed that the virus completely reorganises the DNA loops that form between the MYC and BCL2L11 genes and their enhancers . These loops allow the enhancers to contact their associated gene in order to activate it . Wood et al . found that the Epstein-Barr virus switches on the MYC gene by altering how certain enhancers contact the gene . This may explain how the virus causes particular changes to the MYC gene that are found in Burkitt’s lymphoma . Wood et al . also discovered new enhancers that control the activity of the BCL2L11 gene . The Epstein-Barr virus prevents these enhancers from contacting and switching on BCL2L11 , thus blocking cell death . This “silencing” of BCL2L11 can be reversed by a specific drug that targets the silencing machinery used by the Epstein-Barr virus; such treatment led to the death of the infected cells . It is now important to carry out further studies that determine how the Epstein-Barr virus hijacks enhancers to control other genes that are associated with lymphoma . This will tell us more about how the virus drives lymphoma development and will help to identify new ways of targeting Epstein-Barr virus-infected cancer cells with specific drugs . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cancer",
"biology"
] | 2016 | MYC activation and BCL2L11 silencing by a tumour virus through the large-scale reconfiguration of enhancer-promoter hubs |
The interplay between bacterial antimicrobial susceptibility , phylogenetics and patient outcome is poorly understood . During a typhoid clinical treatment trial in Nepal , we observed several treatment failures and isolated highly fluoroquinolone-resistant Salmonella Typhi ( S . Typhi ) . Seventy-eight S . Typhi isolates were genome sequenced and clinical observations , treatment failures and fever clearance times ( FCTs ) were stratified by lineage . Most fluoroquinolone-resistant S . Typhi belonged to a specific H58 subclade . Treatment failure with S . Typhi-H58 was significantly less frequent with ceftriaxone ( 3/31; 9 . 7% ) than gatifloxacin ( 15/34; 44 . 1% ) ( Hazard Ratio 0 . 19 , p=0 . 002 ) . Further , for gatifloxacin-treated patients , those infected with fluoroquinolone-resistant organisms had significantly higher median FCTs ( 8 . 2 days ) than those infected with susceptible ( 2 . 96 ) or intermediately resistant organisms ( 4 . 01 ) ( p<0 . 001 ) . H58 is the dominant S . Typhi clade internationally , but there are no data regarding disease outcome with this organism . We report an emergent new subclade of S . Typhi-H58 that is associated with fluoroquinolone treatment failure . Clinical trial registration: ISRCTN63006567 .
Enteric ( typhoid ) fever , a systemic infection caused predominantly by the bacterium Salmonella enterica subspecies enterica serovar Typhi ( S . Typhi ) , remains one of the principal bacterial causes of febrile disease in low-income countries ( Parry et al . , 2002 ) . S . Typhi is a distinct , monophyletic lineage of S . enterica that is exquisitely adapted to cause disease only in humans ( Roumagnac et al . , 2006 ) , characterised by a non-specific fever with malaise and asymptomatic convalescent carriage ( Parry et al . , 2002 ) . There are an estimated 20–30 million new cases of enteric fever per year globally ( Crump and Mintz , 2010 ) , with the majority occurring in Asia , but there is an increasingly recognised burden of disease across sub-Saharan Africa . Antimicrobial resistance is a major global health challenge , and resistance against the most commonly used antimicrobials for treating enteric fever has evolved successively over the last 30 years . Ampicillin , chloramphenicol and trimethoprim-sulfamethoxazole were originally standard-of-care for enteric fever . However , multidrug resistance ( MDR ) against these agents began to emerge in the 1970s and 1980s ( Olarte and Galindo , 1973; Wain et al . , 2003 ) . Consequently , third-generation cephalosporins and fluoroquinolones ( FQs ) became the most clinically reliable drugs for treating enteric fever ( Kariuki et al . , 2015 ) , and were formally advocated by the World Health Organization ( WHO ) in 2003 ( World Health Organization , 2003 ) . S . Typhi isolates with acquired resistance against third-generation cephalosporins are rare ( Hendriksen et al . , 2015 ) , but S . Typhi exhibiting reduced susceptibility to FQs , induced by sequential mutations in the gene encoding a target protein ( gyrA ) , now dominate internationally ( Emary et al . , 2012; Kariuki et al . , 2010 ) . The global ascendency of S . Typhi strains with reduced susceptibility to FQs has been partly catalysed by the dissemination of a specific MDR lineage ( H58 ) across Asia and Africa ( Wong et al . , 2015 ) . These H58 strains are rapidly displacing other lineages , and strains with gyrA mutations may have a fitness advantage , even in the absence of antimicrobial exposure ( Baker et al . , 2013 ) . We have previously shown that protracted fever clearance times ( FCTs ) are associated with organisms with higher Minimum Inhibitory Concentrations ( MIC ) against FQs in enteric fever patients treated with ciprofloxacin and ofloxacin ( Parry et al . , 2011 ) . However , whilst the clinical efficacy of the older FQs in enteric fever is contentious , we have shown that the fourth-generation FQ , gatifloxacin , has remained efficacious for uncomplicated disease , even in patients infected with S . Typhi strains with reduced ciprofloxacin susceptibility ( MIC ≥0 . 125 μg/mL ) ( Pandit et al . , 2007; Koirala et al . , 2013; Arjyal et al . , 2011 ) . During a recent randomised controlled trial ( RCT ) comparing ceftriaxone and gatifloxacin , conducted in Nepal , we observed an increased number of treatment failures associated with FQ-resistant ( ciprofloxacin MIC>32 μg/ml ) S . Typhi , prompting the data safety and monitoring board to stop the trial ( Arjyal et al . , 2016 ) . Aiming to assess the molecular epidemiology of the infecting isolates and investigate how genotype may be related to treatment outcome , we performed whole genome sequencing ( WGS ) on the S . Typhi isolated during this trial , and after stratifying by genotype , we assessed clinical presentation and outcome .
We performed WGS on the 78 available S . Typhi isolates from patients in both RCT treatment arms ( gatifloxacin and ceftriaxone ) ( Supplementary file 1 ) . The resulting phylogeny , which incorporated reference sequence CT18 , indicated that the majority of isolates ( 65/78; 83 . 3% ) fell within the H58 lineage , while the remaining 13 ( 16 . 7% ) represented eight different lineages ( Figure 1 ) . All but four of the H58 strains contained the common DNA gyrase ( gyrA ) mutation in codon 83 ( S83F ) , which confers reduced susceptibility to FQs ( ciprofloxacin MIC; 0 . 125–0 5 μg/ml ) ( Parry et al . , 2010 ) . Nested within the S83F H58 group , but separated from the rest of the group by a branch defined by 30 SNPs , was an H58 subclade comprised of 12 isolates containing the S83F gyrA mutation , a mutation in gyrA at codon 87 ( D87N ) , and an additional mutation in the topoisomerase gene , parC ( S80I ) ( H58 triple mutant ) . Notably , these H58 triple mutants shared high MICs against ciprofloxacin ( ≥24 μg/ml ) . Further , an additional two non-H58 RCT isolates with ciprofloxacin MIC≥24 μg/ml had the S83F gyrA mutation , an alternative mutation at codon 87 ( D87V ) , the S80I parC mutation , and an A364V mutation in parE ( Figure 1 , Supplementary file 1 ) . Notably , none of the sequenced isolates harboured plasmid-mediated quinolone resistance genes ( PMQR ) or contained additional antimicrobial resistance genes within the well-described S . Typhi-associated IncH1 family of plasmids . 10 . 7554/eLife . 14003 . 003Figure 1 . The phylogenetic structure of 78 Nepali Salmonella Typhi isolated during a gatifloxacin versus ceftriaxone randomised controlled trial . Maximum likelihood phylogeny based on core-genome SNPs of 78 Salmonella Typhi RCT isolates with the corresponding metadata , including the presence of mutations ( dark grey ) in gyrA ( S83F , D87V and D87N ) , parC ( S80I ) and parC ( A364V ) and susceptibility to ciprofloxacin ( susceptible , light blue; intermediate , mid-blue and non-susceptible , dark blue ) by Minimum Inhibitory Concentration ( MIC ) . The reference strain CT18 was used for context and highlighted by the black boxes . Red lines linking to metadata show isolates belonging to the Salmonella Typhi H58 lineage ( with H58 triple mutants highlighted ) , other lineages ( non-H58 ) are shown with black lines . The scale bar indicates the number of substitutions per variable site ( see methods ) . Asterisks indicate ≥85% bootstrap support at nodes of interest . DOI: http://dx . doi . org/10 . 7554/eLife . 14003 . 003 We stratified clinical data from the RCT by H58 status of the corresponding S . Typhi isolates ( H58; N=65 , non-H58; N=13 ) and compared baseline characteristics between these groups . We found no significant differences in demographics and no association between disease severity at presentation between those infected with an H58 S . Typhi isolate or a non-H58 isolate ( Supplementary file 2A ) . Next , we compared the baseline characteristics of patients stratified by ciprofloxacin susceptibility ( susceptible , intermediate and resistant ) , and found no differences in disease severity or demographics on presentation; the only exception being that FQ-resistant S . Typhi were more frequently isolated from adults ( Supplementary file 2B ) . A significantly lower proportion of H58 S . Typhi ( 4/65; 6 . 2% ) were susceptible to FQs compared to non-H58 isolates ( 6/13; 46% ) ( p=0 . 001 ) ( Table 1 ) and , overall , H58 isolates had significantly higher ( but not resistant ) MICs against the majority of tested antimicrobials than non-H58 isolates ( Table 1 ) . 10 . 7554/eLife . 14003 . 004Table 1 . Comparison of antimicrobial susceptibility by Salmonella Typhi lineage . DOI: http://dx . doi . org/10 . 7554/eLife . 14003 . 004E testNon-H58 ( N=13 ) H58 ( N=65 ) p value*MIC50MIC90GM ( range ) MIC50MIC90GM ( range ) Amoxicillin0 . 510 . 77 ( 0 . 38–38 ) 0 . 75>2561 . 43 ( 0 . 38–>256 ) 0 . 0412Chloramphenicol342 . 7 ( 1 . 5–8 ) 4125 . 7 ( 2–>256 ) 0 . 0147Ceftriaxone0 . 060 . 060 . 06 ( 0 . 05–0 . 13 ) 0 . 090 . 190 . 11 ( 0 . 03–0 . 64 ) 0 . 0004Gatifloxacin0 . 130 . 250 . 06 ( 0 . 01–2 ) 0 . 1320 . 21 ( 0 . 01–3 ) 0 . 1197Nalidixic acid>256>25621 . 6 ( 1–>256 ) >256>256346 . 8 ( 1–>256 ) 0 . 0004Ofloxacin0 . 250 . 750 . 24 ( 0 . 03–>32 ) 0 . 5>32321 . 09 ( 0 . 03–>32 ) 0 . 0240Trimethoprim sulphate0 . 020 . 050 . 03 ( 0 . 02–0 . 05 ) 0 . 050 . 320 . 09 ( 0 . 01–>32 ) 0 . 0016Ciprofloxacin0 . 130 . 750 . 11 ( 0 . 01–>32 ) 0 . 38>320 . 80 ( 0 . 02–>32 ) 0 . 0051Ciprofloxacin susceptibility group0 . 0008#- Susceptible6 ( 46 . 2% ) 4 ( 6 . 2% ) - Intermediate4 ( 30 . 8% ) 48 ( 73 . 8% ) - Resistant3 ( 23 . 1% ) 13 ( 20 . 0% ) *Comparisons between Salmonella Typhi lineage for MICs and ciprofloxacin susceptibility groups were based on the Wilcoxon rank sum test and Fisher’s exact test . respectively . MIC: minimum inhibitory concentration , measured in µg/ml#p value for comparison of susceptible vs . intermediate/resistant combined between groups by Fisher’s exact test is 0 . 001 . GM: geometric mean , the upper range of the values was determined by multiplying the MIC by 2 if the result was >X ( for example , >256 = 256*2 = 512 ) . The primary endpoint of the RCT in which these data were generated was a composite for treatment failure ( see method and previous publication ) ( Arjyal et al . , 2016 ) . Treatment failure with H58 S . Typhi was significantly less common in the ceftriaxone group ( 3/31; 9 . 7% ) than the gatifloxacin group ( 15/34; 44 . 1% ) ( Hazard Ratio ( HR ) of time to failure 0 . 19 , 95%CI 0 . 05–0 . 56 , p=0 . 002 ) ( Table 2 ) . Conversely , there was no significant difference in treatment failure between those infected with non-H58 isolates treated with gatifloxacin ( 0/6; 0% ) or ceftriaxone ( 2/7; 28 . 6% ) ( p=0 . 32 ) . Similarly , time to fever clearance differed significantly between the two treatment groups in H58 infections , with median FCTs of 5 . 03 days ( interquartile range ( IQR ) : 3 . 18–7 . 21 ) in the gatifloxacin group and 3 . 07 days ( IQR: 1 . 89–4 . 52 ) in the ceftriaxone group ( p<0 . 0006 ) . Again , this trend was not mirrored in the non-H58 S . Typhi infections , with FCTs of 2 . 87 ( IQR: 2 . 08–3 . 7 ) and 3 . 12 ( IQR: 2 . 2–4 . 12 ) days for gatifloxacin and ceftriaxone , respectively ( p=0 . 61 ) ( Table 3 ) . Moreover , in the gatifloxacin arm , H58 S . Typhi tended to be associated with a higher risk of treatment failure ( p=0 . 06 ) and a longer fever clearance time ( p=0 . 013 ) ( Figure 2 , Table 2 and Supplementary file 2C ) . 10 . 7554/eLife . 14003 . 005Table 2 . Summary of time to treatment failure by Salmonella Typhi lineage and ciprofloxacin susceptibility . DOI: http://dx . doi . org/10 . 7554/eLife . 14003 . 005Time to treatment failureGatifloxacin ( events/N ) Ceftriaxone ( events/N ) Hazard ratio of time to failure ( 95%CI ) ; p valueHeterogeneity test ( p value ) H58*0 . 020- H5815/343/310 . 19 ( 0 . 05 , 0 . 56 ) ; p=0 . 002- Non-H580/62/73 . 87 ( 0 . 31 , 534 . 24 ) ; p=0 . 32Ciprofloxacin susceptibility group†0 . 08- Susceptible0/51/52 . 40 ( 0 . 13 , 350 . 21 ) ; p=0 . 57- Intermediate7/252/270 . 27 ( 0 . 05 , 0 . 99 ) ; p=0 . 049- Resistant8/102/60 . 27 ( 0 . 05 , 1 . 01 ) ; p=0 . 052*Likelihood ratio test p=0 . 06 and 0 . 40 for comparison of time to treatment failure between H58 vs . non-H58 groups in gatifloxacin arm only and in all patients , respectively†Likelihood ratio test p=0 . 007 for comparison of time to treatment failure between MIC groups in gatifloxacin arm only10 . 7554/eLife . 14003 . 006Figure 2 . The association of Salmonella Typhi lineage and ciprofloxacin susceptibility with treatment failure and fever clearance time in patients randomised to gatifloxacin . ( A ) Kaplan-Meier curve for time to treatment failure by H58 and non-H58 Salmonella Typhi . ( B ) Kaplan-Meier curve for time to treatment failure by Salmonella Typhi susceptibility group ( susceptible , intermediate , resistant to ciprofloxacin ) . ( C ) Non-parametric maximum likelihood estimators for interval-censored fever clearance time ( see methods ) by H58 and non-H58 Salmonella Typhi . ( D ) Non-parametric maximum likelihood estimators for interval-censored fever clearance time by Salmonella Typhi susceptibility group ( susceptible , intermediate , resistant to ciprofloxacin ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14003 . 006 As we identified two non-H58 isolates that were also FQ-resistant ( Figure 1 ) , we additionally stratified outcome for the gatifloxacin arm ( N=40 patients ) by FQ susceptibility of the infecting organism . Those infected with FQ-resistant S . Typhi failed gatifloxacin treatment more frequently ( 8/10; 80% ) than those infected with an intermediately resistant organism ( 7/25; 28% ) or a susceptible organism ( 0/5; 0% ) ( p=0 . 007 ) ( Figure 2 and Table 2 ) . Furthermore , in the gatifloxacin arm , those infected with FQ-resistant organisms had significantly higher median FCTs than those infected with S . Typhi with alternative FQ susceptibility profiles ( median FCTs ( days ) : susceptible , 2 . 96 ( IQR: 2 . 13–3 . 85 ) , intermediate , 4 . 01 ( IQR: 2 . 76–5 . 37 ) and resistant 8 . 2 ( IQR: 5 . 99–10 . 5 ) , respectively [p<0 . 0001] ) ( Table 3 and Supplementary file 2D ) . Comparatively , the median FCT for those infected with an FQ-resistant organism but randomised to ceftriaxone was 3 . 83 days ( IQR: 2 . 96–4 . 7 ) ( p<0 . 0001 for the between-treatment comparison ) . 10 . 7554/eLife . 14003 . 007Table 3 . Summary of fever clearance time by Salmonella Typhi lineage and ciprofloxacin susceptibility . DOI: http://dx . doi . org/10 . 7554/eLife . 14003 . 007Fever clearance timeGatifloxacin median ( IQR ) daysCeftriaxone median ( IQR ) daysAcceleration factor ( 95%CI ) ; p valueHeterogeneity test ( p value ) H58¥0 . 07- H585 . 03 ( 3 . 18 , 7 . 21 ) 3 . 07 ( 1 . 89 , 4 . 52 ) 1 . 59 ( 1 . 22 , 2 . 09 ) ; p=0 . 0006- Non-H582 . 87 ( 2 . 08 , 3 . 7 ) 3 . 12 ( 2 . 2 , 4 . 12 ) 0 . 90 ( 0 . 59 , 1 . 36 ) ; p=0 . 61Ciprofloxacin susceptibility group‡0 . 015- Susceptible2 . 96 ( 2 . 13 , 3 . 85 ) 4 . 78 ( 4 . 01 , 5 . 5 ) 0 . 71 ( 0 . 49 , 1 . 02 ) ; p=0 . 07- Intermediate4 . 01 ( 2 . 76 , 5 . 37 ) 2 . 63 ( 1 . 52 , 4 . 05 ) 1 . 31 ( 0 . 97 , 1 . 76 ) ; p=0 . 07- Resistant8 . 2 ( 5 . 99 , 10 . 5 ) 3 . 83 ( 2 . 96 , 4 . 7 ) 2 . 23 ( 1 . 57 , 3 . 17 ) ; p<0 . 0001¥p=0 . 013 and p=0 . 029 for comparison of interval censored time to fever clearance between H58 vs . non-H58 groups in gatifloxacin arm only and in all patients , respectively‡p<0 . 0001 for comparison of interval censored time to fever clearance between MIC groups in gatifloxacin arm only To measure the pattern of emergence of FQ-resistant S . Typhi in Nepal , we compiled FQ susceptibility data from 837 organisms isolated during enteric fever RCTs conducted at Patan Hospital between 2005 and 2014 ( Figure 3 ) ( Pandit et al . , 2007; Koirala et al . , 2013; Arjyal et al . , 2011 ) . MICs against FQs were generally higher for S . Paratyphi A than for S . Typhi . There was a significant temporal increase in S . Typhi MICs against both ciprofloxacin ( p<0 . 0001 ) and gatifloxacin ( p<0 . 0001 ) , with a sharp increase from 2009 . MICs against gatifloxacin in S . Paratyphi A also significantly increased with time ( p<0 . 0001 ) ; however , MICs against ciprofloxacin showed only weak evidence of an upward trend over time ( p=0 . 06 ) . 10 . 7554/eLife . 14003 . 008Figure 3 . Minimum Inhibitory Concentrations of Nepali Salmonella Typhi and Salmonella Paratyphi against ciprofloxacin and gatifloxacin over ten years . Minimum Inhibitory Concentrations ( μg/ml ) for 568 Nepali Salmonella Typhi ( blue ) and 269 Nepali Salmonella Paratyphi A ( grey ) against ( A ) ciprofloxacin and ( B ) gatifloxacin collected from four randomised controlled trials conducted between 2005–2014 at Patan Hospital in Kathmandu , Nepal ( Pandit et al . , 2007; Koirala et al . , 2013; Arjyal et al . , 2011 ) . The smoothed line derived from the generalized additive model showing a non-linear increase in Minimum Inhibitory Concentrations over time , with shading representing the 95% confidence interval . Lower and upper horizontal lines represent the current CLSI cut-offs for susceptible/intermediate and intermediate/resistant , respectively ( CLSI , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14003 . 008 We hypothesised that the H58 triple mutants represented a contemporary importation into Nepal . To explore this , we compared the genomes of the 78 RCT S . Typhi isolates with those from 58 supplementary S . Typhi isolates from previous studies conducted between 2008 and 2013 in this setting ( Figure 4 , Supplementary file 1 ) ( Wong et al . , 2015 ) . We found that the majority of the local H58 isolates ( 84/121; 69 . 4% ) were closely related; these strains represented an 'endemic' Nepali H58 clade containing a single S83F gyrA mutation . Additionally , we identified a further five Nepali strains isolated in 2013 that belonged to the H58 triple mutant group , and had an MIC ≥24 μg/ml against ciprofloxacin . Incorporating additional genome sequences from a recent international study of the H58 lineage ( Wong et al . , 2015 ) , we found that all the Nepali H58 triple mutants were very closely related ( 5 SNPs to nearest neighbour ) to H58 triple mutants isolated previously in neighbouring India between 2008 and 2012 ( Figure 4 ) . 10 . 7554/eLife . 14003 . 009Figure 4 . The phylogenetic structure of fluoroquinolone resistant Salmonella Typhi in a regional context . Maximum likelihood phylogeny based on core-genome SNPs of 136 ( 78 from the RCT ) Salmonella Typhi isolates from Nepal and neighbouring India ( Supplementary file 1 ) . Main tree shows the overall phylogenetic structure and the presence of specific combinations of mutations in gyrA ( S83F , D87V and D87N ) , parC ( S80I ) and parE ( A364V ) . The inset shows a magnified view of the fluoroquinolone-resistant Salmonella Typhi H58 triple mutants from Nepal and their close association with similarly fluoroquinolone-resistant Salmonella Typhi H58 triple mutants from India ( Wong et al . , 2015 ) . The scale bar on the primary tree indicates the number of substitutions per variable site , while that in the inset indicates genetic distance in number of SNPs ( see methods ) . Asterisks indicate ≥85% bootstrap support at nodes of interest . DOI: http://dx . doi . org/10 . 7554/eLife . 14003 . 009
Our study shows that a new FQ-resistant subclade of H58 S . Typhi has been introduced into Nepal and is associated with a lack of FQ efficacy . This subclade was associated with longer FCTs and treatment failure in patients treated with the FQ , gatifloxacin . For the first time , we can conclusively show how enteric fever patients respond to FQ treatment when infected with a specific subclade of H58 , thereby linking organism genotype with a treatment phenotype . Given the international significance of FQs for the treatment of enteric fever and other bacterial infections , our findings have major global health implications for the long-term use and efficacy of this group of antimicrobials . Our data suggest these FQ-resistant S . Typhi strains circulating in Nepal most likely descended from a single ancestor carrying the triple gyrA/parC mutant , such as that isolated in Nepal in 2011 ( Koirala et al . , 2012 ) . This isolate was also associated with treatment failure , although this organism was not genome sequenced and was assumed to be an isolated case . More significantly , several very closely related strains were genome sequenced during an international study of H58 S . Typhi ( Wong et al . , 2015 ) . These organisms had the same combination of triple FQ resistance mutations as those described here; our analysis shows they belong to the same subclade of H58 . These strains had equivalently high MICs against ciprofloxacin and were isolated in India between 2008 and 2012 . However , there were no associated patient outcome data for these strains and other reports from India have been limited . Our data implies that this lineage was introduced into Nepal from India or elsewhere in South Asia within the last 4–5 years and has subsequently entered in an endemic transmission cycle in Kathmandu . Given the large extent of human movement between India and Nepal , we propose this is the most likely route of introduction . However , there is also a small possibility that multiple strains independently gained resistance against FQs through the same selective pressure . Appropriate antimicrobial therapy is critical in the treatment of enteric fever , as effective drugs curtail symptoms and prevent life threatening complications . Our data has substantial repercussions for enteric fever treatment , and we advocate that FQs should no longer be used for empirical enteric fever therapy on the Indian subcontinent , as we predict these strains are likely to be widespread and are associated with poor outcomes with FQ therapy . Notably , in the RCT from which these data were derived we used the newer generation FQ , gatifloxacin , which binds to a different location on the DNA gyrase than the older FQs and is not as susceptible to the common resistance mutations ( Lu et al . , 1999 ) . The isolates in this study were not generally resistant to gatifloxacin according to the current CLSI guidelines for Enterobacteriaceae ( CLSI , 2012 ) ; we suggest that these guidelines be modified specifically for S . Typhi to reflect these new clinical data . We additionally propose that S . Typhi genotyping , mapping and susceptibility testing is performed routinely and rapidly in reference laboratories inside and outside of South Asia to monitor the international spread of these strains and ensure the provision of alternative efficacious therapies to returning travellers ( Lee et al . , 2013; García-Fernández et al . , 2015 ) . In cases of infection with these FQ-resistant isolates , we suggest that ceftriaxone and azithromycin be used as alternatives , and do not currently recommend a return to the use of first-line drugs without contemporary data on treatment outcome . Whilst none of the isolates in this study were MDR , we predict a rapid return of MDR strains if there is a hasty return to older first-line alternatives . This study has limitations . First , the clinical data was collected from one study in a single location , thus limiting utility outside this setting . Second , the overall sample size ( and the gatifloxacin group subsampling ) of those with culture-positive S . Typhi-associated enteric fever was relatively small , and the analysis presented here was performed in a post hoc manner . Notwithstanding these limitations , we were able to show a highly significant association between disease outcome and susceptibility profile of the infecting organism . Further , by using WGS , we were able to pinpoint causative mutations , identify the subclade responsible for treatment failure and relate these strains to other isolates circulating outside Nepal in other parts of South Asia . The methodologies presented here , in which clinical outcome data are combined with genome sequences and antimicrobial susceptibility data , should become the gold standard for informing empiric treatment for all invasive bacterial infections and understanding the role of bacterial genotype and resistance profile on disease outcome for other bacterial infections . No other combination of methodologies would provide the granularity of data required to understand the epidemiology and clinical impact of this emergent strain in detail . In conclusion , our data , for the first time , show a significant association between S . Typhi genotype , antimicrobial susceptibility and disease outcome for those treated with gatifloxacin in a cohort of Nepali enteric fever patients . A FQ-resistant variant of Typhi H58 has emerged in Nepal and is associated with the clinical failure of FQs . Our data suggest these isolates are likely widespread in the subcontinent and FQs should not be recommended for empirical enteric fever therapy in this setting .
The RCT from which the organisms and corresponding clinical data originated for these analyses was conducted at Patan Hospital and the Civil Hospital in the Lalitpur area of Kathmandu , Nepal , between 2011 and 2014 , as described previously ( Arjyal et al . , 2016 ) . The trial was registered at www . clinicaltrials . gov ( ISRCTN63006567 ) . Briefly , patients were randomly assigned to seven days of treatment with either oral gatifloxacin ( 400 mg tablets , Square Pharmaceuticals Limited , Bangladesh ) at a dose of 10 mg/kg once daily or intravenous ceftriaxone ( Powercef , 1000mg injection vial , Wockhardt Ltd , India ) , injected over 10 min at a dose of 60 mg/kg up to a maximum of two grams ( aged 2 to 13 years ) or two grams ( ≥14 years ) once daily . A detailed description of the RCT from which these data were generated has been previously published ( Arjyal et al . , 2016 ) . The primary endpoint was a composite of treatment failure , defined as the occurrence of at least one of the following events: fever clearance time ( FCT ) ( time from the first dose of a study drug until the temperature dropped to ≤37·5°C and remained there for at least two days ) more than seven days post-treatment initiation; requirement for rescue treatment as judged by the treating physician; blood culture positivity for S . Typhi or S . Paratyphi on day eight of treatment ( microbiological failure ) ; culture-confirmed or syndromic enteric fever relapse within 28 days of initiation of treatment; and the development of any enteric fever-related complication ( e . g . clinically significant bleeding , fall in the Glasgow Coma Score , perforation of the gastrointestinal tract and hospital admission ) within 28 days after the initiation of treatment . Time to treatment failure was defined as the time from the first dose of treatment until the date of the earliest failure event . FCTs were calculated electronically using twice-daily recorded temperatures and treated as interval-censored outcomes . Patients without fever clearance or relapse , respectively , were censored at the time of their last follow-up visit ( additional details regarding study procedures can be found in Arjyal at al . 2016 ( Arjyal et al . , 2016 ) ) . Blood ( 3 ml if aged <14 years; 8 ml if aged ≥14 years ) was taken from all patients for bacterial culture on enrolment . Adult blood samples were inoculated into media containing tryptone soya broth and sodium polyanethol sulphonate , up to a total volume of 50 mL . Bactec Peds Plus culture bottles ( Becton Dickinson , New Jersey , USA ) were used for paediatric blood samples . Culture results were reported for up to seven days , positive bottles were subcultured onto blood , chocolate and MacConkey agar and presumptive Salmonella colonies were identified using standard biochemical tests and serotype-specific antisera ( Murex Biotech , Dartford , England ) . Antimicrobial susceptibility testing was performed by the modified Bauer-Kirby disc diffusion method with zone size interpretation based on CLSI guidelines ( CLSI , 2012 ) . Etests were used to determine MICs , following the manufacturer's recommendations ( bioMérieux , France ) . Ciprofloxacin MICs were used to categorise S . Typhi isolates as susceptible ( ≤0 . 06 μg/mL ) , intermediate ( 0 . 12–0 . 5 μg/mL ) and resistant ( ≥1 μg/mL ) following CLSI guidelines ( CLSI , 2012 ) . Genomic DNA from Nepali S . Typhi organisms originating from this RCT ( 78 isolates ) was extracted using the Wizard Genomic DNA Extraction Kit ( Promega , Wisconsin , USA ) ( Supplementary file 1 ) ( Karkey et al . , 2013 ) . Two μg of genomic DNA was subjected to WGS on an Illumina Miseq platform , following the manufacturer’s recommendations to generate 250bp/100bp paired-end reads . All reads were mapped to the reference sequence of S . Typhi CT18 ( accession no: AL515582 ) using SMALT ( version 0 . 7 . 4 ) . Candidate single nucleotide polymorphisms ( SNPs ) were called against the reference sequence using SAMtools ( Li et al . , 2009 ) and filtered with a minimal phred quality of 30 and a quality cut-off of 0 . 75 . The allele at each locus in each isolate was determined by reference to the consensus base in that genome , using samtools mpileup and removing low confidence alleles with consensus base quality ≤20 , read depth ≤5 or a heterozygous base call . SNPs called in phage regions , repetitive sequences or recombinant regions were excluded , ( Wong et al . , 2015 ) resulting in a final set of 1 , 607 chromosomal SNPs . Strains belonging to haplotype H58 were defined by the SNP glpA-C1047T ( position 2348902 in S . Typhi CT18 , BiP33 ) ( Emary et al . , 2012; Holt et al . , 2008; Parkhill et al . , 2001 ) . A maximum likelihood ( ML ) phylogeny was estimated using a 1440 SNP alignment of the 78 RCT isolates in RAxML ( version 7 . 8 . 6 ) with the generalized time-reversible substitution model ( GTR ) and a gamma distribution , with support for the phylogeny assessed via 1000 bootstrap replicates . The alignment was then compared to a global S . Typhi sequence database , with a particular focus on identifying sequences with a mutational profile suggestive of shared ancestry with a divergent H58 clade identified in the previous phylogeny . A secondary ML phylogenetic tree was then inferred from the SNP alignment of the 136 Nepali Typhi along with 19 recently described Typhi H58 with the aforementioned mutational profile , using the same parameters as above ( 1642 SNPs; Supplementary file 1 ) ( Wong et al . , 2015 ) . Raw sequence data are available in the European Nucleotide Archive ( ENA ) ( Supplementary file 1 ) . Comparison of baseline characteristics within patient groups , stratified by the H58 status or susceptibility category of their corresponding S . Typhi isolates was performed using the Kruskal Wallis test for continuous variables and Fisher’s exact test for categorical variables . Time to treatment failure was analysed using Firth’s penalized maximum likelihood bias reduction method for Cox regression as a solution for the non-convergence of likelihood function in the case of zero event counts in subgroups ( Firth , 1993 ) . For comparisons between treatment arms , H58 status , or ciprofloxacin susceptibility group , the model included treatment arm , H58 status , or susceptibility group as a single covariate . Confidence intervals ( CI ) and p-values were calculated by profile-penalized likelihood . FCT was analysed as an interval-censored outcome , i . e . as the time interval from the last febrile temperature assessment until the first afebrile assessment , using parametric Weibull accelerated failure time models ( Kalbfleisch and Prentice , 2002 ) . Median and inter-quartile range ( IQR ) FCT calculations for subgroups were based on models for each subgroup separately . Acceleration factors were based on models that included treatment arm as the only covariate . The non-parametric maximum likelihood estimator ( NPMLE ) was used to visualize the distribution of FCT between groups . Heterogeneity between subgroups was tested with models that included an interaction between treatment arm and the sub-grouping variable . To study the emergence of FQ resistance , data from previous enteric fever trials from 2005–2014 ( Pandit et al . , 2007; Koirala et al . , 2013; Arjyal et al . , 2011 ) was pooled and generalized additive models ( GAM ) were used to examine potential non-linear trends of ciprofloxacin and gatifloxacin MICs over time . All analyses were performed using R software version 3 . 2 . 2 ( Team , 2012 ) . | People who ingest a type of bacteria called Salmonella Typhi can develop the symptoms of typhoid fever . This disease is common in low-income settings in Asia and Africa , and causes a high rate of death in people who are not treated with antimicrobial drugs . During a study in Nepal , Thanh et al . tried to evaluate which of two antimicrobials was better for treating typhoid fever . One of the drugs – called gatifloxacin – did not work in some of the patients . To understand why this treatment failed , Thanh et al . decoded the entire DNA sequences of all the Salmonella Typhi bacteria isolated during the study . Comparing this genetic data to the clinical data of the patients identified a new variant of Salmonella Typhi . These bacteria have a specific combination of genetic mutations that render them resistant to the family of drugs that gatifloxacin belongs to – the fluoroquinolones . Patients infected with the variant bacteria and treated with gatifloxacin were highly likely to completely fail treatment and have longer-lasting fevers . On further investigation Thanh et al . found these organisms were likely recently introduced into Nepal from India . Fluoroquinolones are amongst the most effective and common antimicrobials used to treat typhoid fever and other bacterial infections . However , the presence of bacteria that are resistant to these compounds in South Asia means that they should no longer be the first choice of drug to treat typhoid fever in this location . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"epidemiology",
"and",
"global",
"health",
"microbiology",
"and",
"infectious",
"disease"
] | 2016 | A novel ciprofloxacin-resistant subclade of H58 Salmonella Typhi is associated with fluoroquinolone treatment failure |
XY systems usually show chromosome-wide compensation of X-linked genes , while in many ZW systems , compensation is restricted to a minority of dosage-sensitive genes . Why such differences arose is still unclear . Here , we combine comparative genomics , transcriptomics and proteomics to obtain a complete overview of the evolution of gene dosage on the Z-chromosome of Schistosoma parasites . We compare the Z-chromosome gene content of African ( Schistosoma mansoni and S . haematobium ) and Asian ( S . japonicum ) schistosomes and describe lineage-specific evolutionary strata . We use these to assess gene expression evolution following sex-linkage . The resulting patterns suggest a reduction in expression of Z-linked genes in females , combined with upregulation of the Z in both sexes , in line with the first step of Ohno's classic model of dosage compensation evolution . Quantitative proteomics suggest that post-transcriptional mechanisms do not play a major role in balancing the expression of Z-linked genes .
In species with separate sexes , genetic sex determination is often present in the form of differentiated sex chromosomes ( Bachtrog et al . , 2014 ) . A sex-specific chromosome can be carried by the male ( such as the Y of mammals and fruit flies , in male heterogamety ) or by the female ( such as the W of birds , in female heterogamety ) . These sex chromosomes originally arise from pairs of autosomes , which stop recombining after they acquire a sex-determining region ( Charlesworth , 1991; Charlesworth et al . , 2005 ) . The loss of recombination between X/Z and Y/W chromosomes is likely driven by selective pressures to link the sex-determining gene and alleles with sexually antagonistic effects , and often occurs through inversions on the sex-specific chromosome ( Rice , 1987; Bergero and Charlesworth , 2009 ) . The inverted Y/W-linked region stops recombining entirely , which hampers the efficacy of selection and leads to its genetic degeneration ( Charlesworth et al . , 2005; Engelstädter , 2008 ) . The appearance of further sexually antagonistic mutations can restart the process and select for new non-recombining regions , creating sex chromosome ‘strata’ of different ages ( Ellegren , 2011; Wang et al . , 2012; Vicoso et al . , 2013a; Vicoso et al . , 2013b ) . Eventually , this suppression of recombination can extend to most of the chromosome , leading to gene-poor , mostly heterochromatic sex chromosomes such as the Y chromosome of mammals ( Lemaitre et al . , 2009 ) . The loss of one gene copy on the Y/W is predicted to result in a two-fold reduction of expression in the heterogametic sex , as gene expression is correlated with gene copy number ( Guo et al . , 1996 ) . This can cause imbalances in gene networks composed of both X/Z-linked and autosomal genes ( Wijchers and Festenstein , 2011 ) . Such imbalances can drive the appearance of dosage compensation mechanisms , which target X/Z chromosomes and regulate their expression to restore optimal dosage ( Ohno , 1967; Gartler , 2014 ) . While X/Z upregulation in the heterogametic sex is required to re-establish balanced levels of expression , global downregulation in the homogametic sex is also observed ( e . g . X-inactivation in mammals ) . Ohno suggested a two-step mechanism , in which the initial upregulation of expression is not sex-specific . This leads to an excess of dosage in the homogametic sex , and secondarily selects for further repressing mechanisms ( ‘Ohno’s hypothesis’ of dosage compensation Ohno , 1967] ) . How relevant this model is to the evolution of mammalian dosage compensation is still under debate ( e . g . Gu and Walters , 2017; Lin et al . , 2012; Nguyen and Disteche , 2006; Vicoso and Bachtrog , 2009a; Mank , 2013 ) . Independent of the underlying mechanisms , balanced gene expression between males and females in species with differentiated sex chromosomes was used as diagnostic of a chromosome-wide ( also referred to as ‘global’ , or ‘complete’ ) mechanism of dosage compensation in many different clades ( Vicoso and Bachtrog , 2009a; Mank , 2013; Gu and Walters , 2017 ) . In ZW systems , the loss of genes on the sex-specific W chromosome is generally accompanied by unequal expression levels of the Z-chromosome between ZZ males and ZW females , as well as reduced expression of the Z relative to the autosomes in females . This has generally been interpreted as a lack of chromosome-wide dosage compensation ( also referred to as ‘partial’ , or ‘incomplete’ ) , with individual dosage-sensitive genes being independently regulated instead . Incomplete dosage compensation was described in a wide range of species , including birds ( Itoh et al . , 2007; Ellegren et al . , 2007; Arnold et al . , 2008; Wolf and Bryk , 2011 ) , fishes ( Chen et al . , 2014 ) and snakes ( Vicoso et al . , 2013a ) . So far , Lepidoptera are the only exception to this observation ( Gu and Walters , 2017; Huylmans et al . , 2017 ) . Why many ZW systems should fail to acquire a global mechanism of dosage compensation is not entirely clear , although several and non-mutually exclusive hypotheses have been put forward ( see Discussion , and Gu and Walters [2017] for a review ) . Another possibility is that the male-bias of the Z is instead caused by an accumulation of genes with male functions due to the male-biased transmission of the Z , which may favor the fixation of sexually antagonistic male-beneficial mutations on this chromosome . While the direct comparison of male and female expression of X/Z-linked and autosomal genes has provided an overview of dosage compensation in many clades , it suffers from several drawbacks ( Gu and Walters , 2017 ) . First , chromosome-wide dosage compensation can lead to strongly sex-biased expression , if only the initial upregulation of expression in both sexes has occurred ( but not the secondary downregulation of Ohno's hypothesis [Ohno , 1967] ) . This has been suggested for the flour beetle ( Prince et al . , 2010 ) and for the young sex chromosomes of the threespine stickleback ( Schultheiß et al . , 2015 ) . Biases in expression levels between sexes and/or chromosomes may also have been present ancestrally , before the present sex-chromosomes evolved , and using a proxy for ancestral expression can yield insights into the direct consequences of sex-linkage ( Julien et al . , 2012; Vicoso and Bachtrog , 2015; Gu et al . , 2017 ) . Finally , the vast majority of studies relied only on microarray or RNA-seq data and did not consider any post-transcriptional regulation that might affect gene dosage at the protein level , but not at the transcript level ( whereas protein dosage is in most cases the functionally relevant measure ) . For instance , a proteomic analysis in birds found that several genes appeared to be partially equalized at the protein level despite being strongly male-biased at the transcript level ( Uebbing et al . , 2015 ) . In humans , post-transcriptional regulation does not appear to play a major role in dosage compensation ( Chen and Zhang , 2015 ) . Here , we combine comparative genomics , transcriptomics and quantitative proteomics to obtain a complete overview of the evolution of gene dosage on the Z-chromosome of parasites of the genus Schistosoma . Schistosomes are a group of blood parasites that can cause schistosomiasis in humans ( Chitsulo et al . , 2004 ) . Their complex life cycle is characterized by a phase of clonal multiplication in an intermediate mollusk host , and a phase of sexual reproduction in the final warm-blooded host . Unlike the other 20 , 000 species of hermaphroditic platyhelminths , schistosomes have separate sexes: sexual reproduction occurs immediately after the primary development of males and females in their definitive host , and mating is compulsory for the sexual maturation of females ( Loker and Brant , 2006; Kunz , 2001 ) . Sex determination is genetic , and relies on a pair of cytogenetically well-differentiated ZW chromosomes ( Grossman et al . , 1981a ) . All schistosomes are thought to share the same ancestral pair of ZW sex chromosomes , but differences in their morphology and in the extent of heterochromatization of the W suggest that different strata were acquired independently by different lineages ( Grossman et al . , 1981a; Lawton et al . , 2011 ) . The model blood fluke Schistosoma mansoni was one of the first ZW clades to be evaluated for the presence of global dosage compensation , through the comparison of male and female microarray data derived from several tissues ( Vicoso and Bachtrog , 2011 ) . It showed reduced expression of Z-linked genes in females relative ( i ) to the autosomes and ( ii ) to males , consistent with a lack of chromosome-wide dosage compensation . Interestingly , the reduction of Z-expression in females was less than two-fold , and the Z:autosome ratio of expression was slightly , but consistently , greater than one in males . Our combined genomic , transcriptomic , and proteomic approaches allow us to fully probe the evolution of the male-biased expression of the Z , and suggest a more complex scenario than previously proposed . We discuss this in light of the different hypotheses put forward to account for the evolution of gene dosage on Z chromosomes .
The difference in morphology of the ZW pair in African and Asian schistosomes suggests that the two lineages may differ in their gene content ( Grossman et al . , 1981a ) . We compared the gene content of the Z-chromosomes of three different species: S . mansoni and Schistosoma haematobium , which belong to African schistosomes , and Schistosoma japonicum , an Asian schistosome ( Figure 1 ) . We first identified syntenic blocks between the S . mansoni genome and the S . haematobium and S . japonicum scaffolds . To this end , we mapped all S . mansoni protein coding sequences to the genome assemblies of the two other species and selected only the hits with the highest scores , yielding 9504 S . mansoni/S . haematobium orthologs and 8555 S . mansoni/S . japonicum orthologs ( Table 1 ) . Scaffolds were then assigned to one of the S . mansoni chromosomes , based on their ortholog content ( Figure 1—source data 1 ) . We further performed a comparative coverage analysis to define the Z-specific regions of the three species . Z-derived sequences are expected to display half the genomic coverage in ZW females as in ZZ males , and as the autosomes . We thus mapped male and female genomic reads ( or only female reads in the case of S . haematobium ) to the reference genome of each species ( Protasio et al . , 2012a; Criscione et al . , 2009; Young et al . , 2012; Zhou et al . , 2009 ) . Publicly available raw reads were used for S . mansoni and S . haematobium ( Wellcome Trust Sanger Institute Bioprojects PREJB2320 and PREJB2425 ) , whereas male and female S . japonicum were sequenced for this study . We then estimated the per base genomic coverage . Median coverage values were 18 . 40 and 18 . 99 for S . mansoni male and female libraries; 23 . 5 and 7 . 43 for S . haematobium female#1 and female#2 libraries; 23 . 77 and 20 . 53 for S . japonicum male and female libraries . Z-specific genomic regions were defined by a maximum value of the female:male ratio of coverage ( S . mansoni: log2 ( female:male ) =−0 . 4; S . japonicum: log2 ( female:male ) =−0 . 84 ) , or a maximum value of female coverage ( S . haematobium: log2 ( female ) =4 . 41 ) . Details of how these cutoff values were obtained are provided in the Materials and methods and Appendix 1 . This analysis resulted in 285 newly described Z-specific genes in S . mansoni that were previously located on 19 unplaced scaffolds longer than 50 kb , and to a refined pseudoautosomal/Z-specific structure of the published ZW linkage group ( Protasio et al . , 2012a; Criscione et al . , 2009 ) ( Figure 1—source data 2 ) . It further allowed us to define 379 Z-specific scaffolds ( containing 1409 annotated genes with orthologs in S . mansoni ) in S . haematobium ( Figure 1—source data 3 for exhaustive list ) and 461 Z-specific scaffolds ( containing 706 annotated orthologs ) in S . japonicum ( Figure 1—source data 4 for exhaustive list ) . While the content of the Z was largely shared between the African S . mansoni and S . haematobium ( Table 1 , Figure 1 ) , large differences were found between the African and Asian lineages: only 476 Z-specific genes were shared by S . mansoni and S . japonicum , while 306 were only Z-specific in S . mansoni and 137 only in S . japonicum ( Table 1 ) . Of all these Z-specific genes , 613 were already mapped to the S . mansoni ZW linkage group ( Protasio et al . , 2012a; Criscione et al . , 2009 ) and , when plotted along the Z-chromosome , outlined three different evolutionary strata: one shared ancestral stratum ( S0: 367 genes ) and two lineage-specific strata ( S1mans , specific to the African schistosomes , with 180 genes; and S1jap , specific to S . japonicum , with 66 genes ) ( Table 1 , Figure 1 , and Figure 1—source data 1 ) . The presence of pseudoautosomal regions throughout the S0 ( Figure 1 ) is likely due to errors in the genome assembly . All further analyses were run using all newly identified Z-specific genes , but hold when only Z-specific genes that were previously mapped to the ZW linkage group are considered ( Appendix 1 ) . In order to test for dosage compensation , the median expression of Z-specific genes in ZW females can be compared to the median autosomal expression ( Z:AA ratio ) and/or to the Z-specific gene expression in ZZ males ( F:M ratio ) . Z:AA or F:M ratio of ~1 supports global dosage compensation , while a ratio between 0 . 5 and 1 suggests partial or local dosage compensation . We performed this analysis in S . mansoni and in S . japonicum , using publicly available RNA-seq reads derived from a sexually undifferentiated stage ( schistosomula , [Picard et al . , 2016; Wang et al . , 2017] ) and a sexually mature stage ( adults , Wellcome Trust Sanger Institute Bioproject PRJEB1237 , [Wang et al . , 2017] ) . The inclusion of a sexually undifferentiated stage ( which lack primary spermatocytes or eggs ) is important , as much of the expression obtained from adults will necessarily come from their well-developed gonads . Sex-linked genes are often sex-biased in the germline , even in organisms that have chromosome-wide dosage compensation ( e . g . due to sex-chromosome inactivation during gametogenesis ) , and the inclusion of gonad expression has led to inconsistent assessments of the status of dosage compensation in other clades ( Gu and Walters , 2017; Huylmans et al . , 2017 ) . Reads were mapped to their respective genomes , and expression values in Reads Per Kilobase Million ( RPKM ) were calculated for each gene ( Figure 2 , Figure 2—source data 1 and 2 ) ; only genes with a minimum RPKM of 1 in both sexes were considered . We consistently observed a strong male bias in the expression of Z-specific genes in both stages and species ( F:M ratio between 0 . 58 and 0 . 69 , Figure 2 and Supplementary file 1 ) , consistent with local or incomplete dosage compensation . While this was generally supported by the lower expression levels of Z-specific genes in females when compared to the autosomes ( Z:AA ratio between 0 . 73 and 0 . 85; Figure 2 , Supplementary file 1 ) , this difference was only apparent for some filtering procedures ( Figure 2—figure supplement 1 to Figure 2—figure supplement 14 ) , and even then was not sufficient to fully account for the strong male-bias of the Z . Instead , the higher expression of the male Z in both stages and species ( ZZ:AA ratio between 1 . 25 and 1 . 46 , Supplementary file 1 ) appeared to also contribute to the male-bias of Z-linked genes . These patterns were qualitatively robust to changes in the methods used to estimate expression ( RPKM or TPM [Transcripts Per Kilobase Million] ) , in the filtering procedure ( RPKM > 0 , RPKM > 1 , TPM >0 or TPM >1 ) , and when only genes that were previously mapped to the ZW linkage group were considered . These analyses were further performed independently in the S0 , S1mans and S1jap strata , which showed no significant difference in the extent of their male bias . All the resulting plots are shown in Figure 2—figure supplement 1 to Figure 2—figure supplement 14 . Finally , Z-specific genes were found to be male-biased even when only genes with broad expression were considered ( RPKM > 1 and RPKM > 3 in all samples , and when genes with strong sex-biases in expression were excluded ( M:F > 2 or F:M > 2 , Figure 2—figure supplement 15 and Figure 2—figure supplement 16 , I–L panels ) , confirming that this pattern does not appear to be driven simply by the presence of genes with sex-specific functions on the Z-chromosome . No further influence of known protein-protein interactions was detected ( Figure 2—figure supplement 17 , Appendix 1 ) . The previous patterns are consistent with an upregulation of the Z-chromosome in both sexes after the degeneration of the W-specific region , and could represent the intermediate step in the evolution of dosage compensation originally postulated by Ohno . However , they could also be due to high expression of the ancestral proto-Z in both sexes , before sex chromosome divergence . To exclude this , we identified one-to-one orthologs between genes annotated in both species using a reciprocal best hit approach ( 7382 orthologs , Figure 3—source data 1 ) . All genes that were classified as Z-specific in one species but as autosomal in the other were considered to be part of the S1 strata ( S1jap if they were Z-specific in S . japonicum or S1mans if they were Z-specific in S . mansoni ) . We then used the pseudoautosomal expression of these lineage-specific strata as a proxy for the ancestral level of expression . For instance , in S . japonicum , we estimated the S1jap:AA ratio , after normalizing the expression data by their respective ( pseudo ) autosomal level in S . mansoni ( Figure 3A and B ) . The reversed analysis was performed for S1mans ( Figure 3C and D ) . Figure 3 confirms that the male-biased expression of Z-specific genes is a consequence of their sex-linkage , and that the Z-chromosome has become under-expressed in females relative to the ancestral expression . However , a full two-fold reduction in female expression is not observed , consistent with partial upregulation , and/or full upregulation of a subset of dosage-sensitive genes ( Z:AA ranging from 0 . 68 to 0 . 83 , Supplementary file 1 ) ( Sangrithi et al . , 2017; Pessia et al . , 2012 ) . Figure 3 also generally supports an increase in expression in males ( ZZ:AA ranging from 0 . 98 to 1 . 35 , Supplementary file 1 ) . Male adults of S . japonicum are the exception , with a ZZ:AA of 0 . 98 . However , given that an excess of expression is observed when ( i ) we do not take into account the ancestral expression ( Figure 2 ) , ( ii ) we focus on genes previously mapped to the ZW pair ( Figure 2—figure supplements 3 , 4 , 10 and 11 ) , and ( iiii ) we consider the schistosomula stage ( with or without the ancestral expression and independent of the classification ) , this is likely due to noise in the sample and not to a true biological difference ( only 58 genes were tested ) . Figure 3 shows the distributions for all genes with a minimum RPKM value of 1 in males and females of both species . We repeated the analysis using the same filters as before ( minimum RPKM of 0 , TPM of 0 , TPM of 1 ) , and with a publicly available list of 1:1 orthologs ( obtained from the Wormbase Biomart , see Methods ) . The resulting plots are shown in Figure 2—figure supplement 1 to Figure 2—figure supplement 14 , and Figure 3—figure supplements 1 and 2 . Gene expression values for the orthologs of each species are provided in Figure 3—source data 2 and 3 . We tested for putative post-transcriptional mechanisms by assessing the dosage compensation pattern at the proteomic level in adult S . mansoni , using a somatic tissue ( head region ) as well as the gonads; three replicates were used for each tissue and sex . Heads and gonads were chosen as they allowed us to compare Z-specific gene dosage in tissues with widespread functional sex-specificity ( ovary and testis ) , and in a tissue where most dosage imbalances are likely to be deleterious . We used a label-free quantitative mass spectrometry approach to obtain a relative quantification of the protein levels in each tissue depending of the sex ( Figure 4—source data 1 to 5 ) . Post-transcriptional dosage compensation mechanisms would be detectable by ( i ) an equalization of the Z expression between sexes at the protein level ( F:M close to one for both the Z and the autosomes ) ; ( ii ) a different correlation between F:M obtained from mRNA and from proteins for Z-linked and autosomal genes . We used publicly available head and gonad microarray data ( Nawaratna et al . , 2011 ) as the transcriptomic reference ( Figure 4—source data 6 ) . A significant and positive correlation was found between the F:M ratio derived from the microarray and from the proteomic data ( Figure 4 ) , and between transcript and protein dosage levels in both males and females ( Figure 4—figure supplements 1 and 2 ) , confirming the validity of the comparison . Similar to what was observed using RNA-seq , the expression of Z-specific genes was strongly male-biased compared to that of autosomal genes in both heads ( F:M of 0 . 68 for the Z chromosome versus 0 . 92 for the autosomes; Figure 4A , Supplementary file 1 ) and gonads ( F:M of 0 . 78 versus 0 . 99; Figure 4B , Supplementary file 1 ) . These F:M ratios are closer to each other than in our RNA-seq analysis ( Figure 2 , Supplementary file 1 ) , or than the microarray data ( Supplementary file 1 ) , which could suggest a potential contribution of post-transcriptional regulation to dosage equalization . However , Figure 4B shows that Z-linked and autosomal genes show a similar correlation between the F:M ratios found for mRNAs and proteins ( p>0 . 05 with a Fisher r-to-z transformation of the correlation coefficients , Figure 4C and D ) , which argues against a major role of post-transcriptional regulation to balance expression . This similarity between Z-linked and autosomal genes holds when only genes with male-biased expression in the microarray data are considered ( Figure 4—figure supplement 3 ) , and when the transcript and protein dosage of Z-linked autosomal genes are compared within each sex ( Figure 4—figure supplements 1 and 2 ) .
S . mansoni , S . haematobium and S . japonicum are the main species responsible for human schistosomiasis and have been the subject of many molecular and genomic studies . Despite the availability of extensive genomic and transcriptomic resources ( e . g . a genome assembly at the near-chromosome level for S . mansoni ( Protasio et al . , 2012a; Criscione et al . , 2009 ) , or sex- and stage-specific transcriptomes ( Picard et al . , 2016; Lu et al . , 2016; Grevelding et al . , 2018; Lu et al . , 2017 ) , many basic questions remain regarding their reproduction and biology . For instance , the master sex-determining gene ( and whether it is located on the W or Z ) is still a mystery ( Lepesant et al . , 2012; Portela et al . , 2010 ) . This is partly due to the inherent challenges of assembling genomes from sequencing data , especially for regions rich in heterochromatin and repetitive sequences , such as sex chromosomes . For instance , 416 scaffolds , including 3893 genes ( 29% of the annotated nuclear genes ) , are still unplaced . By basing our analysis on genomic coverage , we were able to detect a further 285 Z-specific genes in S . mansoni; their role in sex determination can be investigated further . Our comparative approach can also reduce the number of candidates , as any gene involved in sex determination should in principle be found in the ancestral Z-specific stratum; similar analyses in other species can in the future refine the candidate region . Another advantage of basing our Z-assignment purely on coverage patterns is that our results should be largely independent of potential biases in the current version of the genome . It should , however , be noted that many genes are likely still missing from the current assembly ( which has a BUSCO score of 76% complete plus fragmented genes; https://parasite . wormbase . org/index . html [Howe et al . , 2016; Howe et al . , 2017] ) and that repeating these analyses using future improved assemblies will be necessary to obtain the full set of sex-linked genes . A gradient of ZW heteromorphism between schistosome species was revealed by cytogenetic studies ( Grossman et al . , 1981a; Grossman et al . , 1981b; Short and Grossman , 1981 ) ; in particular , African schistosomes were found to have much more extensive ZW differentiation and W heterochromatinization than Asian species ( Hirai et al . , 2000 ) . Our results generally support these cytogenetic data: we confirm the acquisition of independent evolutionary strata in the sex chromosomes of S . mansoni and S . japonicum , and detect a larger number of Z-specific genes in the African species ( 8% to 11% of all annotated orthologs , respectively , in S . mansoni and S . haematobium ) than in S . japonicum ( 5 . 5% of all annotated orthologs ) . Interestingly , although the sex chromosomes of the African S . mansoni and S . haematobium differ morphologically , they are largely similar in their gene content ( Figure 1 ) , consistent with their much closer phylogenetic relationship ( the median synonymous divergence between the two species is around 17% , compared to 65% for S . mansoni/S . japonicum , Appendix 1 ) . This may be comparable to snakes , where ZW pairs with vastly different morphologies were all equally differentiated at the genomic level ( Vicoso et al . , 2013a ) , and highlights the contribution of other factors , such as differential transposable element accumulation , to the large-scale morphology of sex chromosomes . ZW systems ( aside from Lepidoptera ) consistently show male-biased expression of the Z chromosome ( Gu and Walters , 2017 ) . While female-biased expression of the X occurs in a few young XY systems ( Gu and Walters , 2017; Schultheiß et al . , 2015; Howe et al . , 2017; Grossman et al . , 1981b; Short and Grossman , 1981; Hirai et al . , 2000; Mank and Ellegren , 2009 ) , well-established X chromosomes generally show full equalization of gene expression between the sexes . This difference has often been framed as the acquisition of global mechanisms of dosage compensation , which affects the whole X/Z , versus the acquisition of local compensation , in which dosage-sensitive genes become individually regulated ( Mank and Ellegren , 2009 ) . Several parameters should influence this , and favor local compensation in ZW systems: ( i ) The speed of the heterochromosome degeneration: when only a few genes are lost at a time ( because the region of suppressed recombination is small , or because degeneration is slow ) , the establishment of a gene-by-gene dosage compensation may be favored; on the other hand , the loss of many genes at once could favor global mechanisms of dosage compensation ( Gu and Walters , 2017; Vicoso and Charlesworth , 2009b ) . Since more mutations occur during spermatogenesis than oogenesis , female-specific W chromosomes will generally have lower mutation and degeneration rates than male-specific Ys , favoring local compensation; ( ii ) The effective population size of Z ( NeZ ) : NeZ is decreased when the variance in reproductive success of ZZ males is larger than that of ZW females ( e . g . in the presence of strong sexual selection ) . This will impair the adaptive potential of the Z ( Mank , 2009; Mullon et al . , 2015 ) , such that only strongly dosage-sensitive genes can become upregulated in the heterogametic sex , while the others remain uncompensated; ( iii ) More efficient purging of mutations that are deleterious to males: strong sexual selection can also increase the strength of purifying selection on males , by preventing all but the fittest males from contributing to the next generations . If mutations that compensate for the loss of Y/W-linked genes overexpress the X/Z copy in both sexes , they will be under negative selection in the homogametic sex , and may be more efficiently selected against when males are the homogametic sex ( Mullon et al . , 2015 ) . Schistosomes are unusual among female-heterogametic clades in that they appear to have a chromosome-wide upregulation of the Z in both sexes; such an increase in males was not detected in birds ( Julien et al . , 2012 ) or snakes ( Vicoso et al . , 2013a ) , even when ancestral expression was taken into account . They therefore likely represent an intermediate between ZW species with true local compensation , and the chromosome-wide compensation of the ZW Lepidoptera . These results further show that , even if mutations that upregulate gene expression in both sexes are more easily fixed on an evolving X-chromosome than on an evolving Z ( Mullon et al . , 2015 ) , this is not an absolute barrier to the evolution of global dosage compensation . It is however still unclear why the evolutionary dynamics appear to differ between schistosomes and most other ZW clades , as the demographic and population genetics parameters of this group are largely unknown . The observed male biased sex-ratio in adults , combined with a largely monogamous mating system ( Beltran and Boissier , 2009; Beltran and Boissier , 2008 ) , may increase the reproductive variance of males and could reduce the effective population size of the Z . This should also lead to stronger sexual selection in males than in females ( Beltran and Boissier , 2008; Steinauer et al . , 2009 ) , suggesting similar evolutionary dynamics as in other ZW systems . A detailed characterization of the population genetics of the Z chromosome and autosomes will therefore be crucial for understanding what may have driven the evolution of this unusual system . Ohno’s hypothesis predicts that the heterochromosome is initially overexpressed in both sexes , then secondarily downregulated in the homogametic sex ( Ohno , 1967 ) . This theoretical scenario was first formulated to account for the inactivation of the X in mammals . Since then , similar molecular mechanisms to downregulate the X/Z chromosome have been characterized in nematodes and moths ( Kiuchi et al . , 2014; Meyer , 2010 ) . If an initial upregulation of the X did occur in both sexes , then inactivation in the homogametic sex should simply restore the ancestral expression levels , a hypothesis that has been tested in many empirical studies in mammals . Most of them assumed that the X and autosomes must have had similar ancestral levels of expression , and simply compared their expression ( Julien et al . , 2012; Nguyen and Disteche , 2006; Xiong et al . , 2010; Gu and Walters , 2017; Chen and Zhang , 2015; Graves , 2016; Deng et al . , 2011; Kharchenko et al . , 2011; Yildirim et al . , 2011; Lin et al . , 2012; Pessia et al . , 2014 ) . These yielded mixed results , with some ( Xiong et al . , 2010 ) finding reduced expression of the X , while others ( e . g . Deng et al . , 2011; Kharchenko et al . , 2011; Yildirim et al . , 2011 ) found similar levels of expression for X-linked and autosomal genes , in agreement with Ohno’s predictions . Taking ancestral gene expression into account , Julien et al . ( 2012 ) found evidence of an Ohno-like mechanism in the marsupials but not in placental mammals ( Julien et al . , 2012 ) . Pessia et al . ( 2012 ) recently found that while individual dosage-sensitive genes do show evidence of upregulation , the majority does not . The evolution of X-inactivation may therefore have involved a complex scenario under which a few dosage-sensitive genes first became individually upregulated in both sexes ( gene-by-gene compensation ) , followed by the establishment of a chromosome-wide mechanism to downregulate expression in females ( global compensation ) ( Sangrithi et al . , 2017; Pessia et al . , 2012 ) . Our results , which consider ancestral expression and do not indicate a major influence of post-transcriptional regulation , suggest a scenario closer to Ohno’s original hypothesis , with the male Z showing a consistent increase in expression . A similar pattern has been observed in Tribolium castaneum ( Coleoptera , Prince et al . , 2010 ) , where the female X has been found to be over-expressed relative to the autosomes , and to the male X-chromosome . However , an RNA-seq analysis in the same species did not detect this ( Mahajan and Bachtrog , 2015 ) , so it is at this point unclear whether it truly represents an example of Ohno’s model in action . The youngest evolutionary stratum of the young XY pair of threespine sticklebacks also shows overexpression in females ( Schultheiß et al . , 2015 ) , even when ancestral expression is accounted for ( White et al . , 2015 ) . However , the interpretation of these patterns is complicated by the fact that such an overexpression is also detected for the pseudoautosomal region , and that the oldest evolutionary stratum appears to lack dosage compensation altogether . Schistosomes may therefore not only represent an ideal system in which to investigate the evolution of dosage compensation in a ZW system , but also an unparalleled system for understanding the relevance of the model and predictions originally made by Ohno .
Male and female worms preserved in ethanol of S . japonicum were provided by Lu Dabing from Soochow University ( Suzhou , China ) . DNA was extracted from 28 pooled males and 33 pooled females . The worms were lysed using the Tissue Lyser II kit ( QIAGEN ) and DNA was isolated using the DNeasy Blood and Tissue Kit ( QIAGEN ) . DNA was then sheared with Covaris Focused-ultrasonicator . Library preparation and sequencing ( HiSeq 2500 v4 Illumina , 125 bp paired-end reads ) were performed at the Vienna Biocenter Next Generation sequencing facility ( Austria ) . Reads have been deposited at the NCBI Short Reads Archive under accession number SRP135770 . S . mansoni DNA libraries ( 100 bp paired-end reads ) were downloaded from the NCBI Sequence Read Archive , under the accession numbers ERR562989 ( ~6000 male pooled cercariae ) and ERR562990 ( ~6000 female pooled cercariae ) . Female S . haematobium DNA libraries ( 80 bp paired-end reads ) were found under accession numbers ERR037800 and ERR036251 . No male S . haematobium library was available . The reference genome assemblies of S . mansoni ( PRJEA36577 , [Protasio et al . , 2012b] ) , S . haematobium ( PRJNA78265 , [Young et al . , 2012] ) and S . japonicum ( PRJEA34885 , [Zhou et al . , 2009] ) were obtained from the WormBase parasite database ( https://parasite . wormbase . org/index . html [Howe et al . , 2016; Howe et al . , 2017] ) . S . mansoni coding sequences and their respective chromosomal locations were obtained from the WormBase Parasite database ( https://parasite . wormbase . org/index . html , [Howe et al . , 2016; Howe et al . , 2017] ) . This gene set was mapped to the S . haematobium and S . japonicum genome assemblies using Blat ( Kent , 2002 ) with a translated query and dataset ( -dnax option ) , and a minimum mapping score of 50; only the genome location with the best score was kept for each . When more than one gene overlapped by more than 20 base pairs , only the gene that had the highest mapping score was kept . Finally , each scaffold was assigned to one of the S . mansoni chromosomes , depending on the majority location of the genes that mapped to it , or on their total mapping scores if the same number of genes mapped to two separate chromosomes . The final chromosomal assignments are provided in Figure 1—source data 1 . For the S . japonicum DNA reads , adaptors were removed using Cutadapt ( v1 . 9 . 1 [Martin , 2011] ) and the quality of the reads was assessed using FastQC ( v0 . 11 . 2 , https://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) ; no further quality trimming was deemed necessary . For the S . mansoni and S . haematobium reads , potential adaptors were systematically removed , and reads were trimmed and filtered depending on their quality , using Trimmomatic ( v0 . 36 [Bolger et al . , 2014] ) . The resulting read libraries of each species were mapped separately to their reference genomes using Bowtie2 ( --end-to-end --sensitive mode , v2 . 2 . 9 [Langmead and Salzberg , 2012] ) . The resulting alignments were filtered to keep only uniquely mapped reads , and the male and female coverages were estimated from the filtered SAM files with SOAPcoverage ( v2 . 7 . 7 . , http://soap . genomics . org . cn/index . html ) . Coverage values were calculated for each scaffold in S . haematobium and S . japonicum , and for each 10 kb non-overlapping window in S . mansoni . The coverage values for each library are provided in Figure 1—source data 3 and 4 . For each species , we calculated the log2 ( female:male ) coverage of each scaffold or , in the case of S . mansoni , of each 10 Kb window along the genome . Since only female DNA data was available for S . haematobium , the log2 ( female1 +female2 ) was used instead for this species . In order to determine the 95% and 99% percentile of log2 ( female:male ) of Z-linked sequences , which we use as cutoff values for assignment to Z-specific regions , we first excluded scaffolds/windows that fit an autosomal profile . To do so , the 1st and 5th percentile of log2 ( female:male ) were estimated using all 10 kb windows found on the annotated autosomes of S . mansoni ( Protasio et al . , 2012b ) ; in S . haematobium and S . japonicum , all scaffolds that mapped to the S . mansoni autosomes were used for this purpose . By plotting the distribution of log2 ( female:male ) for the scaffolds/windows that map to each chromosome ( Appendix 1 ) , we determined that: ( i ) for S . mansoni the 1st percentile ( log2 ( female:male ) =−0 . 26 ) discriminates effectively between autosomal and Z-specific windows; ( ii ) for S . haematobium and S . japonicum , which have noisier coverage , the 1st percentile included a significant fraction of Z-derived genes , and the 5st percentile was used instead ( respectively log2 ( female ) = 4 . 57 and log2 ( female:male ) =−0 . 40 ) . All scaffolds with higher log2 ( female:male ) or log2 ( female ) were excluded . The 95th and 99th quantiles of coverage were then calculated for the remaining , putatively Z-linked , sequences . By plotting all the coverage values along the S . mansoni Z-chromosome ( Appendix 1 and Figure 1 ) , we determined that: ( i ) for S . mansoni and S . haematobium , the 95th quantile ( log2 ( female:male ) =−0 . 4 and log2 ( female ) = 4 . 41 , respectively ) was an effective cut-off for discriminating pseudoautosomal and Z-specific sequences; ( ii ) for S . japonicum , using the 95th percentile lead to the exclusion of many genes in Z-specific regions , and the 99st percentile ( log2 ( F:M ) =−0 . 84 ) was used instead . The Z-specific or autosome assignation was finally attributed as follows: ( i ) in S . mansoni , windows were classified as Z-linked if they displayed log2 ( F:M ) <−0 . 4 and as autosomal if log2 ( F:M ) >−0 . 4; ( ii ) for S . haematobium , scaffolds with log2 ( femalesum ) <4 . 41 were classified as Z-linked , scaffolds with log2 ( femalesum ) >4 . 57 as autosomal , and all others as ambiguous; ( iii ) for S . japonicum , scaffolds with log2 ( F:M ) <−0 . 84 were classified as Z-linked , scaffolds with log2 ( F:M ) >−0 . 40 as autosomal , and all others as ambiguous . For S . haematobium and S . japonicum , we considered only scaffolds with at least a coverage of 1 in each library ( n . a . ) . In S . mansoni , unplaced scaffolds shorter than 50 kb were excluded; five consecutive 10 kb windows with consistent coverage patterns were required for a region to be classified as either Z-specific or autosomal . Smaller regions , as well as the two 10 Kb windows surrounding them , were excluded ( see Appendix 1 ) . The final classifications for the three species are provided in Figure 1—source data 2 , 3 and 4 , and summarized for orthologs in Figure 1—source data 1 . The Z-specific gene content of S . mansoni and S . japonicum was compared in order to define Z-chromosome strata . Genes that were located on Z-specific scaffolds/windows in both species were assigned to the shared stratum ‘S0’ . Genes that were assigned to Z-specific regions in one species but not in the other were assigned to lineage-specific strata: ‘S1mans’ genes were Z-specific in S . mansoni and autosomal in S . japonicum , while ‘S1jap’ genes were Z-specific in S . japonicum and autosomal in S . mansoni . While the main figures consider all the genes that were classified as Z-linked or autosomal based on coverage ( referred to as the ‘exhaustive classification’ in Appendix 1 and Figures ) , independent of their original genomic location , we repeated the analyses using only the Z-specific genes that were already assigned to the ZW linkage map of S . mansoni ( the ‘stringent classification’ in Appendix 1 ) . All genes belonging to the categories ‘excluded’ , ‘ambiguous’ , ‘n . a . ’ or that did not have orthologs on S . japonicum scaffolds were not further considered ( Table 1 ) . ‘PSA_shared’ and ‘Aut_shared’ are common to the two classifications and correspond to genes that were classified as autosomal in both species using coverage and that were previously mapped to the ZW linkage group or to the autosomes of S . mansoni , respectively ( Protasio et al . , 2012b ) . S . mansoni and S . japonicum RNA-seq libraries were obtained from SRA ( NCBI ) . Accession numbers are: S . mansoni adult females: ERR506076 , ERR506083 , ERR506084; S . mansoni adult males: ERR506088 , ERR506082 , ERR506090; S . mansoni schistosomula females: SRR3223443 , SRR3223444; S . mansoni schistosomula males: SRR3223428 , SRR3223429; S . japonicum adult females: SRR4296944 , SRR4296942 , SRR4296940; S . japonicum adult males: SRR4296945 , SRR4296943 , SRR4296941; S . japonicum schistosomula females: SRR4279833 , SRR4279491 , SRR4267990; S . japonicum schistosomula males: SRR4279840 , SRR4279496 , SRR4267991 . Raw reads were cleaned using trimmomatic ( v 0 . 36 [Bolger et al . , 2014] ) , and the quality of the resulting reads was assessed using FastQC ( v0 . 11 . 2 , https://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) . Reads were mapped to their respective reference genomes used Tophat2 ( Trapnell et al . , 2009 ) . Read counts were obtained with H ( Anders et al . , 2015 ) and expression values ( in Reads Per Kilobase of transcript per Millionmapped reads , RPKM ) were calculated for each gene in each of the RNA-seq libraries ( Figure 2—source data 1 ) . TPM ( Transcripts Per Kilobase Million ) values were also calculated using Kallisto ( Bray et al . , 2016 ) against set of coding sequences of the respective species . All expression values are provided in Figure 3—source data 2 and 3 . A Loess Normalization ( R library Affy ) was performed on the schistosomulum data and on the adult data separately and all analyses were performed using different thresholds ( RPKM > 0 , RPKM > 1 , TPM >0 or TPM >1 in all the libraries of the studied stage ) . The Loess normalization was applied to all conditions at once when we filtered for minimum expression in all stages and sexes ( RPKM > 1 and RPKM > 3 , Figure 2—figure supplements 15–17 , Figure 3—figure supplements 1 and 2 ) . Correlation analyses were performed for each developmental stage , considering libraries from both males and females , and both species . As shown in Appendix 1 , two S . mansoni libraries ( ERR506076 and ERR506082 ) were not well correlated with the other samples , and were excluded from our study . Expression values were averaged for each stage and sex . The significance of differences between medians of expression was tested with Wilcoxon rank sum tests with continuity correction . S . japonicum coding DNA sequences and their respective location on the genome scaffolds were obtained from the WormBase Parasite database ( https://parasite . wormbase . org/index . html [Howe et al . , 2016; Howe et al . , 2017] ) . The S . mansoni set of coding sequences ( see above ) was mapped to the S . japonicum gene set using Blat ( Kent , 2002 ) with a translated query and dataset ( -dnax option ) , and a minimum mapping score of 50; only reciprocal best hits were kept . This reciprocal best hit ortholog list is provided in Figure 3—source data 1 and 2 . A second list of orthologs was obtained from the Biomart of WormBase Parasite , excluding paralogues , and requiring a gene stable ID for both S . mansoni ( PRJEA36577 ) and S . japonicum ( PRJEA34885 ) ( in Figure 3—source data 1 and 3 ) . In subsequent transcriptomic analyses , each list was used independently to ensure that the results were independent of the method used to assign orthology . Microarray data for male and female heads and gonads ( Nawaratna et al . , 2011 ) were obtained from the Gene Expression Omnibus ( GEO ) database ( NCBI , ftp://ftp . ncbi . nlm . nih . gov/geo/series/GSE23nnn/GSE23942/matrix/ ) . A Loess Normalization ( R library Affy ) was performed on the head and gonad data separately . When different probes corresponded to one gene , their expression values were averaged . Gene expression was available for a total of 6925 genes . The normalized data are available in Figure 4—source data 6 . Male and female adult S . mansoni gonads were sampled using the whole-organ isolation approach described previously ( Hahnel et al . , 2013 ) . Twenty ovaries and 20 testes , as well as five heads of each sex , were sampled , in triplicate , from paired worms . All biological samples were resuspended in Laemmli buffer , denatured and frozen at −20°C until further processing . Subsequent protein treatment and analyses were performed at the ‘EDyP-service’ – proteomic platform ( Grenoble , France ) . The extracted proteins were digested by modified trypsin ( Promega , sequencing grade ) . The resulting peptides were analyzed by nanoLC-MS/MS ( Ultimate 3000 RSLCnano system coupled to Q-Exactive Plus , both Thermo Fisher Scientific ) . Separation was performed on a 75 µm x 250 mm C18 column ( ReproSil-Pur 120 C18-AQ 1 . 9 µm , Dr . Maisch GmbH ) after a pre-concentration and desalting step on a 300 µm × 5 mm C18 precolumn ( Pepmap , Thermo Fisher Scientific ) . MS and MS2 data were acquired using Xcalibur ( Thermo Fisher Scientific ) . Full-scan ( MS ) spectra were obtained from 400 to 1600 m/z at a 70 , 000 resolution ( 200 m/z ) . For each full-scan , the most intense ions ( top 10 ) were fragmented in MS2 using high-energy collisional dissociation ( HCD ) . The obtained data were processed in MaxQuant 1 . 5 . 8 . 3 against the database loaded from Uniprot ( taxonomy Schistosoma mansoni , October 26th , 2017 , 13 . 521 entries ) and the MaxQuant embedded database of frequently observed contaminants . The resulted iBAQ values ( Tyanova et al . , 2016 ) were loaded into ProStaR ( Wieczorek et al . , 2017 ) for statistical analysis . Contaminant and reverse proteins were removed and only the proteins with three quantified values in at least one condition were taken into account . After log2 transformation , the iBAQ values were normalized by overall-wise median centering followed by imputation using detQuantile algorithm with quantile set to 1 ( Figure 4—source data 1 and 2 ) . An alternative set of data without imputed values is available in Figure 4—source data 3 and 4 . 1988 and 2750 Schistosoma mansoni proteins were identified in heads and in gonads , respectively ( see Figure 4—source data 1 and 2 for statistical testing of differential abundance between male and female samples ) . Among them , 1741 and 2516 could be attributed unambiguously to a Schistosoma mansoni gene and were represented by more than one peptide; these were subsequently analyzed ( See Figure 4—source data 5 ) . DNA reads of male and female S . japonicum are available on the SRA database under study number SRP135770 . Sex and tissue-specific S . mansoni label-free proteomic data are provided in Figure 4—source data 1 to Figure 4—source data 4 . The full bioinformatic pipeline used in this study is provided in Appendix 1 . | The DNA inside cells is organized in structures called chromosomes , some of which can control whether individuals develop as males or females . For instance , female mammals have two X chromosomes , whereas male mammals have one X and one Y chromosome . A mechanism called ‘dosage compensation’ makes sure that females do not produce double the number of transcripts from genes on the X-chromosome as males . In other organisms , including the parasitic flatworms called Schistosomes , females have ZW sex chromosomes , whereas males have two Z chromosomes . In these parasites , males do create more transcripts from genes on the Z chromosome than females do , suggesting they do not have the same kind of compensation mechanisms as mammals . Among Schistosome parasites , the Z chromosome has only been studied in detail in the model organism Schistosoma mansoni . Investigating other closely related species can shed light on how the Z and W chromosomes evolved . Picard et al . studied the Z chromosome in two additional species of Schistosome parasites: the African S . haematobium and the Asian S . japonicum . Using a technique called DNA sequencing , Picard et al . were able to analyse their genes , focusing on a part of the Z chromosome known to have been lost from the W chromosome . The results revealed that this region was different in the African and Asian species . In addition , females of both species expressed genes on their single Z chromosome at fairly high levels . The males did not need to express these genes at a high level because they have two copies – but they did so anyway . This could be because this high expression is a by-product of the way the females have evolved to boost their Z chromosome gene expression . A next step will be to investigate the molecular mechanisms underlying this regulation . Schistosomiasis – a disease caused by this type of flatworm parasite – is one of the deadliest neglected tropical diseases , according to the US Centers of Disease Control . It kills more than 200 , 000 people a year . Better understanding of the reproductive biology of this parasite could eventually help to develop ways to control it by interfering with its reproduction . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"genetics",
"and",
"genomics"
] | 2018 | Evolution of gene dosage on the Z-chromosome of schistosome parasites |
Cortical microinfarcts are linked to pathologies like cerebral amyloid angiopathy and dementia . Despite their relevance for disease progression , microinfarcts often remain undetected and the smallest scale of blood flow disturbance has not yet been identified . We employed blood flow simulations in realistic microvascular networks from the mouse cortex to quantify the impact of single-capillary occlusions . Our simulations reveal that the severity of a microstroke is strongly affected by the local vascular topology and the baseline flow rate in the occluded capillary . The largest changes in perfusion are observed in capillaries with two inflows and two outflows . This specific topological configuration only occurs with a frequency of 8% . The majority of capillaries have one inflow and one outflow and is likely designed to efficiently supply oxygen and nutrients . Taken together , microstrokes bear potential to induce a cascade of local disturbances in the surrounding tissue , which might accumulate and impair energy supply locally .
As the brain’s energy storage is limited , a sustained supply of oxygen and nutrients is crucial to avoid local tissue damage . Accordingly , flow disturbances even at the level of individual vessels can result in cortical tissue lesions , so called microinfarcts ( Nishimura et al . , 2007; Shih et al . , 2013; Zhang et al . , 2015; Taylor et al . , 2016; Summers et al . , 2017 ) . In recent decades , it has become evident that such microinfarcts are linked to various pathologies , for example cerebral amyloid angiopathy ( CAA ) , Alzheimer’s disease ( AD ) , and dementia ( van Veluw et al . , 2021; Shih et al . , 2018; Pétrault et al . , 2019; van Veluw et al . , 2017; Smith et al . , 2012 ) . Depending on the severity of flow disturbance , the dimension of documented microinfarcts ranges between 50 µm to a few millimeters ( Shih et al . , 2013; Taylor et al . , 2016; Summers et al . , 2017; Shih et al . , 2018; van Veluw et al . , 2017; Smith et al . , 2012; Brundel et al . , 2012 ) . This small size renders the quantification of the brain’s microinfarct burden challenging and makes it difficult to gain insights on their role for local blood supply . Animal models allow a more refined study of the etiology of microinfarcts ( Shih et al . , 2018 ) . By occluding vessels via photothrombosis ( Nishimura et al . , 2007; Shih et al . , 2013; Zhang et al . , 2015; Taylor et al . , 2016; Summers et al . , 2017; Underly et al . , 2017; Nishimura et al . , 2006; Zhang et al . , 2020; Nishimura et al . , 2010; Enright et al . , 2007 ) or by injecting microemboli ( Wang et al . , 2017; Wang et al . , 2012; Nozari et al . , 2010; Silasi et al . , 2015; Zhu et al . , 2012; Lam et al . , 2010 ) , microstrokes can be induced and their impact on blood flow and on the surrounding tissue can be studied . Here , most studies focus on the occlusion of penetrating vessels , which because of their one-dimensional topology ( Schmid et al . , 2019a; Duvernoy et al . , 1981; Blinder et al . , 2013 ) have been identified as the ‘bottleneck of perfusion’ ( Nishimura et al . , 2007 ) . Less attention has been given to occlusions of descending arteriole ( DA ) offshoots and capillaries . While the occlusion of DA offshoots causes a maximal infarct volume of 0 . 8 nl ( 275 times smaller than for DA occlusions ) , no tissue damage could be detected for the occlusion of capillaries > 2 branches apart from the DA ( Shih et al . , 2013 ) . However , the effect of anesthesia on these results remains unknown . This is because anesthesia can act as a vasodilator and tends to increase red blood cell ( RBC ) flux and tissue oxygenation ( Lyons et al . , 2016; Roche et al . , 2019 ) . Importantly , it has also been shown that single-capillary occlusion causes flow reversals and RBC speed reductions of up to 90% in the vessels downstream of the occluded capillary ( Nishimura et al . , 2006 ) . Additionally , singlecapillary occlusion can induce the formation and alter the morphology of amyloid-beta ( Aβ ) plaques ( Zhang et al . , 2020 ) , which are related to AD and CAA . Taken together , even if single-capillary occlusion might not directly cause local tissue damage , it disturbs blood flow locally and might impair tissue clearance . As such , single-capillary occlusions could play an important role in the development and progression of larger disturbances and pathologies . Further studies have investigated the effect of simultaneously occluding multiple capillaries or multiple microvessels of larger caliber ( Underly et al . , 2017; Lam et al . , 2010 ) . Underly et al . , 2017 showed that the occlusion of ~10 proximal capillaries via photothrombosis leads to deterioration of the blood brain barrier ( BBB ) . The accumulation of occluded capillaries can also influence the total perfusion of the cortical vasculature . In a simulation study related to in vivo investigations in an AD mouse model , Cruz Hernández et al . , 2019 showed that with 2% of capillaries stalled , cortical cerebral blood flow is reduced by ~5% . Capillary stalls have also been identified as an important factor for incomplete reperfusion after stroke ( El Amki et al . , 2020; Erdener et al . , 2021 ) . These aspects underline the need for an in-depth quantification of blood flow changes in response to single-capillary occlusion , which will allow us to better understand the role of these small disturbances on local tissue perfusion . We will identify factors influencing the severity of micro-occlusions and quantify the area of impact . Additionally , by looking at the smallest possible scale of occlusion , valuable insights on the robustness of perfusion within the capillary bed can be gained and our knowledge of topological characteristics of cortical capillary beds can be extended . In this context , we will also analyze the arrangement of arteriole- and venule-sided capillaries . Moreover , identifying the smallest scale of disturbance is a prerequisite for the correct interpretation of disturbances and changes observed on a larger scale . To address these questions , we employed blood flow simulations in realistic microvascular networks ( MVNs ) from the mouse cortex ( Blinder et al . , 2013; Schmid et al . , 2017; Schmid et al . , 2019b ) , in which individual capillaries have been occluded . Using an in silico approach comes with several advantages . First of all , it is challenging to monitor blood flow changes in vivo with single vessel resolution in multiple vessels or even entire vascular networks simultaneously . This problem is even more pronounced , if the focus is on blood flow changes in the capillary bed , which is highly interconnected ( Schmid et al . , 2019a; Blinder et al . , 2013; Schmid et al . , 2017; Lauwers et al . , 2008; Hirsch et al . , 2012; Smith et al . , 2019 ) and in which the flow field is highly heterogeneous and fluctuating ( Schmid et al . , 2017; Schmid et al . , 2019b; Kleinfeld et al . , 1998; Villringer et al . , 1994; Guibert et al . , 2010; Parpaleix et al . , 2013 ) . Secondly , in silico studies allow us to investigate the impact of single-capillary occlusions in an isolated manner . This is in contrast to in vivo analyses , where a capillary occlusion will always be accompanied by a response from directly neighboring cells ( e . g . endothelial cells , mural cells , microglia ) . By studying flow changes in response to the occlusion of 167 different capillaries , we reveal that the severity of a microstroke strongly depends on the local vascular topology and the baseline flow rate in the occluded capillary . More precisely , in the worst-case scenario , the flow rate dropped by as much as 70% in the direct vicinity of the occluded capillary . As well as this , a microstroke locally reduces the number of available flow paths between DAs and ascending venules ( AVs ) . This aspect might play an important role for the upregulation of blood flow during neural activation , since the ability of the microvasculature to adapt to local changes in energy demand might be impaired in the disturbed flow field . The fact that the worst-case scenario only occurs with a frequency of 8% across all capillaries and the re-routing of blood to neighboring vessels suggests that the capillary bed offers an inherent robustness toward single-capillary occlusions . These aspects as well as compensatory oxygen supply from neighboring capillaries likely help to avoid severe hypoxic conditions in response to single-capillary occlusion . Our results further indicate that the different vascular topologies are not only relevant for the severity of the microstroke , but that they might in fact fulfill distinct functional tasks . We postulate that there is a topological difference between capillaries responsible for the distribution of blood and capillaries responsible for supplying oxygen and nutrients to the cortical tissue . In summary , our work provides an in-depth quantification of flow changes in response to single-capillary occlusions and reveals novel topological characteristics of the cortical microvasculature . Our results give valuable insights into the role of microinfarcts , which are relevant for future in vivo studies on the robustness of oxygen and nutrient supply and on pathological flow disturbances .
Microvascular bifurcations are either divergent or convergent . Thus , depending on the bifurcation types at the source and the target vertex of the MSC , four topological configurations are possible at the MSC ( Figure 1a–d ) . To identify the MSC type , the flow directions in all five capillaries need to be known . An identification based purely on the topological arrangement and appearance of the vessel is not possible . To investigate the impact of the topological configuration on the severity of the microstroke , we performed ≥20 microstroke simulations per configuration . Based on the time-averaged flow field before and after stroke , we computed the thresholded relative flow change Δqij for each vessel ( Materials and methods ) . Note that , the response to long-lasting stalls ( >20 s ) and permanent occlusions is equivalent for the observation period considered in this work . In a first step , we analyzed the relative flow changes Δqij in the vessels in up to five generations up- and downstream of the MSC . As more than 80% of all capillaries in the vicinity of the MSC experienced a decrease in flow ( Figure 1—figure supplement 1e–h ) , the subsequent analyses focus on capillaries with a reduction in flow . Figure 1e–g shows that the relative changes are larger for MSCs fed by two upstream vessels and for MSCs feeding two downstream vessels . For the worst-case scenario , that is MSCs with a convergent bifurcation upstream and a divergent bifurcation downstream ( 2-in-2-out , Figure 1a , e ) , the median relative decrease is still <−30% at generation ±2 from the site of occlusion . In contrast for the best-case scenario , that is MSCs with a divergent bifurcation upstream and convergent bifurcation downstream ( 1-in-1-out , Figure 1d , h ) , the median relative change is ~15% at generation ±2 ( p<0 . 001 at generation ±2 , Supplementary file 1c ) . The differences between 2-in-2-out and 1-in-1-out are even more pronounced for vessels of generation ±1 . Here , the median relative drop in blood flow is as large as 70% for 2-in-2-out , while it is only 22% for 1-in-1-out ( p<0 . 001 at generation ±1 , Supplementary file 1c ) . 2-in-1-out and 1-in-2-out are intermediate MSC types ( Figure 1b , c and f , g ) . For example , on the upstream side , 2-in-1-out experienced relative changes comparable to 2-in-2-out , while on the downstream side , the trends correspond to the ones observed for 1-in-1-out ( Supplementary file 1c ) . Analyzing the flow direction changes reveals that the occlusion of a 2-in-2-out lead to a flow reversal in one of the two vessels at generation −1 and 1 for most MSCs ( Figure 1—figure supplement 2a ) . This is plausible because from a fluid dynamical point of view , there are only two possible outcomes for a 2-in-2-out occlusion . The first is the observed flow reversal at generation ±1 . The second would be the complete cessation of flow in generation ±1 , which would lead to a larger infarct volume and thus would increase the severity of the microstroke . The cessation of flow in generation ±1 has only been observed for a small number of MSCs of type 2-in-2-out , 2-in-1-out or 1-in-2-out . For some of these scenarios , a very specific topology was identified at generation ±1 , where the two generation ±1 vessels are connected to the same generation ±2 vessel ( Figure 1—figure supplement 2b ) . We conclude that the cessation of flow in vessels adjacent to the MSC is rare and that the less severe flow reversal is more common . At 1-in-1-out capillaries , flow direction changes were rare , and our results across the MSC types show that flow reversal mostly occur if the MSC is fed by two inflow/feeds two outflow capillaries . To predict oxygen and nutrient supply in the tissue around the MSC , it is important to account for changes in capillaries , which are in the direct vicinity of the MSC but not directly upstream or downstream of the MSC . Therefore , we defined an analysis box around the MSC and compute its total inflow before and after stroke ( Figure 2a–e , Materials and methods ) . The initial analysis box volume is set to 0 . 2 nl and was chosen such that each MSC fits into the initial box volume and that the box has at least five inflows . The box volume was increased progressively and the relative inflow difference has been recomputed ( Figure 2a , Materials and methods ) . This analysis allows us to comment on the reduction in perfusion of a tissue volume around the MSC capillary . Moreover , it provides an estimate of the tissue volume , which is affected by a reduced perfusion in response to the microstroke . In line with the relative flow changes in the upstream and downstream vessels ( Figure 1e–h ) , we observed the largest inflow reduction for 2-in-2-out ( Figure 2b ) . For the initial box volume , that is a volume factor of 1 . 0 , the median inflow reduction is as large as −14% . For a volume factor of 1 . 75 , the median inflow reduction already drops to −5 . 1% ( p=0 . 005 , pairwise t-test with Bonferroni correction ) . However , it is not until a volume of 0 . 6 nl ( volume factor: 3 . 0 ) that the median inflow reduction approaches 0% . For MSC-type 2-in-1-out , the median inflow reduction for a volume factor of 1 . 0 is −13% ( Figure 2c ) . The tissue around MSC-types 1-in-2-out and 1-in-1-out did not experience significant changes in total inflow . Here , for all volume factors , the median inflow difference is smaller than 2 . 5% . Importantly , the resulting inflow reduction in the analysis box is also affected by the topological connectivity around the MSC and the redistribution of flow in response to a microstroke . These aspects become apparent if we compare the relative flow changes in vessels with different topological positions with respect to the MSC . We discern three topological positions: ( 1 ) vessels that are directly upstream and downstream of the MSC , ( 2 ) vessels that run parallel to the MSC , and ( 3 ) vessels that do not belong to the first two categories , that is distant vessels ( Figure 2f , Figure 2—figure supplement 1e , Materials and methods ) . Figure 2—figure supplement 1a shows that for up to a volume factor of 2 , more than 50% of the vessels in the analysis box are directly upstream or downstream of the MSC . In these vessels , we had a significant flow reduction ( Figure 2g ) . In contrast , in the vessels that run parallel to the MSC , we observed an increase in flow ( Figure 2h ) . This clearly indicates that during a microstroke the flow is redistributed to pathways parallel to the MSC . However , only ~15–20% of all capillaries in the analysis box are parallel ( Figure 2—figure supplement 1b ) , and consequently , we still observed an overall flow reduction in the analysis box . In the third vessel category , the distant vessels , the median relative flow difference is <2% for all volume factors ( Figure 2i ) . This result confirms that the impact of a microstroke is most pronounced in vessels that are directly connected to the MSC . Worthy of note is that for a tissue volume of 0 . 4 nl ( i . e . volume factor = 2 ) ~50% of vessels in the analysis box are distant and parallel capillaries ( Figure 2—figure supplement 1a–d ) . This topological configuration might be beneficial for the robustness of perfusion of the tissue volume around the MSC . Because even if the total inflow decreases in the tissue volume around the MSC , there is always a fraction of vessels within the analysis box that are not significantly affected by the microstroke ( distant vessels ) or that experience an increase in flow in response to the microstroke ( parallel vessels ) . Therewith , an even larger drop in overall perfusion can be avoided and a minimum remaining perfusion can be ensured . This likely is beneficial to avoid a significant drop in oxygen partial pressure ( pO2 ) in the tissue around the MSC . The most relevant hemodynamic quantity for local pO2 is the RBC flux ( Lücker et al . , 2017 ) . Thus , in addition to investigating changes in flow , we repeated our analyses for changes in RBC flux ( Figure 2—figure supplement 2 ) . As RBC flux and flow rate are related the general trends are comparable for both quantities . Interestingly , the reduction in RBC flux in the analysis box around the MSC is larger than for the flow rate ( Figure 2—figure supplement 2c , d ) . This indicates that a single-capillary occlusion also affects the distribution of RBCs , which might further increase the risk of local tissue hypoxia . As our study is limited to changes in perfusion within the vasculature , further investigations resolving oxygen transport within the tissue are necessary to answer the question if a single-capillary occlusion significantly affects local tissue pO2 . Our results demonstrate that the local vascular topology plays a crucial role in the severity of a microstroke . To identify further structural and functional characteristics relevant to the level of flow change in response to microstroke , we repeated our analysis for eight additional cases . We looked at the impact of the baseline flow rate in the MSC ( case 5 ) , the cortical depth ( cases 8–12 ) , and the distance to the penetrating vessels ( cases 6–7 ) . An overview of the selection criteria for each vessel subset is provided in Supplementary file 1a . In this study , we focused on 2-in-2-out MSCs because we expect the largest changes here . For the case with a higher baseline flow rate , we observed that the relative change tends to be larger at generations ±3 , ±4 , and ±5 ( paired t-test: ±3: p<0 . 015 , ±4: p<0 . 04 , ±5: p<0 . 04 , Figure 2—figure supplement 3a , b ) . However , no significant two-way interaction between baseline flow rate category and generation and no main effect of the baseline flow rate could be detected ( likely because the relative changes do not differ at generations ±1 and ±2 ) . The impact of the baseline flow rate on the changes in the vicinity of the MSC is further supported by the analysis of the total inflow change into the analysis box around the MSC ( Figure 2—figure supplement 3c–d ) , where we found a significant main effect of the baseline flow rate on the relative inflow change ( F ( 1 , 45 ) = 13 . 97 , p<0 . 001 , two-way mixed ANOVA ) . Here , the most relevant difference is that the occlusion of a capillary with a high baseline flow rate increases the volume in which a significant decrease in inflow can be observed . No significant differences could be observed for the relative changes in upstream and downstream capillaries for occlusions at different cortical depths or with varying distance to the penetrating vessels ( Figure 2—figure supplement 4 and Figure 2—figure supplement 5a–b ) . Regarding the analysis over cortical depth , it is important to note that the baseline flow rate of the chosen MSC has to be between 0 . 1 and 7 . 0 µm3/ms . This selection criterion might cancel out the potential effects of the decrease in flow rate over depth ( Schmid et al . , 2017; Guibert et al . , 2010; El-Bouri and Payne , 2018; Li et al . , 2019 ) . We observed a significant effect of the position of the MSC along the capillary path for inflow changes in the analysis box around the MSC ( Figure 2—figure supplement 5c–d ) . We hypothesize that these differences are affected by the relative frequency of upstream and downstream , parallel , and distant capillaries in the analysis box ( Figure 2—figure supplement 1g–h ) . Our results show that single-capillary occlusion affects the flow rate in multiple capillaries in the direct vicinity of the MSC . Moreover , in vivo observations suggest that capillary stalls are more likely in low flow capillaries ( Erdener et al . , 2019; Hartmann et al . , 2021 ) , and thus , microstrokes might accumulate in the direct vicinity of the MSC . To investigate the impact of an accumulation of capillary occlusions around the MSC , we performed simulations in which three , five , seven , and nine capillaries have been occluded in the analysis box around the MSC with a volume of 0 . 3 nl ( volume factor = 1 . 5 , Figure 3a ) . The simulations have been performed sequentially and in each step the two capillaries with the lowest time-averaged flow rate have been occluded for the subsequent simulation . Figure 3b shows that the relative flow difference in the analysis box increases with the number of occluded capillaries . This is most apparent for the occlusion of nine capillaries , where we observed a flow decrease of ~20% in an analysis box 1 . 6 nl ( volume factor = 8 ) . To further analyze the perfusion changes within the analysis box , we counted the number of vessels with a flow decrease within the analysis box ( Figure 3c ) . The number of vessels with a flow decrease increased with the volume factor , which underlines that the capillary occlusions also affect the perfusion in neighboring vessels . Interestingly , the number of vessels with flow decrease is smaller if more capillaries are occluded . This indicates that single-capillary occlusion causes a small flow reduction in a larger number of capillaries . In contrast , if multiple proximal capillaries are occluded a re-routing of flow occurred and a smaller number of vessels experienced a flow decrease . This is consistent with the observations in Figure 2g–i , where we describe a flow increase in vessels parallel to the MSC and shows how the perfusion in the capillary bed adapts to local disturbances of increasing severity . It is well established that pO2 in capillaries shortly downstream of DAs is higher than in capillaries just upstream of AVs ( Li et al . , 2019; Sakadžić et al . , 2014 ) . Moreover , it has been suggested that the tissue supplied by venule-sided capillaries might be more susceptible to hypoxia in the case of blood flow disturbances or during neural activation ( Sakadžić et al . , 2014; Lücker et al . , 2018a; Devor et al . , 2011 ) . Consequently , the spatial arrangement of arteriole-sided and venule-sided capillaries with respect to each other might be an important topological feature for the robustness of oxygen and nutrient supply . A convenient way to avoid local hypoxia could be obtained by a topological structure where arteriole-sided and venule-sided capillaries are positioned in close proximity to each other . To investigate the spatial arrangement of arteriole-sided and venule-sided capillaries with respect to each other , we introduce the AV-factor . The AV-factor for each capillary has been computed by identifying all paths leading from the capillary to all possible DA end points and to all possible AV end points . The AV-factor has subsequently been calculated from the median distance to all DA/AV end points ( Figure 4a , Materials and methods ) . The AV-factor is close to 0 if the capillary is close to a DA and is close to one if the capillary is adjacent to an AV . We define arteriole-sided capillaries as capillaries with an AV-factor < 0 . 5 and venule-sided capillaries with an AV-factor ≥ 0 . 5 . The following investigations have been performed in MVN1 and MVN2 . The precise analysis approaches are described in more details in the Materials and methods . In an initial analysis we computed the shortest distance between a discretization point along a venule-sided capillary and an arteriole-sided capillary . As a reference we additionally calculated the shortest distance to any vessel around a venule-sided capillary ( Figure 4b ) and the average distance between two capillaries . The median shortest distance from a venule-sided capillary to any vessel is 18 µm and 15 µm for MVN1 and MVN2 , respectively . The average distance between two capillaries is 32 µm ( MVN1 ) and 31 µm ( MVN2 ) . The shortest distance from a venule-sided capillary to an arteriole-sided capillary is 2 . 73 ( MVN1 ) and 2 . 78 ( MVN2 ) times larger than the shortest distance to any vessel ( Figure 4c ) . This corresponds to a distance of 46 µm ( MVN1 ) and 41 µm ( MVN2 ) to the closest arteriole-sided capillary , which are only factor 1 . 4 ( MVN1 ) and 1 . 3 ( MVN2 ) larger than the average distance between two capillaries . Subsequently , we analyzed the average AV-factor in analysis spheres of 50 µm surrounding venule-sided capillary points ( Figure 4d ) . For 68% ( MVN1 ) and 60% ( MVN2 ) of all venule-sided capillaries , the average AV-factor in the analysis sphere is smaller than the AV-factor of the venule-sided capillary under investigation ( Figure 4e ) . This implies that these capillaries have multiple arteriole-sided points nearby , which potentially act as backup for oxygen and nutrient supply . The relative difference between the AV-factor of the venule-sided capillary and the mean of all points within the analysis sphere was −5 . 9% and −4 . 2% for MVN1 and MVN2 , respectively . In a further analysis , we computed the average AV-factor for analysis cubes of different sizes ( side length 30–120 µm , Figure 4—figure supplement 1 ) . The range of the average AV-factor per analysis cube goes from almost 0 to 1 . Nonetheless , the median AV-factor across all analysis cubes is independent of the cube size and equal to 0 . 52 and 0 . 54 for MVN1 and MVN2 , respectively . Taken together , the shortest distance of 46 µm ( MVN1 ) and 41 µm ( MVN2 ) to an arterial-sided capillary and the frequent decrease of the average AV-factor in the 50 µm analysis spheres around venule-sided capillaries suggest that arteriole-sided capillaries are well distributed throughout the network . Nonetheless , further studies are necessary to estimate whether proximal arterial-sided capillaries help to avoid hypoxic tissue areas in the vicinity of venule-sided capillaries . Moreover , it has to be kept in mind that in the rodent cortical vasculature , AVs outnumber DAs ( Schmid et al . , 2019a ) . This is in contrast to the primate vasculature where DAs are approximately twice as frequent as AVs ( Schmid et al . , 2019a; Guibert et al . , 2010; Weber et al . , 2008; Adams et al . , 2015 ) . As such , these results might be species dependent . The significant impact of the different topological configurations on the severity of the microstroke raises questions about the frequency of occurrence and the distribution of the different MSC types in realistic MVNs . The following investigations are based on the time-averaged flow field in two realistic MVNs from the mouse somatosensory cortex acquired by Blinder et al . , 2013 , which jointly encompass a tissue volume of ~3 . 6 mm3 and which contain 31 , 400 vessels ( Materials and methods ) . Interestingly , the worst-case scenario , that is MSC-type 2-in-2-out , only occurs with a frequency of 11% ( MVN1 ) and 6% ( MVN2 ) , while the best-case scenario , that is MSC-type 1-in-1-out , represents 44% ( MVN1 ) and 43% ( MVN2 ) of all possible MSCs ( Figure 5d , e ) . Moreover , the median-supplied tissue volume of 1-in-1-out is 52% ( MVN1 ) and 119% ( MVN2 ) larger than the supplied tissue volume of 2-in-2-out ( Figure 5f , g , Materials and methods ) . Consequently , a total of 51% ( MVN1 ) and 55% ( MVN2 ) of the tissue is supplied by 1-in-1-out capillaries and only 8% ( MVN1 ) and 4% ( MVN2 ) is supplied by 2-in-2-out capillaries . This also becomes apparent in Figure 5b where the tissue volume of realistic MVN1 is color-coded based on the MSC type by which it is supplied . The differences in the median-supplied tissue volume are partly caused by the larger median vessel length of 1-in-1-out capillaries ( Figure 5—figure supplement 1b , c ) . The small number of 2-in-2-out capillaries and the small flow reduction for the frequent MSC-type 1-in-1-out suggest that the cortical microvasculature is inherently robust to the occlusion of a single capillary . The significant differences in the supplied tissue volume further underline this aspect . Figure 5h–i shows that the median flow rate in a 2-in-2-out capillary is 2 . 6 ( MVN1 ) and 2 . 2 ( MVN2 ) times larger than in a 1-in-1-out . As a higher baseline flow rate increases the area of impact of a microstroke , we conclude that these differences further contribute to the severity of a microstroke in a 2-in-2-out configuration . We hypothesize that different local topological configurations might fulfill different tasks in microvascular blood supply . While 2-in-2-out capillaries might be more relevant for distributing blood in the cortical microvasculature , 1-in-1-out capillaries are likely designed to robustly deliver oxygen and nutrients to the cortical tissue . This hypothesis is strengthened by the number of unique paths going from DA to AV through the different MSC types ( Figure 5j–k , Materials and methods ) . While for 1-in-1-out we only have 37 ( MVN1 ) and 15 ( MVN2 ) unique paths connecting DA and AV , for 2-in-2-out , we have 297 ( MVN1 ) and 115 ( MVN2 ) unique paths . Generally , the described trends are consistent for MVN1 and MVN2 . However , it is noteworthy that the overall average flow rate is larger in MVN2 and the number of paths per vessel is larger in MVN1 . The latter is likely caused by a higher density of penetrating vessels in MVN2 , which reduces the number of highly interconnected flow paths through the capillary bed . Subsequently , we asked whether the frequency of MSC types varied over cortical depth ( Supplementary file 1b ) or along the pathway between DA and AV ( Figure 5c , Figure 5—figure supplement 1d–k ) . The latter investigation was performed by analyzing the frequency of occurrence of the four MSC types for different AV-factors ( Figure 5c ) . With respect to cortical depth , the frequency of the different MSC types showed the same characteristics . For the distribution of the MSC types along the pathway between DA and AV , we observed that 1-in-2-out capillaries are more frequent on the arterial side of the capillary bed , while 2-in-1-out are more common toward the AVs . This is plausible because at the arterial end blood is distributed to the capillary bed , while it is re-collected close to the AVs . No significant differences could be observed for the distribution of 1-in-1-out and 2-in-2-out capillaries along the capillary path . Notably , 93% ( MVN1 ) and 64% ( MVN2 ) of all paths between DA and AV contain each MSC type at least once . MSC-type 2-in-2-out is most frequently missing along a path between DA and AV . As previously mentioned , the median flow rate decreases significantly over cortical depth ( −66% and –80% for MVN1 and MVN2 , respectively ) . This is consistent for all MSC types ( Figure 5—figure supplement 2d–f ) . No consistent trend as to how the supplied tissue volume changes over depth could be identified ( Figure 5—figure supplement 2g–i ) . Important to note , is that the largest supplied tissue volume is found for 1-in-1-out capillaries in all ALs and the relative supplied tissue volume ( Figure 5—figure supplement 2j–l ) does not vary significantly with depth . Consequently , our conclusion holds that 1-in-1-out capillaries might be key capillaries for nutrient and oxygen discharge . To further investigate the redistribution of flow during a microstroke , we analyzed the number of flow paths leading from DA to AV . To this end , we followed the flow downstream from the main branch of a DA until it reaches an AV main branch ( Materials and methods ) . Importantly , due to the finite size of the MVN , various flow paths do not start at a DA or do not end at an AV . These flow paths are not considered ( Materials and methods ) . In a first step , we computed the total number of unique flow paths between DA and AV main branch ( Figure 6b , e ) . The total number of flow path during baseline in MVN1 is 139 , 399 and does not change significantly for single or multi-capillary occlusion . This large number highlights the interconnectivity of the capillary bed . Nonetheless , it is important to keep in mind that some flow paths only differ by one or a few vessel segments . Subsequently , we investigated if single-capillary occlusion reduces the number of unique DA-AV-endpoint-pairs , that is the total number of fluid dynamically connected pairs between DA and AV end points . Our results show that in response to a microstroke new DA-AV-endpoint-pairs have been connected and existing DA-AV-endpoint-pairs have been lost ( Figure 6—figure supplement 1b ) . However , with respect to the total number of DA-AV-endpoint-pairs , these changes are small ( ~0 . 2% ) . As such , our results suggest that single-capillary occlusion and the occlusion of up to nine proximal capillaries do not reduce the overall number of flow paths and the number of unique DA-AV-endpoint-pairs in MVN1 . To investigate changes in flow paths in the vicinity of the MSC , we count the number of paths going through a predefined capillary ( Figure 6—figure supplement 1c ) and analyzed the change in the number of unique flow paths going through: ( 1 ) capillaries upstream and downstream of the MSC ( up to generation 3 ) , ( 2 ) parallel to the MSC , and ( 3 ) distant to the MSC ( Figure 2f , Materials and methods ) . Based on the results presented in Figure 2g–i , where we detected an increased flow in the parallel vessels , we expected to see an increase in the number of paths going through the parallel vessels . However , no consistent trend could be observed for the three different vessel categories ( Figure 6—figure supplement 1d ) . This suggests that an increase in flow does not necessarily cause an increase in the number of flow paths through the respective capillary . In our last analysis , we examined the number of possible flow paths between given DA-AV-endpoint-pairs . Therefore , the DA-AV-endpoint-pairs are assigned to two categories ( Figure 6c , Materials and methods ) : ( 1 ) before stroke there is at least one path that leads from the DA to the AV end point through the MSC and ( 2 ) none of the paths between the given DA-AV-endpoint-pair go through the MSC . For DA-AV-endpoint-pairs of category 1 , we note a decrease in the number of available flow paths between the respective DA-AV-endpoint-pairs for all MSC types ( Figure 6d ) . The mean ratio of unique flow paths before and after microstroke at the respective DA-AV-endpoint-pair is between 0 . 67 and 0 . 75 . If more than five capillaries are occluded in the vicinity of the MSC , the relative frequency of DA-AV-endpoint-pairs with a decrease in available flow paths rises ( Figure 6f ) . Interestingly , the ratio of unique flow paths before and after microstroke is not affected by the number of occluded capillaries and remains at ~0 . 7 . For the majority of DA-AV-endpoint-pairs of category 2 , we do not observe a change in the number of unique flow path in response to single- or multi-capillary occlusion . Nonetheless , ~10% of DA-AV-endpoint-pairs of category 2 experience an increase or decrease in the number of connecting pathways , which underlines that the impact of capillary occlusion goes beyond directly connected vessels . Taken together , the decrease in the number of pathways between DA-AV-endpoint-pairs of category 1 shows that a microstroke locally reduces the number of available paths . Based on the results shown in the preceding sections ( Figure 2h ) , we suggest that the average flow rate likely increases along some of the remaining paths .
By performing blood flow simulations in realistic MVNs for 167 of single-capillary occlusions , we show that the severity of a microstroke depends on the local vascular topology and on the baseline flow rate in the occluded capillary . The largest impact is observed if capillaries with two inflows and two outflows ( 2-in-2-out ) are occluded . Here , a drop in flow rate > 30% is still observed two generations away from the MSC . In contrast , flow rate changes remain below 25% for all capillaries at a MSC with a divergent bifurcation upstream and a convergent bifurcation downstream ( 1-in-1-out ) . In accordance , the occlusion of a 2-in-2-out capillary reduces perfusion by 14% in a tissue volume of 0 . 2 nl . For the occlusion of a capillary with a more than three times higher baseline flow rate , a 14% drop in perfusion can still be observed for a tissue volume of 0 . 55 nl . Besides a local decrease in perfusion , single-capillary occlusion also causes a decrease in the number of available flow paths in the vicinity of the MSC . Our observation that the severity of a microstroke is affected by the baseline flow rate of the occluded vessels is in agreement with previous in vivo and in silico observations for the occlusion of penetrating vessels ( Shih et al . , 2013; Taylor et al . , 2016; Nishimura et al . , 2006; Blinder et al . , 2013 ) . Additionally , Nishimura et al . , 2006 report an RBC velocity reduction in response to single-capillary occlusion of 93% and 55% in downstream vessels of generation 1–2 and 3–4 , respectively . Although the in vivo velocity reductions are slightly higher , they generally compare well with our results for the occlusion of a 2-in-2-out and a 1-in-2-out MSC . However , Nishimura et al . , 2006 do not observe flow reversals and velocity reductions in upstream and parallel vessels . These differences are likely due to the fact that many of the occluded vessels in Nishimura et al . , 2006 are direct offshoots of the DA main branch . Due to significantly larger flow velocities in the DA main branch , velocity reductions and reversals upstream of the site of occlusion are not to be expected . To be highlighted is that for all MSC types the effects of single-capillary occlusions are spatially constrained . To be more precise , no significant reduction in flow rate is visible five generations away from the MSC and in a tissue volume of 0 . 3 nl around the MSC the perfusion drops maximally by 10% for all MSC types . This is in contrast with the occlusion of DAs where the flow rate does not fully recover until the 10th downstream vessel and where the infarct volume is as large as 220 nl ( Nishimura et al . , 2007; Shih et al . , 2013; Blinder et al . , 2013 ) . For scenarios in which multiple proximal capillaries are occluded , the volume of the affected area increases and the local drop in perfusion is more severe . As the supplied tissue volume of a 1-in-1-out MSC is 0 . 055 nl , which is approximately 15 times smaller than the infarct volume observed for the occlusion of a DA offshoot ( Shih et al . , 2013 ) , we postulate that the occlusion of single-capillaries does not directly cause tissue damage . This hypothesis is in line with the results of Shih et al . , 2013 . Nonetheless , our results show that single-capillary occlusion has a strong impact on the local flow field . As such , it seems plausible that the altered flow field is a possible mechanism by which to affect Aβ deposition ( Zhang et al . , 2020 ) or solute clearance via the perivascular space in general ( Arbel-Ornath et al . , 2013 ) . The local disturbances in the flow field might increase the susceptibility for vessel ruptures ( van Veluw et al . , 2021 ) or additional occlusions , which subsequently might further impede clearance ( Hawkes et al . , 2014; Carare et al . , 2008 ) as well as oxygen and nutrient supply in an increasing area around the MSC . For the occlusion of larger caliber vessels , it has been shown that proximal microinfarcts are likely to coalesce ( Shih et al . , 2013 ) and that BBB leakage and intravascular platelet aggregation ( Taylor et al . , 2016 ) , as well as deficits in neuronal activity and functional vasodynamics ( Summers et al . , 2017 ) , are also observed beyond the microinfarct border . However , whether comparable effects can be triggered by the occlusion of a single-capillary remains unknown . Likewise , we do not know whether single-capillary occlusion induces local tissue hypoxia or if proximal vessels and increased gradients in tissue pO2 compensate for the lack of perfusion in the occluded capillary . Due to the dense capillary bed and the redistribution of flow to neighboring vessels , a single-capillary occlusion likely only causes hypoxic conditions if tissue pO2 is already low during baseline and if the oxygen saturation in proximal capillaries is too low to compensate for the drop in perfusion . Nonetheless , because of the significant impact on the flow field , single-capillary occlusions might lead to a local drop in tissue oxygenation and oxygen saturation within RBCs , which might provoke a cascade of consecutive responses in the affected tissue around the MSC or downstream of the occluded area . Furthermore , the impact of a microstroke on tissue oxygenation can be affected by the level of oxygen within the occluded capillary and by the local arrangement of arteriole- and venule-side capillaries with respect to each other . We hypothesize that arteriole-sided capillaries with a high oxygen content might be distributed in a convenient fashion throughout the vasculature to enhance the robustness of oxygen delivery throughout the tissue . Indeed , Nishimura et al . , 2010 provided supporting evidence for this hypothesis by showing that capillaries with varying topological distances to the DA can be in spatial proximity . However , further studies resolving oxygen partial pressure within capillaries in MVNs are necessary to confirm this hypothesis . The remaining unknowns clearly underline the need for an in-depth in vivo quantification of the impact of single-capillary occlusion . Based on our results , we suggest that the focus of future in vivo microstroke studies should be on tissue oxygenation , Aβ deposition and long-term changes in the vicinity of the MSC . In these investigations , it is important to keep in mind that the severity of the microstroke is affected by the local vascular topology and the baseline perfusion of the MSC . Consequently , care should be taken that effects are analyzed in an MSC-type-specific manner . In addition to in vivo approaches , in silico studies resolving oxygen discharge from individual RBCs ( Lücker et al . , 2015 ) are a convenient tool to improve our understanding of the impact of single-capillary occlusions on tissue oxygenation . As previously stated , the severity of a microstroke depends on the local vascular topology . Worthy of note , we observe significant differences in the frequency and the characteristics of the local vascular topologies ( MSC types ) . 1-in-1-out is by far the most frequent MSC type and supplies the largest tissue volume . At the same time , it is characterized by having the smallest average flow rate and by containing the smallest number of unique paths connecting DAs and AVs . In contrast , 2-in-2-out is the rarest MSC type and contains the largest number of flow paths connecting DAs and AVs . We postulate that MSC-type 2-in-2-out is responsible for distributing blood within the capillary bed and that MSC-type 1-in-1-out is designed to enable efficient oxygen and nutrient discharge to the tissue . In vivo evidence supporting this hypothesis is not yet available . Here , the first step would be to confirm the characteristics of the different MSC types in vivo and to subsequently study the role of these differences on oxygen and nutrient supply . The frequency of 2-in-2-out MSCs is low , and in a volume of 0 . 2 nl around the MSC , 27% of vessels show no flow decrease after a microstroke . These two features suggest that the capillary bed offers an inherent level of robustness toward single-capillary occlusion and they agree well with the reported highly interconnected nature of the capillary bed that allows efficient re-routing of blood flow ( Schmid et al . , 2019a; Blinder et al . , 2013; Lauwers et al . , 2008; Hirsch et al . , 2012; Smith et al . , 2019; Cassot et al . , 2006; Lorthois and Cassot , 2010 ) . Nonetheless , the significant differences between the characteristics of the MSC types raise further questions regarding the origin and the severity of microstrokes . First of all , due to the larger supplied tissue volume , might the occlusion of a 1-in-1-out MSC be more severe for oxygen and nutrient supply , while the occlusion of a 2-in-2-out MSC has a larger impact on the flow field ? Here , in vivo studies monitoring tissue oxygenation in response to capillary occlusion or combined blood flow and oxygen transport simulations could provide insights into the most critical MSC type for oxygen and nutrient supply . Secondly , Would a microstroke be more probable in a 1-in-1-out MSC ? This idea is based on the lower average flow rate in 1-in-1-out MSCs , which implies that the vessel might be blocked more easily ( Erdener et al . , 2019; Hartmann et al . , 2021 ) . However , to answer this question , we need to improve our understanding of the mechanisms that cause capillary occlusions . In healthy mice , stalls occur with a frequency < 1% ( Cruz Hernández et al . , 2019; El Amki et al . , 2020; Erdener et al . , 2019; Reeson et al . , 2018 ) . This number increases to 1 . 8% in AD ( Cruz Hernández et al . , 2019 ) and to 30% in the core of the stroke ( El Amki et al . , 2020 ) . For both pathological conditions , the majority of stalls are caused by neutrophils adhering to the vessel wall and occurred across all capillary diameters ( ~4–10 µm ) ( Cruz Hernández et al . , 2019; El Amki et al . , 2020 ) . Besides this aspect , no capillary phenotype could be identified in which stalls are more prominent ( Erdener et al . , 2019; Reeson et al . , 2018 ) . Next to the phenotype of individual capillaries , the type of bifurcation might be an important factor for the likelihood of occlusion . For example , convergent bifurcations are likely more susceptible to blood particles getting stuck , which might cause an occlusion in capillaries up- and downstream of the bifurcation . However , if the origin of occlusions does not come from blocked particles but from plaque deposits or mural cell activity , then the situation is less clear and all MSC types are likely affected to similar extent . Studying the effect of single-capillary occlusions in an isolated manner in our in silico model is advantageous on the one hand , but limitated on the other hand . For example , our simulation model does not account for dynamic responses of the vasculature . It has been shown that single DA occlusion induces a heterogeneous response in the capillary bed comprising capillary dilations and constrictions ( Taylor et al . , 2016; Nishimura et al . , 2010 ) . Nonetheless , the in silico approach enables us to perform an in-depth study of fluid dynamical changes in response to single-capillary occlusion detached from external and internal influences . Moreover , our observations can be conveniently linked to the surrounding vascular topology . These insights can subsequently be used to distinguish changes observed in in vivo studies . Taken together , we show that for 57% of all capillaries an occlusion significantly reduces the flow rate in the directly neighboring capillaries . Consequently , we conjecture that a single-capillary occlusion can be the starting point of a cascade of small consecutive disturbances , which might be relevant for the development of larger microinfarcts and for the progression of pathologies . In addition , we reveal novel features of the capillary bed , which are relevant for the robustness of perfusion and for advancing our understanding of topological characteristics of this highly interconnected network . Importantly , resolving the smallest scale of disturbance is not only essential to improve our understanding of microinfarct development , but might eventually offer novel possibilities for therapeutic treatment and prevention .
The MVN is represented as a graph structure , that is it consists of a set of nodes ni connected by a set of edges eij . The subscript ij indicates that edge eij is connecting node ni and nj . Anatomically accurate MVNs have been acquired by Blinder et al . , 2013 from the mouse somatosensory cortex by two-photon laser scanning microscopy . They are embedded in a tissue volume of ~1 . 6 mm3 ( MVN1 ) and ~2 . 2 mm3 ( MVN2 ) and contain ~12 , 100 and ~19 , 300 vessels , respectively . The vessels are labeled as pial arteries ( PAs ) , DAs , capillaries ( Cs ) , AVs , and pial veins . For the penetrating vessels , that is DAs and AVs , we additionally differ between the main branch and the offshoot vessels . The vessel type is assigned by following the vessels from the cortical surface and by applying a diameter criterion which requests that two subsequent vessels have a diameter smaller than 6 µm in order to change the vessel type from DA to C ( Schmid et al . , 2017 ) . The equivalent criterion is applied on the venule side . To differentiate between the main branch of the penetrating vessels and the offshoots , we use a criterion that is based on the branching angle and the length of the resulting main branches . This approach ensures that short offshoots are not labeled as main branch . To compute the pressure field and the blood flow rates in the realistic MVN , we employ the continuity equation at every node and Poiseuille’s law along the vessels . This approach is valid due to the small Reynolds numbers in the cortical microvasculature ( Re < 1 . 0 for all vessels ) . To account for the presence of RBCs , the vessel resistance is multiplied by the relative effective viscosity μrele . Taken together , Poiseuille’s law readsqij=pi-pjRije=πDij4128Lijμμrele ( pi-pj ) where Dij and Lij are the vessel diameter and the length and pi and pj are the pressure at node i and j , respectively . μ is the dynamic plasma viscosity and μrele the relative effective viscosity , which is computed as a function of the hematocrit and the vessel diameter as described in Pries et al . ( in vitro formulation ) ( Pries and Secomb , 2005 ) . The hematocrit of individual vessels is computed from the discretely tracked RBCs . In order to correctly model the motion of RBCs , we account for the Fahraeus effect ( Pries and Secomb , 2005; Fåhraeus , 1929 ) and the phase separation . The phase separation in vessels with a diameter larger than 10 µm is described based on the empirical relation by Pries and Secomb , 2005 . In vessels with a diameter < 10 µm , single file flow can be assumed , and consequently , we postulate that the RBC follows the path of the largest pressure force ( Schmid et al . , 2017; Schmid et al . , 2015; Fung , 1973; Yen and Fung , 1978 ) . The unequal partitioning of RBCs at divergent bifurcations and their impact on the vessel resistance cause a fluctuating flow and pressure field . In the current study , we focus on the analysis of the time-averaged flow field of the statistical steady state . Our average is computed over 10 turnover times ( 15 . 4 s ) , where one turnover time is defined as the time until 85% of all vessels have been completely perfused at least once . The pressure boundary conditions are assigned as described in Schmid et al . , 2017 . In brief , at the pial vessels , we make use of existing experimental data and assign a diameter-dependent pressure value . The pressure values at capillary in- and outlets are set based on the simulation results of the hierarchical boundary conditions approach . Here , the realistic MVN is implanted into a large artificial MVN . Subsequently , the flow and pressure field for a constant hematocrit is computed and the resulting pressure values are assigned as boundary conditions . The pressure boundary conditions are kept constant for each microstroke scenario . The microstroke simulations are performed in MVN1 . To mimic a microstroke , the diameter of a single capillary is set to 0 . 01 µm . For all investigated scenarios , the resulting flow rate in the MSC is <10−10μm3ms−1 . The average flow rate in a capillary in MVN1 is 4 . 2μm3ms-1 . This proves that the flow rate in the MSC approaches 0μm3ms-1 and therewith confirms the validity of our microstroke model . In total , there are 11 , 386 capillaries in realistic MVN1 . To ensure that we choose representative MSCs the following selection criteria have to be fulfilled: It should be noted that the ‘number of possible MSCs’ is computed by subsequently considering an additional selection criterion . One of our goals is to comment on factors influencing the severity of a microstroke . To analyze the impact of different factors , for example the baseline flow rate in the MSC , additional selection criteria may be prescribed . These criteria are defined in more detail in the related results sections . Supplementary file 1a provides an overview of all selection criteria . In total , we analyzed 12 different cases . For each case , at least 12 microstroke simulations have been performed . A major challenge in comparing blood flow simulations in realistic MVNs is the large variety in flow rates , which ranges from 0 . 09μm3ms-1 to values as large as 26 . 76μm3ms-1 in the capillary bed ( minimum and maximum of 95% of all flow rates in the capillary bed , median:1 . 99μm3ms-1 ) . In such a flow field a large relative change in a vessel with a small baseline flow rate might be negligible , while a small relative change in a vessel with a large baseline flow rate might be significant . To improve the comparability of simulation results , we introduce an absolute threshold , thabs . If the absolute change is smaller than thabs the relative change is set to 0% . To choose an appropriate threshold value , we compare the average flow rate at two points in time for three different averaging intervals ( 10 , 5 , and 3 turnover times ) . The absolute flow change between the two time points is characteristic for the fluctuations of the baseline flow field . Thus , it can be used as a reference of how large the absolute flow change needs to be such that it is likely to be caused by the microstroke and not by baseline fluctuations . The difference between the two time points is 20 s . It becomes apparent that for each of the three averaging intervals > 87% of the capillaries change their flow rate by less than 0 . 1μm3ms-1 ( Supplementary file 1f ) . Consequently , in the microstroke simulations a flow rate change >0 . 1μm3ms−1 is likely caused by the impact of the microstroke and not by baseline fluctuations . Accordingly , we set the absolute threshold thabs to 0 . 1μm3ms-1 . The relative change in flow rate can either be computed directly from the flow rate in the vessel . Δdirqij=qijstroke-qijbaseqijbaseor from the absolute flow rates in the vessel , Δqij=qijstroke-|qijbase||qijbase| , where qijbase and qijstroke are the flow rates in vessel ij for the baseline and the simulation with microstroke , respectively . Even though the second formulation neglects changes in flow direction , it is more suitable for comparing the total perfusion of individual capillaries . As the total perfusion is more relevant for oxygen and nutrient supply , we employ the second expression and analyze flow direction changes separately . Taken together the thresholded relative change is computed asΔdirqij={|qijstroke|−|qijbase||qijbase|for|qijstroke|−|qijbase|≥thabs0 . 0for|qijstroke|−|qijbase|<thabs . The same approach is used to compute the relative change in RBC flux . Here , the threshold is set to 0 . 2 RBCs/s . To analyze differences over cortical depth the realistic MVN is divided into five ALs each 200 µm thick ( Figure 2—figure supplement 4 ) . This analysis approach was first introduced by Schmid et al . , 2017 . To assign a vessel to an AL , either the source or the target vertex of the vessel has to be within the upper and lower bound of the AL ( Supplementary file 1b ) . The second end point of the vessel is required to be within ±50 µm of the bounds of the AL . To comment on the blood supply of a tissue volume around the MSC , we compute the total inflow into an analysis box around the MSC . The volume of the smallest analysis box is chosen such that each MSC fits into the smallest analysis box ( Figure 2a ) . This results in an initial box volume of 0 . 2 nl ( 200 , 000 µm3 ) , which would be equivalent to a cube with a side length of 58 . 48 µm . Moreover , for each MSC we have at least six capillaries in the initial analysis box and at least five capillaries crossing the border of the analysis box . The chosen initial box volume is a compromise between having the smallest possible analysis box around the MSC and ensuring at the same time that sufficient capillaries are within the analysis box to perform a quantitative investigation . The side lengths of the analysis box vary for the different MSC capillaries . To increase the box volume , the side lengths of the smallest analysis box are increased by the same distance in all three directions until we reach the desired box volume . To compute the relative inflow change in response to a microstroke , we add up all inflows during baseline and during stroke and calculate the relative difference between the total inflow during baseline and during stroke . It is important to note that due to flow reversals in response to a microstroke , the number of inflow vessels can change for the baseline and the microstroke case . The equivalent analysis is repeated for increasing box volumes . The relative inflow change per analysis box is depicted in Figure 2b–e , Figure 2—figure supplement 2c , d , Figure 2—figure supplement 3c , d , and Figure 2—figure supplement 5c , d . The change in total flow rate per analysis box is computed comparably to the inflow change in an analysis box . Here , instead of computing the total inflow during baseline and during stroke , we add up the length-weighted total flow rates in the analysis box for baseline and during stroke by summing up the flow rate of all vessels in the analysis box . We consider the vessel tortuosity to compute the vessel length within the analysis box . The total flow rate change per analysis box is used in Figure 2g–i . To study the redistribution of flow in an analysis box , we introduce three vessel categories: ( 1 ) Vessels upstream and downstream of the MSC . ( 2 ) Vessels that branch off/into an upstream/downstream vessel of generation 1 or 2 of the MSC ( parallel vessels ) . Here , we follow each parallel vessel of generation 1 and 2 three segments downstream/upstream to create the entire set of parallel vessels . ( 3 ) All other vessels in the analysis box ( i . e . neither upstream , downstream , nor parallel vessels ) are called distant vessels . A schematic drawing of these vessel categories is provided in Figure 2f , Figure 2—figure supplement 1e . As the whole MVN is connected , distant vessels are also connected to the MSC . However , for this vessel category , the point of connection is relatively far upstream or downstream . This approach allows us to study changes in response to a microstroke with respect to the topological distance from the MSC . Note that the concept of parallel vessels has also been used in Nishimura et al . , 2006 . However , their definition of parallel vessels is different from that used in our analysis . Nishimura et al . , 2006 consider parallel vessels to be only those vessels that have the same source vertex as the occluded vessel . As we hypothesize that the occlusion of a single-capillary might trigger an accumulation of additional microstrokes around the initial MSC , we designed a simulation approach where we sequentially occluded more capillaries around the MSC . From the 27 2-in-2-out simulations , eight MSCs have more than 12 capillaries in the analysis box around the MSC with a volume of 0 . 3 nl ( volume factor = 1 . 5 ) . From this subset we randomly picked six MSC capillaries to study the effect of multi-capillary occlusions . Based on the time-averaged blood flow rates for the simulation with N occlusions , we chose the two capillaries with the lowest blood flow rate in the analysis box around the MSC . These two capillaries will subsequently be occluded and the time-averaged flow field will be recomputed for the new setup with N + 2 occlusions . This sequential approach is repeated until we reach nine occlusions in the analysis box of 0 . 3 nl around the MSC ( Figure 3a ) . The AV-factor is computed to distinguish between capillaries that are close to DAs ( arteriole-sided capillaries ) and those that are close to AVs ( venule-sided capillaries ) . For each capillary ij , we computed all paths to all DA end points and all paths to all AV end points ( Figure 4a ) . This results in a set of path lengths on the arteriole side of the capillary ( distijDA ) and a set of path lengths on the venule side ( distijAV ) . To compute the AV-factor for capillary ij , we use the median of each set of path lengths , that is , AV-factorij=mediandistijDAmediandistijDA+mediandistijAV . The resulting AV-factor lies between 0 and 1 and approaches 0 on the arterial side and one on the venule side of the capillary path . To investigate the distribution of arteriole- and venule-sided capillaries within the vascular network , the following investigations have been performed . First for each venule-sided capillary , we computed the shortest distance to any vessel ( Figure 4b ) and to the closest arteriole-sided capillary ( Figure 4c ) . For this analysis , each vessel is split into multiple discretization points , which are on average 1 . 3 µm apart and which are used to compute the shortest distance . Note that the AV-factor can only be assigned to capillaries along a flow path from DA to AV . Due to the finite size of our simulation domain , the AV-factor is only assigned to 60% and 63% of all capillaries in MVN1 and MVN2 , respectively . In consequence , the calculated shortest distance to an arteriole-sided capillary might be slightly overestimated . In the second investigation , an analysis sphere with a radius of 50 µm is moved along the discretization points of every venule-sided capillary ( Figure 4d ) . All capillary points within these analysis spheres are identified and used to calculate the average AV-factor within the analysis spheres of the venule-sided capillary . To guarantee a representative analysis , only venule-sided capillaries are considered if at least 50% of all capillary points within the analysis spheres have been assigned an AV-factor . For the third study on the distribution of AV-factors within the vascular network , we employ analysis cubes of varying size ( side length 30–120 µm ) . For each cube size , the vascular network is discretized by the analysis cubes and the average AV-factor per analysis cube is computed ( Figure 4—figure supplement 1 ) . Here , an overlap of half the cube side length is used between neighboring analysis cubes . The analysis cube is only considered if it contains at least four capillaries with AV-factor and if at least 50% of the capillaries within the analysis cube could be assigned an AV-factor . The analysis cube with a side length of 30 µm contains on average 4 . 6 ± 0 . 9 ( MVN1 ) and 4 . 7 ± 1 . 0 ( MVN2 ) capillaries with AV-factor . For the analysis cube with a side length of 120 µm , we find 35 . 7 ± 8 . 9 ( MVN1 ) and 51 . 0 ± 11 . 6 ( MVN2 ) capillaries with AV-factor within the cube . This results in 7336 ( MVN1 ) and 17 , 915 ( MVN2 ) analysis cubes with a side length of 30 µm and in 1316 ( MVN1 ) and 1700 ( MVN2 ) analysis cubes for a side length of 120 µm . To comment on the infarct volume of a microstroke and for further topological studies , we compute the supplied tissue volume for each vessel . To do this , the tissue is discretized on a Cartesian grid , in which the realistic MVNs are embedded . One grid cell spans 4 × 4 × 4 µm3 , which results in ~11 . 6 million grid cells for MVN1 and ~15 . 3 million grid cells for MVN2 . Each cell center is assigned to the closest vessel . By summing over all cells assigned to one vessel , we obtain the topological supplied tissue volume per vessel . It is important to note that the topological and the effective supplied tissue volume can differ significantly ( Sakadžić et al . , 2014; Lücker et al . , 2018a; Lücker et al . , 2015; Lücker et al . , 2018b ) . This is because of different oxygen levels along the capillary path . Consequently , for vessels with high oxygen levels , the effective supplied tissue volume is likely larger than the topological supplied tissue volume and vice versa . Nonetheless , we believe that the topological supplied tissue volume is a representative characteristic for the study of topology and perfusion-related aspects of the cortical microvasculature . Please note that for simplicity the topological supplied tissue volume is called supplied tissue volume throughout this manuscript . Flow paths between the penetrating vessels are computed by following the flow from the DA to the AV . For this investigation , a DA end point is defined as the first branch point after the main branch of the DA arteriole . The equivalent definition is used for AVs , that is the end point of an AV is the point proximal to the capillary bed and the start point of the AV is the root of the penetrating tree at the cortical surface . To compute all paths between DA and AV , we first identify all DA and AV end points . Subsequently , for each DA-AV-endpoint-pair , we compute all unique connecting flow paths . Note that some DA and AV end points are not fluid dynamically connected . Additionally , multiple paths enter/leave the MVN across its boundaries . As these paths do not connect a DA with an AV , they are not considered any further for this analysis . The resulting flow path data allows for various investigations: Depending on the underlying data , different statistical tests have been employed . The statistical tests have been performed in R or with the Python Library Stats . The statistical output is summarized in the figure legends and in Supplementary file 1c , d and e . For the relative changes presented in Figures 1–3 , Figure 2—figure supplement 3 and Figure 2—figure supplement 4 , we use a two-way mixed ANOVA with Bonferroni correction ( Supplementary file 1c-d ) . As the data in Figure 2—figure supplement 2 is paired , we compare the grouped data with the paired Wilcoxon test ( grouping based on generation and volume factor , respectively ) . For differences in the characteristics of the four MSC types , we use the Kruskal–Wallis and the Mann–Whitney U test ( Figure 5 , Supplementary file 1e ) . To test for a statistical significant differences in the total number of unique flow paths ( Figure 6b , e ) , we employ the Kruskal–Wallis test . | A blockage in one of the tiny blood vessels or capillaries of the brain causes a ‘microstroke’ . Microstrokes do not cause the same level of damage as a major stroke , which is caused by a blockage in a larger blood vessel that completely cuts off oxygen to a part of the brain for a period . But microstrokes do increase the risk of developing conditions like dementia – including Alzheimer’s disease – later in life . People with these neurodegenerative conditions have fewer capillaries in their brains . The capillaries make up a mesh-like network of millions of vessels that supply most of the energy and oxygen to the brain . Repeated microstrokes may contribute to progressive loss of capillaries over time . Reduced numbers of capillaries may increase memory loss and other brain difficulties . To better understand how microstrokes affect blood flow in the brain , Schmid et al . created a computer model to simulate blood flow in capillaries in the mouse brain . Then , they modeled what happens to the blood flow when one capillary is blocked . The experiments showed that the configuration of the blocked capillary determines how much blood flow in neighboring capillaries changes . Blockages in capillaries with two vessels feeding in and two vessels feeding out caused the greatest blood flow disturbances . But these 2-in-2-out vessels only make up about 8% of all brain capillaries . Blockages in capillaries with different configurations with respect to feeding vessels had less effect . The experiments suggest that most microstrokes have limited effects on blood flow on the scale of the entire brain because of redundancies in the capillary network in the brain . However , the ability of the capillary network to adapt and reroute blood flow in response to small blockages may decrease with aging . Over time , ministrokes in a single capillary may set off a chain reaction of disturbed blood flow and more blockages . This may decrease energy and oxygen supplies explaining age- and disease-related brain decline . Better understanding the effects of microstrokes on blood flow may help scientists develop new ways to prevent such declines . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2021 | The severity of microstrokes depends on local vascular topology and baseline perfusion |
Insulin resistance ( IR ) contributes to the pathophysiology of diabetes , dementia , viral infection , and cardiovascular disease . Drug repurposing ( DR ) may identify treatments for IR; however , barriers include uncertainty whether in vitro transcriptomic assays yield quantitative pharmacological data , or how to optimise assay design to best reflect in vivo human disease . We developed a clinical-based human tissue IR signature by combining lifestyle-mediated treatment responses ( >500 human adipose and muscle biopsies ) with biomarkers of disease status ( fasting IR from >1200 biopsies ) . The assay identified a chemically diverse set of >130 positively acting compounds , highly enriched in true positives , that targeted 73 proteins regulating IR pathways . Our multi-gene RNA assay score reflected the quantitative pharmacological properties of a set of epidermal growth factor receptor-related tyrosine kinase inhibitors , providing insight into drug target specificity; an observation supported by deep learning-based genome-wide predicted pharmacology . Several drugs identified are suitable for evaluation in patients , particularly those with either acute or severe chronic IR .
Systemic insulin resistance ( IR ) is a multi-organ pathophysiological state and an early characteristic of type 2 diabetes mellitus ( T2DM ) . IR contributes to the pathobiology of neurodegeneration ( Norambuena et al . , 2017 ) , heart failure ( Wamil et al . , 2021 ) and viral infections , such as COVID-19 ( Ceriello et al . , 2020; Donath , 2021 ) . Several T2DM drug treatments indirectly reduce IR following improved metabolic homeostasis , making them candidate treatments for various diseases ( Donath , 2021; Everett et al . , 2018; Norambuena et al . , 2017 ) . Drug repurposing ( DR ) aims to accelerate the discovery and reduce the costs of new treatments . Multiple evolving strategies are being trialled , including mining of medical records , development of large databases of drug-gene interactions ( Subramanian et al . , 2017 ) and virtual compound screening ( Himmelstein et al . , 2017 ) . Drug transcriptome responses in cells represent one of the most extensive resources ( Subramanian et al . , 2017 ) , while transcriptomics is also an ideal technology to capture complex biological processes in human tissues ( Jenkinson et al . , 2016; Timmons et al . , 2018; Timmons et al . , 2005 ) . Effectiveness at reversing of the molecular responses to disease ( Wagner et al . , 2015 ) helps to predict drug efficacy in cancer ( Brown and Patel , 2018; Iorio et al . , 2018; Karatzas et al . , 2017; Wang et al . , 2016 ) . Successful application of DR specifically to oncology may reflect that drug profiles are typically generated in tumour cell lines ( Subramanian et al . , 2017 ) and that barriers to clinical validation can be lower compared with many other diseases . Identifying informative disease signatures for DR in cells is challenging ( Chen et al . , 2020; Karatzas et al . , 2017; Regan-Fendt et al . , 2019 ) , particularly when no positive controls exist ( Williams et al . , 2019 ) . Currently , there are no reliable human cellular models for systemic IR , while it remains unclear if multi-gene assays can capture quantitative pharmacological relationships suitable for optimising drug design . However , clinically effective drugs typically target several proteins , many of which are unknown ( Keenan et al . , 2018 ) , highlighting the limitations of single-target drug development programmes . Network modelling and deep learning ( DL ) have been utilised to connect the pharmacological properties of active drugs to their protein targets ( Woo et al . , 2015; Zeng et al . , 2020 ) . For IR we also have effective non-drug treatments ( Nakhuda et al . , 2016; Slentz et al . , 2016; Timmons et al . , 2018; Phillips et al . , 2017 ) , and this enabled production of a novel human-based IR-DR assay – using more than 2000 tissue profiles generated in our laboratories ( Nakhuda et al . , 2016; Slentz et al . , 2016; Timmons et al . , 2018 ) and one other ( Civelek et al . , 2017 ) . Performance of the present RNA-based multi-gene assays was judged against positive control in vivo drug signatures ( Stathias et al . , 2020 ) , genome-wide association ( Lotta et al . , 2017; Vujkovic et al . , 2020 ) and blood proteome-based assays ( Gudmundsdottir et al . , 2020 ) . Validation of the in vitro results for >2500 drugs ( Subramanian et al . , 2017 ) relied on a variety of protein , drug- and disease-centric ( Parisi et al . , 2020 ) criteria: DL-based modelling of drug-to-protein interactions; targeted gene knock down; and published evidence that the drug reduced IR in vivo ( Figure 1 ) .
Homeostasis model assessment version 2 ( HOMA2-IR ) was used to quantify IR ( Wallace et al . , 2004 ) . RNA biomarkers consistently related to fasting IR ( ‘disease’ ) across tissues were combined with those regulated in common across tissues following lifestyle-based reversal of IR ( ‘treatment’ ) . Biomarkers were ranked based on consistent direction and strength of association across two major human organs targeted by insulin ( human adipose and muscle ) because most orally dosed drugs will act systemically . Quantitative network modelling ( Song and Zhang , 2015 ) was used to rank genes for their tissue-based hub connectedness ( Appendix 1—figure 1 ) . In the present study , we considered the performance of only four RNA-based IR-DR assays ( Appendix 1—figure 1; Ganter et al . , 2006 ) ; testing their ability to match the in vivo directionality of positive controls , thiazolidinedione ( TZD ) and oestrogen , expression signatures correctly ( Hevener et al . , 2018; Sears et al . , 2009 ) . The top-scoring RNA signature ( Signature 3A from Appendix 1—figure 1 ) was a statistically ranked combination of disease- and treatment-associated genes ( n = 120 genes ) outranked selection by hub connectedness , recapitulated cellular gene expression patterns indicative of TZD treatment responses in vivo in muscle ( moderated Z-score , p<0 . 0000008 , Appendix 1—figure 2 ) and is referred to as the IR-DR signature/assay hereafter ( Appendix 1—figure 1 ) . Lack of superiority for the assay designed using hub connectedness may be considered at odds with other studies ( Cheng et al . , 2012 ) but could reflect that inclusion of multi-tissue treatment response biomarkers supersedes any benefit of using network weighting . Several Genome-wide Association Study ( GWAS ) IR and T2DM ( Lotta et al . , 2017; Vujkovic et al . , 2020 ) -derived signatures ( e . g . Signature 4 , Appendix 1—figure 1 ) were considered but were unable to match positive control drugs in vivo . The T2DM blood proteome signature ( Gudmundsdottir et al . , 2020 ) had a weak association with one positive control drug . Protein-level network interactions formed by each list ( Appendix 1—figure 1 ) were distinct ( Appendix 1—figure 2 ) and were only possible to partially recreate from existing databases ( Li et al . , 2018a ) . The largest available database of in vitro drug signatures ( Subramanian et al . , 2017 ) was used to identify cell-type agnostic drug responses . To achieve this , we utilised aggregated scores ( the maximum quantile statistic from the within-cell line-normalised scores ) from across nine human cell lines . This approach also increases the sample size per drug by at least ninefold , making any inferences more reliable ( Subramanian et al . , 2017 ) . At the request of a reviewer , we provide results from individual cells ( Table S3 Appendix 1—figure 3 ) ; however , we caution that these within-cell rank-order values are known to be less robust ( Subramanian et al . , 2017; Xu et al . , 2018 ) . Critically , we noted that members of each drug ‘class’ ( drugs sharing a nominal primary protein target in common ) were segregated with either active or neutral IR-DR scores , with extremely few drug classes having both positive- and negative-scoring compounds . Only 10% of the database matched the IR-DR signature ( n = 254 , Appendix 1—figure 3 and Table S3 ) , and 138 compounds ( after excluding assay codes with ambiguous compound labels ) positively regulated the IR-DR signature ( potential treatments ) , 45% of which were kinase inhibitors ( Appendix 1—figure 4 ) . Most negative acting drugs targeted tubulin and cell cycle proteins or were pro-inflammatory agents ( Appendix 1—figure 5 ) . Positive and negative acting compounds did not differ in average physiochemical properties ( Appendix 1—figure 6 ) , while assays based on GWAS-selected genes for IR ( Lotta et al . , 2017; Vujkovic et al . , 2020 ) and T2DM produced no discernible pattern of in vitro hits . The pharmacology of the 138 positive compounds indicated that a substantial number of targeted aspects of insulin signalling were known , empirically , to reverse IR in vivo ( Table 1 ) . Compounds identified varied in nature from inhibitors of glucosylceramide synthase , which reverses IR and fatty liver disease ( Aerts et al . , 2007; Herrera Moro Chao et al . , 2019 ) , to 10 mTOR inhibitors . The mTOR complex , mTORC1 , coordinates a negative feedback loop on insulin signalling , for example , through activation of GRB10 or via S6K1 ( Um et al . , 2004 ) . mTORC1 signalling is also regulated by protein kinase C ( PKC ) ( Zhan et al . , 2019 ) , and specific PKC isoforms are dysregulated in ageing , metabolic , neurodegenerative and inflammatory diseases ( Li et al . , 2015; Sajan et al . , 2018; Sharma et al . , 2019 ) . We observed that the broad-spectrum PKC inhibitor , bisindolylmaleimide I , induced a strong positive IR signature score ( +87 ) and the related compound , ruboxistaurin , reverses IR in vivo ( Guo et al . , 2020; Naruse et al . , 2006 ) . In contrast , bisindolylmaleimide IX , a 20-fold more potent broad-spectrum PKC inhibitor , was inactive in the IR-DR assay , probably reflecting its greater non-specific pharmacology ( against other kinase families ) . Three so-called PKC activators ( French et al . , 2020; Lee et al . , 2020 ) induced negative IR-DR scores ( phorbol-12-myristate-13-acetate = –87 , ingenol = –96 and prostratin = –97 ) . RNAi targeting of individual PKC isoforms ( clue . io ) demonstrated that the IR-DR assay was sensitive to specific PKC isoformactivity . While >95% of all RNAi assays produced no significant scores , knock-down ( KD ) of PKC-beta ( +74 ) and PKC-theta ( +97 ) yielded positive IR-DR scores , while loss of PKC-alpha ( –75 ) and -eta ( –75 ) produced negative IR-DR scores and overexpression ( OE ) of PKC-alpha was positive scoring ( +85 ) . The multi-gene IR-DR assay therefore identifies numerous true-positive drugs ( Table 1 ) and reflects isoform-specific activity , strongly validating the cell-agnostic aggregation methodology . The 138 positive scoring IR-DR drugs target 1007 proteins ( Mendez et al . , 2019; Moret et al . , 2019 ) . Of these , 465 genes had single-gene KD or OE scores , aggregated across 6–9 cell lines ( Appendix 1—figure 7 ) . Seventy-three targets ( 15 . 7% ) yielded a significant IR-DR score; double the assay hit rate ( p<0 . 0001 , see Methods and Figure 2 ) . Predictably , due to input bias ( Timmons et al . , 2015 ) , these targets regulated ‘peptidyl-serine phosphorylation-related processes’ ( q-value <1 × 10–23 ) . None belonged to the IR-DR gene signature ( Appendix 1—figure 8 ) , but they did belong to numerous common pathways ( Appendix 1—figure 8 ) . These observations are consistent with the idea that an effective DR signature captures the pathway biology of the disease and/or treatment ( Brown and Patel , 2018; Chen et al . , 2020; Karatzas et al . , 2017; Keenan et al . , 2018; Regan-Fendt et al . , 2019; Wagner et al . , 2015; Woo et al . , 2015 ) but does not necessarily include the nominal drug targets ( Figure 2 ) . Network representation of how these 73 independently validated target proteins of active drugs ( Appendix 1—figure 4 ) interact with IR-DR signature at a pathway level is presented in Figure 2A ( Benjamini-Hochberg corrected p-values ) . The GO terms are scaled by the total number of significant terms and labelled by the top-level category ( Figure 2B ) . Coloured orange ( Figure 2B ) , a module of the ‘carboxylic acid biosynthetic process’ pathway contains genes that , when more ‘active , ’ positively modulate the IR-DR signature ( Z = 8 . 3 , p=1 × 10–9 ) . Each pathway is also coloured ( Figure 2C ) to indicate whether it contains a known drug target or was part of the IR-DR signature . The IR-DR assay genes formed eight pathway clusters , of which the majority directly contain some RNAi validated protein targets , for example , ‘negative regulation of phosphate metabolic process’ ( coloured brown , q-value 1 × 10–7 ) . As with the analysis of the individual PKC isoforms , there are compelling examples of proteins contributing to metabolic disease , for example , SMAD3 ( +87 IR-DR score from OE and –95 IR-DR score from RNAi ) is an in vivo-validated IR pathway ( Budi et al . , 2019; Sun et al . , 2015; Tan et al . , 2011 ) . However , if multi-gene DR assays are to be used for optimising drug properties , it is critical to establish that they can produce quantitative pharmacological feedback when comparing related drugs ( Hopkins , 2008 ) . The relationship between in vitro drug potency and IR-DR assay score for 37 compounds ( Appendix 1—figure 5 ) , nominally targeting epidermal growth factor receptor ( EGFR or HER1 ) tyrosine kinase , was investigated . This stress-induced inflammatory protein has recently emerged as a target for treating metabolic disease and neurodegeneration ( Donath , 2021; Menden et al . , 2019; Norambuena et al . , 2017 ) . One endogenous EGFR ligand , amphiregulin , is induced by high-fat feeding to drive TNF-mediated IR ( Skurski et al . , 2020 ) , while EGFR is overexpressed in astrocytes of Alzheimer’s disease ( AD ) . The EGFR inhibitor afatinib ( IR-DR score = 77 ) attenuates astrocyte activation ( Chen et al . , 2019 ) while inhibition of EGFR can reduce FOXK1 and FOXK2 phosphorylation ( Klaeger et al . , 2017 ) to normalise mTORC1-regulated autophagy and reduce IR ( Bowman et al . , 2014; Jahng et al . , 2019 ) . Nine EGFR inhibitors have been screened against >300 kinases ( Davis et al . , 2011; Klaeger et al . , 2017 ) , which enabled us to directly contrast laboratory-derived kinase selectivity with the IR-DR score . Potency versus EGFR directly related to IR-DR assay score ( Figure 3A ) , yet this could not fully explain why certain compounds were inactive . Cluster analysis of the most targeted proteins ( <300 nM potency for at least one compound ) illustrated that alisertib , the only potent EGFR inhibitor with a negative IR-DR score ( –90 ) , inhibited PLK4 , AURKB and AURKA ( Figure 3B ) . Orantinib , a 24 nM inhibitor of AURKB , also had a negative IR-DR score ( –77 ) , as did MK-5108 , an inhibitor of AURKB and AURKA ( –90 ) , indicating that alisertib’s profile reflects pharmacology beyond EGFR ( probably AURKB as AURKB KD had an IR-DR score of –83 , Table S3 ) . A broader exploration of ‘EGFR’ inhibitor targets provides a better understanding of the activity of this group of compounds in the IR-DR assay . For example , neutral scoring bosutinib and neratinib target several mitogen-activated protein kinase ( MAPK ) family members , and some of these oppose positive IR-DR scoring , for example , MAP2K2 ( –78 , Table S4 ) . Neutral scoring , yet potent EGFR inhibitors ( e . g . neratinib , bosutinib and lapatinib ) also inhibit the related proteins , ERBB2 , ERBB3 and ERBB4 ( <5 nM , HER2-4 ) . Some of these family members may represent beneficial ‘off targets’ while others may be detrimental ( Moret et al . , 2019 ) . For example , hyperglycaemia induces erbb4 in mice and erbb4 expression is increased in AD ( Huh et al . , 2016; Woo et al . , 2011 ) , where OE increases tau phosphorylation via mTOR activation ( Nie et al . , 2018b ) . Gefitinib ( an EGFR inhibitor ) reduces IR-mediated glucose excursions in vivo in a RIPK2-dependent manner ( Duggan et al . , 2020 ) and rescues memory deficits in mice at a very low chronic dose of 0 . 01 mg/kg ( Wang et al . , 2012 ) . Loss of RIPK2 ( or a dominant-negative mutant of RIPK2 ) prevents excessive NFκB activation ( Chin et al . , 2002 ) , and RIPK2 is a downstream effector of innate immunity ( TLR signalling ) . Some EGFR targeting drugs also potently inhibit the tyrosine kinase ABL1 , and loss of adipose ABL1 reduces obesity-induced IR in the mouse ( Wu et al . , 2017 ) . Erlotinib ( IR-DR score = +72 ) , which also inhibits ABL1 , reduces kidney inflammation and preserves pancreas function and insulin sensitivity in a mouse model of diabetes ( Li et al . , 2018b ) . Furthermore , Aβ activates neuronal ABL1 in vitro , intra-hippocampal injection of Aβ fibrils increases expression of ABL1 in vivo and imatinib ( STI571 ) , a 90 nM inhibitor of ABL1 , inhibits ABL1-mediated Aβ neurodegenerative pathways ( Cancino et al . , 2011; Cancino et al . , 2008; Gutierrez et al . , 2019 ) . However , imatinib does not yield a significant IR-DR score ( nor inhibit EGFR ) , indicating that targeting ABL1 alone might be insufficient to treat human IR . Importantly , potency of the EGFR inhibitors against ABL1 also correlates with their potency against several ephrin receptors ( EPHA5 R = 0 . 74 , EPHA6 R = 0 . 95 and EPHA8 R = 0 . 76 ) , as well as with RIPK2 ( R = 0 . 77 ) . Oral dosing of the ephrin A receptor inhibitor , UniPR500 , reverses high-fat feeding-induced glucose intolerance without changes in plasma insulin ( Giorgio et al . , 2019 ) , and these benefits likely reflect UniPR500 targeting proteins in common with our top-ranked ‘EGFR’ kinase inhibitors . Thus , while drug potency against EGFR quantitatively tracks with the IR-DR score ( Figure 3A ) , this may reflect binding affinity at other related protein kinases , and identification of these additional targets is important ( Redhead et al . , 2021 ) . Thus we find that ( Figure 3A–C ) the best scoring potent ‘EGFR’ kinase inhibitors reflect a balance of activities against positively and negatively acting kinases and that this is partly interpretable versus the extensive in vitro screening data for those drugs ( Davis et al . , 2011; Klaeger et al . , 2017 ) . Genome-wide pharmacological profiles are however prohibitively costly and thus often unavailable . Predicted drug-protein interactions , using emerging techniques from graph machine learning ( Sturm et al . , 2020 ) , aim to overcome this lack of laboratory data . Using the DeepPurpose DL suite of algorithms ( Huang et al . , 2021 ) , we modelled EGFR family kinase inhibitors as simplified molecular-input line-entry system ( SMILES ) string and all proteins by their amino acid sequence ( a strategy that obfuscates the need for 3D structures obtained by costly experimental models ) . This expanded the scope of drug target information of the EGFR inhibitors discussed above to a genome-wide level ( Tables S6-S8 ) . Each of the EGFR inhibitors was scored against 19 , 211 proteins using 14 pre-trained models ( Table S9 ) , and we relied on a fusion of ranking scores across models to identify the top protein targets of each compound . The nine EGFR inhibitors described above inhibit 25 proteins with nanomolar potency ( <300 nM ) , and the DL model accurately ranked these proteins in the top 0 . 1–1 . 7% of all 2 , 420 , 586 predictions ( median = 0 . 15% , Table S6 ) . The DL-predicted rank score ( ‘potency’ ) against EGFR strongly related to the measured IR-DR score ( Figure 3D ) , replicating the potency-activity relationship noted using laboratory data ( Figure 3A ) correctly clustering the nine compounds ( Figure 3C , Appendix 1—figure 9A ) . DL also ranked several proteins that we already identified may compromise a positive IR-DR score ( Figure 3E , Appendix 1—figure 9 ) . For example , AURKB was a top-ranked predicted target for alisertib ( 26/19211 ) , aligning with the data that inhibition of AURKB drives a negative IR-DR score . Additional predicted protein targets ( Table S8 ) will also influence the IR-DR score , independently of EGFR , for example , MERTK , KCNH6 and PTK2B ( Appendix 1—figure 9 ) . Loss of PTK2B ( focal adhesion kinase 2 [FAK2] ) , a risk gene for the development of tauopathy in AD ( Tan et al . , 2021 ) , can promote the development of IR in vivo and in adipocytes ( Luk et al . , 2017; Yu et al . , 2005 ) while inhibition of KCNH6 should probably be avoided , as it regulates insulin secretion ( Yang et al . , 2018 ) . In contrast , a genetic loss-of-function variation in MERTK appears protective against IR , fatty liver disease and pro-inflammatory mediators in humans ( Musso et al . , 2017 ) , and thus it represents a potential protein target against which current ‘EGFR’ inhibitors should be screened against . Therefore , we applied the same modelling strategy to 16 of the 28 less well-characterised EGFR inhibitors with proven sub-micromolar activity . Each was ranked highly by the model against EGFR ( Table S7 ) , with MAP3K19 being one of the highest ranked additional targets ( Appendix 1—figure 10 ) , and there was a negative correlation between predicted MAP3K19 binding and IR-DR score ( Appendix 1—figure 10 ) . Little is known about MAP3K19 ( a ‘dark’ kinase ) other than that it may contribute to ERK pathway activation ( Hoang et al . , 2020 ) – and some ERK inhibitors proved positive scoring in the IR-DR assay ( Appendix 1—figure 3 ) – and thus MAP3K19 may be a novel positive effector of insulin signalling . MAP3K19 is not abundantly expressed in adipose or muscle tissue ( lowest 10th percentile of gene expression in our studies ) such that net compound efficacy in vivo may also reflect tissue-specific patterns of protein activity . We illustrate that a cell-line transcriptome-based high-throughput DR assay yields interpretable and quantitative pharmacological data when designed around robust clinical RNA signatures , and that DL-based drug target predictions can be used to interpret assay scores . Some of the drugs we identified may be suitable for treating acute IR , such as occurring during infection ( Ceriello et al . , 2020; Donath , 2021 ) , and encouragingly several positive scoring drugs appear tolerable in longer-term preclinical models of metabolic or neurogenerative disease ( Li et al . , 2018b; Wang et al . , 2012 ) . The present approach could be extended to include a stratified medicine component , where evaluation of positively acting compounds is first trialled in T2DM patients with extreme IR ( Choi et al . , 2019 ) . A number of positively acting IR-DR compounds , including selected mTOR inhibitors ( Appendix 1—figure 3 ) , are able to mimic a longevity-related RNA signature ( Timmons et al . , 2019 ) and thus may be potential geroprotectors ( Fuentealba et al . , 2019 ) . A more extensive multi-disease signature approach could ultimately help tailor the DR process to the individual patient . We do acknowledge that some IR-DR assay negative scores may be false negatives , for example , selective HDAC inhibition ( HDACi ) can , through lower and shorter daily exposure ( Sartor et al . , 2019; Volmar et al . , 2017 ) , be beneficial ( although positive attributes of HDACi on IR appear to reflect non-specific actions; Martins et al . , 2019 ) . Furthermore , despite correctly matching with nuclear receptor-induced muscle transcriptome signatures in vivo , there was a dearth of matches in vitro indicating that further optimisation of the IR-DR assay format is merited . It can be the case that certain classes of ligand require more sophisticated assay condition or require the use of primary cells . In conclusion; human transcriptome signatures , classic pharmacological assays , drug action in vivo and DL-based target prediction consistently link with drug transcriptional profiles in cell lines , establishing that expansion of such resources represents an important strategy for future DR efforts .
The disease and the treatment genes lists represent the pool of features from which we selected IR-DR signatures . Quantitative network modelling ( Song and Zhang , 2015 ) was applied to tissue expression values of these genes , as previously described ( Timmons et al . , 2019 ) , to identify hub genes; genes with greater connectivity ( Appendix 1—figure 1 ) . Four alternative similarly sized sets of genes were selected for validation , comprising 60 positively associated and 60 negatively associated RNAs . The final models shared only two genes named as candidates from genome-wide IR association studies ( Lotta et al . , 2017 ) , and those genes ( INSR and GRB14 ) were not essential for our analysis . To rank the performance of each of our four RNA assays ( Table S1 ) , we utilised DrugMatrix , a database of in vivo rodent tissue drug-response signatures ( http://www . ilincs . org/ilincs/ ) that includes TZD and oestrogen-related molecules known to reverse IR in vivo ( Hevener et al . , 2018; Sears et al . , 2009 ) and target IR pathways in cell models ( Sood et al . , 2016 ) . Signatures validated at this stage were considered suitable for further use . Lists that failed this step ( lists 4 and 5 ) included genes inferred from genome-wide IR association studies , and T2DM proteome biomarkers ( Gudmundsdottir et al . , 2020; Lotta et al . , 2017; Vujkovic et al . , 2020 ) were modelled in vitro only to produce summary statistics for Table S1 . The in vivo-validated IR signatures were screened using the largest public database of in vitro drug signatures ( Subramanian et al . , 2017 ) via the clue . io resource ( version 1 , 2020 ) . These drug transcriptional signatures were generated in nine cell lines , and while each cell line captures some unique signals from each compound ( Baillif et al . , 2020 ) , part of this will be noise , reflecting the small sample size ( typically n = 3 ) . Therefore , we used aggregated signature matching across the nine human cell lines ( PC3 , VCAP , A375 , A549 , HA1E , HCC515 , HT29 , MCF7 and HEPG2 ) to both deliberately reduce the influence of cell line-specific effects ( Xu et al . , 2018 ) and to increase the sample size ninefold ( Subramanian et al . , 2017 ) . Active compounds were those with scores exceeding ~10th percentile of positive and negative scores ( Appendix 1—figure 3 ) , a value that represents the mean threshold ( ±1 standard deviation ) of scores exceeding the assay scoring threshold ( Subramanian et al . , 2017 ) . The use of aggregated scores across cell types was validated using an extensive validation process ( Figure 1 ) . In addition to our IR-DR assay , we considered two additional in vivo RNA models . One is a novel 141-gene human-derived muscle growth signature ( Stokes et al . , 2020 ) , which demonstrated – as expected – that the clue . io database contains a sizeable number of compounds known to inhibit cell growth ( negative-scoring compounds , Appendix 1—figure 2 and Appendix 1—figure 3 ) . The second was an IR-adjusted longevity-associated signature ( Timmons et al . , 2019 ) , which identified relevant drug matches from the cell-line perturbagen database ( Appendix 1—figure 2 and ‘Discussion’ ) . The output from each assay was a list of >2500 DR scores , each assigned to a particular assay ID and drug name . There then followed a laborious manual annotation process , reflecting that study of drugs is challenging ( Christmann-Franck et al . , 2016 ) , and annotation errors populate all databases , including iLINCS . For all of the active compounds , we carried out a manual check to ensure that compound labels in CLUE were verifiable in Chembl ( Mendez et al . , 2019 ) , and that both were consistent with the data deposited in the small molecule suit ( Moret et al . , 2019 ) . The manually confirmed data progressed to the next phase of the analysis . Active compounds belonged to a wide range of distinct pharmacological classes ( Appendix 1—figures 4 and 5 ) . To identify if positive and negatively acting compounds ( from IR-DR score ) could be easily distinguished from each other , we calculated simple molecular descriptors . A set of 2837 compounds for which in vitro results were available was considered , and chemical structures were extracted from the CLUE database ( https://clue . io/ ) as SMILES strings . These were parsed with RDKit ( version 2020 . 03 . 6 ) , with 14 compounds failing to parse correctly . For each compound , a set of 13 physicochemical descriptors was calculated with RDKit including molecular weight , heavy atom count , number of heteroatoms , LogP , number of rotatable bonds , topological polar surface area ( TPSA ) , number of rings , number of aromatic rings , number of saturated rings , number of aliphatic rings , Balaban’s J index , number of hydrogen bond donors and number of hydrogen bond acceptors . Positive acting compounds were coloured in orange , and negative acting in blue ( Appendix 1—figure 6 ) . For each compound active against list 3A IR-DR signature ( Appendix 1—figure 3 ) , we identified their protein targets using a variety of resources ( Christmann-Franck et al . , 2016 , https://www . ebi . ac . uk/chembl/ , https://clue . io/ , and https://pubchem . ncbi . nlm . nih . gov/ ) . The small-molecule suite was used to extract laboratory-derived potency values against each protein ( https://labsyspharm . shinyapps . io/smallmoleculesuite/ ) . Very few compounds are profiled against the majority ( >300 ) kinases . For those that had values , in vitro potency values ( log10 ) were plotted against the IR-DR scores and Pearson’s correlation coefficients calculated in Excel , allowing us to establish the relationship between potency against a protein and the IR-DR score . Very extensive manual PubMed searches were then undertaken to characterise the positive and negative acting drugs using the terms ‘drug name’ ‘alternative drug name’ with the following terms used in sequence until either relevant publications were identified or no hits were obtained; insulin , diabetes , obesity , dementia , Alzheimer’s disease , COVID-19 and inflammation ( during 2020 ) . The in vitro drug signatures in clue . io are the most robust data from that project ( Subramanian et al . , 2017 ) . However , the database also contains gene KD ( n = 3799 genes ) and OE ( n = 2161 genes ) data , making it possible to link the IR-DR signature with specific protein targets . The KD or OE activity score was again averaged across 6–9 cell lines , and we considered protein targets derived from the known targets of the active compound list ( to limit the known higher false-positive rate with the KD/OE data; Subramanian et al . , 2017 ) . As stated in the results , >15% of IR-active drug targets yielded a significant IR-DR score . In comparison , a total of 459 genes were associated with an IR-DR score at a threshold of >70 or <-70 , representing 265 positive associations ( mean = 85 ) and 194 negative IR-DR scores ( mean = –83 ) , equalling only ~7% of all 5954 genomic assays . The pathway biology of the 73 independently validated protein targets was analysed using Metascape ( Zhou et al . , 2019 ) , along with the IR-DR signature ( Table 1 , Appendix 1—figure 2 ) , where ‘Combo_sig’ represents the IR-DR pathways , and the 73 genes split into two lists depending on how KD or OE impacted on the IR-DR signature match ( Table S4 ) . For the data plot , edges represent connected GO biological processes ( >0 . 3 ) , and nodes within each cluster are coloured/named by their most statistically enriched GO term ( Figure 3B ) . Each node is presented as a pie chart , scaled in size by the total number of terms represented by that ( top-scoring ) ontology , and with the ‘slices’ coloured to indicate from which gene list the terms originate . The network structure is separately colour-coded ( Figure 3C ) by list membership to identify when drugs directly target IR-DR assay pathways ( red ) ; when KD or OE genes negatively correlated ( blue ) with the IR-DR score ( inhibition yields a positive or OE yields a negative IR-DR score ) or positively correlated ( green ) with the IR-DR score . To investigate the mechanisms of action of positively acting IR-DR compounds , we extended the existing pharmacological data with computationally derived drug-target interaction ( DTI ) predictions . Publicly available chemogenomic databases are very far from complete , and therefore , ML modelling approaches can be used to provide estimates for missing data . DL-based models have shown promise in this context ( Gaudelet et al . , 2021; James et al . , 2020 ) . We used DeepPurpose , a DL library for DTI prediction ( Huang et al . , 2021 ) that takes as an input SMILES of the small molecules of interest and the amino acid sequences of the protein-coding genome . Different encoders were implemented to provide a compound and a protein embedding . The small molecule and protein embeddings are concatenated and fed to a multi-layer perceptron that predicts the binding affinity as a dissociation constant ( Kd ) . DeepPurpose provides a set of pre-trained models that can be used ‘off-the-shelf’ . We used 14 pre-trained models that were available as of 01/12/2020 . Those models differ from one another depending on the encoders and on the DTI training set . Drug encoders included convolutional neural network ( CNN ) , daylight fingerprints , Morgan fingerprints and message-passing neural network ( MPNN ) , while protein encoders amino acid composition ( AAC ) and CNN were used . The training sets were BindingDB ( Liu et al . , 2007 ) , DAVIS ( Davis et al . , 2011 ) and KIBA ( Tang et al . , 2014 ) . A list of the models used is shown in Table S9 . All DTI models described above were applied to obtain 14 rankings of 19 , 211 human proteins as potential targets for each compound . A final score was obtained by an average ranking of each protein across 14 models , with the final top-ranking targets predicted to be the most likely protein targets of the input drug list . Comparable consensus-oriented strategies are often applied in virtual screening to exploit the strengths of multiple models ( Gaudelet et al . , 2021; James et al . , 2020 ) and achieve improved performance ( Palacio-Rodríguez et al . , 2019; Perez-Castillo et al . , 2017 ) . DeepPurpose models showed promising performance in various testing scenarios , and we refer to the original publication for further details . The code used for the entire analysis can be located in the supplemental document . | Developing a new drug that is both safe and effective is a complex and expensive endeavor . An alternative approach is to ‘repurpose’ existing , safe compounds – that is , to establish if they could treat conditions others than the ones they were initially designed for . To achieve this , methods that can predict the activity of thousands of established drugs are necessary . These approaches are particularly important for conditions for which it is hard to find promising treatment . This includes , for instance , heart failure , dementia and other diseases that are linked to the activity of the hormone insulin becoming modified throughout the body , a defect called insulin resistance . Unfortunately , it is difficult to model the complex actions of insulin using cells in the lab , because they involve intricate networks of proteins , tissues and metabolites . Timmons et al . set out to develop a way to better assess whether a drug could be repurposed to treat insulin resistance . The aim was to build a biological signature of the disease in multiple human tissues , as this would help to make the findings more relevant to the clinic . This involved examining which genes were switched on or off in thousands of tissue samples from patients with different degrees of insulin resistance . Importantly , some of the patients had their condition reversed through lifestyle changes , while others did not respond well to treatment . These ‘non-responders’ provided crucial new clues to screen for active drugs . Carefully piecing the data together revealed the molecules and pathways most related to the severity of insulin resistance . Cross-referencing these results with the way existing drugs act on gene activity , highlighted 138 compounds that directly bind 73 proteins responsible for regulating insulin resistance pathways . Some of the drugs identified are suitable for short-term clinical studies , and it may even be possible to rank similar compounds based on their chemical activity . Beyond giving a glimpse into the complex molecular mechanisms of insulin resistance in humans , Timmons et al . provide a fresh approach to how drugs could be repurposed , which could be adapted to other conditions . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"medicine",
"computational",
"and",
"systems",
"biology"
] | 2022 | A human-based multi-gene signature enables quantitative drug repurposing for metabolic disease |
The neuronal PAS domain proteins NPAS1 and NPAS3 are members of the basic helix-loop-helix-PER-ARNT-SIM ( bHLH-PAS ) family , and their genetic deficiencies are linked to a variety of human psychiatric disorders including schizophrenia , autism spectrum disorders and bipolar disease . NPAS1 and NPAS3 must each heterodimerize with the aryl hydrocarbon receptor nuclear translocator ( ARNT ) , to form functional transcription complexes capable of DNA binding and gene regulation . Here we examined the crystal structures of multi-domain NPAS1-ARNT and NPAS3-ARNT-DNA complexes , discovering each to contain four putative ligand-binding pockets . Through expanded architectural comparisons between these complexes and HIF-1α-ARNT , HIF-2α-ARNT and CLOCK-BMAL1 , we show the wider mammalian bHLH-PAS family is capable of multi-ligand-binding and presents as an ideal class of transcription factors for direct targeting by small-molecule drugs .
The mammalian bHLH-PAS transcription factors share a common protein architecture consisting of a conserved bHLH DNA-binding domain , tandem PAS domains ( PAS-A and PAS-B ) , and a variable activation or repression domain . These factors can be grouped into two classes based on their heterodimerization patterns ( McIntosh et al . , 2010; Bersten et al . , 2013 ) ( Figure 1A and Figure 1—figure supplement 1 ) . Class I includes the three hypoxia-inducible factor ( HIF ) -α proteins ( HIF-1α , HIF-2α and HIF-3α ) , four neuronal PAS domain proteins ( NPAS1-4 ) , aryl hydrocarbon receptor ( AHR ) , AHR repressor ( AHRR ) , single-minded proteins ( SIM1 , SIM2 ) and clock circadian regulator ( CLOCK ) ; while class II includes aryl hydrocarbon receptor nuclear translocator ( ARNT , also called HIF-1β ) , ARNT2 , brain and muscle ARNT-like protein 1 ( BMAL1 , also called ARNTL ) and BMAL2 ( ARNTL2 ) . Heterodimerization between class I and class II members produces functional transcription factors capable of DNA binding and target gene regulation . 10 . 7554/eLife . 18790 . 003Figure 1 . Comparison of bHLH-PAS proteins . ( A ) Heterodimerization patterns between bHLH-PAS proteins . ( B ) Protein domain arrangements of ARNT , NPAS1 and NPAS3 . Percent amino-acid identities between corresponding domains of NPAS1 and NPAS3 are in red . ( C ) Overall crystal structure of the NPAS1-ARNT complex shown in two views . ( D ) Superposition of NPAS1-ARNT and HIF-2α-ARNT heterodimers . The arrows on top show the shift in position for the PAS-B domain of ARNT . DOI: http://dx . doi . org/10 . 7554/eLife . 18790 . 00310 . 7554/eLife . 18790 . 004Figure 1—figure supplement 1 . Phylogenetic tree of all mouse bHLH-PAS family members based on protein sequences at their bHLH-PAS-A-PAS-B regions . DOI: http://dx . doi . org/10 . 7554/eLife . 18790 . 00410 . 7554/eLife . 18790 . 005Figure 1—figure supplement 2 . Comparative protein sequence analysis of mouse bHLH-PAS proteins . The bHLH , PAS-A and PAS-B segments , and the secondary structure components of NPAS1 are shown above the alignment . The NPAS1 residues interacting directly with ARNT ( at interface 1–4 ) are marked using cyan round dots , while those at intra-molecular interfaces ( 5 and 6 ) are marked with brown round dots . Triangles show pocket amino-acids in NPAS1 , HIF-2α , CLOCK , ARNT and BMAL1 ( in cyan , magenta , brown , green and lime colors , respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18790 . 00510 . 7554/eLife . 18790 . 006Figure 1—figure supplement 3 . Comparison of the overall structures of NPAS1-ARNT and CLOCK-BMAL1 complexes . The two complexes are superimposed by aligning the bHLH domains ( A ) or PAS-B domains ( B ) of ARNT with BMAL1 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 18790 . 006 Many bHLH-PAS proteins are significantly involved in human disease processes and would be ideal drug targets if their architectures could accommodate binding and modulation by small-molecules . Members of the HIF-α subgroup mediate the response to hypoxia and regulate key genetic programs required for tumor initiation , progression , invasion and metastasis ( Semenza , 2012 ) . AHR controls T cell differentiation ( Stevens and Bradfield , 2008 ) and maintains intestinal immune homeostasis ( Leavy , 2011 ) , making it a potential target for alleviating inflammation and autoimmune diseases . Loss-of-function mutations in SIM1 have been linked to severe obesity in human populations ( Bonnefond et al . , 2013 ) , and defects in SIM2 are associated with cancers ( Bersten et al . , 2013 ) . CLOCK and BMAL1 together establish molecular circadian rhythms and their functional disruption can lead to a variety of metabolic diseases ( Gamble et al . , 2014 ) . The NPAS genes are highly expressed in the nervous system ( Zhou et al . , 1997; Brunskill et al . , 1999; Ooe et al . , 2004 ) . In mice , genetic deficiencies in NPAS1 and NPAS3 are associated with behavioral abnormalities including diminished startle response , social recognition deficit and locomotor hyperactivity ( Erbel-Sieler et al . , 2004; Brunskill et al . , 2005 ) . NPAS2 is highly related to CLOCK in protein sequence ( Reick et al . , 2001 ) , and altered patterns of sleep and behavioral adaptability have been observed in NPAS2-deficient mice ( Dudley et al . , 2003 ) . NPAS4 deficiency is associated with impairment of long-term contextual memory formation ( Ramamoorthi et al . , 2011 ) . In humans , alterations in all four NPAS genes have been linked to neuropsychiatric diseases including schizophrenia , autism spectrum disorders , bipolar disease and seasonal depression disorders ( Kamnasaran et al . , 2003; Pieper et al . , 2005; Partonen et al . , 2007; Pickard et al . , 2009; Huang et al . , 2010a; Bersten et al . , 2014; Stanco et al . , 2014 ) . Structural information has not been available for any NPAS proteins to show if they could bind drug-like molecules for treating psychiatric diseases . A crystal structure was previously reported for the CLOCK-BMAL1 heterodimer ( Huang et al . , 2012 ) , and we recently reported crystal structures for both HIF-2α-ARNT and HIF-1α-ARNT heterodimers ( Wu et al . , 2015 ) . In all these complexes , the conserved bHLH-PAS-A-PAS-B protein segments were visualized . While no internal cavities were reported within the CLOCK-BMAL1 architecture; we identified multiple hydrophobic pockets within HIF-1α-ARNT and HIF-2α-ARNT heterodimers . Discrete pockets were encapsulated within each of the four PAS domains of their heterodimers ( two within ARNT and two within each HIF-α protein ) ( Wu et al . , 2015 ) . Beyond the first structural characterizations of NPAS1-ARNT and NPAS3-ARNT complexes presented here , we further addressed if ligand-binding cavities are a common feature of mammalian bHLH-PAS proteins . A comparison of these two structures with those of CLOCK-BMAL1 , HIF-1α-ARNT and HIF-2α-ARNT heterodimers unveils the larger mammalian bHLH-PAS family as ligand binding transcription factors with internal pockets appropriate for the selective binding of lipophilic molecules and drug-like compounds .
We employed the contiguous bHLH-PAS-A-PAS-B segments of NPAS1 , NPAS3 and ARNT proteins for our crystallographic studies ( Figure 1B ) . For the NPAS1-ARNT heterodimer , we obtained crystals that diffracted to 3 . 2 Å resolution ( Table 1 ) . The quaternary organization of the NPAS1-ARNT complex is shown in Figure 1C . We found that the bHLH , PAS-A and PAS-B domains of ARNT twist along the outside surface of the NPAS1 protein . Figure 1D shows that NPAS1-ARNT and HIF-2α-ARNT heterodimers are very similar in overall architectures , but the PAS-B domain of ARNT is slightly displaced in the NPAS1 heterodimeric complex . This observation indicates that the ARNT architecture can display flexibility in accommodating its different class I partners . 10 . 7554/eLife . 18790 . 007Table 1 . Data collection and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 18790 . 007NPAS-ARNT NPAS3-ARNT-DNA Data collection Space groupP 1P 43Cell dimensionsa , b , c ( Å ) 69 . 9 , 81 . 2 , 138 . 164 . 8 , 64 . 8 , 249 . 1α , β , γ ( ° ) 90 . 4 , 95 . 1 , 107 . 490 . 0 , 90 . 0 , 90 . 0Resolution ( Å ) 50 . 0–3 . 20 ( 3 . 26–3 . 20 ) *50 . 0–4 . 20 ( 4 . 27–4 . 20 ) Rmerge5 . 5 ( 76 . 4 ) 6 . 0 ( 84 . 8 ) CC* ( highest resolution shell ) 0 . 7930 . 971CC1/2 ( highest resolution shell ) 0 . 4590 . 893I/σI 14 . 0 ( 1 . 2 ) 20 . 0 ( 1 . 1 ) Completeness ( % ) 98 . 6 ( 98 . 3 ) 94 . 6 ( 72 . 3 ) Redundancy2 . 1 ( 2 . 2 ) 5 . 2 ( 3 . 9 ) Refinement Resolution ( Å ) 37 . 7–3 . 20 ( 3 . 32–3 . 20 ) 36 . 9–4 . 20 ( 5 . 28–4 . 20 ) No . reflections39 , 096 ( 802 ) 5868 ( 2120 ) Rwork/ Rfree ( % ) 19 . 2/24 . 9 ( 28 . 3/40 . 4 ) 29 . 5/36 . 2 ( 27 . 1/34 . 8 ) No . atoms Protein/DNA13 , 3035313 Water00B-factors Protein/DNA49 . 666 . 9 Water--R . m . s deviationsBond lengths ( Å ) 0 . 0160 . 004Bond angles ( ° ) 1 . 460 . 77One crystal was used for each structure . *Highest resolution shell is shown in parenthesis . We could not obtain crystals of apo NPAS3-ARNT and instead pursued its DNA complex . The response element for NPAS1-ARNT is known to match the consensus hypoxia response element ( HRE ) ( Ohsawa et al . , 2005; Teh et al . , 2007 ) , but the response element for NPAS3-ARNT was not previously characterized . NPAS1 and NPAS3 closely share amino-acids within their bHLH domain ( 75% identity , see Figure 1B and Figure 1—figure supplement 2 ) , including conservation of residues that recognize DNA base-pairs based on our observations of the HIF-2α-ARNT-DNA complex ( Wu et al . , 2015 ) . Therefore , we tested if NPAS3-ARNT could efficiently bind to the same consensus HRE sequence used by NPAS1-ARNT and HIF-α-ARNT heterodimers . We measured the dissociation constants ( Kd ) using a DNA duplex containing a central HRE sequence ( 5’-TACGTG-3’ ) and found similar Kd values of ~20 nM for both NPAS1-ARNT and NPAS3-ARNT ( Figure 2A ) . These binding constants indicate a relatively higher affinity than that of the HIF-2α-ARNT heterodimer ( Kd~40 nM ) ( Wu et al . , 2015 ) . 10 . 7554/eLife . 18790 . 008Figure 2 . Comparison of NPAS1-ARNT and NPAS3-ARNT complexes . ( A ) DNA-binding affinities measured using fluorescence anisotropy . The Kd values of NPAS1-ARNT and NPAS3-ARNT binding to the same HRE element were 16 . 3 ± 1 . 1 nM and 17 . 5 ± 1 . 1 nM , as calculated from three technical replicates respectively . ( B ) Overall structures of the NPAS3-ARNT-DNA , with the hexameric HRE site colored in cyan and blue . ( C ) Structure comparison of NPAS3-ARNT-DNA and HIF-2α-ARNT-DNA complexes aligned on their DNA . ( D ) Superposition of NPAS3-ARNT and NPAS1-ARNT structures . DNA was omitted from the NPAS3-ARNT complex . Arrows show the extension of α1 helices likely associated with its DNA binding . DOI: http://dx . doi . org/10 . 7554/eLife . 18790 . 008 We then obtained crystals for NPAS3-ARNT bound to this HRE element and solved the structure at 4 . 2 Å resolution ( Figure 2B and Table 1 ) . The resolution made possible an architectural comparison , at the quaternary level , between the NPAS3-ARNT , NPAS1-ARNT , HIF-1α-ARNT and HIF-2α-ARNT heterodimers . All of these complexes share a similar overall architecture ( Figure 2C , D ) stabilized by the same six domain-domain junctions ( see below ) . Furthermore , the cooperation between the two subunits of NPAS3-ARNT also creates a DNA-reading head that is similar to that of the HIF-α-ARNT-DNA complexes , allowing direct readout of the HRE site ( Figure 2C ) . We next tested if the six domain-domain interfaces observed in NPAS1-ARNT and NPAS3-ARNT ( Figure 3A ) are important for maintaining the stability of their full-length heterodimers within cells . For our study , we used co-immunoprecipitation ( co-IP ) experiments in HEK293T cells , together with a series of single or double mutations positioned within ARNT ( Figure 3B ) . Mutations in interfaces 1–4 were found to significantly destabilize ARNT’s heterodimeric interactions with both NPAS1 and NPAS3 , as predicted from their crystal structures . Point mutations within NPAS1 at interfaces 5 and 6 also destabilized heterodimerization with ARNT ( Figure 3C ) . We additionally tested the effects of these destabilizing mutations on the transcriptional function of NPAS1-ARNT . NPAS1 has been shown to function as a transcriptional repressor on its target gene tyrosine hydroxylase ( TH ) , since it lacks a functional transactivation domain ( Teh et al . , 2006 ) . We confirmed this repression activity in both HRE-driven and TH-driven reporter assays , and further found that mutations that destabilized the heterodimer also compromised the transcriptional repression ( Figure 3D ) . 10 . 7554/eLife . 18790 . 009Figure 3 . Testing of domain interfaces in NPAS1-ARNT and other ARNT heterodimers using mutagenesis . ( A ) Detailed interactions at each of six interfaces in NPAS1-ARNT . Each junction is circled in the context of the overall structure of NPAS1-ARNT on the left , and each interface is further listed in the table . ( B ) Co-IP experiments showing the effects of ARNT mutations at interfaces 1–4 on the cellular stabilities of heterodimers formed with NPAS1 , NPAS3 , SIM1 , NPAS4 and AHR , respectively . ( C ) Co-IP experiments showing the effects of NPAS1 mutations at interfaces 5 and 6 on the stabilities of NPAS1-ARNT heterodimer . ( D ) Luciferase reporter assay testing the effects of NPAS1 ( wide-type and three mutants ) on HRE-driven ( left ) and tyrosine hydroxylase ( TH ) promoter-driven ( right ) transactivation . Each sample represents the average reading of cells from three wells . ( E ) Co-IP experiments showing the effects of ARNT2 mutations at interfaces 1–4 ( corresponding to ARNT ) on the stabilities of heterodimers formed with SIM1 and NPAS4 . The residues and mutants of ARNT , NPAS1 and ARNT2 are labelled in green , cyan and brown , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 18790 . 00910 . 7554/eLife . 18790 . 010Figure 3—figure supplement 1 . Amino-acid sequence alignment of full-length mouse ARNT and ARNT2 proteins . Secondary structures and positions of the bHLH , PAS-A and PAS-B domains of ARNT are indicated on top . The ARNT residues participating at dimerization interfaces are indicated with green dots . DOI: http://dx . doi . org/10 . 7554/eLife . 18790 . 010 In discovering that the NPAS1-ARNT , NPAS3-ARNT and HIF-α-ARNT heterodimers utilize the same six domain-domain interfaces to create a shared quaternary structure ( Figures 1D and 2D ) , we also found that this type of architecture is clearly distinct from that of CLOCK-BMAL1 complex ( Huang et al . , 2012 ) ( Figure 1—figure supplement 3 ) . Therefore , we conclude that there are at least two distinct architectural forms within the mammalian bHLH-PAS family . Since the ARNT heterodimers form a larger group than the BMAL1 heterodimers ( Figure 1A ) , we asked if other ARNT heterodimers rely on the same six domain-domain junctions we observed in NPAS1-ARNT , NPAS3-ARNT and HIF-α-ARNT heterodimers . We could experimentally interrogate the degree of architectural variation within the ARNT heterodimer class by using our panel of ARNT mutations that were able to block its heterodimerization with NPAS1 and NPAS3 . Using co-IP studies , we tested if these same ARNT mutations would disrupt its heterodimeric complexes with SIM1 , NPAS4 and AHR ( Figure 3B ) . The SIM1-ARNT heterodimer stability was indeed compromised by these ARNT mutations , indicating that this heterodimer shares the same overall architecture as NPAS1-ARNT , NPAS3-ARNT and the two HIF-α-ARNT heterodimers . However , NPAS4-ARNT and AHR-ARNT heterodimer stabilities were not impacted in the same manner by these ARNT mutations , particularly when the mutations were located to interfaces 3 and 4 . Therefore , we believe there is greater architectural variation within ARNT heterodimer group than what has been crystallographically observed to date . A protein amino-acid sequence alignment further indicates that residues observed to stabilize domain-domain junctions in the NPAS1/3-ARNT and HIF-α-ARNT heterodimers are much more conserved in SIM1 than in NPAS4 and AHR ( Figure 1—figure supplements 1 and 2 ) . The ARNT protein is ubiquitously expressed in mammalian cells , but its paralog ARNT2 is more specifically enriched in brain and kidney tissues ( Hirose et al . , 1996 ) . ARNT and ARNT2 have both unique and overlapping cellular functions ( Keith et al . , 2001 ) , and ARNT2 has been suggested as the preferred physiological partner for Class I members SIM1 ( Michaud et al . , 2000 ) and NPAS4 ( Bersten et al . , 2014 ) . Since the amino-acid sequence identity at the bHLH-PAS region is nearly 80% between ARNT and ARNT2 , and since all the ARNT residues observed to be participating at dimerization interfaces are fully conserved in ARNT2 ( Figure 3—figure supplement 1 ) , we predict that ARNT2 heterodimers will display similar overall quaternary architectures as their ARNT counterparts . To test this prediction , we mutated two ARNT2 residues at each of the four dimer interfaces ( Figure 3E ) . Compared with the ARNT mutations ( Figure 3B ) , these corresponding ARNT2 mutations ( both single and double ones ) indeed had similar effects on the dimerization with not only SIM1 but also NPAS4 ( Figure 3E ) , indicating that ARNT2 indeed dimerizes with SIM1 and NPAS4 in the same way as ARNT does . In both HIF-1α-ARNT and HIF-2α-ARNT heterodimers , we previously identified hydrophobic pockets encapsulated within the two PAS domains of ARNT , and within the two PAS domains of each HIF-α protein ( Wu et al . , 2015 ) . Here we asked if NPAS1-ARNT and NPAS3-ARNT harbored similarly positioned pockets . NPAS1 and NPAS3 are genetically associated with a wide range of human neuropsychiatric disorders ( Kamnasaran et al . , 2003; Pieper et al . , 2005; Stanco et al . , 2014 ) . Thus , the discovery of ligand-binding cavities could lead to the future discovery of therapeutic molecules for these illnesses . We show that NPAS1 protein’s PAS-A and PAS-B domains do contain internal cavities with volumes measuring 190 Å3 and 180 Å3 , respectively ( Figure 4A , B and Table 2 ) . Similarly positioned pockets are seen in the NPAS3 protein , measuring 100 Å3 and 230 Å3 , respectively . Each of these PAS domains resembles a baseball catcher’s mitt , with the beta strands forming the palm and short alpha-helices forming the opposing thumb to enclose a central pocket . Through heterodimerization , ARNT further brings its own two pockets to join each of its class I partners . We examined if the two ARNT pockets alter their shape when ARNT forms different heterodimeric complexes . We could not detect any major change in the cavity size or shape of ARNT’s PAS-A and PAS-B pockets , whether this protein was in a complex with HIF-2α or NPAS1 ( Figure 4C ) . ARNT is ubiquitously expressed in many cell types and is predominantly nuclear ( Reyes et al . , 1992 ) ; whereas its class I heterodimerization partners are signal activated and/or tissue restricted . Therefore , the pockets located in NPAS1 , NPAS3 , HIF-1α and HIF-2α should allow more selective actions of therapeutic drugs than the pockets within ARNT ( Figure 4B–D ) . Our findings of two similarly positioned pockets within each of NPAS1 , NPAS3 , HIF-1α , HIF-2α and ARNT led us to consider if any other bHLH-PAS family members could also harbor putative ligand-binding cavities . The crystal structure of CLOCK-BMAL1 has been reported , but no internal cavities were described in its published analysis ( Huang et al . , 2012 ) . Therefore , we closely examined this structure and found that both CLOCK and BMAL1 proteins contained discrete pockets within each of their PAS-A and PAS-B domains ( Figure 4E–G ) . The BMAL1’s PAS-A and PAS-B pockets ( measuring 200 Å3 and 220 Å3 , respectively ) are larger than those of CLOCK ( measuring 120 Å3 and 140 Å3 , respectively ) ( Table 2 ) . Crystal structures are unavailable for SIM1 and SIM2 proteins; however , the close evolutionary and amino-acid sequence similarity with the NPAS1 protein led us to generate plausible models for their PAS-A and PAS-B pockets ( Figure 4H , Figure 1—figure supplements 1 and 2 ) . 10 . 7554/eLife . 18790 . 011Figure 4 . Comparison of ligand-binding pockets among multiple bHLH-PAS proteins . ( A and E ) Relative positions of the four pockets ( red circles ) within the structures of NPAS1-ARNT ( A ) and CLOCK-BMAL1 ( E ) complexes . ( B–D and F–H ) Empty pockets in each of PAS-A and PAS-B domains of NPAS1 ( B ) , ARNT ( C ) , HIF-2α ( D ) , CLOCK ( F ) , BMAL1 ( G ) and SIM1 ( H ) . The accessible cavities are shown in red meshes , together with the amino-acid residues lining each pocket . For NPAS1 , the surrounding residues are further covered with 2FO – FC map contoured at 1 . 0σ ( B ) . The NPAS1 residues with identical counterparts in NPAS3 are in cyan and labelled in black , while those with non-conserved NAPS3 counterparts ( indicated in parentheses ) are in magenta and labelled in red ( B ) . ARNT pocket residues from apo HIF-2α-ARNT complex ( PDB: 4ZP4 ) ( in wheat ) and those from NPAS1-ARNT complex ( in green ) are superposed for comparison ( C ) . The HIF-2α residues with identical counterparts in HIF-1α are in magenta , while those with non-conserved HIF-1α counterparts ( indicated in parentheses ) are in green ( D ) . Structures of CLOCK ( F ) and BMAL1 ( G ) are from the CLOCK-BMAL1 complex ( PDB: 4F3L ) in brown and lime , respectively . The SIM1 structure in orange was modelled from NPAS1 and HIF-2α; and the pocket residues not conserved in SIM2 are in green with their SIM2 counterparts indicated in parentheses ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18790 . 01110 . 7554/eLife . 18790 . 012Table 2 . Volumes of ligand-binding pockets of bHLH-PAS proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 18790 . 012Location NPAS1 NPAS3 HIF-1α HIF-2α HIF-3α ARNT CLOCK BMAL1 NPAS2 SIM1 SIM2 PAS-A190100100150170110120200230210210PAS-B180230160370590210140220170370310Pocket volumes ( Å3 ) were calculated using CASTp program ( Dundas et al . , 2006 ) using the default probe sphere radius of 1 . 4 Å . The PDB coordinate files used for NPAS1 and NPAS3 proteins were from the NPAS1-ARNT and NPAS3-ARNT-DNA complexes , the coordinates for ARNT and HIF-2α were from the apo HIF-2α-ARNT complex ( PDB: 4ZP4 ) , and the coordinates for CLOCK and BMAL1 were from the CLOCK-BMAL1 complex ( PDB: 4F3L ) . The values of PAS-B domains of HIF-1α and HIF-3α ( fatty acid bound ) were from high-resolution single domain structures ( PDBs: 4H6J and 4WN5 ) , respectively . The PAS-A domain of HIF-3α , and both PAS domains of NPAS2 , SIM1 and SIM2 were modeled using the SWISS-MODEL server ( Biasini et al . , 2014 ) . The pockets of AHR and its repressor AHRR are more difficult to model based on our existing crystal structures , because of greater sequence and evolutionary divergence of these proteins ( Figure 1—figure supplement 1 ) . Interestingly , multiple classes of ligands , including halogenated aromatic hydrocarbons , tetrapyrroles and several tryptophan derivatives were previously identified as direct AHR binding ligands ( Denison et al . , 2011; Stejskalova et al . , 2011 ) . These molecules display significant variations in their chemical structures and sizes , but still bind with high affinities to AHR ( Kd values of 0 . 1–100 nM ) . High-affinity , multi-ligand binding can best be accounted for by the use of four discrete pockets within AHR-ARNT , as shown here for other ARNT heterodimers , than just one pocket as suggested previously ( Pandini et al . , 2007 ) . Figure 4 shows side-by-side comparisons of the PAS-A and PAS-B pockets from multiple bHLH-PAS proteins . Importantly , each PAS domain relies on a constellation of different amino-acids to form its interior cavity , allowing the pockets to bind selectively to different repertoires of endogenous ligands . Moreover , within subgroups such as HIF-1α/2α , NPAS1/3 , and SIM1/2 , the proteins share pocket residues more closely with each other than they do between subgroups . This observation indicates close ligand preferences within subgroups , suggesting they could recognize similar metabolites or signaling molecules derived from the same biosynthetic pathway . Importantly , in all the PAS-A and PAS-B pockets , the amino-acid residues lining interior cavities are predominantly hydrophobic . This property would allow favorable van der Waals interactions with lipophilic ligands . The desolvation of hydrophobic ligands can further contribute to favorable energetics of ligand-binding inside these cavities . The internal pocket volumes in the bHLH-PAS family ( 100–600 Å3 ) ( Table 2 ) are in-line with pocket sizes observed in other classes of ligand-binding proteins ( 100–1000 Å3 ) ( Liang et al . , 1998 ) . Moreover , we found that ligand binding can increase the size of pocket significantly . For example , HIF-2α specific inhibitor 0X3 enlarged its PAS-B pocket volume from 370 Å3 to 560 Å3 ( Wu et al . , 2015 ) . These observations suggest a high degree of adaptability in PAS domain pockets , and should encourage future drug-discovery efforts seeking multiple classes of synthetic modulators for bHLH-PAS proteins . An analogy can be made with the nuclear receptor family , where diverse classes of small-molecules can bind to the same pocket , and ligand binding can expand pocket sizes ( Huang et al . , 2010b ) .
Previously , the nuclear receptors were believed to form the only transcription factor family in higher eukaryotes with conserved ligand binding capabilities . We showed here that the mammalian bHLH-PAS proteins constitute a second independent family of transcription factors with the appropriate and conserved molecular frameworks required for multi-ligand binding . Our findings are based on the direct crystallographic analysis of seven distinct bHLH family members ( ARNT , HIF-1α , HIF-2α , NPAS1 , NPAS3 , CLOCK and BMAL1 ) . In all these proteins , internal hydrophobic cavities were clearly observed within both their PAS-A and PAS-B domains . These PAS domains become interweaved in bHLH-PAS heterodimers through at least two highly distinct forms of architecture . Aside from the seven members we crystallographically examined , three additional members: AHR ( Stejskalova et al . , 2011 ) , HIF-3α ( Fala et al . , 2015 ) and NPAS2 ( Dioum et al . , 2002 ) are known to have ligand-binding capabilities associated with at least one of their PAS domains , bringing the experimentally confirmed number of ligand pocket containing members to ten . Our sequence-structure comparative analyses further implicate three more members: ARNT2 , SIM1 and SIM2 , as likely to harbor pockets due to considerable amino-acid conservations with ARNT and NPAS1 . Our unmasking of the wider bHLH-PAS protein family as a group of transcription factors with multi-ligand-binding capabilities should fuel the future search for their physiological and endogenous ligands . The protein pockets , in each case , make these factors ideally suited for applications of high-throughput screening campaigns to find small-molecule therapeutics for a variety of human diseases , including psychiatric illnesses , cancers and metabolic diseases . The distinctive amino-acid residues inside their pockets predict successful outcomes for screens seeking highly selective modulators for each protein . These pockets are also highly pliable and can accommodate significantly larger ligands than their empty volumes alone suggest . The precise mechanism through which endogenous and/or pharmacologic ligands can modulate transcriptional activities of bHLH-PAS proteins has not yet been fully revealed , and could differ for each family member . For example , ligand binding to AHR can displace heat shock protein 90 and initiate the nuclear translocation of AHR ( Soshilov and Denison , 2011 ) . Some small-molecule HIF-2α antagonists that bind to the PAS-B domain can disrupt the dimerization between HIF-2α and ARNT ( Scheuermann et al . , 2013 ) . Beyond these initial observations to date , and without the availability of small-molecule tools to probe other bHLH-PAS proteins , the mechanistic links between ligand binding and transcriptional regulation remain to be discovered . It is interesting that while dimeric nuclear receptors harbor two pockets in their functional architectures , the bHLH-PAS proteins present four distinct pockets . Notably a fifth pocket was also observed in HIF-2α-ARNT heterodimers , located between two subunits , allowing proflavine to promote subunit dissociation ( Wu et al . , 2015 ) . The latter finding further suggests that crevices formed between the PAS domains within the quaternary architectures of family members could form additional ligand binding and modulation sites . The availability of so many distinctive pockets within bHLH-PAS proteins could allow a variety of biosynthetic or metabolic signals derived from cellular pathways to be integrated into a single unified functional response . Alternatively , each pocket may allow a bHLH-PAS protein to be the site of unique ligand within each cell-type . The identification and validation of endogenous ligands for this family will help define the genetic programs controlled by each bHLH-PAS member .
For protein overexpression in Escherichia coli , mouse NPAS1 ( GenBank accession: AAI32114 . 1 , residues 43–423 ) and NPAS3 ( GenBank accession: AAI67248 . 1 , residues 56–455 ) were cloned into the pSJ2 vector , respectively . For the co-immunoprecipitation studies , full-length mouse NPAS1 , NPAS3 , SIM1 and NPAS4 were cloned into the pCMV-Tag4 vector ( C-terminal Myc-tagged ) , and mouse ARNT2 was cloned into the pCMV-Tag1 vector ( C-terminal Flag-tagged ) . The cloning of ARNT and AHR constructs has been described previously ( Wu et al . , 2013 , 2015 ) . Site-directed mutagenesis was confirmed in each case by DNA sequencing . The recombinant plasmids pSJ2-NPAS1 and NPAS3 were co-transformed along with pMKH-ARNT into BL21-CodonPlus ( DE3 ) -RIL competent cells ( Agilent Technologies , Santa Clara , CA , #230245 ) . Proteins were expressed and purified as previously described ( Wu et al . , 2015 ) . To prepare NPAS3-ARNT DNA-bound complexes , synthetic 21mer double-strand DNA ( forward: 5’- GGCTGCGTACGTGCGGGTCGT-3’ and reverse: 5’-CACGACCCGCACGTACGCAGC-3’ ) was mixed with the heterodimeric proteins . Crystallization of the NPAS1-ARNT complex was carried out using the sitting drop vapor diffusion method at 16°C , by mixing equal volume of protein ( 4 mg/ml ) and reservoir solution containing 2% Tacsimate pH 7 . 0 , 3% PEG3350 . Before flash frozen in liquid nitrogen , crystals were soaked in reservoir plus 30% glycerol as the cryoprotectant . NPAS3-ARNT-DNA crystals were grown at 16°C in sitting drops formed by equal volume of complex ( 4 mg/ml ) and reservoir consisting of 100 mM NH4F , 9% PEG 3350 , and then transferred stepwise to cryoprotectant of 100 mM NH4F , 10% PEG 3350 and up to 30% PEG 400 ( 5% increase each step ) prior to flash freezing . Diffraction data were collected at the Argonne National Laboratory SBC-CAT 19ID beamline at 100 K . The structures of NPAS1-ARNT and NPAS3-ARNT-DNA complexes were solved by molecular replacement with Phaser ( RRID: SCR_014219 ) ( McCoy et al . , 2007 ) , using the HIF-2α-ARNT structures ( PDB: 4ZP4 and 4ZPK ) as the search models . Further manual model building was facilitated using Coot ( RRID: SCR_014222 ) ( Emsley et al . , 2010 ) , combined with the structure refinement using Phenix ( RRID: SCR_014224 ) ( Adams et al . , 2010 ) . The diffraction data and final statistics are summarized in Table 1 . The Ramachandran statistics , calculated by Molprobity ( RRID: SCR_014226 ) ( Chen et al . , 2010 ) , are 93/0 . 06% and 89/0 . 19% ( favored/outliers ) for NPAS1-ARNT and NPAS3-ARNT-DNA complexes , respectively . All the structural figures were prepared using PyMol ( The Pymol Molecular Graphics System , RRID: SCR_000305 ) . Coordinates and structure factors have been deposited in Protein Data Bank under accession numbers 5SY5 ( NPAS1-ARNT ) and 5SY7 ( NPAS3-ARNT-DNA ) . The 21-mer fluoresceinated double-strand DNA was prepared by annealing 6-FAM labelled forward strand ( 5’-GGCTGCGTACGTGCGGGTCGT-3’ ) with the unlabeled reverse strand ( 5’-ACGACCCGCACGTACGCAGCC-3’ ) in the buffer consisting of 10 mM Tris pH 7 . 5 , 1 mM EDTA and 2 mM MgCl2 . For the binding assay , 2 nM DNA was incubated with purified proteins for 30 min , and final protein concentrations were varied by serial dilution in binding buffer ( 20 mM Tris pH 8 . 0 , 50 mM NaCl and 10 mM DTT ) . The fluorescence polarization signals were recorded and processed as previously ( Wu et al . , 2015 ) . HEK293T cells ( ATCC CRL-3216 , RRID: CVCL_0063 ) were seeded in 10 cm dishes and cultured in DMEM containing 10% FBS ( Thermo Fisher Scientific , #11995 and #16000 ) at 37°C with 5% CO2 . One day later , cells were transfected with 2 μg pCMV-Tag4-NPAS1 , NPAS3 , SIM1 , NPAS4 or AHR ( WT or mutants ) and 6 μg pCMV-Tag1-ARNT or ARNT2 ( WT or mutants ) plasmids using 16 μL jetPRIME regent ( Polyplus-transfection , #114–07 ) . After overnight incubation , medium was refreshed ( 10 nM 2 , 3 , 7 , 8-tetrachlorodibenzo-p-dioxin added in the case of AHR ) . Another 24 hr later , cells were harvested and immunoprecipitation was performed similarly to our previous work ( Wu et al . , 2015 ) . HEK293T cells were seeded in 24-well plates , and one day later transfected with 200 ng of pCMV-Tag4-NPAS1 ( WT , mutants or empty plasmid ) , 1 ng of pRL ( control Renilla luciferase ) , 100 ng of HRE-luc reporter ( Ao et al . , 2008 ) or TH-luc reporter containing the tyrosine hydroxylase promoter sequence ( Thiel et al . , 2005 ) using 0 . 6 μL jetPRIME regent for each well . Medium was refreshed after overnight transfection , and luciferase activity was measured another 24 hr later using the Dual-Glo Luciferase Assay System ( Promega , #E2920 ) . Final data were normalized by the relative ratio of firefly and Renilla luciferase activity . | Transcription factors are proteins that can bind to DNA to regulate the activity of genes . One family of transcription factors in mammals is known as the bHLH-PAS family , which consists of sixteen members including NPAS1 and NPAS3 . These two proteins are both found in nerve cells , and genetic mutations that affect NPAS1 or NPAS3 have been linked to psychiatric conditions in humans . Therefore , researchers would like to discover new drugs that can bind to these proteins and control their activities in nerve cells . Understanding the three-dimensional structure of a protein can aid the discovery of small molecules that can bind to these proteins and act as drugs . Proteins in the bHLH-PAS family have to form pairs in order to bind to DNA: NPAS1 and NPAS3 both interact with another bHLH-PAS protein called ARNT , but it is not clear exactly how this works . In 2015 , a team of researchers described the shapes that ARNT adopts when it forms pairs with two other bHLH-PAS proteins that are important for sensing when oxygen levels drop in cells . Here , Wu et al . – including many of the researchers involved in the earlier work – have used a technique called X-ray crystallography to determine the three-dimensional shapes of NPAS1 when it is bound to ARNT , and NPAS3 when it is bound to both ARNT and DNA . The experiments show that each of these structures contains four distinct pockets that certain small molecules might be able to bind to . The NPAS1 and NPAS3 structures are similar to each other and to the previously discovered bHLH-PAS structures involved in oxygen sensing . Further analysis of other bHLH-PAS proteins suggests that all the members of this protein family are likely to be able to bind to small molecules and should therefore be considered as potential drug targets . The next step following on from this work is to identify small molecules that bind to bHLH-PAS proteins , which will help to reveal the genes that are regulated by this family . In the future , these small molecules may have the potential to be developed into new drugs to treat psychiatric conditions and other diseases in humans . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2016 | NPAS1-ARNT and NPAS3-ARNT crystal structures implicate the bHLH-PAS family as multi-ligand binding transcription factors |
The molecular roles of HOX transcriptional activity in human prostate epithelial cells remain unclear , impeding the implementation of new treatment strategies for cancer prevention and therapy . MEIS proteins are transcription factors that bind and direct HOX protein activity . MEIS proteins are putative tumor suppressors that are frequently silenced in aggressive forms of prostate cancer . Here we show that MEIS1 expression is sufficient to decrease proliferation and metastasis of prostate cancer cells in vitro and in vivo murine xenograft models . HOXB13 deletion demonstrates that the tumor-suppressive activity of MEIS1 is dependent on HOXB13 . Integration of ChIP-seq and RNA-seq data revealed direct and HOXB13-dependent regulation of proteoglycans including decorin ( DCN ) as a mechanism of MEIS1-driven tumor suppression . These results define and underscore the importance of MEIS1-HOXB13 transcriptional regulation in suppressing prostate cancer progression and provide a mechanistic framework for the investigation of HOXB13 mutants and oncogenic cofactors when MEIS1/2 are silenced .
Prostate cancer ( PrCa ) is the fifth leading cause of cancer-related death in men worldwide and is responsible for the highest incidence of male cancer in the United States ( Ferlay et al . , 2015; Siegel et al . , 2019 ) . While PrCa can progress slowly and remain relatively asymptomatic for years , some patients present with aggressive metastatic PrCa and a poor prognosis ( Barlow and Shen , 2013; Lin et al . , 2009 ) . Further , it can be difficult to distinguish which men harbor indolent or aggressive tumors ( Culig , 2014 ) , particularly in patients with intermediate Gleason scores ( Gearman et al . , 2018 ) . These features of PrCa pose a significant clinical problem . One novel pathway to understand tumor etiology and disease progression as well as develop new treatments is through HOXB13 , which exhibits germline mutation in a subset of familial PrCa . HOXB13 is the predominant HOX factor that drives development and differentiation of prostate epithelial cells ( Brechka et al . , 2017 ) . Germline mutations of HOXB13 confer a substantial risk of PrCa , but mutation frequency is rare within the general population ( Brechka et al . , 2017 ) . On the other hand , our prior studies show that prostate tumors frequently harbor downregulation of the transcription factors and HOX binding partners MEIS1 and MEIS2 ( myeloid ecotropic viral integration site 1/2 ) ( Bhanvadia et al . , 2018; Chen et al . , 2012 ) . MEIS proteins function as critical transcriptional co-factors during development and within adult tissues to bind HOX proteins and specify HOX gene targeting ( Merabet and Mann , 2016 ) . Most PrCa HOXB13 mutations ( including the original G84E mutation ) are located within the MEIS-interacting domain , emphasizing the importance of MEIS/HOX interactions in prostate tumor biology . We originally demonstrated that increased mRNA expression of MEIS1 and MEIS2 in PrCa is correlated with significantly longer overall survival in a large cohort of watchful waiting patients with mid-range Gleason scores ( Chen et al . , 2012 ) . More recently , we and others demonstrated that patients harboring MEIS-positive tumors have a significantly favorable outcome; there is a step-wise decrease in both MEIS1 and MEIS2 expression as tumors progress to metastatic ( Bhanvadia et al . , 2018; Jeong et al . , 2017; Nørgaard et al . , 2019 ) . These correlative findings provide support to a tumor-suppressive role for MEIS1 and MEIS2 in PrCa . However , there remain significant gaps in our understanding of how MEIS proteins suppress tumor progression and the role of HOXB13 in MEIS-mediated tumor suppression . The function of MEIS proteins is critical but distinct among normal and malignant tissues . Further , the oncogenic vs . tumor-suppressive functions of MEIS proteins depend upon tissue of origin ( Brechka et al . , 2017 ) . MEIS proteins belong to the three amino-acid loop extension ( TALE ) protein family ( Longobardi et al . , 2014 ) and are critical for multiple components of normal human development and maintenance , including hematopoiesis ( Argiropoulos et al . , 2007; Ariki et al . , 2014; Hisa et al . , 2004 ) , vascular patterning ( Azcoitia et al . , 2005 ) , limb patterning ( Graham , 1994 ) , and anterior-posterior axis determination in combination with Homeobox ( HOX ) genes ( Choe et al . , 2014; Shanmugam et al . , 1999; Williams et al . , 2005 ) . Increased expression of MEIS proteins is associated with tumorigenesis in certain cancers , including leukemia ( Kumar et al . , 2009 ) , ovarian carcinoma ( Crijns et al . , 2007 ) , and neuroblastoma ( Geerts et al . , 2005 ) . However , in colorectal carcinoma ( Crist et al . , 2011 ) , gastric carcinoma ( Song et al . , 2017 ) , renal cell carcinoma ( Zhu et al . , 2017 ) , and non-small cell lung cancer ( Li et al . , 2014 ) , increased MEIS expression is associated with tumor suppression . In some instances , MEIS1 expression results in reduced proliferation by inducing cell cycle arrest at the G1/S phase transition ( Song et al . , 2017; Zhu et al . , 2017 ) . HOX transcription factors play a key role in anterior-posterior axis formation , proliferation , and differentiation ( McGinnis and Krumlauf , 1992; Seifert et al . , 2015 ) but require co-factors to help specify DNA binding ( Mann et al . , 2009 ) , stabilize interactions at the genome level ( Shen et al . , 1997a ) , and regulate transcription factor activation or repression ( Bürglin , 1998; Huang et al . , 2005; Hyman-Walsh et al . , 2010; Longobardi et al . , 2014; Mann et al . , 2009; Shanmugam et al . , 1999; Williams et al . , 2005; Zandvakili and Gebelein , 2016 ) . Anterior HOX1-8 paralogs prefer to heterotrimerize with MEIS and PBX family proteins ( Ladam and Sagerström , 2014; Moens and Selleri , 2006; Penkov et al . , 2013; Slattery et al . , 2011 ) . In the prostate , however , the dominant HOX genes expressed are Abd-B-like HOX genes and include paralogs 9–13 ( Brechka et al . , 2017; Huang et al . , 2007 ) . Notably , HOX11–13 paralogs , including HOXB13 , prefer to heterodimerize with MEIS1 ( Shen et al . , 1997a ) and exclude PBX proteins ( Shen et al . , 1997b ) . Thus , MEIS/HOX interactions are likely key in prostate development and cancer . Indeed , these combined studies implicate interaction between MEIS1 and the Abd-B-like HOX proteins of the prostate in regulating organ homeostasis . However , the phenotypic impact of MEIS/HOX interactions in PrCa cell gene expression and behavior remains unknown , as do the critical drivers of MEIS/HOX-mediated tumor suppression . Here , we report a phenotypic and mechanistic determination that MEIS proteins promote indolent and non-metastatic prostate cancer via the HOXB13-dependent regulation of extracellular proteoglycans , in particular the multi-RTK inhibitor Decorin . These studies establish critical mechanisms for future utilization of MEIS proteins and predictive biomarkers of indolent prostate cancer and will enable mechanistic studies to define the roles of HOXB13 mutants and oncogenic HOXB13 cofactors in prostate cancer progression .
Our previous studies demonstrated that expression of both MEIS1 and MEIS2 is frequently decreased in PrCa patients and that MEIS-positive tumors confer an overall lower risk of biochemical recurrence and metastasis ( Bhanvadia et al . , 2018; Chen et al . , 2012 ) . Analysis of MEIS1 and MEIS2 expression in a panel of PrCa cell lines compared to primary prostate epithelial cell ( PrEC ) cultures revealed significantly decreased MEIS1 and MEIS2 mRNA ( p<0 . 05 , Figure 1A ) . Similarly , western blot analysis in all PrCa cell lines documented low protein levels of MEIS1—with the exception of androgen-receptor ( AR ) -negative lines Du145 and PC3—as well as low levels of MEIS2 ( Figure 1B ) . We previously demonstrated that depletion of both MEIS1 and MEIS2 in LAPC4 cells was necessary to promote tumor xenograft growth ( Bhanvadia et al . , 2018 ) . To determine whether increased MEIS expression is sufficient to block PrCa cell growth , we ectopically expressed either MEIS1 or MEIS2 via lentiviral constructs in CWR22Rv1 and LAPC4 PrCa cells ( LV-MEIS1 and LV-MEIS2; Figure 1C ) . MEIS2 is known to have several isoforms that differ mainly in the exons used at the C-terminus , as well as one homeodomain-less variant known as MEIS2E ( Figure 1—figure supplement 1A; Geerts et al . , 2005 ) . Analyses of mRNA documented that PrECs and prostate cancer cell lines express multiple detectable MEIS2 transcript isoforms , albeit at low levels of expression ( Figure 1; Figure 1—figure supplement 1B ) . The homeodomain-less and putative dominant-negative MEIS2E isoform was undetectable across multiple lines ( Figure 1—figure supplement 1C ) . Further , ectopic expression of either MEIS2A or MEIS2D isoforms are sufficient to inhibit cell growth , while expression of MEIS2E did not impact cell growth ( Figure 1—figure supplement 1D and E ) . These data support the necessity of MEIS DNA binding to suppress cell growth . Decreased cell number over time with either MEIS1 or MEIS2A was not associated with increased cell death but was associated with significant accumulation of cells in the G1 phase and fewer cells in G2 ( p<0 . 05 , Figure 1E; Figure 1—figure supplement 2 ) . Subcutaneous xenografting of CWR22Rv1 cells with exogenous MEIS expression into nude mice also resulted in MEIS-mediated tumor suppression in vivo ( p<0 . 05 , Figure 1F&G ) . For the ensuing studies we narrowed our analyses to MEIS1 due to the 72 . 12% sequence similarity shared by MEIS1 and MEIS2 , the phenocopied growth suppression in vitro and in vivo ( Figure 1; Figure 1—figure supplement 1D and E ) , the complexity of experimental design around multiple MEIS2 isoforms , and their similarly reduced expression in prostate tumors and cancer cells ( Bhanvadia et al . , 2018 ) . Given the correlation between MEIS expression and metastasis within annotated tumor specimens , we investigated the migratory capacity of CWR22Rv1 and LAPC4 cells expressing exogenous MEIS1 . To minimize the impact of differences in proliferation between LV-MEIS1 and control lines , 3 µM aphidicolin was used to inhibit proliferation in all assays , and is not reported to affect migration ( Müller et al . , 2002 ) . MEIS1-expressing cells showed significantly decreased migration in vitro compared to control cells ( p<0 . 05 , Figure 1H , I ) . Thus , both MEIS1 and full-length MEIS2 are sufficient to slow PrCa cell proliferation and tumor growth via reduced G1/S phase transition and decreased migratory capacity . These data are consistent with findings from other histological tumor types , where MEIS1 functions as a tumor suppressor , slows G1/S phase transition , and reduces migration and invasion capacity in vitro ( Song et al . , 2017; Zhu et al . , 2017 ) . HOXB13 has critical roles in normal prostate secretory function , differentiation , and response to androgens in rodent prostate models and is implicated in human PrCa ( Chen et al . , 2018; Economides and Capecchi , 2003; Hamid et al . , 2014; Huang et al . , 2007; Jung et al . , 2004a; Jung et al . , 2004b; Kim et al . , 2014a; Kim et al . , 2010a; Kim et al . , 2014b; Kim et al . , 2010b; Navarro and Goldstein , 2018; Pomerantz et al . , 2015 ) . Comparative analyses of HOX gene mRNA expression using publicly-available RNA-Seq datasets of adult human prostate tissues demonstrated that HOXB13 is the highest-expressed ( Pflueger et al . , 2011; Robinson et al . , 2015 ) HOX gene across benign epithelium , tumor , and metastatic tissue; HOXA10 is the next-highest ( FPKMs HOXB13 vs . HOXA10: benign , 167 . 69 vs 38 . 53; primary tumor , 197 . 40 vs 35 . 63; metastasis , 149 . 44 vs 28 . 03 , Figure 2A; Bhanvadia et al . , 2018 ) . Notably , this result is consistent with observations in both rat and murine prostate , which also have high HOXA13 and D13 expression levels though still below the level of HOXB13 ( Brechka et al . , 2017; Huang et al . , 2007 ) . While expression of HOXB13 remains high throughout prostate tumors and metastases , we previously demonstrated a step-wise decrease in MEIS1 and MEIS2 expression from benign epithelium to tumor and metastasis and that MEIS-positive prostate tumors confer an overall favorable patient outcome , thus implying that MEIS-HOXB13 interactions are tumor-suppressive ( Bhanvadia et al . , 2018 ) . We thus sought to confirm a nuclear interaction between MEIS1 and HOXB13 within normal PrECs when both proteins are present and in prostate cancer cells when MEIS1 expression is increased . To accomplish this , we used in situ proximity ligation assays ( PLA ) and co-immunoprecipitation with antibodies specific to MEIS1 and HOXB13 in benign PrECs and in our CWR22Rv1-Control , CWR22Rv1-LV-MEIS1 , LAPC4-Control , and LAPC4-LV-MEIS1 cell line models ( Figure 2B ) . While PLA does not enable absolute quantitation of interactions within a cell , it does permit quantification of relative differences between samples ( Söderberg et al . , 2006; Söderberg et al . , 2008; Weibrecht et al . , 2010 ) . Interactions between MEIS1 and HOXB13 were detectable in normal PrECs; such interactions decreased with lower MEIS expression in prostate cancer cells . Importantly , ectopic MEIS1 expression leading to growth suppression in CWR22Rv1- and LAPC4 PrCa cells was associated with increased MEIS-HOXB13 interactions compared to the respective cell line controls ( Figure 2C ) . Furthermore , co-immunoprecipitation of MEIS1 with HOXB13 was observed in PrECs , and increased MEIS1 pulldown was observed in CWR22Rv1 and LAPC4 cells ectopically expressing MEIS1 ( Figure 2D ) . Together , these results document HOXB13 as the highest-expressed HOX gene in the adult human prostate , thus prioritizing the functional importance of HOXB13 in this tissue . These data also suggest that loss of normal MEIS-HOXB13 interactions by decreased MEIS expression could enable non-canonical HOXB13 binding partners to partner with HOXB13 to promote oncogenesis and tumor progression . Having established the tumor-suppressive capability of MEIS1 and its ability to act as a putative HOXB13 cofactor in PrECs , we next sought to test whether MEIS-mediated growth suppression is HOXB13-dependent . To accomplish this , we used CRISPR/Cas9 to knock-out HOXB13 in CWR22Rv1 and LAPC4 PrCa cells . We then ectopically expressed MEIS1 in the resulting HOXB13ko lines to create HOXB13ko-LV-MEIS1 cells ( Figure 3A ) . With HOXB13ko and HOXB13ko-LV-MEIS1 lines , we were able to determine the impact of exogenous MEIS1 expression in the absence of HOXB13 . Cell proliferation assays demonstrated distinct but agreeable phenotypes between CWR22Rv1 and LAPC4 cell lines . In CWR22Rv1 cells , neither HOXB13ko alone nor HOXB13ko-LV-MEIS1 lines significantly differed from control cells , while LV-MEIS1 ( with HOXB13 present ) remained growth-suppressed ( p<0 . 05 , Figure 3B ) . Somewhat surprisingly , in LAPC4 cells , deletion of HOXB13 significantly decreased proliferation to approximately the same rate as the LV-MEIS1 line compared to controls ( p<0 . 05 ) . Further analyses demonstrated that loss of HOXB13 in LAPC4 cells was not associated with increased cell death ( Figure 3—figure supplement 1 ) . However , in keeping with a requirement for HOXB13 expression with MEIS1 to enact a tumor suppressive effect , ectopic expression of LV-MEIS1 in LAPC4-HOXB13ko cells significantly increased proliferation compared to HOXB13ko ( p<0 . 05 vs . HOXB13ko , Figure 3C ) , rather than further decreasing proliferation or remaining growth suppressed when HOXB13 was present . In this instance , loss of HOXB13 may enable MEIS1 to interact with an alternate HOX protein and differentially regulate cell proliferation . The interaction between MEIS1 and HOXA9 , for example , is known to be tumorigenic in leukemia ( Kelly et al . , 2011 ) . However , given the infrequency of HOXB13 loss in prostate tumor specimens , this scenario is not expected to be observed clinically ( Bhanvadia et al . , 2018; Brechka et al . , 2017 ) . The HOXB13-dependency of MEIS1 tumor suppressive function in PrCa cells was further demonstrated with the analyses of cell migration phenotypes . In both CWR22Rv1 and LAPC4 cells , HOXB13ko and HOXB13ko-LV-MEIS1 lines demonstrated significantly greater migration than LV-MEIS1 cells , while LV-MEIS1 continued to show significantly reduced migration compared to controls ( p<0 . 05 , Figure 3D , E ) . Taken together , these results indicate that MEIS-mediated suppression of cell proliferation and migration in PrCa cells requires HOXB13 expression . Due to previously published clinical data identifying an increased risk of metastasis with loss of MEIS1/2 expression ( Bhanvadia et al . , 2018 ) , the results of MEIS-mediated suppression of in vitro migration , and the putative role for HOXB13 in PrCa progression ( Brechka et al . , 2017 ) , we sought to determine the role of MEIS1 expression and/or HOXB13 deletion using a clinically-relevant in vivo metastasis model . We thus performed intracardiac ( IC ) injection of luciferase-expressing versions of CWR22Rv1-Control , -LV-MEIS1 , -HOXB13ko , and -HOXB13ko-LV-MEIS1 in castrated male athymic nude mice ( Figure 4A ) and monitored metastatic dissemination and growth via in vivo bioluminescent imaging as previously described ( Figure 4B; Kregel et al . , 2016 ) . Overall survival post-injection was significantly increased in LV-MEIS1 cells compared to control cells naturally lacking detectable MEIS1 expression ( p<0 . 05 ) . Further , in accordance with our in vitro data , there was no statistically significant difference in overall survival between Control vs . HOXB13ko and HOXB13ko-LV-MEIS1 cells ( Figure 4C ) . Several clinically relevant organ sites of distant metastases were also observed , including pelvis , vertebrae , skull ( maxilla and mandible ) , kidney , lymph nodes , and lungs ( Figure 4D ) . Notably , LV-MEIS1 cells were the only condition in which bone metastases were not observed . These data demonstrate that MEIS1 suppresses prostate tumor growth and metastatic colonization in vivo , and the tumor-suppressive capability of MEIS1 in vivo is dependent upon expression and interaction with HOXB13 . To identify direct gene targets and pathways regulated by MEIS1 in prostate cells and identify mechanisms of tumor suppression , we performed chromatin immunoprecipitation and sequencing ( ChIP-seq ) of MEIS1 in the CWR22Rv1-LV-MEIS1 line . Additionally , given the dependence on HOXB13 and to enable determination of HOXB13-dependent vs . HOXB13-independent MEIS1 DNA binding , we performed parallel MEIS1 ChIP-seq in the CWR22Rv1-HOXB13ko-LV-MEIS1 line ( Figure 5A ) . In the LV-MEIS1 line , where both MEIS1 and HOXB13 are present , we observed 7559 peaks that were annotated to 4161 unique gene targets ( Figure 5A and Supplementary file 1 ) . In the HOXB13ko-LV-MEIS1 line , where MEIS1 is present but HOXB13 is absent , we observed only 2048 peaks that were annotated to 1617 unique gene targets ( Figure 5A and Supplementary file 2 ) . The reduction in the number of peaks as well as the shift toward new peak locations of the MEIS1 cistrome in the absence of HOXB13 ( Figure 5A ) are supported by previous literature describing that HOX proteins stabilize MEIS1 on the DNA and that heterodimers of HOX and MEIS proteins develop latent motif specificity that is not demonstrated by either protein individually ( Slattery et al . , 2011 ) . As further validation of an interaction between MEIS1 and HOXB13 , we performed spaced motif ( SpaMo ) analysis on MEIS1 peaks from both ChIP-seq experiments to identify conserved motifs associating with MEIS1 . Unsurprisingly , when both MEIS1 and HOXB13 were present in the LV-MEIS1 line , the HOXB13 motif was significantly conserved in MEIS1 peaks at a distance of 1 bp from the MEIS1 motif itself , strongly supporting a direct interaction between these two proteins ( p=2 . 84×10−12 , Figure 5B top ) . On the other hand , when HOXB13 was absent , the only HOX motif associated with MEIS1 was HOXA9 , which is conserved at a distance of 3 bp from the MEIS1 motif ( p=1 . 14×10−6 , Figure 5B bottom ) . Notably , due to the highly conserved DNA binding domain of many HOX genes , the HOXA9 motif potentially associated with MEIS1 in the absence of HOXB13 also showed significant similarity to the Abd-B-like HOX general core motif ( E-value = 1 . 3373×10−6 ) and PBX1 motif ( E-value = 1 . 4640×10−6 ) ( Figure 5—figure supplement 1A ) . Thus , it is feasible that MEIS1 associates with HOXA10 in the absence of HOXB13 , since HOXA10 is the next-highest-expressed HOX in the prostate ( Figure 2A ) . We next conducted RNA-seq to identify MEIS1-mediated gene regulation . These analyses compared global gene expression of CWR22Rv1-Control and CWR22Rv1-LV-MEIS1 as well as HOXB13ko and HOXB13ko-LV-MEIS1 cells to precisely delineate MEIS1- and HOXB13-regulated genes ( Figure 5—figure supplement 1B ) . Importantly , inclusion of HOXB13ko lines enabled determination of HOXB13-associated gene regulation as well as identification of significant changes between LV-MEIS1 and control cells that were HOXB13-independent and thus unrelated to tumor suppression ( Supplementary file 3 ) . Gene regulation was defined as direct MEIS1 binding via ChIP , increased mRNA expression when MEIS1 was ectopically expressed , and loss of MEIS1 genome binding and mRNA expression when MEIS1 was expressed but HOXB13 was deleted . Integration of RNA-seq and ChIP-seq data revealed 745 differentially expressed genes ( DEGs ) ( edgeR , fold-change >1 . 5 and FDR < 0 . 05 ) in CWR22Rv1-LV-MEIS1 compared to controls ( Supplementary file 4 ) . Of those 745 DEGs , 186 were directly targeted by MEIS1; of these targets , 29 genes were also bound by MEIS1 in the HOXB13ko condition and were thus removed since they would not be expected to be critical mediators of MEIS1–HOXB13-mediated tumor suppression . The resulting 157 DEGs ( Supplementary file 5 ) represent genes that are direct targets of MEIS1 only when HOXB13 is present and therefore represent prioritized candidates to elucidate the mechanism of MEIS1-dependent tumor suppression ( Figure 5C ) . Pathway analyses of these 157 genes prioritized multiple putative pathways , of which ‘proteoglycans in cancer’ was the most enriched pathway associated with MEIS1 and HOXB13 expression ( Figure 5D ) . Further analyses documented that the majority of proteoglycans targeted by MEIS1 were upregulated by MEIS1 expression ( Figure 5E ) . Of particular interest was elevated expression of DCN , which was one of the most increased of the significant DEGs in the dataset ( fold-change = 11 . 38 , FDR = 7 . 37×10−12 ) . DCN belongs to the small-leucine-rich-proteoglycan ( SLRP ) family of proteins that has been well-documented to decrease tumor growth and progression ( Bi and Yang , 2013; Csordás et al . , 2000; Edwards , 2012; Goldoni et al . , 2009; Hildebrand et al . , 1994; Iozzo et al . , 1999; Järvinen and Prince , 2015; Khan et al . , 2011; Santra et al . , 2002; Schönherr et al . , 1998; Schönherr et al . , 2005; Zhang et al . , 2018; Zhu et al . , 2005 ) . Lumican ( LUM ) , which also increased with LV-MEIS1 expression ( fold-change = 2 . 88 , FDR = 1 . 79×10−9 ) , is another member of the tumor-suppressive SLRP protein family and has been shown to increase integrin B1 ( ITGB1 ) -mediated adhesion as well as regulate expression of ITGB1 ( D'Onofrio et al . , 2008; Jeanne et al . , 2017; Zeltz et al . , 2010 ) . In parallel , TGFBR3 ( also known as betaglycan ) significantly increased with MEIS1 expression ( fold-change = 1 . 67 , FDR = 4 . 13×10−4 ) and has been shown to inhibit TGFβ signaling and decrease prostate tumor growth and progression in a manner similar to DCN ( Ajiboye et al . , 2010; Eickelberg et al . , 2002; Sharifi et al . , 2007; Turley et al . , 2007 ) . Analyses of MEIS1 ChIP-Seq demonstrated binding in the DCN , LUM , and TGFBR1 genomic region ( Figure 5F ) , and independent ChIP-qPCR validated MEIS1 binding ( Figure 5G ) . Importantly , MEIS1 binding was significantly diminished when HOXB13 was deleted ( Figure 5F and G ) . Increased mRNA expression of DCN , TGFBR3 , LUM , and ITGB1 were validated at the protein level in both CWR22Rv1 and LAPC4 cell lines ( Figure 5H ) . Additionally , increased protein expression was not observed when HOXB13 was absent , thus verifying the dependency of HOXB13 interaction to regulate expression of these targets . The observed MEIS-mediated increase in DCN mRNA and protein expression was also observed in a third prostate cancer cell line , VCAP ( Figure 5—figure supplement 1C ) . Moreover , we previously demonstrated that dual MEIS1/MEIS2 knockdown in LAPC4 cells increased tumor xenograft growth Bhanvadia et al . , 2018; analysis of DCN protein in these cells showed decreased DCN expression when MEIS1 , MEIS2 , and both MEIS1 and MEIS2 were depleted using shRNAs ( Figure 5—figure supplement 2A ) . DCN is a multi-RTK inhibitor and likely has the broadest functional impact of these proteoglycans on cancer-associated pathways and response to growth factors . The most well-established role for DCN is as an inhibitor of TGFβ signaling ( Baghy et al . , 2012; Harper et al . , 1994; Yamaguchi et al . , 1990; Zhu et al . , 2007 ) . DCN can also exert tumor-suppressive functions via affecting multiple other signaling pathways , including EGFR , IGFR1 , AKT , and cMYC . DCN inhibits EGFR signaling after transient activation , leading to increased p21 expression ( Csordás et al . , 2000; Hu et al . , 2009; Moscatello et al . , 1998; Santra et al . , 1997; Seidler et al . , 2006 ) , and DCN binds and inhibits IGF1R and downstream AKT signaling in cancer cells ( Iozzo et al . , 2011; Morrione et al . , 2013; Schönherr et al . , 2005 ) . DCN also antagonizes the c-MET receptor , which can lead to decreased non-canonical β-catenin and decreased cMYC ( by way of increased phospho-T58 , which destabilizes cMYC and leads to degradation ) ( Goldoni et al . , 2009 ) . To test activity of these various pathways , we used MSigDb curated gene sets for oncogenic signatures to perform gene set enrichment analysis ( GSEA ) . The results in Figure 5I indicate that the LV-MEIS1 condition , where DCN expression is high , has significantly decreased pathway activation for TGFβ , ( FDR: 0 . 039 ) , EGFR ( FDR: 0 . 036 ) , WNT ( FDR: 0 . 057 ) , and MYC ( FDR: 0 . 045 ) gene sets . These four specific gene sets represent genes normally suppressed by oncogene activation , and their enrichment in the LV-MEIS1 condition suggests decreased activity of the specified oncogenic pathway . This also correlates with increased DCN expression as an established inhibitor of these oncogenic pathways . We then used RNA-seq data from the four CWR22Rv1 cell line variants ( control , LV-MEIS1 , HOXB13ko , and HOXB13ko-LV-MEIS1 ) to conduct a leading edge of enrichment analysis to further solidify that decreased activity of these oncogenic pathways is a direct result of regulation by MEIS1 and HOXB13 ( Figure 5J ) . GSEA for LV-MEIS1 vs . control was also performed on the MSigDb curated gene sets for ‘gene ontology: biological processes’ , which further supported and expanded our findings with enrichment in pathways including: regulation of epithelial to mesenchymal transition , growth factor binding , integrin mediated signaling , and regulation of response to transforming growth factor beta stimulus ( Supplementary file 6 ) . These data prioritize proteoglycan-mediated tumor suppression , particularly DCN expression , as key mediators of MEIS1-HOXB13-induced tumor suppression in PrCa cells . Given the potential role of DCN as a critical gene target of MEIS1–HOXB13-mediated tumor suppression , we sought to functionally validate the ability of DCN to regulate tumor suppression in MEIS1-expressing CWR22Rv1 and LAPC4 PrCa cells . We thus depleted DCN expression using an siRNA pool to knock-down DCN in LV-MEIS1 lines ( Figure 6A ) . In comparison with the decreased proliferation observed with LV-MEIS1 expression , knockdown of DCN partially abrogated the growth suppression by LV-MEIS1 in CWR22v1 cells ( p<0 . 05 ) but did not have a significant effect in LAPC4 cells ( Figure 6B , C ) . However , siRNA knockdown of DCN in both LV-MEIS1 lines was sufficient to partially restore migratory capacity ( p<0 . 05 , Figure 6D , E ) . We also investigated the effect of DCN knockdown on some of the previously identified pathways under DCN control ( Figure 5G ) . Western blot analyses revealed increased DCN in the presence of MEIS1 , with concomitant decreases in EGFR signaling ( decreased phospho-EGFR and increased total EGFR ) , increased p21 , decreased cMYC signaling ( decreased total cMYC and increased degradation signal at phospho-T58 ) , and decreased TGFβ signaling ( decreased phospho-SMAD2 and increased total SMAD2/3 ) ( Figure 6F ) . Knockdown of DCN in MEIS-expressing cells resulted in partial or complete restoration of EGFR activation , cMYC expression , SMAD2 activation , and decreased p21 expression . Concordant results were also observed in LAPC4 cells with DCN knockdown ( Figure 6—figure supplement 1 ) . The ability of DCN knockdown to restore the activity of EGFR , MYC , and TGFβ pathways , which decreased as a result of LV-MEIS1 expression , along with the partial restoration of cellular migration and proliferation in CWR22Rv1 cells with DCN knockdown , strongly supports a critical role for DCN as a key mediator of MEIS1/HOXB13-dependent metastasis suppression . Previous analysis of DCN in prostate tissues indicated high stromal expression compared to epithelial staining ( Henry et al . , 2018 ) . Protein analyses of DCN , MEIS1 , and HOXB13 in primary PrECs indicated expression of both MEIS1 and HOXB13 and detectable DCN expression ( Figure 7A ) . We subsequently analyzed publicly available RNA-seq datasets from human prostate tumors for associations between MEIS- and DCN-regulated pathways ( Abeshouse et al . , 2015 ) . These analyses of patient-derived datasets agree with our cell line data , whereby MEIS expression demonstrated a positive correlation ( p<0 . 05 ) with DCN , TGFBR3 , LUM , and ITGB1 ( Figure 7B ) .
Within a normal PrECs , MEIS1 complexes with HOXB13 to maintain expression of proteoglycans such as DCN , LUM , and TGFBR3 and repress growth factor and migration/invasion signaling through RTKs ( Figure 7C ) . As a cell transforms to a malignant state , MEIS1/2 are epigenetically silenced in more aggressive prostate tumors ( Bhanvadia et al . , 2018 ) and expression of tumor-suppressive proteoglycans is suppressed , leading to decreased regulation of oncogenic signaling through pathways such as TGFβ , EGFR , cMYC , WNT , and IGF1R ( Figure 7C ) . While this is likely not the only mechanism of MEIS-mediated tumor suppression , it does appear to have clinical significance . However , validation in larger clinical datasets is needed . Loss of MEIS1/2 expression also opens the possibility of new , non-canonical cofactors interacting with HOXB13 and further driving oncogenic signaling . An exciting possibility in this regard is the documented interaction between AR and HOXB13 that arises in malignant cells , and whether pharmacologic restoration of MEIS1 expression blocks oncogenic AR–HOXB13 interaction and thus impedes metastasis and castration-resistance . Further , a mechanism for increased risk and aggressiveness of PrCa observed with HOXB13G84E and HOXB13G135E mutations remains undefined . However , their locations in the MEIS-interacting domains of HOXB13 point toward changes to MEIS–HOXB13 complexes and/or transcriptional regulation as leading to decreased tumor suppression and enabling malignant transformation .
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact and Project PI , Donald J . Vander Griend ( dvanderg@uic . edu ) . To derive cell lines overexpressing lentiviral constructs of MEIS1 or MEIS2 , all lentiviral vectors used in this study contain the gene-of-interest in the pReceiver-LV105 backbone ( GeneCopoeia ) . High titer lentivirus was made by separately co-transfecting the LV105 constructs with ViraPower Lentiviral packaging mix ( #K497500 , Thermo Fisher Scientific ) in HEK-293T cells using Lipofectamine 2000 ( #11668019 , Thermo Fisher Scientific ) according to manufacturer's instructions . After 48 and 72 hr , media containing the lentivirus was collected , spun down , and filtered using a 0 . 45 mm filter and used to infect target CWR-22rv1 or LAPC-4 cells with 5 mg/mL polybrene for 48 hr . Complete media were then replaced followed by selection and maintenance with puromycin ( 1 mg/mL , Invitrogen ) . Confirmation of MEIS1 or MEIS2 expression was confirmed using both qRT-PCR and Western blotting ( anti-MEIS1 #ab19867 , abcam ) ; anti-MEIS2 ( #TA337288 , OriGene ) . To achieve CRISPR generated knockout of HOXB13 , parental CWR22rv1 and LAPC4 cells were seeded at 1 × 106 cells in a 10 cm dish . Cells were co-transfected with a 1:1 ratio of pT2-EF1a-Cas9-P2A-puro and pCMV ( CAT ) T7-SB100 ( #34879 , Addgene ) using Lipofectamine 2000 following manufacturer guidelines . After 48 hr , cells with EF1a-Cas9-P2A-puro integrated into the genome by the SB100 transposase were selected for and maintained with puromycin ( 1 mg/mL , Invitrogen ) . Following 1 week of puromycin selection , Cas9 expression was confirmed by western blot ( #14697 , Cell Signaling Technologies ) . After confirmation of constitutive Cas9 expression , a custom crRNA ( IDT ) targeting the N-terminus of HOXB13 was selected using CHOPCHOP software ( HOXB13 crRNA: 5’- TTGACAGCAGGCATCAGCGT-3’ ) and was annealed with the tracrRNA-ATTO 550 ( #1075927 , IDT ) according to manufacturer guidelines . A final concentration of 10 nM of the crRNA-tracrRNA duplex was then transfected into CWR22Rv1-Cas9 or LAPC4-Cas9 cells using siLentFect Lipid Reagent for RNAi ( #1703360 , Bio-Rad ) according to manufacturer guidelines . Successful knockout of HOXB13 was confirmed by western blot ( anti-HOXB13 ( F-9 ) , #sc-28333 , SCBT ) . Limiting dilution was then performed to establish 6 clonal knockout lines , each verified by western blot . The 6 knockout clones were then re-combined into one knockout pool deemed CWR22rv1- or LAPC4-HOXB13ko . The HOXB13ko lines were then infected with LV-MEIS1 as described above to generate the HOXB13ko-LV-MEIS1 lines for both CWR22Rv1 and LAPC4 . Whole-cell lysates of 100 , 000 or more cells were used . Cells were rinsed with cold PBS and scraped into protein lysis buffer ( 20 mM Tris , pH 7 . 5; 150 mM NaCl; 1 mM EDTA; 1 mM EGTA; 2 . 5 mM sodium pyrophosphate; 1 mM sodium glycerophosphate; 1 mM sodium orthovanadate; 1% Triton-X 100 ) supplemented with cOmplete Mini Protease Inhibitor Cocktail ( #11873580001 , Sigma-Aldrich ) , sonicated on ice for 10 s at 30% amplitude . The Pierce BCA Protein Assay Kit ( #23227 , Thermo Fisher Scientific ) was used to determine protein concentration according to manufacturer directions . Forty micrograms of protein per lane was resuspended in 5x Laemmli Sample Buffer supplemented with 10% b-mercaptoethanol and boiled at 95°C for 5 min . Samples were loaded on a 10% SDS-polyacrylamide gel , and SDS-PAGE was carried out in Tris/Glycine/SDS Buffer for 1 hr at 120 V . Protein was transferred to nitrocellulose membranes with Tris/Glycine Buffer containing 20% methanol for 1 . 5 hr at 4°C with 400 mA . Nitrocellulose membranes were blocked with TBS+5% nonfat milk for 1 hr at room temperature . Primary antibodies were applied in TBST+5% nonfat milk overnight at 4°C at the noted dilutions . ( Anti-β-actin 1:10000; Anti-MEIS1 1:1500; Anti-MEIS2 1:1000; Anti-HOXB13 ( F-9 ) 1:100; Anti-DCN 1:500; Anti-TGFBR3 1:200; Anti-LUM 1:300; Anti-ITGB1 1:1000; Anti-EGFR 1:1000; Anti-EGFR-phos 1:500; Anti-c-MYC 1:1000; Anti c-MYC-pT58 1:500; Anti-SMAD2/3 1:500; Anti-SMAD2-phos 1:200; Anti-p21 1:500 ) . Membranes were washed in TBST 3 times for 5 min . Secondary antibodies were applied in TBST+5% nonfat milk at 1:10 , 000 for 1 hr at room temperature . Membranes were washed in TBST 3 times for 5 min . Blots were scanned with an Odyssey imaging system ( LI-COR Biosciences ) and analyzed with LI-COR Image Studio software . Whole cell lysates from PrECs , CWR22Rv1-Cas9 , CWR22Rv1-LV-MEIS1 , LAPC4-Cas9 and LAPC4-LV-MEIS1 cells were prepared in protein lysis buffer ( 20 mM Tris , pH 7 . 5; 150 mM NaCl; 1 mM EDTA; 1 mM EGTA; 2 . 5 mM sodium pyrophosphate; 1 mM sodium glycerophosphate; 1 mM sodium orthovanadate; 1% Triton-X 100 ) supplemented with cOmplete Mini Protease Inhibitor Cocktail ( #11873580001 , Sigma-Aldrich ) . To pull down MEIS1 , 1000 µg of precleared cell lysates were incubated with 2 μg of Anti-HOXB13 ( EPR17371; Cat #ab201682 , Abcam ) antibody overnight at 4°C . The lysates were then incubated with immobilized Protein A/G Agarose beads ( #20421 , Thermo Scientific , USA ) overnight at 4°C . Finally , the beads were washed three time with protein lysis buffer and centrifuged at 600 g for 5 min . Co-immunoprecipitated proteins were then eluted from the beads by adding 40 μL of RIPA buffer to 5 × SDS PAGE sample buffer and heating at 95°C for 5 min . Samples were further size-fractionated on 10% SDS-polyacrylamide gels . The resolved gels were electro-transferred onto nitrocellulose membranes and probed for Anti-MEIS1 ( 1:1000 , Cat #0T12A3 , Origene ) and Anti-HOXB13 ( 1:1000 ) with respective secondary antibodies ( 1:10000 ) . The membranes were scanned on Odyssey imaging system ( LI-COR Biosciences ) and analyzed with LI-COR Image Studio software . Cultured cells ( 300 , 000 ) were lysed in Buffer RLT containing 1% 2-mercaptoethanol and homogenized with a 28-gauge needle syringe . Subsequent total RNA extraction and DNase treatment of samples was performed using the RNeasy Mini Kit ( #74106 , Qiagen ) and the RNase-Free DNase Set ( #79254 , Qiagen ) according to manufacturer directions . Purified RNA was quantified on a Synergy LX Multi-mode Reader with a Take 3 plate ( BioTek ) and quality tested for RIN score >7 using an Agilent Bioanalyzer 2100 ( Agilent Technologies ) . Reverse transcription of RNA to cDNA was carried out using qScript cDNA SuperMix ( #95048–100 , QuantaBio ) according to manufacturer protocol starting with 1 µg total RNA per reaction . qRT-PCR was done on a Roche LightCycler 96 using Power SYBR Green Master Mix ( #4368702 , Life Technologies ) . Reactions were performed in 20 µL volumes ( 10 µL 2x Power SYBR Green Master Mix; 1 µL of 10 µM forward primer; 1 µL of 10 µM reverse primer; 50 ng cDNA; 7 µL nuclease free water ) . Relative expression of cDNA was normalized by the ΔΔCt method using RPL13A as a housekeeping gene . Total RNA was purified as described above . Illumina sequencing libraries were prepared with the KAPA mRNA-Seq Kit ( #KK8420 , KAPA Biosystems ) following manufacturer’s protocol starting from 2 µg total RNA and aiming for fragment size of 100–200 bp before addition of adapters . All libraries received 9 cycles of amplification . Quality of the enriched libraries was validated using the 2100 TapeStation System aiming for an average final fragment size of approximately 300–350 bp . The libraries were quantified using the Library Quantification Kit – Illumina/Universal Kit ( #KK4824 , KAPA Biosystems ) and evenly pooled together by molarity for 12 libraries per lane . Sequencing was performed by the Functional Genomics Core Facility at the University of Chicago on a HiSeq 4000 Sequencing System ( Illumina ) with 50 bp single-end reads . The quality of raw reads was accessed by FastQC ( v0 . 11 . 4 ) . Adapter sequences and low-quality reads were trimmed using Trimmomatic-0 . 38 . All reads were pseudo-aligned to the human transcriptome built from ENSEMBL Human GRCh38 . p12 cDNA and ncRNA using kallisto ( v0 . 43 . 1 ) with default settings for single-end reads with fragment length of 180 and standard deviation of 20 . Estimated transcript counts were summarized to the gene level using tximport ( v1 . 8 . 0 ) and filtering of lowly-expressed genes ( <10 counts in half the samples of one group ) , library normalization , and differential expression analysis was carried out in edgeR ( v3 . 22 . 2 ) using the glmTreat method with a log fold-change threshold of log2 ( 1 . 5 ) and FDR < 0 . 05 . Further biological insights were gained by performing Gene Set Enrichment Analysis ( GSEA 3 . 0 ) from Broad Institute on MSigDB collections for Oncogenic Signaling and for Gene Ontology: Biological Function gene sets . Pathway analyses were performed using Enrichr online tool . Chromatin isolation of 25 million cells per cell line and chromatin immunoprecipitation ( ChIP ) was performed with the iDeal ChIP-seq Kit for Transcription Factors ( #C01010055 , Diagenode ) according to manufacturer’s guidelines . Chromatin was sheared to a size of 200–500 bp using a Bioruptor Pico ( Diagenode ) and shearing efficiency was verified by agarose gel electrophoresis . The ChIP was performed with 6 µg of ChIP-grade antibody against MEIS1 ( #ab19867 , abcam ) or IgG control antibody ( #3900 , Cell Signaling Technology ) . Anti-MEIS1 ChIP was performed in biological triplicate in CWR22Rv1 LV-MEIS1 cells and in duplicate from CWR22Rv1 HOXB13ko-LV-MEIS1 cells . Verification of immunoprecipitation of target protein and chromatin was verified by western blot . Final immunoprecipitated chromatin concentration was determined with a Qubit dsDNA HS Assay Kit ( #Q32851 , Thermo Fisher Scientific ) . ChIP sequencing libraries were generated using the Low Throughput Library Prep Kit ( #KK8230 , Kapa Biosystems ) according to manufacturer protocol . 5 ng of either cell line specific Input chromatin or MEIS1-ChIP’d chromatin were used to begin the protocol . All libraries underwent 13 cycles of amplification . Quality of the enriched libraries was validated using the 2100 TapeStation System . The libraries were quantified using the Library Quantification Kit – Illumina/Universal Kit and evenly pooled together by molarity for six libraries per lane . Sequencing was performed by the Functional Genomics Core Facility at the University of Chicago on a HiSeq 4000 Sequencing System ( Illumina ) with 50 bp single-end reads . The quality of raw reads was accessed by FastQC ( v0 . 11 . 4 ) . Adapter sequences and low-quality reads were trimmed using Trimmomatic-0 . 38 . Duplicate reads were marked and removed using Picard Tools ( v2 . 18 . 10 ) . Processed reads were then aligned to human reference genome hg38 ( ENSEMBL release 93 , GRCh38 . p12 ) using Bowtie2 with default settings . Samtools was then used to change . sam files to . bam files , sort , and index the . bam files . Peaks were called against sequenced input chromatin using MACS2 and pooling immunoprecipitated samples for either LV-MEIS1 or HOXB13ko-LV-MEIS1 cells . Called peaks were then annotated to hg38 using the annotatePeaks . pl command in HOMER ( v4 . 9 . 1 ) . DeepTools ( v3 . 2 . 0 ) was also used for generation of enrichment heatmaps and bigwig files . SpaMo analysis was performed with MEME Suite online tool , and motif similarity assessment was performed using STAMP online tool . Input chromatin or MEIS1-ChIP chromatin from CWR22Rv1-LV-MEIS1 cells and CWR22Rv1-HOXB13ko-LV-MEIS1 cells were used with 500 pg DNA per reaction . qPCR was done on a LightCycler 96 using Power SYBR Green Master Mix ( Cat #4368702 , Life Technologies ) . Reactions were performed in 20 µL volumes ( 10 µL 2x Power SYBR Green Master Mix; 0 . 5 µL of 10 µM forward primer; 0 . 5 µL of 10 µM reverse primer; 500 pg of Input chromatin or MEIS1-ChIP chromatin; 8 µL nuclease free water ) . Relative binding of MEIS1 to DCN , LUM , and TGFβR3 was normalized to expression levels of Negative and Positive Contorl ACTB-1 ChIP Primer sets ( Active Motif; Carlsbad , CA ) . Cells were seeded at 5 × 105 cells per well in an 8-well glass chamber slide and incubated at 37°C and 5% CO2 for 24 hr . Next , cells were fixed with 4% paraformaldehyde in PBS for 20 min on ice , with gentle shaking . Cells were then quenched with 50 mM ammonium chloride in PBS for 10 min . Cells were then washed 3 times for 5 min each at room temperature with PBS . Cells were then quickly rinsed with MilliQ ddH2O to remove any salts . Plastic chambers were then removed from the slides and reaction areas were delimited with a grease pen on the boarder of each well . Next , cells were permeabilized with 0 . 3% TritonX-100 in PBS for 20 min at room temperature before being washed 3 times for 5 min each with PBS . At this point , the manufacturer protocol for Duolink In Situ Red Starter Kit Mouse/Rabbit ( #DUO92101 , Sigma-Aldrich ) was followed as directed , starting at adding 1 drop of Duolink Blocking Solution to each well of the 8-well slide and incubating for 30 min at 37°C . Primary antibodies against proteins-of-interest used were anti-HOXB13 ( F-9 ) ( 1:100 , #sc-28333 , Santa Cruz Biotechnology ) and anti-MEIS1 ( 1:1000 , #ab19867 , Abcam ) . Secondary-antibody-only controls as well as controls individually replacing each of the primary antibodies with the species matched IgG ( # 3900 , Cell signaling technologies; # 5415 , Cell signaling technologies ) were used to ensure signals were not background . Imaging of slides were done on the Keyence BZX-800 with a 60x-oil objective . All images were taken as 10-micron thick z-stacks with a 0 . 4-micron step size and max projections of each stack were used for image analysis . Image analysis was done using Fiji ( Schindelin et al . , 2012 ) to count the number of foci observed per individual nucleus . Foci that did not overlap with the nuclear DAPI signal were considered to be background and ignored . The presence or absence of each MEIS2 isoform was determined by sequencing individual transcripts via TOPO-TA cloning . MEIS2 transcripts were amplified using pan-MEIS2 PCR primers and fragments inserted into a pCR4-TOPO-TA cloning vector ( Invitrogen ) . Each bacterial colony thus represented a single isoform . 100 colonies per cell line were sequenced using T3 and T7 sequencing primers and analyzed using standard Sanger sequencing . A 5% cutoff was considered to indicate the presence of a particular isoform , whereas below that threshold the presence of the isoform is not definitive . MEIS2E specific primers were designed and used to further confirm the absence of MEIS2E in PrECs . Three different measurement techniques were used to assay cell proliferation in this paper . The first method used was physical counting of cells with the assistance of a Cellometer Auto T4 Bright Field Cell Counter . Cells were plated at 250 k cells per dish into three 60 mm dishes per line and per timepoint ( i . e . for a five-day experiment with cells counted every 24 hr , one cell line would start with fifteen 60 mm dishes – three dishes per timepoint ) . Every 24 hr three dishes per cell line were trypsinized , re-suspended , sent through a filter cap to achieve single cell suspension , mixed with trypan blue to avoid counting dead cells , and counted on Cellometer Auto T4 Bright Field Cell Counter . Counts were performed in triplicate per dish , per timepoint and averaged . The second and third methods for assessing proliferation both used the CyQUANT Direct Cell Proliferation Assay ( #C35011 , Thermo Fisher Scientific ) and either 1 ) measured relative fluorescence in a well using a SpectraMax i3x Multi-Mode Microplate Reader ( #i3x , Molecular Devices ) at 480/535 nm excitation/emission wavelengths; or 2 ) imaging green fluorescence in entire wells at 10x , stitching images , and performing automated cell counting on the Keyence BZX-800 all-in-one fluorescence microscope . The switch to imaging rather than relative fluorescence was made due to lack of availability of the SpectraMax i3x plate reader with bottom-read capabilities at later dates . The relative fluorescence measurement method is dependent on a bottom-read capable plate reader . For both methodologies using the CyQuant direct cell proliferation kit , 1 , 500 cells per well were plated in black-wall , clear-bottom 96-well plates ( #353219 , Corning ) in complete growth medium and incubated at 37C with 5% CO2 . Plates were then read according to the manufacturer protocol with either the SpectraMax i3x or Keyence BZ X-800 every 24 hr post seeding of the cells . Cell cycle was determined on a Cellometer Spectrum by way of propidium iodide ( PI ) fluorescence intensity according to manufacturer protocol ( #CSK-0112 , Nexcelom Bioscience ) ( Chan et al . , 2011 ) . Briefly , cells were plated 48 hr before assay and grown to ~80% confluence . Cells were then trypsinized , filtered to a single cell solution , fixed with ice cold ethanol , treated with PI and RNAse A , washed , and loaded into Cellometer spectrum for imaging and counting . Cell count , size , and PI intensity data were then exported to FCS Express six cytometry software where proper gating was determined for G0/G1 , S phase , and G2/M cell cycle phases based on cell size , count , and PI intensity . Each cell line was performed in triplicate . Cell death due to MEIS1 or MEIS2 exogenous expression was determined by Click-iT TUNEL Alexa Fluor 647 Imaging Assay , for microscopy and HCS kit according to manufacturer protocol ( # C10247 , Thermo Fisher Scientific ) . Briefly , Control , MEIS1 , or MEIS2 expressing cells were plated in clear-bottom , black-wall , 96-well plates at 2 , 500 cells per well and allowed to grow for 48 hr in normal growth media . An extra group of Control expressing cells was plated to be used as a positive control for the assay by treating this group with DNAse 1 . Cells were fixed , permeabilized and treated with TdT reaction buffer followed by the Click-it reaction buffer . AlexaFluor 647 was used as a secondary antibody to visualize successful TdT reactions marking dead cells , and DNA was counterstained with Hoechst to visualize nuclei . Complete wells were imaged on the Keyence BZ X-800 microscope with both a DAPI and a far red filter at 4x . Images were stitched together for each well and double positive nuclei ( Hoechst and AF647 ) were counted manually with the aid of Fiji software . Cell viability over time in LAPC4-Control , -LV-MEIS1 , -HOXB13ko , and -HOXB13ko-LV-MEIS1 was assessed using the ViaStain AOPI Staining Solution ( #CS2-0106 , Nexcelom ) and Cellometer spectrum according to manufacturer guidelines . Briefly , cells were plated at 1 . 0 × 105 in 12 well dishes in complete media and allowed to incubate at 37C and 5% CO2 until specified timepoints at 24 , 48 , and 96 hr . At each time point , the conditioned media , PBS wash , and trypsinized cells were all collected and spun down at 300 x g for four mins . Cells were resuspended in 1 mL of media and mixed 1:1 with Viastain AOPI Staining solution . Live and dead cells were then counted on a Cellometer Spectrum with with the manufacturer defined program for AO/PI Viability that uses Red and green fluorescence as well as brightfield . Four counts from each cell line were performed at each timepoint . Corning Transwell polycarbonate membrane cell culture inserts , 8 µm pore size ( #CLS3422 , Corning ) were used to measure cell migration . Cells were serum starved for 24 hr prior to seeding . Cells were seeded at 1 . 5 × 105 per insert to the top half of each insert in serum free media . The attracting media on the underside of each insert was complete growth media with 10% FBS . All media ( top and bottom ) also contained 3 µM Aphidicolin ( #14007 , Cayman Chemical Company ) to inhibit proliferation in order to decrease the confounding effect of differences seen in proliferation rates seen between Control and MEIS1 or MEIS2 expressing cells . Aphidicolin has been shown to have no impact migration ( Müller et al . , 2002 ) . After 48 hr , non-migrated cells were removed from the top of the insert using a cotton swab . Transwell membranes were then fixed and stained with Crystal Violet staining solution ( 50 mg Crystal Violet; 2 . 7 mL 37% formaldehyde; 1 mL methanol; 96 . 3 mL 1x PBS ) for 20 mins at room temp . After staining , excess stain was washed away by dunking inserts into 6 consecutive , 100 mL aliquots of ddH2O . Inserts were then allowed to air dry for 30mins before membranes were cut out using a razor blade and mounted on imaging slides with CytoSeal60 ( #8310–4 , Thermo Fisher Scientific . ) . Each complete insert was imaged at 10x brightfield on the Keyence BZ X-800 microscope and the Hybrid cell counter function of Keyence software was used to automate the counting of migrated cells stained by crystal violet . For siRNA-mediated knockdown of DCN , we utilized a commercially available pre-validated pool of siRNAs targeting DCN ( # L-021491-00-0005 , Dharmacon ) , as well as a pool of non-targeting control siRNAs ( # D-001810-10-05 , Dharmacon ) . Cells were plated 24 hr prior to transfection and allowed to reach 50–60% confluence . siRNA ( 10 nM per plate ) and siLentFect lipid transfection reagent ( # 1703360 , BioRad ) were prepared according to manufacturer instructions in OptiMEM transfection media . Cells were exposed to transfection media containing siRNAs and siLentFect reagent for 18 hr before switching to normal growth media . DCN knockdown was confirmed at 72 hr post transfection by RT-PCR and western blot . Further experiments were performed as described above . Statistical analyses are as noted in each figure legend and were performed using GraphPad Prism 7 or R . For comparison of two groups , p values were calculated with a one-sided unpaired Student’s t-test or Welch’s two-tailed t-test for samples with unequal variance . For comparison of overall survival of mice in intracardiac study , the survdiff functionality of the R package , survival , was used . Survdiff makes use of log-rank and Chi square tests to determine significance between groups . Adjusted p-values or FDR for all sequencing data was done using Benjamini-Hochberg method . All error bars represent standard error of the mean ( SEM ) . Asterisks ( ∗ ) always indicate significant differences as *=p < 0 . 05; ns = not significant , and n = number of replicates , unless otherwise specified . | Decisions regarding the treatment of patients with early-stage prostate cancer are often based on the risk that the cancer could grow and spread quickly . However , it is not always straightforward to predict how the cancer will behave . Studies from 2017 and 2018 found that samples of less aggressive prostate cancer have higher levels of a group of proteins called MEIS proteins . MEIS proteins help control the production of numerous other proteins , which could affect the behavior of prostate cancer cells in many ways . VanOpstall et al . – including some of the researchers that performed the 2017 and 2018 studies – have investigated how MEIS proteins affect prostate cancer . When prostate cancer cells are implanted into mice , they result in tumors . VanOpstall et al . found that tumors that produced MEIS proteins grew more slowly . Next , MEIS proteins were extracted from the prostate cancer cells and were found to interact with another protein called HOXB13 , which regulates the activity of numerous genes . When the cells were genetically modified to prevent HOXB13 being produced , the protective effect of MEIS proteins was lost . MEIS proteins work with HOXB13 to regulate the production of several other proteins , in particular a protein called Decorin that can suppress tumors . When MEIS proteins and HOXB13 are present , the cell produces more Decorin and the tumors grow more slowly and are less likely to spread . VanOpstall et al . found that blocking Decorin production rendered MEIS proteins less able to slow the spread of prostate cancer . These results suggest that MEIS proteins and HOXB13 are needed to stop tumors from growing and spreading , and some of this ability is by prompting production of Decorin . This study explains how MEIS proteins can reduce prostate cancer growth , providing greater confidence in using them to determine whether aggressive treatment is needed . A greater understanding of this pathway for tumor suppression could also provide an opportunity for developing anti-cancer drugs . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cancer",
"biology"
] | 2020 | MEIS-mediated suppression of human prostate cancer growth and metastasis through HOXB13-dependent regulation of proteoglycans |
Although mutations in HNF4A were identified as the cause of Maturity Onset Diabetes of the Young 1 ( MODY1 ) two decades ago , the mechanisms by which this nuclear receptor regulates glucose homeostasis remain unclear . Here we report that loss of Drosophila HNF4 recapitulates hallmark symptoms of MODY1 , including adult-onset hyperglycemia , glucose intolerance and impaired glucose-stimulated insulin secretion ( GSIS ) . These defects are linked to a role for dHNF4 in promoting mitochondrial function as well as the expression of Hex-C , a homolog of the MODY2 gene Glucokinase . dHNF4 is required in the fat body and insulin-producing cells to maintain glucose homeostasis by supporting a developmental switch toward oxidative phosphorylation and GSIS at the transition to adulthood . These findings establish an animal model for MODY1 and define a developmental reprogramming of metabolism to support the energetic needs of the mature animal .
The global rise in the prevalence of diabetes has prompted increased efforts to advance our understanding of metabolic systems and how they become disrupted in the diseased state . Although genetics and environment have a significant impact on diabetes susceptibility , severity , and care , the causal factors are often complex and unclear . Several cases of familial diabetes have been identified , however , that show clear patterns of heritability due to monogenic disease alleles , highlighting these genes as critical factors for glycemic control . To date , mutations in 13 genes have been shown to cause autosomal dominant inheritance of Maturity Onset Diabetes of the Young ( MODY1-13 ) , representing the most common forms of monogenic diabetes . MODY patients typically present with hyperglycemia and impaired glucose-stimulated insulin secretion ( GSIS ) by young adulthood , while having normal body weight and lacking β-cell autoimmunity ( Fajans and Bell , 2011 ) . Consistent with this , several genes associated with MODY have well-characterized functions in glucose homeostasis , including the glycolytic enzyme Glucokinase ( GCK/MODY2 ) , and Insulin ( INS/MODY10 ) . Mechanistic insight into the anti-diabetic roles of other MODY genes , however , remains limited . The genetic basis for the first MODY subtype was reported two decades ago , identifying loss-of-function mutations in Hepatocyte Nuclear Factor 4A ( HNF4A ) as responsible for MODY1 ( Yamagata et al . , 1996 ) . HNF4A is a member of the nuclear receptor superfamily of ligand-regulated transcription factors , which play important roles in the regulation of growth , development , and metabolic homeostasis . Studies in mice demonstrated a critical requirement for Hnf4A in early development , with null mutants dying during embryogenesis due to defects in gastrulation ( Chen et al . , 1994 ) . Heterozygotes , however , show no apparent phenotypes . As a result , tissue-specific genetic studies were used to investigate the functions of Hnf4A in key tissues where it is expressed , including the liver , kidney , intestine , and pancreatic β-cells . Two groups generated adult mice deficient for Hnf4A in β-cells with the goal of modeling MODY1 ( Gupta et al . , 2005; Miura et al . , 2006 ) . Although both studies reported impaired glucose tolerance in Hnf4A deficient mice , along with defects in GSIS , neither study observed sustained hypoinsulinemic hyperglycemia – the defining symptom that brings MODY1 patients to the clinic . As a result , we still have a limited understanding of the mechanisms by which Hnf4A maintains carbohydrate homeostasis and the molecular basis for MODY1 . Studies in Drosophila have revealed a high degree of conservation with major pathways that regulate cellular metabolism and systemic physiology in humans ( Diop and Bodmer , 2015; Owusu-Ansah and Perrimon , 2014; Padmanabha and Baker , 2014; Teleman et al . , 2012 ) . This includes a central role for the insulin-signaling pathway in maintaining proper levels of circulating sugars through nutrient-responsive secretion of Drosophila insulin-like peptides ( DILPs ) from neuroendocrine cells in the fly brain ( Nassel et al . , 2015 ) . Destruction of these insulin-producing cells ( IPCs ) results in elevated levels of circulating sugars , analogous to type 1 diabetes ( Rulifson et al . , 2002 ) . In addition , nutrient-sensing mechanisms for insulin release are conserved in adult Drosophila , including roles for the Glut1 glucose transporter , mitochondrial metabolism , and ATP-sensitive potassium channels in the IPCs , which respond to the anti-diabetic sulfonylurea drug glibenclamide ( Fridell et al . , 2009; Kreneisz et al . , 2010; Park et al . , 2014 ) . Consistent with these similarities , an increasing number of studies in Drosophila have proven relevant to mammalian insulin signaling and metabolic homeostasis , highlighting the potential to provide insight into human metabolic disorders such as diabetes ( Alfa et al . , 2015; Owusu-Ansah and Perrimon , 2014; Park et al . , 2014; Ugrankar et al . , 2015; Xu et al . , 2012 ) . Here we describe our functional studies of Drosophila HNF4 ( dHNF4 ) with the goal of defining its roles in maintaining carbohydrate homeostasis . dHNF4 is a close ortholog of human HNF4A , with 89% amino acid identity in the DNA-binding domain and 61% identity in the ligand-binding domain . The spatial expression patterns of the fly and mammalian receptors are also conserved through evolution , raising the possibility that they share regulatory activities ( Palanker et al . , 2009 ) . In support of this , our previous studies of dHNF4 mutant larvae demonstrated a critical role in fatty acid catabolism , leading to defects in lipid homeostasis that are similar to those caused by liver-specific HNF4A deficiency in mammals ( Palanker et al . , 2009 ) . Here we report the first functional study of dHNF4 mutants at the adult stage of development . Our studies show that adult dHNF4 mutants display the hallmark symptoms of MODY1 , including hyperglycemia , glucose intolerance and impaired GSIS . Metabolomic analysis of dHNF4 mutants revealed coordinated changes in metabolites that are indicative of diabetes , along with an unexpected effect on mitochondrial activity . This was further evident in our RNA-seq and ChIP-seq studies , which indicate that dHNF4 is required for the proper transcription of both nuclear and mitochondrial genes involved in oxidative phosphorylation ( OXPHOS ) . A homolog of mammalian GCK , Hex-C , is also under-expressed in mutants . dHNF4 appears to act through these pathways to promote GSIS in the IPCs and glucose clearance by the fat body . In addition , we show that dHNF4 expression increases dramatically at the onset of adulthood , along with its downstream transcriptional programs . These studies suggest that dHNF4 triggers a developmental transition that establishes the metabolic state of the adult fly , promoting GSIS and OXPHOS to support the energetic needs of the mature animal .
All genetic studies used a transheterozygous combination of dHNF4 null alleles ( dHNF4Δ17/dHNF4Δ33 ) and genetically-matched controls that were transheterozygous for precise excisions of the EP2449 and KG08976 P-elements , as described previously ( Palanker et al . , 2009 ) . Consistent with this earlier study , dHNF4 null mutants die as young adults , with most mutants failing to emerge properly from the pupal case when raised under standard lab conditions ( Figure 1A ) ( Palanker et al . , 2009 ) . While testing for potential dietary effects on dHNF4 mutant viability , we discovered that sugar levels have a dramatic influence on their survival . When reared on either standard cornmeal food or a medium containing 15% sugar ( 2:1 glucose to sucrose , 8% yeast ) , less than 30% of mutant animals survive though eclosion , and the rest die primarily during the first day of adulthood ( Figure 1A , B ) . In contrast , a five-fold reduction in dietary sugar content is sufficient to rescue most dHNF4 mutants through eclosion and allow them to survive as adults for several weeks ( Figure 1B , C ) . Sugar intolerance persists through adulthood , indicating that dHNF4 plays a critical role in carbohydrate metabolism at this stage ( Figure 1C ) . Notably , this dietary response is specific to alterations in carbohydrate levels , as calorically matched changes in dietary protein did not affect mutant viability ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 11183 . 003Figure 1 . dHNF4 mutants are sugar intolerant and display hallmarks of diabetes . ( A ) Percent survival of genetically-matched controls and dHNF4 mutants at each stage of development when raised on standard media . Adult viability represents survival past the first day of adulthood . ( B ) Percent of control and dHNF4 mutants that successfully eclose when reared on the 15% , 9% , or 3% sugar diet . ( C ) Controls and dHNF4 mutants were reared on the 3% sugar diet until 5 days of adulthood , transferred to the indicated diet , and scored for survival . ( D ) Free glucose levels measured from whole animal lysates of controls and dHNF4 mutants raised on the 3% sugar diet and transferred to the indicated diet for three days . ( E ) Circulating free glucose levels were measured from hemolymph extracted from control and dHNF4 mutant adults raised on the 3% sugar diet and transferred to the 15% sugar diet for 1 day prior to analysis . ( F ) Trehalose levels measured from whole animal lysates of controls and dHNF4 mutants raised on the 3% sugar diet and transferred to the 15% sugar diet for three days . ( G ) Oral-glucose tolerance test performed on adults raised on the 3% sugar diet , fasted overnight , fed on 15% glucose media for 1 hr , and then re-fasted for either 2 or 4 hr . Data represents relative free glucose levels from whole animal homogenates . ( H ) Relative ATP levels in control and dHNF4 mutant adults raised on the 3% sugar diet and transferred to sugar-only medium ( 10% sucrose ) for 1 day prior to analysis . Data is plotted as the mean ± SEM . ***p≤0 . 001 , **p≤0 . 01 , *p≤0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 00310 . 7554/eLife . 11183 . 004Figure 1—figure supplement 1 . Dietary sugar , but not protein , correlates with reduced dHNF4 mutant survival . Newly hatched first instar larvae were placed in vials ( ~60 larvae per vial ) containing either the 3% sugar diet , the 3% sugar diet + 240 kcal/L extra sugar ( 2:1 glucose to sucrose , 9% final concentration ) , or the 3% sugar diet + 240 kcal/L of additional protein ( peptone ) . All diets contained 8% yeast , 1% agar , 0 . 05% MgSO4 , and 0 . 05% CaCl2 , , as described in Materials and methods . Animals were reared at 25˚C and successful eclosion was scored as complete emergence from the pupal case . Data is plotted as the mean ± SEM . ***p≤0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 00410 . 7554/eLife . 11183 . 005Figure 1—figure supplement 2 . Profiling of major metabolites in dHNF4 mutant adults fed different levels of dietary sugar . Controls and dHNF4 mutants were raised to adulthood on the 3% sugar diet under density-controlled conditions . Five days after eclosion , mature adult males were transferred to the 3% , 9% , or 15% sugar diets for three days prior to analysis . Whole-animal lysates were analyzed for concentrations of total triacylglycerol ( TAG ) ( A ) , glycogen ( B ) , or protein ( C ) . All data was normalized to the level in controls on the 9% sugar diet . Data represents six biological replicates each with five animals , and results were consistent between at least three independent experiments . Data is plotted as the mean ± SEM . ***p≤0 . 001 , **p≤0 . 01 , *p≤0 . 05 Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 005 To examine the effects of sugar consumption on the metabolic state of dHNF4 mutants , major metabolites were measured in adult males raised on the low 3% sugar diet and transferred to the 3% , 9% or 15% sugar diet for three days . Although dHNF4 mutants display elevated levels of triglycerides , similar to our observations in mutant larvae , these levels are not affected by the different sugar diets ( Figure 1—figure supplement 2A ) . Similarly , while dHNF4 mutants have reduced glycogen stores and a modest decrease in total protein , the severity of these phenotypes does not correlate with the improved viability due to decreasing dietary sugar ( Figure 1—figure supplement 2B , C ) . In contrast , the abundance of free glucose is greatly elevated in dHNF4 mutants on the 15% sugar diet , but is progressively reduced in mutants exposed to decreasing amounts of dietary sugar , similar to the response of diabetics to a low carbohydrate diet ( Figure 1D ) . As expected , the accumulation of free glucose in dHNF4 mutants represents increased levels in circulation and is accompanied by elevated levels of the glucose disaccharide trehalose ( Figure 1E , F ) . Taken together , these results demonstrate that Drosophila HNF4 is required for proper glycemic control . To assess whether the hyperglycemia in dHNF4 mutants arises due to impaired glucose clearance , adult flies were subjected to an oral glucose tolerance test . Control and mutant animals were reared on the low sugar diet , fasted overnight , transferred to a glucose diet for one hour , and then re-fasted for 2 or 4 hr . Although dHNF4 mutants display a normal postprandial spike in free glucose levels after feeding , glucose clearance is significantly impaired in mutant animals at both 2 and 4 hr , indicating glucose intolerance ( Figure 1G ) . Taken together , these data demonstrate that dHNF4 mutant adults display hallmarks of diabetes and may provide an animal model of MODY1 . Small-molecule gas chromatography/mass spectrometry ( GC/MS ) metabolomic analysis was used to further characterize the metabolic state of dHNF4 mutants fed a 3% or 15% sugar diet ( Figure 2 ) . This study confirmed and extended our observations of their diabetic phenotype and revealed underlying defects in glucose homeostasis that are independent of dietary sugar content . Consistent with hyperglycemia , dHNF4 mutants accumulate glycolytic metabolites on both diets . These include elevated glucose-6-phosphate , dihydroxyacetone phosphate ( DHAP ) , and serine , which is produced from 3-phosphoglycerate , although the increased DHAP was only observed on the 15% sugar diet ( Figure 2 ) . Several other glucose-derived metabolites are aberrantly increased in dHNF4 mutants , including sorbitol and fructose , which are intermediates in the polyol pathway ( Figure 2 and Figure 2—figure supplement 1 ) . This pathway provides an alternate route for cellular glucose uptake under conditions of sustained hyperglycemia . As a result , these metabolites can accumulate to high levels in diabetics and correlate with neuropathy and nephropathy ( Gabbay , 1975 ) . dHNF4 mutants also display increased levels of inosine , adenine , xanthine , hypoxanthine , and uric acid , which are purine metabolites that are associated with increased diabetes risk and diabetic nephropathy ( Figure 2 ) ( Johnson et al . , 2013 ) . Taken together , these findings reveal additional similarities between the dHNF4 mutant phenotype and the metabolic complications of diabetes in humans . Finally , in addition to elevated carbohydrates , we observed increased levels of pyruvate and lactate accompanied by decreased levels of ATP , suggesting a potential defect in mitochondrial respiration ( Figure 1H , 2 ) . 10 . 7554/eLife . 11183 . 006Figure 2 . dHNF4 mutants display defects in glycolysis and mitochondrial metabolism . GC/MS metabolomic profiling of controls and dHNF4 mutants raised to adulthood on the 3% sugar diet , transferred to the indicated diet for 3 days , and subjected to analysis . Data were obtained from three independent experiments consisting of 5–6 biological replicates per condition and values were normalized to control levels on the 15% sugar diet . Box plots are presented on a log scale , with the box representing the lower and upper quartiles , the horizontal line representing the median , and the error bars representing the minimum and maximum data points . ***p≤0 . 001 , **p≤0 . 01 , *p≤0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 00610 . 7554/eLife . 11183 . 007Figure 2—figure supplement 1 . dHNF4 mutants show broad defects in carbohydrate homeostasis . GC/MS metabolomic profiling reveals that a range of sugars and sugar alcohols are increased in dHNF4 mutants compared to genetically matched controls , indicating that dHNF4 is required for proper carbohydrate metabolism . Animals were raised to adulthood on the 3% sugar diet prior to being transferred to the indicated diet for 3 days . Data was obtained from three independent experiments consisting of five to six biological replicates per condition and values were normalized to control levels on the 15% sugar diet . Data are graphically represented as a box plot , with the box representing the lower and upper quartiles , the horizontal line representing the median , and the error bars denoting the minimum and maximum data points . ***p≤0 . 001 , **p≤ 0 . 01 , *p≤0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 007 To further assess mitochondrial metabolism , dHNF4 mutant adults were maintained on 10% sucrose medium for three days and analyzed for TCA cycle intermediates using GC/MS metabolomics . This approach was aimed at restricting the ability of dietary amino acids to replenish TCA cycle intermediates by anapleurosis to provide more robust detection of underlying defects in this pathway . Interestingly , dHNF4 mutants display specific alterations in these metabolites , with increased abundance of citrate , aconitate , isocitrate , fumarate and malate , along with decreased levels of alpha-ketoglutarate and succinate , suggesting a specific block in TCA cycle progression ( Figure 3—figure supplement 1 ) . Taken together , these metabolite changes suggest that mitochondrial function is impaired in dHNF4 mutants , providing a possible primary cause for their glucose intolerance . As a first step toward identifying transcriptional targets of dHNF4 that mediate its effects on glucose homeostasis , we performed RNA-seq profiling in control and mutant adults . A total of 1370 genes are differentially expressed in dHNF4 mutants ( ≥1 . 5-fold change , 1% FDR ) , with just over half of these genes showing reduced abundance ( 726 down , 644 up ) ( Supplementary file 1 ) . Gene ontology analysis revealed that the majority of the down-regulated genes correspond to metabolic functions , with the most significant category corresponding to oxidoreductases ( Supplementary file 2 ) . In contrast , the up-regulated genes largely correspond to the innate immune response , reflecting a possible inflammatory response in dHNF4 mutants ( Supplementary file 2 ) . Interestingly , most of the transcripts encoded by the mitochondrial genome ( mtDNA ) are expressed at greatly reduced levels in mutant animals ( Supplementary file 1 ) . In Drosophila , as in humans , the mitochondrial genome contains 13 protein-coding genes , all of which encode critical components of the electron transport chain ( ETC ) that contribute to oxidative phosphorylation ( OXPHOS ) . Further analysis of these transcripts by northern blot hybridization confirmed their reduced expression in dHNF4 mutants , corresponding to mtDNA genes involved in Complex I ( mt:ND1 , mt:ND2 , mt:ND4 , mt:ND5 ) , Complex IV ( mt:Cox1 , mt:Cox2 , mt:Cox3 ) and Complex V/ATP synthase ( mt:ATPase6 , mt:ATPase8 ) , along with reduced levels of the mitochondrial large ribosomal RNA ( mt:lrRNA ) ( Figure 3A , Supplementary file 1 ) . Importantly , not all mtDNA genes are misregulated , as the expression of mt:Cyt-b is consistently unaltered in mutants ( Figure 3A ) . In addition , the copy number of mtDNA is unaffected in dHNF4 mutants , suggesting that mitochondrial abundance is normal in these animals ( Figure 3—figure supplement 2A ) . 10 . 7554/eLife . 11183 . 008Figure 3 . dHNF4 regulates nuclear and mitochondrial gene expression . ( A ) Validation of RNA-seq data by northern blot using total RNA extracted from control and dHNF4 mutant adults . Affected transcripts include those involved in glucose homeostasis ( Hex-C , pdgy ) , the electron transport chain ( Sdhaf4 , mt:ND1 , mt:ND2 , mt:ND4 , mt:ND5 , mt:CoxI , mt:Cox2 , mt:Cox3 , mt:ATPase6/8 , mt:Cyt-b and mt:lrRNA ) , the TCA cycle ( Scsalpha , dSdhaf4 ) , and insulin signaling ( 4EBP , InR ) . rp49 is included as a control for loading and transfer . Mitochondrial-encoded transcripts are indicated by the prefix 'mt' . Depicted results were consistent across multiple experiments . ( B–C ) ChIP-seq analysis performed on adult flies for endogenous dHNF4 genomic binding shows direct association with both nuclear ( B ) and mitochondrial-encoded ( C ) genes involved in OXPHOS . Data tracks display q value FDR ( QValFDR ) significance values ( y-axis ) compared to input control , where QValFDR 50 corresponds to P=10–5 and 100 corresponds to P=10–10 . Gene names in bold represent those expressed at reduced levels in dHNF4 mutants by RNA-seq and/or northern blot analysis . Gene names in red ( ND6 , Cyt-B ) denote the mtDNA-encoded transcriptional unit confirmed to show no change in dHNF4 mutants . ( D ) Whole-mount immunostaining of adult fat body tissue for ATP5A ( green ) to detect mitochondria and DAPI ( blue ) to mark nuclei , showing fragmented mitochondrial morphology in dHNF4 mutants . ( E ) Analysis of dHNF4 mutant MARCM clones ( GFP+ ) shows reduced mitochondrial membrane potential by TMRE staining of live fat body tissue from adult flies maintained on the 15% sugar diet . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 00810 . 7554/eLife . 11183 . 009Figure 3—figure supplement 1 . dHNF4 mutants display changes in TCA cycle intermediates that correlate with changes in gene expression . GC/MS metabolomic profiling of adult flies fed a sugar only diet ( 10% sucrose + 1% agar ) reveals defective TCA cycle metabolism in dHNF4 mutants . Intermediates of the TCA cycle are presented along with genes that are underexpressed in dHNF4 mutants ( in italics ) . The mRNA abundance for each gene in dHNF4 mutants is indicated ( red text ) relative to matched controls from RNA-seq data . These genes are predicted to affect enzymatic complexes for isocitrate dehydrogenase ( IDH1/2 ) , alpha-ketoglutarate dehydrogenase ( α-KGDH ) , succinyl-CoA synthetase ( SCS ) , or succinate dehydrogenase ( SDH ) . Consistent with impaired IDH and SCS activity , levels of alpha-ketoglutarate and succinate are reduced , while upstream citrate , aconitate , and isocitrate accumulate . The modest effect on succinate abundance is possibly due to a combined effect of impaired succinate production balanced by decreased succinate oxidation resulting from impaired SDH activity . In addition , intermediates of the urea cycle are broadly accumulated in dHNF4 mutants including fumarate , which lies at the interface of these two pathways . This may contribute to the elevated levels of fumarate and downstream malate independent of impaired SDH activity . Metabolites lacking graphical data were not detected or could not be measured by GC/MS . Animals were reared on the 3% sugar diet until 5 days of adulthood and then transferred to medium containing 10% sucrose with 1% agar ( sugar only ) for 3 days prior to analysis . Data represents six biological replicates consisting of 20 males per replicate . ***p≤0 . 001 , **p≤0 . 01 , *p≤0 . 05 . Inset - western blot analysis of whole animal lysates revealed reduced levels of SDHB protein , but no effect on SDHA , in dHNF4 mutants . ATP5A was detected as a control for loading and transfer . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 00910 . 7554/eLife . 11183 . 010Figure 3—figure supplement 2 . dHNF4 mutants display mitochondrial defects . ( A ) mtDNA abundance relative to nuclear DNA quantified by qPCR in control and dHNF4 mutant adult males on the 15% sugar diet . ( B ) Adult fat body tissue from control or dHNF4 mutants immunostained for ATP5A ( green ) to detect mitochondria , dHNF4 protein ( red ) , and DAPI ( blue ) to mark nuclei . ( C ) Analysis of dHNF4 mutant MARCM clones ( GFP+ ) shows reduced mitochondrial membrane potential by TMRE staining of live fat body tissue from adult flies maintained on the 15% sugar diet . ( D ) MARCM clonal analysis of adult dHNF4 mutant fat body cells ( GFP+ ) for reactive oxygen species ( ROS ) by DHE staining shows no detectable difference in ROS levels in mutant cells . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 01010 . 7554/eLife . 11183 . 011Figure 3—figure supplement 3 . Predicted functions of dHNF4 target genes . Forty-seven genes identified as high confidence targets for direct regulation by dHNF4 are depicted . These genes fit the criteria of showing proximal dHNF4 binding along with reduced transcript abundance in mutant animals as determined by RNA-seq ( ≥1 . 5 fold change , 1% FDR ) ( Supplementary Tables 1 , 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 011 Several nuclear-encoded OXPHOS genes also require dHNF4 for their maximal expression , including genes that encode the alpha and beta subunits of the electron transfer flavoprotein ( ETFA and ETFB ) , ETF-ubiquinone oxidoreductase ( ETF-QO ) , and the Complex II ( succinate dehydrogenase , SDH ) assembly factor dSdhaf4 ( Figure 3A , Supplementary file 1 ) . Similar to flies lacking dSdhaf4 , dHNF4 mutants display reduced steady-state levels of SDH complex as assayed by western blot ( Figure 3—figure supplement 1 ) ( Van Vranken et al . , 2014 ) . These observations are thus consistent with impaired mitochondrial SDH function , and suggest that dSdhaf4 is a critical functional target of dHNF4 . Additional genes involved in the TCA cycle are misexpressed in dHNF4 mutants , including Succinyl-CoA synthetase alpha ( Scsalpha ) , CG5599 ( which encodes a protein with homology to the E2 subunit of the α-ketoglutarate dehydrogenase complex ( α-KGDHC ) as well as the E2 subunit of the branched-chain alpha-ketoacid dehydrogenase complex ) , CG1544 ( which encodes a homolog of α-KGDHC E1 ) , as well as Isocitrate dehydrogenase ( IDH , NADP+-dependent ) ( Figure 3A , Supplementary file 1 ) . These changes in gene expression are consistent with the observed changes in the levels of TCA cycle intermediates in dHNF4 mutants , suggesting that they are functionally relevant to the mutant metabolic phenotype ( Figure 3—figure supplement 1 ) . Notably , dHNF4 mutants also have decreased expression of the GCK homolog Hexokinase-C ( Hex-C ) ( Figure 3A ) . GCK is a tissue-specific glycolytic enzyme that is required for glucose sensing by pancreatic β-cells and glucose clearance by the liver . These activities , combined with the association of GCK mutations with MODY2 , make Hex-C a candidate for mediating the effects of dHNF4 on carbohydrate metabolism . The glucose transporter CG1213 is also down-regulated in dHNF4 mutants , along with phosphoglucomutase ( pgm ) , which is involved in glycogen metabolism , and transaldolase and CG17333 , which are involved in the pentose phosphate shunt . Additionally , the gluconeogenesis genes Pyruvate carboxylase ( CG1514 ) and Phosphoenolpyruvate carboxykinase ( Pepck , CG17725 ) show reduced expression in mutant animals , similar to their dependence on Hnf4A for expression in the mammalian liver ( Supplementary file 1 ) ( Chavalit et al . , 2013; Yoon et al . , 2001 ) . Finally , dHNF4 mutants display transcriptional signatures of reduced insulin signaling , including up-regulation of the dFOXO-target genes 4EBP and InR ( Figure 3A , Supplementary file 1 ) . Taken together , these findings indicate an important role for dHNF4 in mitochondrial OXPHOS and glucose metabolism , and suggest that it acts through multiple pathways to maintain glycemic control . Chromatin immunoprecipitation followed by high-throughput sequencing ( ChIP-seq ) was performed to identify direct transcriptional targets of the receptor . Through this analysis , forty-seven genes were identified as high confidence targets by fitting the criteria of showing proximal dHNF4 binding along with reduced transcript abundance in mutant animals ( ≥1 . 5 fold change , 1% FDR ) ( Figure 3—figure supplement 3 , Supplementary file 3 ) . These include nuclear-encoded OXPHOS genes such as ETFB , ETF-QO , dSdhaf4 , and genes that encode TCA cycle factors Scsalpha and CG5599 ( Figure 3B ) . We also observed abundant and specific binding of dHNF4 within the control region of the mitochondrial genome ( Figure 3C and Supplementary file 3 ) . Taken together with our other results , these data suggest that dHNF4 is required to maintain normal mitochondrial function . Consistent with this , mitochondrial morphology is severely fragmented in mutant animals , and MARCM clonal analysis in the adult fat body shows reduced mitochondrial membrane potential in dHNF4 mutant cells ( Figure 3D , E and Figure 3—figure supplement 2B , C ) . In contrast , we were unable to detect changes in reactive oxygen species ( ROS ) in dHNF4 mutant clones by DHE staining ( Figure 3—figure supplement 2D ) . This might be due to the decreased levels of ROS-generating ETC complexes in dHNF4 mutants , along with no detectable effect on the transcripts that encode ROS-scavenging enzymes , such as catalase and SOD ( Supplementary file 1 ) . Taken together , these data support the model that dHNF4 regulates both nuclear and mitochondrial gene expression to promote OXPHOS and maintain mitochondrial integrity . Tissue-specific RNAi was used to disrupt dHNF4 expression in the IPCs , fat body , and intestine to examine the contributions of dHNF4 in these tissues to systemic glucose homeostasis . This revealed a requirement in both the IPCs and fat body for glucose homeostasis , consistent with the well-established roles of these tissues in insulin signaling and the regulation of circulating sugar levels ( Figure 4A , Figure 4—figure supplement 1B ) . Our initial functional analysis of dHNF4 target genes supports these tissue-specific activities and provides insights into the molecular mechanisms of dHNF4 action . Tissue-specific inactivation of Hex-C by RNAi demonstrates that it is required in the fat body , but not the IPCs , to maintain normal levels of circulating glucose ( Figure 4B , C ) . This is consistent with the important role of mammalian GCK for glucose clearance by the liver as well as its association with MODY2 ( Postic et al . , 1999 ) . In contrast , both fat body and IPC-specific RNAi for the direct target of dHNF4 , CG5599 , significantly impaired glucose homeostasis ( Figure 4B , C ) . This indicates that CG5599 is required in each of these tissues for glycemic control , similar to dHNF4 , suggesting that it is a key downstream target of the receptor . Although technical limitations prevent us from performing tissue-specific RNAi studies of mitochondrial-encoded transcripts , disruption of ETC Complex I by targeting a critical assembly factor , CIA30 ( Complex I intermediate-associated protein 30 kDa ) ( Cho et al . , 2012 ) in either the IPCs or fat body produced elevated levels of free glucose ( Figure 4B , C ) . Fat body-specific RNAi for dHNF4 , CIA30 , or CG5599 resulted in fragmented mitochondrial morphology , consistent with previous reports of CIA30 loss of function and the onset of mitochondrial dysfunction ( Figure 4D ) ( Cho et al . , 2012 ) . In contrast , RNAi for Hex-C had no detectable effect on mitochondrial morphology ( Figure 4D ) . 10 . 7554/eLife . 11183 . 012Figure 4 . dHNF4 acts through multiple tissues and pathways to control glucose homeostasis . ( A ) Circulating glucose levels in adult males expressing tissue-specific RNAi against mCherry ( TRiP 35785 , grey bars ) or dHNF4 ( TRiP 29375 , dark red bars ) in the fat body ( r4-GAL4 ) , IPCs ( dilp2-GAL4 ) , or midgut ( mex-GAL4 ) . ( B–C ) Relative free glucose levels in adult males on the 15% sugar diet expressing fat body ( r4-GAL4 , B ) or IPC ( dilp2-GAL4 , C ) -specific RNAi compared to mCherry RNAi controls ( light grey bars ) . RNAi lines directed against Hex-C , CG5599 , Scsα , CIA30 , Cox5a , and ATPsynβ were obtained from the TRiP RNAi collection . Blue and orange bars depict significant changes in glucose levels . Dark grey bars are not significant . Data represents the mean ± SEM . ***p≤0 . 001 , **p≤0 . 01 , *p≤0 . 05 . ( D ) Confocal imaging of mitochondrial morphology ( marked by ATP5A immunostaining , red ) in the adult fat body from animals expressing fat-body specific RNAi ( r4-GAL4 ) . The extended network of mitochondria seen in controls is disrupted and appears more punctate upon RNAi for dHNF4 , CIA30 , or CG5599 , indicative of mitochondrial fragmentation . No effect is seen upon RNAi for Hex-C . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 01210 . 7554/eLife . 11183 . 013Figure 4—figure supplement 1 . dHNF4 is required in the insulin-producing cells and fat body to maintain glucose homeostasis . ( A ) Tissue-specific RNAi directed against dHNF4 effectively reduces steady-state levels of dHNF4 protein . Immunostaining was used to detect dHNF4 in organs dissected from third instar larvae that express either the r4-GAL4 driver alone as a control or fat body-specific RNAi for dHNF4 ( r4>dHNF4RNAi ) using UAS-RNAi constructs from either the TRiP or VDRC stock collections ( fb = fat body , mt = Malpighian tubules ) . ( B ) dHNF4 function is required in both the IPCs and fat body for glucose homeostasis . This data reproduces that shown in Figure 4A using a distinct dHNF4 UAS-RNAi construct . Circulating glucose levels were measured in animals with tissue-specific RNAi against dHNF4 ( VDRC ) in the IPCs ( dilp2-GAL4 ) or fat body ( r4-GAL4 ) compared to GAL4 and UAS lines alone as controls . Controls are represented by grey bars while dHNF4 RNAi is represented by blue bars . Data represents the mean ± SEM , ***p≤0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 01310 . 7554/eLife . 11183 . 014Figure 4—figure supplement 2 . Fat body-specific disruption of the electron transport chain causes sugar intolerance . Newly hatched first instar larvae were placed in vials ( ~60 per vial ) containing the 3% or 15% sugar diet and scored for puparium formation when raised at either 25˚C ( A ) or 18˚C ( B ) . Disruption of ETC complex IV ( Cox5a RNAi ) or complex V ( ATPsynβ RNAi ) in the fat body ( r4-GAL4 ) results in sugar intolerance as seen by a pronounced developmental delay and reduced survival to puparium formation on the 15% sugar diet . These defects are partially suppressed when reared on the low sugar diet at either temperature; however , the 18˚C condition allows for more robust suppression of the sugar-dependent developmental delay ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 01410 . 7554/eLife . 11183 . 015Figure 4—figure supplement 3 . Additional RNAi lines confirming the importance of Hex-C in the fat body for glycemic control . Tissue-specific RNAi directed against Hex-C in the fat body , but not midgut , results in elevated levels of free glucose . RNAi lines against Hex-C were obtained from the VDRC RNAi collection ( UAS-Hex-C RNAi#1: VDRC 35337 , and UAS-Hex-C RNAi#2: VDRC 35338 ) . Grey bars represent the indicated GAL4 driver line crossed to w1118 as a negative control . These results are consistent with those in Figure 4B showing hyperglycemia upon fat body-specific RNAi for Hex-C using a line from the TRiP collection ( Bloomington stock #57404 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 015 While RNAi for Complex V ( ATPsynβ RNAi ) in the fat body caused lethality prior to eclosion , IPC-specific RNAi produced viable adults that appeared normal but displayed significant hyperglycemia ( Figure 4C ) . In contrast , disruption of ETC Complex IV in the IPCs ( Cox5a RNAi ) failed to produce hyperglycemia , while RNAi in the fat body caused lethality prior to adulthood , similar to ATPsynβ . This premature lethality was accompanied by severe developmental delay and more than 50% of the animals dying prior to puparium formation when raised on the 15% sugar diet . Interestingly , we discovered that these animals are sugar intolerant , similar to dHNF4 mutants , such that rearing them on the 3% sugar diet allowed for 100% survival to puparium formation while also alleviating the developmental delay ( Figure 4—figure supplement 2 ) . Although adult viability was not achievable through dietary intervention , these findings demonstrate that ETC function in the fat body is important for sugar tolerance during development , similar to the requirement for dHNF4 . Taken together , these data reveal important roles for dHNF4 in both the IPCs and fat body to maintain glucose homeostasis , likely in part by promoting Hex-C expression in the fat body for glucose clearance and supporting mitochondrial function and OXPHOS in both the IPCs and fat body . The requirement for dHNF4 function in the IPCs for systemic glucose homeostasis fits with the important roles for Hnf4A in mouse pancreatic β-cells as well as the contribution of β-cell physiology to the onset of MODY1 . Accordingly , we examined if dHNF4 mutants display defects in GSIS . We used an experimental approach developed for this purpose in Drosophila larvae , assaying for the steady-state levels of DILP2 peptide in the IPCs using a fasting/refeeding paradigm ( Geminard et al . , 2009 ) . As expected , DILP2 accumulates in the IPCs of fasted control animals and is effectively released into circulation in response to glucose feeding ( Figure 5A , B ) . In contrast , while DILP2 accumulates normally in fasted dHNF4 mutants , it fails to respond to dietary glucose stimulation , despite these animals having normal IPC number and morphology ( Figure 5A , B ) . Peripheral insulin signaling is also reduced in dHNF4 mutants relative to controls , consistent with their reduced GSIS ( Figure 5C ) . This defect in GSIS is due to a tissue-specific requirement for dHNF4 in the IPCs since IPC-specific RNAi for dHNF4 resulted in impaired DILP2 secretion into the hemolymph , along with reduced peripheral insulin signaling ( Figure 5D , E and Figure 5—figure supplement 1 ) ( Park et al . , 2014 ) . Taken together , these data demonstrate that impaired GSIS plays a central role in the diabetic phenotype of dHNF4 mutants . 10 . 7554/eLife . 11183 . 016Figure 5 . dHNF4 is required for glucose-stimulated DILP2 secretion by the insulin-producing cells . ( A ) Whole-mount staining for DILP2 peptide in brains dissected from adult control and dHNF4 mutants that were either fasted overnight or re-fed glucose for two hours . ( B ) Quantification of relative DILP2 fluorescent intensity in the IPCs of fasted and glucose-fed controls and dHNF4 mutants . Data is plotted as a Tukey boxplot with outliers denoted as individual data points ( n= 11 ± 3 brains per-condition ) . Results were consistent between three independent experiments . ( C ) Western blot analysis to detect phosphorylated AKT ( pAKT ) , total AKT , and Tubulin in extracts from controls and dHNF4 mutants that were either fasted overnight or re-fed glucose for two hours . ( D ) Levels of circulating HA-FLAG-tagged DILP2 ( DILP2HF ) were assayed in animals with IPC-specific RNAi ( TRiP ) against either mCherry as a control ( blue ) or dHNF4 ( red ) using the dilp2-GAL4 driver ( dilp2>RNAi ) . Data is combined from five independent experiments , each containing 5–6 biological replicates per genotype . The horizontal lines depict the mean value . ( E ) Western blot analysis to detect phosphorylated AKT ( pAKT ) , total AKT , and Tubulin in extracts from ad libitum fed adult males with IPC-specific RNAi against either mCherry as a control or dHNF4 using the dilp2-GAL4 driver ( dilp2>RNAi ) . ***p≤0 . 001 , **p≤0 . 01 , *p≤0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 01610 . 7554/eLife . 11183 . 017Figure 5—figure supplement 1 . dHNF4 RNAi in the IPCs causes reduced levels of circulating DILP2-HF . This data reproduces that shown in Figure 5D using a distinct dHNF4 UAS-RNAi construct ( VDRC collection ) . Levels of circulating HA-FLAG-tagged DILP2 ( DILP2HF ) were assayed in ad libitum fed animals with IPC-specific RNAi against dHNF4 ( VDRC 12692 ) using the dilp2-GAL4 driver compared to animals carrying the dilp2-GAL4 driver alone as a control . Dot plot depicts individual sample values from two independent experiments and the midline represents the mean ( n=6 biological replicates per genotype per experiment ) . ***p≤0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 017 As we reported in our prior study of dHNF4 , the receptor is not expressed in the larval IPCs ( Figure 6A ) ( Palanker et al . , 2009 ) . It is , however , expressed in the IPCs of the adult fly , consistent with its central roles at this stage in GSIS , insulin signaling , and glucose homeostasis ( Figure 6B ) . Interestingly , this cell-type specific switch in dHNF4 expression correlates with a developmental change in IPC physiology . Unlike mammalian β-cells , larval IPCs fail to secrete DILPs in response to dietary glucose ( Geminard et al . , 2009 ) . Adult IPCs , however , display calcium influx , membrane depolarization , and DILP2 secretion in response to glucose , analogous to β-cells ( Alfa et al . , 2015; Fridell et al . , 2009; Kreneisz et al . , 2010; Park et al . , 2014 ) . Along with the temporal induction of dHNF4 expression in adult IPCs , these results suggest that there is a developmental switch in the response to glucose at the onset of adulthood . Consistent with this , glucose feeding activates insulin signaling in adult flies , but not in larvae ( Figure 6C ) . This correlates with a ~ten-fold increase in the basal circulating levels of glucose in adults compared to larvae , which first becomes apparent during the final stages of pupal development ( Figure 6D ) ( Tennessen et al . , 2014a ) . Moreover , dHNF4 mutants maintain euglycemia on a normal diet during larval and early pupal stages , but display hyperglycemia just prior to eclosion ( Figure 6D ) . Taken together , these observations point to a switch in IPC physiology and glucose homeostasis as Drosophila transition into maturity . 10 . 7554/eLife . 11183 . 018Figure 6 . dHNF4 supports a developmental transition toward GSIS and OXPHOS in adult Drosophila . ( A–B ) Whole-mount immunostaining of larval ( A ) or adult ( B ) brains to detect dHNF4 protein ( magenta ) or GFP , which marks the IPCs ( dilp2>GFP , green ) . ( C ) Western blot analysis to detect phosphorylated AKT ( pAKT ) and total AKT in extracts from w1118 third-instar larvae or mature adults that were fasted overnight and re-fed 10% glucose for two hours . ( D ) Relative levels of free glucose in controls and dHNF4 mutants staged as either feeding third-instar larvae ( larva ) , white prepupae ( wpp ) , pharate adults ( ~4 day-old pupae ) , or mature adults . Data is plotted as the mean ± SEM . ***p≤0 . 001 , **p≤0 . 01 , *p≤0 . 05 . ( E ) Northern blot analysis of RNA extracted from feeding third-instar larvae ( larva ) , white prepupae ( wpp ) , pupae at one-day intervals , mid-eclosion ( eclo ) , and mature adults . rp49 is included as a control for loading and transfer . DOI: http://dx . doi . org/10 . 7554/eLife . 11183 . 018 The induction of dHNF4 in the adult IPCs and the adult onset of hyperglycemia in dHNF4 mutants raise the interesting possibility that this receptor may play a role in coordinating the metabolic switch toward GSIS and OXPHOS at this stage . Indeed , northern blot analysis of RNA samples isolated from staged wild-type larvae , pupae , and young adults , demonstrate that dHNF4 expression increases dramatically at the onset of adulthood ( Figure 6E ) . Moreover , this temporal pattern of expression is accompanied by increased expression of both nuclear and mitochondrial-encoded dHNF4 target genes that contribute to OXPHOS as well as Hex-C . The expression of these target genes is also reduced in staged dHNF4 mutants , consistent with our earlier findings that their maximal expression depends on receptor function ( Figure 6E ) . Taken together , our results support the model that dHNF4 contributes to a metabolic switch in glucose homeostasis at the onset of adulthood that promotes GSIS and OXPHOS to meet the energy demands of the adult fly .
Drosophila HNF4 mutants display late-onset hyperglycemia accompanied by sensitivity to dietary carbohydrates , glucose intolerance , and defects in GSIS – hallmarks of MODY1 . These defects arise from roles for dHNF4 in multiple tissues , including a requirement in the IPCs for GSIS and a role in the fat body for glucose clearance . The regulation of GSIS by dHNF4 is consistent with the long-known central contribution of pancreatic β-cells to the pathophysiology of MODY1 ( Fajans and Bell , 2011 ) . Similarly , several MODY-associated genes , including GCK , HNF1A and HNF1B , are important for maintaining normal hepatic function . These distinct tissue-specific contributions to glycemic control may explain why single-tissue Hnf4A mutants in mice do not fully recapitulate MODY1 phenotypes and predict that a combined deficiency for the receptor in both the liver and pancreatic β-cells of adults would produce a more accurate model of this disorder . We used metabolomics , RNA-seq , and ChIP-seq to provide initial insights into the molecular mechanisms by which dHNF4 exerts its effects on systemic metabolism . These studies revealed several downstream pathways , each of which is associated with maintaining homeostasis and , when disrupted , can contribute to diabetes . These include genes identified in our previous study of dHNF4 in larvae that act in lipid metabolism and fatty acid β-oxidation , analogous to the role of Hnf4A in the mouse liver to maintain normal levels of stored and circulating lipids ( Hayhurst et al . , 2001; Palanker et al . , 2009 ) . Extensive studies have linked defects in lipid metabolism with impaired β-cell function and peripheral glucose uptake and clearance , suggesting that these pathways contribute to the diabetic phenotypes of dHNF4 mutants ( Prentki et al . , 2013; Qatanani and Lazar , 2007 ) . An example of this is pudgy , which is expressed at reduced levels in dHNF4 mutants and encodes an acyl-CoA synthetase that is required for fatty acid oxidation ( Figure 3A ) ( Xu et al . , 2012 ) . Interestingly , pudgy mutants have elevated triglycerides , reduced glycogen , and increased circulating sugars , similar to dHNF4 mutants , suggesting that this gene is a critical downstream target of the receptor . It is important to note , however , that our metabolomic , RNA-seq , and ChIP-seq studies were conducted on extracts from whole animals rather than individual tissues . As a result , some of our findings may reflect compensatory responses between tissues , and some tissue-specific changes in gene expression or metabolite levels may not be detected by our approach . Further studies using samples from dissected tissues would likely provide a more complete understanding of the mechanisms by which dHNF4 maintains systemic physiology . Notably , the Drosophila GCK homolog encoded by Hex-C is expressed at reduced levels in dHNF4 mutants ( Figure 3A ) . The central role of GCK in glucose sensing by pancreatic β-cells as well as glucose clearance by the liver places it as an important regulator of systemic glycemic control . Our functional data supports these associations by showing that Hex-C is required in the fat body for proper circulating glucose levels , analogous to the role of GCK in mammalian liver ( Figure 4B ) ( Postic et al . , 1999 ) . Unlike mice lacking GCK in the β-cells , however , we do not see an effect on glucose homeostasis when Hex-C is targeted by RNAi in the IPCs . This is possibly due to the presence of a second GCK homolog in Drosophila , Hex-A , which could act alone or redundantly with Hex-C to mediate glucose sensing by the IPCs . In mammals , GCK expression is differentially regulated between hepatocytes and β-cells through the use of two distinct promoters , and studies in rats have demonstrated a direct role for HNF4A in promoting GCK expression in the liver ( Roth et al . , 2002 ) . Our findings suggest that this relationship has been conserved through evolution . In addition , the association between GCK mutations and MODY2 raise the interesting possibility that defects in liver GCK activity may contribute to the pathophysiology of both MODY1 and MODY2 . Interestingly , gene ontology analysis indicates that the up-regulated genes in dHNF4 mutants correspond to the innate immune response pathways in Drosophila ( Supplementary file 2 ) . This response parallels that seen in mice lacking Hnf4A function in enterocytes , which display intestinal inflammation accompanied by increased sensitivity to DSS-induced colitis and increased permeability of the intestinal epithelium , similar to humans with inflammatory bowel disease ( Ahn et al . , 2008; Babeu and Boudreau , 2014; Cattin et al . , 2009 ) . Disruption of Hnf4A expression in Caco-2 cells using shRNA resulted in changes in the expression of genes that act in oxidative stress responses , detoxification pathways , and inflammatory responses , similar to the effect we see in dHNF4 mutants ( Marcil et al . , 2010 ) . Moreover , mutations in human HNF4A are associated with chronic intestinal inflammation , irritable bowel disease , ulcerative colitis , and Crohn’s disease , suggesting that these functions are conserved through evolution ( UK IBD Genetics Consortium et al . , 2009; Marcil et al . , 2012; van Sommeren et al . , 2011 ) . Taken together , these results support the hypothesis that dHNF4 plays an important role in suppressing an inflammatory response in the intestine . Future studies are required to test this hypothesis in Drosophila . In addition , further work is required to better define the regulatory functions of HNF4 that are shared between Drosophila and mammals . Although our work described here suggests that key activities for this receptor have been conserved in flies and mammals , corresponding to the roles of HNF4 in the IPCs ( β-cells ) for GSIS , fat body ( liver ) for lipid metabolism and glucose clearance , and intestine to suppress inflammation , there are likely to be divergent roles as well . One example of this is the embryonic lethality of Hnf4A mutant mice , which is clearly distinct from the early adult lethality reported here for dHNF4 mutants . Further studies are required to dissect the degree to which the regulatory functions of this receptor have been conserved through evolution . It is also important to note that mammalian Hnf4A plays a role in hepatocyte differentiation and proliferation in addition to its roles in metabolism ( Bonzo et al . , 2012; Li et al . , 2000 ) . This raises the possibility that early developmental roles for dHNF4 could impact the phenotypes we report here in adults . Indeed , all of our studies involve zygotic dHNF4 null mutants that lack function throughout development . In an effort to address this possibility and distinguish developmental from adult-specific functions , we are constructing a conditional dHNF4 mutant allele using CRISPR/Cas9 technology . Future studies using this mutation should allow us to conduct a detailed phenotypic analysis of this receptor at different stages of Drosophila development . It is also interesting to speculate that our functional studies of dHNF4 uncover more widespread roles for MODY-associated genes in glycemic control , in addition to the link with MODY2 described above . HNF1A and HNF1B , which are associated with MODY3 and MODY5 , respectively , act together with HNF4A in an autoregulatory circuit in an overlapping set of tissues , with HNF4A proposed to be the most upstream regulator of this circuit ( Boj et al . , 2001; Nagaki and Moriwaki , 2008 ) . The observation that Drosophila do not have identifiable homologs for HNF1A and HNF1B raises the interesting possibility that dHNF4 alone replaces this autoregulatory circuit in more primitive organisms . The related phenotype of these disorders is further emphasized by cases of MODY3 that are caused by mutation of an HNF4A binding site within the HNF1A promoter ( Gragnoli et al . , 1997 ) . Consistent with this link , MODY1 , MODY3 and MODY5 display similar features of disease complication and progression , and studies of HNF1A and HNF4A in INS-1 cells have implicated roles for these transcription factors in promoting mitochondrial metabolism in β-cells ( Wang et al . , 2002 ) . In line with this , mitochondrial diabetes is clearly age progressive , as are MODY1 , 3 , and 5 , but not MODY2 , which represents a more mild form of this disorder . Furthermore , the severity and progression of MODY3 is significantly enhanced when patients carry an additional mutation in either HNF4A or mtDNA ( Forlani et al . , 2010 ) . Overall , these observations are consistent with the well-established multifactorial nature of diabetes , with multiple distinct metabolic insults contributing to disease onset . Our RNA-seq analysis supports a role for dHNF4 in coordinating mitochondrial and nuclear gene expression ( Supplementary file 1 and Figure 3A ) . This is represented by the reduced expression of transcripts encoded by the mitochondrial genome , along with effects on nuclear-encoded genes that act in mitochondria . In addition , ChIP-seq revealed that several of the nuclear-encoded genes are direct targets of the receptor . Mitochondrial defects have well-established links to diabetes-onset , with mutations in mtDNA causing maternally-inherited diabetes and mitochondrial OXPHOS playing a central role in both GSIS and peripheral glucose clearance ( Sivitz and Yorek , 2010 ) . Consistent with this , our functional studies indicate that dHNF4 is required to maintain normal mitochondrial function and that defects in this process contribute to the diabetic phenotypes in dHNF4 mutants . It is important to note that the number of direct targets for dHNF4 in the nucleus is difficult to predict with our current dataset . A relatively low signal-to-noise ratio in our ChIP-seq experiment allowed us to identify only 37 nuclear-encoded genes as high confidence targets by fitting the criteria of proximal dHNF4 binding along with reduced expression in dHNF4 mutants ( Figure 3B and Supplementary file 3 ) . Future ChIP-seq studies will allow us to expand this dataset to gain a more comprehensive understanding of the scope of the dHNF4 regulatory circuit and may also reveal tissue-restricted targets that are more difficult to detect . Nonetheless , almost all of the genes identified as direct targets for dHNF4 regulation correspond to genes involved in mitochondrial metabolism , including the TCA cycle , OXPHOS , and lipid catabolism , demonstrating that this receptor has a direct impact on these critical downstream pathways that influence glucose homeostasis ( Figure 3—figure supplement 3 , Supplementary file 3 ) . An unexpected and significant discovery in our studies is that dHNF4 is required for mitochondrial gene expression and function . Several lines of evidence support the model that dHNF4 exerts this effect through direct regulation of mitochondrial transcription , although a number of additional experiments are required to draw firm conclusions on this regulatory connection . First , most of the 13 protein-coding genes in mtDNA are underexpressed in dHNF4 mutants ( Figure 3A , Supplementary file 1 ) . Our lab and others have conducted RNA-seq studies of Drosophila nuclear transcription factor mutants and , to our knowledge , similar effects on mitochondrial gene expression have not been reported previously . Second , dHNF4 protein is abundantly bound to the control region of the mitochondrial genome , representing the fifth strongest enrichment peak in our ChIP-seq dataset ( Figure 3C ) . Although the promoters in Drosophila mtDNA have not yet been identified , the site bound by dHNF4 corresponds to a predicted promoter region for Drosophila mitochondrial transcription and coincides with the location of the major divergent promoters in human mtDNA ( Garesse and Kaguni , 2005; Roberti et al . , 2006 ) . It is unlikely that the abundance of mtDNA relative to nuclear DNA had an effect on our ChIP-seq peak calling because the MACS2 platform used for our analysis accounts for local differences in read depth across the genome ( including the abundance of mtDNA ) . In addition , although the D-loop in mtDNA has been proposed to contribute to possible false-positive ChIP-seq peaks in mammalian studies ( Marinov et al . , 2014 ) , the D-loop structure is not present in Drosophila mtDNA ( Rubenstein et al . , 1977 ) . Nonetheless , additional experiments are required before we can conclude that this apparent binding is of regulatory significance for mitochondrial function . Third , the effects on mitochondrial gene expression do not appear to be due to reduced mitochondrial number in dHNF4 mutants ( Figure 3—figure supplement 2A ) . This is consistent with the normal expression of mt:Cyt-b in dHNF4 mutants ( Figure 3A ) , which has a predicted upstream promoter that drives expression of the mt:Cyt-b and mt:ND6 operon ( although we could not detect mt:ND6 RNA in our northern blot studies ) ( Berthier et al . , 1986; Roberti et al . , 2006 ) . Fourth , immunostaining for dHNF4 shows cytoplasmic protein that overlaps with the mitochondrial marker ATP5A , in addition to its expected nuclear localization ( Figure 3—figure supplement 2B ) . Some of the cytoplasmic staining , however , clearly fails to overlap with the mitochondrial marker , making it difficult to draw firm conclusions from this experiment . Multiple efforts to expand on this question biochemically with subcellular fractionation studies have been complicated by abundant background proteins that co-migrate with the receptor in mitochondrial extracts . We are currently developing new reagents to detect the relatively low levels of endogenous dHNF4 protein in mitochondria , including use of the CRISPR/Cas9 system for the addition of specific epitope tags to the endogenous dHNF4 locus . Finally , we observe multiple hallmarks of mitochondrial dysfunction , including elevated pyruvate and lactate , specific alterations in TCA cycle metabolites , reduced mitochondrial membrane potential , reduced levels of ATP , and fragmented mitochondrial morphology . These phenotypes are consistent with the reduced expression of key genes involved in mitochondrial OXPHOS ( Figure 3A and Supplementary file 1 ) , and studies showing that decreased mitochondrial membrane potential and ATP production are commonly associated with mitochondrial fragmentation ( Mishra and Chan , 2014; Toyama et al . , 2016 ) . Although unexpected , our proposal that dHNF4 may directly regulate mitochondrial gene expression is not unprecedented . A number of nuclear transcription factors have been localized to mitochondria , including ATFS-1 , MEF2D , CREB , p53 , STAT3 , along with several nuclear receptors , including the estrogen receptor , glucocorticoid receptor , and the p43 isoform of the thyroid hormone receptor ( Leigh-Brown et al . , 2010; Nargund et al . , 2015; Szczepanek et al . , 2012 ) . The significance of these observations , however , remains largely unclear , with few studies demonstrating regulatory functions within mitochondria . In addition , these factors lack a canonical mitochondrial localization signal at their amino-terminus , leaving it unclear how they achieve their subcellular distribution ( Marinov et al . , 2014 ) . In contrast , one of the five mRNA isoforms encoded by dHNF4 , dHNF4-B , encodes a predicted mitochondrial localization signal in its 5’-specific exon , providing a molecular mechanism to explain the targeting of this nuclear receptor to this organelle . Efforts are currently underway to conduct a detailed functional analysis of dHNF4-B by using the CRISPR/Cas9 system to delete its unique 5’ exon , as well as establishing transgenic lines that express a tagged version of dHNF4-B under UAS control . Future studies using these reagents , along with our dHNF4 mutants , should allow us to dissect the nuclear and mitochondrial functions of this nuclear receptor and their respective contributions to systemic physiology . Finally , it is interesting to speculate whether the role for dHNF4 in mitochondria is conserved in mammals . A few papers have described the regulation of nuclear-encoded mitochondrial genes by HNF4A ( Rodriguez et al . , 1998; Wang et al . , 2000 ) . In addition , several studies have detected cytoplasmic Hnf4A by immunohistochemistry in tissue sections , including in postnatal pancreatic islets ( Miura et al . , 2006; Nammo et al . , 2008 ) and hepatocytes ( Bell and Michalopoulos , 2006; Soutoglou et al . , 2000; Sun et al . , 2007; Yanger et al . , 2013 ) . Moreover , the regulation of nuclear/cytoplasmic shuttling of HNF4A has been studied in cultured cells ( Soutoglou et al . , 2000 ) . The evolutionary conservation of the physiological functions of HNF4A , from flies to mammals , combined with these prior studies , argue that more effort should be directed at defining the subcellular distribution of HNF4A protein and its potential roles within mitochondria . Taken together with our studies in Drosophila , this work could provide new directions for understanding HNF4 function and MODY1 . Physiological studies by George Newport in 1836 noted that holometabolous insects reduce their respiration during metamorphosis leading to a characteristic “U-shaped curve” in oxygen consumption ( Needham , 1929; Newport , 1836 ) . Subsequent classical experiments in Lepidoptera , Bombyx , Rhodnius and Calliphora showed that this reduction in mitochondrial respiration during metamorphosis and dramatic rise in early adults is seen in multiple insect species , including Drosophila ( Bodine , 1925; Merkey et al . , 2011 ) . Consistent with this , the activity of oxidative enzyme systems and the levels of ATP also follow a “U-shaped curve” during development as the animal transitions from a non-feeding pupa to a motile and reproductively active adult fly ( Agrell , 1953 ) . Although first described over 150 years ago , the regulation of this developmental increase in mitochondrial activity has remained undefined . Here we show that this temporal switch is dependent , at least in part , on the dHNF4 nuclear receptor . The levels of dHNF4 expression increase dramatically at the onset of adulthood , accompanied by the expression of downstream genes that act in glucose homeostasis and mitochondrial OXPHOS ( Figure 6E ) . This coordinate transcriptional switch is reduced in dHNF4 mutants , indicating that the receptor plays a key role in this transition . Importantly , the timing of this program correlates with the onset of dHNF4 mutant phenotypes in young adults , including sugar-dependent lethality , hyperglycemia , and defects in GSIS , indicating that the upregulation of dHNF4 expression in adults is of functional significance . It should also be noted , however , that dHNF4 target genes are still induced at the onset of adulthood in dHNF4 mutants , albeit at lower levels , indicating that other regulators contribute to this switch in metabolic state ( Figure 6E ) . Nonetheless , the timing of the induction of dHNF4 and its target genes in early adults , and its role in promoting OXPHOS , suggest that this receptor contributes to the end of the “U-shaped curve” and directs a systemic transcriptional switch that establishes an optimized metabolic state to support the energetic demands of adult life . Interestingly , a similar metabolic transition towards OXPHOS was recently described in Drosophila neuroblast differentiation , mediated by another nuclear receptor , EcR ( Homem et al . , 2014 ) . Although this occurs during early stages of pupal development , prior to the dHNF4-mediated transition at the onset of adulthood , the genes involved in this switch show a high degree of overlap with dHNF4 target genes that act in mitochondria , including ETFB , components of Complex IV , pyruvate carboxylase , and members of the α-ketoglutarate dehydrogenase complex . This raises the possibility that dHNF4 may contribute to this change in neuroblast metabolic state and play a more general role in supporting tissue differentiation by promoting OXPHOS . To our knowledge , only one other developmentally coordinated switch in systemic metabolic state has been reported in Drosophila and , intriguingly , it is also regulated by a nuclear receptor . Drosophila Estrogen-Related Receptor ( dERR ) acts in mid-embryogenesis to directly induce genes that function in biosynthetic pathways related to the Warburg effect , by which cancer cells use glucose to support rapid proliferation ( Tennessen et al . , 2011; Tennessen et al . , 2014b ) . This switch toward aerobic glycolysis favors lactate production and flux through biosynthetic pathways over mitochondrial OXPHOS , supporting the ~200-fold increase in mass that occurs during larval development . Taken together with our work on dHNF4 , these studies define a role for nuclear receptors in directing temporal switches in metabolic state that meet the changing physiological needs of different stages in development . Further studies should allow us to better define these regulatory pathways as well as determine how broadly nuclear receptors exert this role in coupling developmental progression with systemic metabolism . Although little is known about the links between development and metabolism , it is likely that coordinated switches in metabolic state are not unique to Drosophila , but rather occur in all higher organisms in order to meet the distinct metabolic needs of an animal as it progresses through its life cycle . Indeed , a developmental switch towards OXPHOS in coordination with the cessation of growth and differentiation appears to be a conserved feature of animal development . Moreover , as has been shown for cardiac hypertrophy , a failure to coordinate metabolic state with developmental context can have an important influence on human disease ( Lehman and Kelly , 2002 ) . In addition to promoting a transition toward systemic oxidative metabolism in adult flies , dHNF4 also contributes to a switch in IPC physiology that supports GSIS . dHNF4 is not expressed in larval IPCs , but is specifically induced in these cells at adulthood ( Figure 6A , B ) . Similarly , the fly homologs of the mammalian ATP-sensitive potassium channel subunits , Sur1 and Kir6 , which link OXPHOS and ATP production to GSIS , are not expressed in the larval IPCs but are expressed during the adult stage ( Fridell et al . , 2009; Kim and Rulifson , 2004 ) . They also appear to be active at this stage as cultured IPCs from adult flies undergo calcium influx and membrane depolarization upon exposure to glucose or the anti-diabetic sulfonylurea drug glibenclamide ( Kreneisz et al . , 2010 ) . In addition , reduction of the mitochondrial membrane potential in adult IPCs by ectopic expression of an uncoupling protein is sufficient to reduce IPC calcium influx , elevate whole-animal glucose levels , and reduce peripheral insulin signaling ( Fridell et al . , 2009 ) . This switch in IPC physiology is paralleled by a change in the nutritional signals that trigger DILP release . Amino acids , and not glucose , stimulate DILP2 secretion by larval IPCs ( Geminard et al . , 2009 ) . Rather , glucose is sensed by the corpora cardiaca in larvae , a distinct organ that secretes adipokinetic hormone , which acts like glucagon to maintain carbohydrate homeostasis during larval stages ( Kim and Rulifson , 2004; Lee and Park , 2004 ) . Interestingly , this can have an indirect effect on the larval IPCs , triggering DILP3 secretion in response to dietary carbohydrates ( Kim and Neufeld , 2015 ) . Adult IPCs , however , are responsive to glucose for DILP2 release ( Park et al . , 2014 ) ( Figure 5A , D ) . In addition , dHNF4 mutants on a normal diet maintain euglycemia during larval and early pupal stages , but display hyperglycemia at the onset of adulthood , paralleling their lethal phase on a normal diet ( Figure 6D ) . Taken together , these observations support the model that the IPCs change their physiological state during the larval-to-adult transition and that dHNF4 contributes to this transition toward GSIS . The observation that glucose is a major circulating sugar in adults , but not larvae , combined with its ability to stimulate DILP2 secretion from adult IPCs , establishes this stage as an experimental context for genetic studies of glucose homeostasis , GSIS , and diabetes . Functional characterization of these pathways in adult Drosophila will allow us to harness the power of model organism genetics to better understand the regulation of glucose homeostasis and the factors that contribute to diabetes .
All genetic studies used a transheterozygous combination of dHNF4 null alleles ( dHNF4Δ17/dHNF4Δ33 ) and genetically-matched controls that were transheterozygous for precise excisions of the EP2449 and KG08976 P-elements , as described previously ( Palanker et al . , 2009 ) . Sugar diets were made using 8% yeast , 1% agar , 0 . 05% MgSO4 , 0 . 05% CaCl2 , and either 3% , 9% or 15% dietary sugar ( 2:1 ratio of glucose to sucrose , percentages represent weight/final food volume . 10 ml/L tegosept and 6 ml/L propionic acid were added just prior to pouring ) . Fasting was achieved by using 1% agar as a medium . For adult studies , 8–10 day old males were selected for all studies unless otherwise indicated . The following GAL4 driver lines were used for tissue-specific expression experiments: Fat body: r4-GAL4 ( Lee and Park , 2004 ) , midgut: mex-GAL4 ( Phillips and Thomas , 2006 ) , IPCs: yw; UAS-Dicer2; dilp21 dilp2HF dilp2-GAL4 ( Park et al . , 2014 ) . RNAi lines used in this study include: UAS-dHNF4RNAi ( TRiP 29375 used in Figure 4A , Figure 4—figure supplement 1A , Figure 5D , E; VDRC 12692 used in Figure 4—figure supplement 1A–B , Figure 5—figure supplement 1 ) , UAS-mCherryRNAi ( TRiP 35785 ) , UAS-Cox5aRNAi ( TRiP 27548 ) , UAS-CIA30RNAi ( TRiP 55660 ) , UAS-ATPsynβRNAi ( TRiP 28056 ) , UAS-Scs-αRNAi ( TRiP 51807 ) , UAS-CG5599RNAi ( TRiP 32876 ) , and UAS-HexCRNAi ( TRiP 57404 used in Figure 4B–D , and VDRC 35337 and VDRC 35338 used in Figure 4—figure supplement 3 ) . The dHNF4 mutant lethal phase was assessed by raising 50–60 newly hatched first instar larvae in vials of standard laboratory media at 25˚C and scoring for survival through embryonic hatching , wandering third instar , puparium formation , eclosion , and survival through the first day of adulthood ( Figure 1A ) . Eclosion rates were scored in a similar manner for dHNF4 mutants and genetically matched controls raised on the 3% , 9% or 15% sugar diets . For adult survival studies on different diets ( Figure 1C ) , 50–60 newly hatched first instar larvae were placed in vials with the 3% sugar diet and raised until five days of adulthood . These mature adults were then transferred to fresh vials of 3% , 9% , or 15% dietary sugar with approximately 25 males and 25 females per vial at 25˚C and scored daily for lethality . Flies were transferred to fresh vials every 2–3 days . At least three replicate vials were analyzed per condition and each experiment was repeated at least three times with similar results . For the developmental time course northern blot , we collected 0–12 hr feeding third instar larvae , pupae at 24 , 48 or 72 hr after puparium formation , stage P12 males ( newly formed pharate adult with visible sex combs ) for 4 day pupae , and males at mid-eclosion from the pupal case ( to ensure that all dHNF4 mutants were alive ) ( n=15 animals per sample ) . Animals collected for developmental RNA and glycemia measurements ( Figure 6D–E ) were raised on the 15% sugar diet , except for the adults , which were reared on the 3% sugar diet and transferred to the 15% sugar diet after eclosion . Glycogen , trehalose , and free glucose levels were determined using the Hexokinase ( HK ) and/or Glucose Oxidase ( GO ) assay kits ( Sigma GAHK20 , GAGO20 ) as previously described , with approximately six biological replicates ( n=5 animals per sample ) assayed per condition ( Tennessen et al . , 2014a ) . Total protein levels were determined in parallel by Bradford assay to control for potential variations in sample homogenization and/or animal size . Hemolymph glucose measurements were determined by centrifuging 60–100 adult males in a prepared zymogen barrier column ( Zymo Research Corporation C1006 ) at 9000 g for 5 min at 4˚C , as described ( Park et al . , 2014 ) . The hemolymph flow-though was diluted 1:100 in 1xPBS prior to heat treatment , followed by analysis using the HK assay kit . Mature adult males were placed on 10% sucrose + 1% agar overnight prior to analysis . Six biological replicates per genotype ( n=5 males per sample ) were analyzed for ATP levels by using a luminescence assay kit ( Molecular Probes ATP kit; A22066 , ) as described ( Tennessen et al . , 2014a ) . Animals were raised on the 3% sugar diet until five days of adulthood and transferred to a fresh vial with either the 3% or 15% sugar diet for three days prior to collection . Twenty adult males per sample were snap-frozen in liquid nitrogen and prepared for analysis by gas chromatography-mass spectrometry ( GC/MS ) as described ( Tennessen et al . , 2014a ) . Data is presented from three independent experiments , each consisting of 5–6 biological replicates per condition . Sample preparation and GC/MS analysis were performed by the Metabolomics Core Research Facility at the University of Utah School of Medicine . Tissues were dissected in cold 1xPBS , fixed in 4% paraformaldehyde for 20-30 min at room temperature , and washed once quickly followed by three washes for 20 min each in PBS , 0 . 2% Triton-X100 . Samples were blocked using 5% normal goat serum for 1–2 hr at room temperature and incubated with primary antibodies for a minimum of 24 hr at 4˚C , washed , and incubated with secondary antibody at 4˚C for 24 hr . Images were acquired using an Olympus FV1000 confocal microscope and analyzed by Volocity software to generate Z-stack projections and overlay images . The following antibodies were used for immunofluorescence studies: rat anti-dHNF4 ( Palanker et al . , 2009 ) , rat anti-DILP2 ( a gift from P . Leopold ) , rabbit anti-GFP ( Molecular Probes A-6455 ) , chicken anti-GFP ( Abcam 13970 ) , and mouse anti-ATP5A ( Abcam 14748 ) . Dissected brains stained for DILP2 peptide were mounted in SlowFade® Gold ( Life Technologies ) and imaged using an Olympus FV1000 confocal microscope and 60X water-immersion objective . Scan settings were fixed to be identical for all captured images , with Z-stack limits set to encompass the entire depth of DILP2 fluorescence for each brain . Maximum projection Z-stacks were analyzed using Volocity software to calculate mean pixel intensity of DILP2 fluorescence within the IPCs ( selected by ROI tool ) for each set of IPCs analyzed . Total RNA was extracted from groups of 10–20 animals using TriPure isolation reagent ( Roche ) . RNA was fractionated by 1% formaldehyde gel electrophoresis at a constant voltage ( 70V ) for ~3 . 5 hr prior to transfer to a nylon membrane overnight . After transfer , RNA was UV cross-linked to the membrane and placed in hybridization buffer ( 5 mls formamide , 2 mls 10x PIPES buffer , pH 6 . 5 , 2 mls ddH20 , 1 ml 10% SDS , 100 μl sheared herring sperm ssDNA ) for two hours at 42˚C prior to the addition of radiolabeled probe . Probes were generated by Klenow-mediated random primer amplification of purified template DNA corresponding to approximately 100–500 bp of the transcript of interest , allowing incorporation of 32P radiolabeled-nucleotide ( dCTP , PerkinElmer ) . Probes were incubated with membranes overnight , and the hybridized membranes were washed and exposed to X-ray film at -80˚C using two intensifying screens . dHNF4 mutants and genetically-matched controls were reared on the 3% sugar diet for five days of adulthood and transferred to the 15% sugar diet for 3 days prior to extracting total RNA . Eight biological replicates per genotype ( n=20 males per sample ) were collected using TriPure isolation reagent ( Roche ) . Pairs of biological replicates were pooled to obtain four biological replicates for further purification using an RNeasy kit ( QIAGEN ) . RNA quality was subsequently analyzed using an Agilent Bioanalyzer RNA 6000 . PolyA-selected RNAs from each biological replicate were then assembled into barcoded libraries and pooled into a single-flow cell lane for Illumina HighSeq2000 50-cycle single-read sequencing , which produced ≥21 . 9 million reads per sample . Standard replicate RNA-seq analysis was performed using USeq and DESeq analysis packages with alignment to the Drosophila melanogaster dm3 genome assembly . Transcripts meeting a cutoff of ≥1 . 5 fold difference in mRNA abundance and FDR ≤1% were considered as differentially expressed genes . RNA quality control , library preparation , sequencing , and data analysis were performed at the University of Utah High Throughput Genomics and Bioinformatics Core Facilities . Although mt:Cyt-b and mt:ND6 are included among the down-regulated genes in our RNA-Seq dataset ( Supplementary file 1 ) , these represent false positives as demonstrated by subsequent repeated northern blot hybridization studies for mt:Cyt-b ( Figure 3A ) . We are also unable to detect mt:ND6 mRNA in either mutant or control flies , consistent with a previous report ( Berthier et al . , 1986 ) . Chromatin isolation and immunoprecipitation were performed as described ( Schwartz et al . , 2003 ) . w1118 flies were reared on standard cornmeal-based lab medium and 1–1 . 5 g of mature adults were homogenized in ice-cold buffer A ( 0 . 3 M sucrose , 2 mM MgOAc , 3 mM CaCl2 , 10 mM Tris-Cl [pH 8] , 0 . 3% Triton X-100 , 0 . 5 mM dithiothreitol , 1 Roche protease inhibitor tablet per 10 ml ) for 1 . 5 min using a Brinkmann Homogenizer Polytron PT 10/35 . The homogenate was filtered through a layer of 100 μm-pore nylon mesh into a pre-chilled glass-Teflon homogenizer , stroked on ice 35 times using a B pestle , and filtered through two layers of 40-μm-pore nylon mesh prior to adding one volume of warm cross-linking buffer ( 0 . 1 M NaCl , 1 mM EDTA , 0 . 5 mM EGTA , 50 mM Tris [pH 8] , pre-warmed to 40°C ) to bring the sample to room temperature for crosslinking ( 0 . 3% formaldehyde for 3 min ) . 2 . 5 M glycine was added to a final concentration of 125 mM to stop the crosslinking after 3 min . The mixture was pelleted for resuspension into 10 ml of RIPA buffer ( 140 mM NaCl , 10 mM Tris-Cl [pH 8 . 0] , 1 mM EDTA , 1% Triton , 0 . 3% SDS , 0 . 1% sodium deoxycholate , and Roche protease inhibitor cocktail ) for sonication . The sonicated material was centrifuged at 20 , 000 g , and the supernatant was distributed into 500 μl aliquots that were snap frozen in liquid nitrogen . To each aliquot , 1 ml of cold RIPA buffer ( without SDS ) was added prior to removing 150 μl for an input control , and then 5 µl of polyclonal affinity-purified anti-dHNF4 antibody was added to each sample for immunoprecipitation overnight at 4˚C ( Palanker et al . , 2009 ) . 50 μl of prepared Protein G Dynabeads ( Life Technologies ) were added to each sample and incubated for 4 hr at 4˚C prior to wash and elution using a magnetic stand . Washes were performed for 3 min each at 4˚C with 1 ml of the following ice-cold solutions: three times in low salt wash buffer ( 0 . 1% SDS , 1% Triton X-100 , 2 mM EDTA , 20 mM Tris-Cl [pH 8 . 0] , 150 mM NaCl ) , one time in high salt wash buffer ( 0 . 1% SDS , 1% Triton X-100 , 2 mM EDTA , 20 mM Tris-Cl [pH 8 . 0] , 500 mM NaCl ) , once in LiCl wash buffer ( 0 . 25 M LiCl , 1% NP-40 , 1% deoxycholate , 1 mM EDTA , 10 mM Tris-Cl [pH 8 . 0] ) , and twice with Tris-EDTA . Protein-DNA complexes were eluted in 1 . 5 ml DNA-low bind tubes ( Eppendorf ) using two 15-minute washes with 250 μl elution buffer ( 1% SDS , 0 . 1 M NaHCO3 ) . NaCl was added to a final concentration of 0 . 2 M to reverse the crosslinks of IP and input control samples , followed by an overnight incubation at 65°C . dHNF4-bound DNA was purified by using PCR-purification columns ( Qiagen ) , and pooled to acquire four replicates of dHNF4-IP chromatin and input controls . Barcoded libraries for dHNF4-IP and input control samples were generated by the University of Utah High Throughput Genomics core facility and sequenced in a single lane Illumina HiSeq 50-cycle single-read sequencing . Data analysis was performed by the Bioinformatics Core at the University of Utah School of Medicine using USeq ScanSeqs ( FDR80 ) as well as Model-based Analysis for ChIP-seq 2 . 0 ( MACS2 ) ( Zhang et al . , 2008 ) with an FDR cutoff of 1% ( FDR20 ) , identifying 68 enrichment regions . Nearest neighboring genes were compiled using USeq FindNeighboringGenes and UCSC dm3 EnsGenes gene tables and were compared to our RNA-seq dataset to identify proximal genes that are also misexpressed in dHNF4 mutants , highlighting these as direct targets of dHNF4 . Samples were homogenized in Laemmli sample buffer with protease and phosphatase inhibitor cocktails , resolved by SDS-PAGE ( 10% acrylamide ) , transferred to PVDF membrane overnight at 4˚C , and blocked with 5% BSA prior to immunoblotting . Antibodies used for western blots include rabbit anti-pAKT ( Cell Signaling #4054 ) , rabbit anti-AKT ( Cell Signaling #4691 ) , mouse anti-SDHA ( Abcam 137756 ) , rabbit anti-SDHB ( a generous gift from D . Winge , generated against yeast SDH2/SDHB ) , mouse anti-ATP5A ( Abcam 14748 ) , and mouse anti-βTubulin ( Chemicon MAB380 ) . Transgenic lines expressing HA-FLAG tagged DILP2 ( DILP2-HF ) peptide were used for ELISA assays , essentially as described ( Park et al . , 2014 ) . The indicated UAS-RNAi line or w1118 controls ( Figure 5D and Figure 5—figure supplement 1 ) were crossed to yw; UAS-Dcr2; dilp21 DILP2HF dilp2-GAL4 flies . Adult male progeny were either fed ad libitum ( Figure 5—figure supplement 1 ) or fasted overnight and then re-fed the 15% sugar diet for 45 min prior to analysis ( Figure 5D ) . The posterior end of the abdomen was removed using micro-dissection scissors and placed in 60 µl of PBS , using 10 males per biological replicate . Tubes were vortexed for 20 min , after which 50 μl of supernatant was collected in a fresh tube . Samples were homogenized in 600 μl of PBS , 1% Triton X-100 . HA-FLAG peptide standards from 0 , 20 , 40 80 , 160 , 320 and 640 pg/ml were generated for a linear standard curve . 50 μl of circulating DILP2HF , total DILP2HF , or peptide standards were added to 50 μl of anti-HA-peroxidase ( 5 ng/ml , Roche 3F10 ) in PBS ( with or without 1% Triton X-100 ) and then pipetted into a 96-well ELISA plate ( Thermo Scientific MaxiSorp Immulon 4 HBX , Cat# 3855 ) coated with mouse anti-FLAG antibody ( Sigma F1804 , M2 monoclonal ) . Samples were incubated at 4˚C overnight and washed 6 times with PBS , 0 . 2% Tween-20 . 100 μl of 1-Step Ultra TMB ELISA Substrate ( Thermo Scientific 34029 ) was added to the plate and incubated at room temperature for 30 min . 100 μl of 2M sulfuric acid was added to stop the reaction and the absorbance was measured immediately at 450 nm . Circulating DILP2HF ( pg/fly ) versus total remaining peptide was calculated to determine the percent secretion relative to controls ( n≥4 biological replicates per condition ) . Virgin female flies ( yw , hsFLP , UAS-GFP; tub-GAL80 , FRT40A;tub-GAL4 ) were crossed to w;HNF4∆33/CyO twi>GFP males for MARCM analysis . Flies were reared on the 3% sugar diet and heat treated at 37°C for one hour as pharate adults , and again as newly-eclosed adults . Adults were maintained on the 15% sugar diet for 8–10 days prior to analysis . Adult flies were dissected in room temperature Schneider’s medium ( ThermoFisher 21720024 ) with microdissection scissors to separate the abdomen from the thorax . The intestine was removed and the abdominal cavity was cut open to allow maximal contact of fat body with the TMRE solution ( ThermoFisher T669 ) . Dissected abdomens were placed in freshly-prepared solution of 20 nM TMRE in Schneider’s medium and incubated in the dark for 5–7 min . Samples were then washed briefly twice in Schneider’s medium , mounted on glass slides in fresh Schneider’s medium , and immediately imaged on an Olympus FV1000 confocal microscope with a 60x water objective . DHE staining was performed as previously described ( Owusu-Ansah and Banerjee , 2009 ) . Briefly , adult abdomens were dissected as described for TMRE staining , but instead placed in freshly prepared 30 μM DHE ( ThermoFisher D11347 ) in Schneider’s medium and incubated in the dark for 5–7 min . Samples were washed rapidly three times with Schneider’s medium , followed by fixation in 4% formaldehyde in 1xPBS for seven minutes . Samples were then washed twice in 1xPBS , mounted in SlowFade Gold ( ThermoFisher S36936 ) , and imaged on an Olympus FV1000 confocal microscope with a 20X objective . GraphPad PRISM 6 software was used to plot data and perform statistical analysis . Pairwise comparison P values were calculated using a two-tailed Student’s t-test , multiple comparison P values were calculated using one-way ANOVA with Šídák multiple test correction ( except Figure 4B , C which included Dunnett’s correction ) . Error bars are ± 1xSEM unless otherwise noted . Box plots display the full range of data ( error bars ) , the 25–75th quartiles ( box ) , and the median ( midline ) . | Diabetes is a complex disease that is caused by a combination of factors , including the person’s habits and environment , as well as their genetic make-up . However , there are some rare forms of diabetes that are caused simply by mutations in single genes and are directly inherited . For example , it has been known for twenty years that a type of diabetes called “Maturity Onset Diabetes of the Young type 1” ( or MODY1 for short ) occurs when a gene called HNF4 is mutated or deleted . The symptoms of MODY1 usually appear during early adulthood and include abnormally high levels of sugar in the blood , as well as the pancreas not being able to release the hormone insulin properly in response to these sugars . Previous studies in mice have tried to understand how losing the HNF4 gene leads to MODY1 . However , these mouse models did not fully recreate the symptoms of this disorder and the precise role of HNF4 in preventing diabetes remains unclear . Barry and Thummel have now used the fruit fly , because it is a model organism with simple genetics , to help shed light on this question . Furthermore , flies and mammals use many of the same pathways to control metabolism , making the fly a good model for the disease in humans . Barry and Thummel deleted the HNF4 gene in fruit flies and observed that the flies had all the symptoms that are typical in people with MODY1 . These symptoms included high sugar levels and decreased production of insulin-like hormones . The experiments also showed that HNF4 normally supports the proper expression of another gene called Hex-C; this gene encodes a protein that senses how much sugar is available and helps to keep the amount of sugar circulating the body within normal levels . Barry and Thummel went on to discover that the HNF4 gene is required for the expression of some genes in structures called mitochondria , which provide most of the energy used by animal cells . Lastly , the HNF4 gene became more active as the flies matured , and appeared to help the metabolism of a developing fruit fly transition towards that of an adult . Together these findings show that HNF4 protects against MODY1 by influencing several components of sugar metabolism in fruit flies . In the future , more studies are needed to understand how exactly HNF4 acts in mitochondria and to explore if similar results are seen in mammals . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"developmental",
"biology"
] | 2016 | The Drosophila HNF4 nuclear receptor promotes glucose-stimulated insulin secretion and mitochondrial function in adults |
The molecular mechanisms underlying the trade-off between plant innate immunity and steroid-mediated growth are controversial . Here , we report that activation of the transcription factor BZR1 is required and sufficient for suppression of immune signaling by brassinosteroids ( BR ) . BZR1 induces the expression of several WRKY transcription factors that negatively control early immune responses . In addition , BZR1 associates with WRKY40 to mediate the antagonism between BR and immune signaling . We reveal that BZR1-mediated inhibition of immunity is particularly relevant when plant fast growth is required , such as during etiolation . Thus , BZR1 acts as an important regulator mediating the trade-off between growth and immunity upon integration of environmental cues .
The trade-off between plant growth and immunity needs to be finely regulated to ensure proper allocation of resources in an efficient and timely manner upon effective integration of environmental cues ( Pieterse et al . , 2012 ) . A key aspect of plant immunity is the perception of pathogen-associated molecular patterns ( PAMPs ) by surface-localized pattern-recognition receptors ( PRRs ) , leading to PAMP-triggered immunity ( PTI ) ( Dodds and Rathjen , 2010 ) . PRRs of the leucine-rich repeat receptor kinases ( LRR-RKs ) class rely on the regulatory LRR-RK BAK1 ( BRASSINOSTEROID INSENSITIVE 1-ASSOCIATED KINASE 1 ) for signaling ( Monaghan and Zipfel , 2012 ) ; that is the case of FLS2 ( FLAGELLIN SENSITIVE 2 ) and EFR ( EF-TU RECEPTOR ) , which perceive bacterial flagellin ( or the active peptide flg22 ) and EF-Tu ( or the active peptide elf18 ) respectively . BAK1 also interacts with the LRR-RK BRI1 ( BRASSINOSTEROID INSENSITIVE 1 ) , the main receptor for the growth-promoting steroid hormones brassinosteroids ( BR ) , and is a positive regulator of BR-mediated growth ( Kim and Wang , 2010 ) . Hence , a crosstalk between the BR- and PAMP-triggered signaling pathways resulting from competition for BAK1 was hypothesized . While a unidirectional antagonism between BR and PTI signaling has been recently described in Arabidopsis ( Albrecht et al . , 2012; Belkhadir et al . , 2012 ) , the exact underlying mechanisms are still controversial . Activation of the BR signaling pathway via either transgenic overexpression of BRI1 or the BR biosynthetic gene DWF4 or expression of the activated BRI1 allele BRI1sud suppresses PTI outputs ( Belkhadir et al . , 2012 ) . One such output , the PAMP-triggered callose deposition , could be restored by over-expression of BAK1-HA , suggesting that BAK1 is a limiting factor ( Belkhadir et al . , 2012 ) . However , exogenous BR treatment of wild-type plants does not affect the FLS2-BAK1 complex formation upon FLS2 activation , while it results in decreased PTI responses ( Albrecht et al . , 2012 ) .
In order to clarify the role of BAK1 in the BR-PTI crosstalk , we investigated FLS2-BAK1 complex formation in the transgenic Arabidopsis lines overexpressing BRI1 or DWF4 or expressing BRI1sud ( Belkhadir et al . , 2012 ) . Upon treatment with flg22 , FLS2 associated normally with BAK1 in these transgenic plants , and neither FLS2 nor BAK1 accumulation was altered ( Figure 1—figure supplement 1A ) . Moreover , these plants displayed a weaker reactive oxygen species ( ROS ) burst in response to chitin ( Figure 1—figure supplement 1B ) , whose signaling pathway is BAK1-independent ( Shan et al . , 2008; Ranf et al . , 2011 ) . This result is consistent with the previous finding that exogenous BR treatment can also inhibit the chitin-induced ROS burst ( Albrecht et al . , 2012 ) . BAK1-HA is not fully functional in BR signaling and exerts a dominant-negative effect on the endogenous BAK1 ( Figure 1—figure supplement 1C ) , which may explain that introduction of the BAK1-HA transgene can override the suppression of immunity triggered by overexpression of BRI1 ( Belkhadir et al . , 2012 ) ; BAK1-HA does not exert such a dominant negative effect , however , on PTI signaling ( Figure 1—figure supplement 1D ) . Taken together , these results indicate that the BR-mediated suppression of PTI is triggered independently of a competition between BRI1 and PRRs for BAK1 . We sought to determine at which level of the BR signaling pathway the antagonism initiates . After BR perception by BRI1 and activation of the BRI1-BAK1 complex , the BR signal transduction cascade includes inactivation of BIN2 ( BR INSENSITIVE 2 ) and BIN2-like kinases , a family of GSK3-like kinases acting as negative regulators of the pathway ( Vert and Chory , 2006 ) . This leads to dephosphorylation of BZR1 ( BRASSINAZOLE RESISTANT 1 ) and BES1/BZR2 ( BRI1-EMS-SUPPRESSOR 1/BRASSINAZOLE RESISTANT 2 ) , two bHLH transcription factors acting as major regulators of BR-induced transcriptional changes , which then become active ( Wang et al . , 2002; Yin et al . , 2002 ) . Treatment with the chemicals LiCl and bikinin , which inhibit GSK3-like kinases ( De Rybel et al . , 2009; Yan et al . , 2009 ) , resulted in impaired flg22-triggered ROS burst ( Figure 1A , B ) , as observed upon genetic or ligand-induced activation of the BR pathway . Furthermore , a triple mutant in BIN2 and the two closest related GSK3-like kinases , BIL1 ( BIN2-LIKE 1 ) and BIL2 ( triple GSK3 mutant; Vert and Chory , 2006 ) , shows a similar impairment in response to either flg22 or chitin ( Figure 1C ) . Interestingly , in spite of regulating MAPKs involved in stomata development ( Kim et al . , 2012; Khan et al . , 2013 ) , neither BR treatment nor loss of function of BIN2 affect flg22-triggered MAPK activation ( Figure 1—figure supplement 2 ) , contrary to what has been recently suggested ( Choudhary et al . , 2012; Zhu et al . , 2013 ) . These results indicate that the BR-PTI crosstalk occurs downstream of BIN2 . 10 . 7554/eLife . 00983 . 003Figure 1 . Activation of BZR1 is sufficient to inhibit the PAMP-triggered ROS burst . ( A ) and ( B ) Flg22-triggered ROS burst after LiCl ( A ) or bikinin ( B ) treatment . Leaf discs were pre-treated with a 10 mM LiCl solution for 5 hr or with a 50 μM bikinin solution for 16 hr . ( C ) Flg22- or chitin-triggered ROS burst in Col-0 and the triple GSK3 mutant plants . ( D ) Flg22- or chitin-induced ROS burst in Col-0 and BZR1Δ plants . ( E ) Elf18-triggered ROS burst in bri1-5 and bri1-5/BZR1Δ plants . In all cases , bars represent SE of n = 28 rosette leaf discs . Asterisks indicate a statistically significant difference compared to the corresponding control ( mock treatment [A and B] , Col-0 [C and D] or bri1-5 [E] ) , according to a Student’s t-test ( p<0 . 05 ) . Leaf discs of four- to five-week-old Arabidopsis plants were used in these assays . Flg22 and elf18 were used at a concentration of 50 nM; chitin was used at a concentration of 1 mg/ml . Total photon counts were integrated between minutes two and 40 after PAMP treatment . All experiments were repeated at least three times with similar results . DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 00310 . 7554/eLife . 00983 . 004Figure 1—figure supplement 1 . The BR-mediated suppression of PTI can be triggered independently of a competition for BAK1 . ( A ) Co-IP of BAK1 and FLS2 in Col-0 , 35S:BRI1-cit , 35S:BRI1sud-cit and 35S:DWF4 seedlings treated with 1 μM flg22 for 10 min . Coimmunoprecipitated proteins were analyzed by using anti-FLS2 or anti-BAK1 antibodies . ( B ) Chitin-triggered ROS burst in Col-0 , 35S:BRI1-cit and 35S:DWF4 plants . Chitin was used at a concentration of 1 mg/ml . Total photon counts were integrated between minutes two and 40 after PAMP treatment . Bars represent SE of n = 28 rosette leaf discs . ( C ) Root length of seven-day-old Col-0 , BAK1p:BAK1-HA ( in Col-0 WT background ) or bak1-3 seedlings grown on medium supplemented or not with 10 nM BL . Bars represent SE of 12 ≤ n ≤ 17 . Asterisks indicate a statistically significant difference between treatments according to a Student's t-test ( p<0 . 05 ) . ( D ) Flg22-triggered ROS burst in Col-0 , BAK1p:BAK1-HA ( in Col-0 WT background ) or bak1-3 plants . Leaf discs of four- to five-week-old Arabidopsis plants were used in these assays . Flg22 was used at a concentration of 50 nM . Total photon counts were integrated between minutes two and 40 after PAMP treatment . Bars represent SE of n = 28 rosette leaf discs . All experiments were repeated at least twice with similar results . DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 00410 . 7554/eLife . 00983 . 005Figure 1—figure supplement 2 . PAMP-triggered MAPK activation is not impaired upon activation of BR signaling . ( A ) MAPK activation in Col-0 seedlings upon treatment with 1 μM flg22 ( F ) and/or epiBL ( B ) for 10 min ( with or without a 90-min or 5-hr BL pre-treatment ) . ( B ) Quantification of total MAPK activation in the experiment shown in ( A ) , measured as pixel intensity using ImageJ . Results are the average of two independent blots , corresponding to two independent biological replicated . ( C ) MAPK activation in Col-0 and Triple GSK3 mutant seedlings upon treatment with 1 μM flg22 . ( D ) Quantification of total MAPK activation in the experiment shown in ( C ) , measured as pixel intensity using ImageJ . Results are the average of two independent blots , corresponding to two independent biological replicated . Proteins were separated in a 10% acrylamide gel and transferred to PVDF membranes . Membranes were blotted with phospho-p44/42 MAPK ( Erk1/2; Thr202/Tyr204 ) rabbit monoclonal antibodies . Bars represent SD . DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 00510 . 7554/eLife . 00983 . 006Figure 1—figure supplement 3 . Activation of BZR1 , but not BES1 , is sufficient to inhibit the PAMP-triggered ROS burst . ( A ) Flg22- or chitin-triggered ROS burst in BZR1S173A plants . ( B ) Flg22-triggered ROS burst in BES1S171A plants . ( C ) Flg22-triggered ROS burst in mock- or bikinin-treated Col-0 or bri1-5 plants . Leaf discs were pre-treated with a 50 μM bikinin solution for 16 hr . ( D ) Flg22-triggered ROS burst in mock- or BRZ-treated Col-0 or BZR1Δ plants . Leaf discs were pre-treated with a 2 . 5 μM BRZ solution for 16 hr . In all cases , bars represent SE of 21 ≤ n ≤ 28 . Asterisks indicate a statistically significant difference compared to Col-0 ( A and B ) or mock-treatment ( C and D ) according to a Student's t-test ( p<0 . 05 ) . Flg22 was used at a concentration of 50 nM; chitin was used at a concentration of 1 mg/ml . Total photon counts were integrated between minutes two and 40 after PAMP treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 006 Transgenic expression of two different constitutively active versions of BZR1 , BZR1Δ ( Gampala et al . , 2007 ) and BZR1S173A ( Ryu et al . , 2007 ) , results in impaired flg22- or chitin-triggered ROS burst ( Figure 1D , Figure 1—figure supplement 3A ) . Consistent with previous results ( Albrecht et al . , 2012; Figure 1—figure supplements 1A and 2 ) , plants expressing BZR1Δ or BZR1S173A display normal FLS2-BAK1 complex formation and MAPK activation upon flg22 treatment ( Figure 2A , B , Figure 2—figure supplement 1A , B ) , but are impaired in PAMP-triggered marker gene expression , seedling growth inhibition ( SGI ) ( Figure 2C–E ) and induced resistance to P . syringae pv . tomato ( Pto ) DC3000 ( Figure 2F , Figure 2—figure supplement 1C ) , and are more susceptible to the non-host strain Pseudomonas syringae pv . cilantro ( Pci ) 0788-9 ( Lewis et al . , 2010 ) ( Figure 2G ) . Notably , transgenic expression of a constitutively active form of BES1 , BES1S171A ( Gampala et al . , 2007 ) , does not impact the flg22-triggered ROS burst ( Figure 1—figure supplement 3B ) . We then tested if activation of BZR1 is sufficient to inhibit PTI signaling . Induction of BR signaling by bikinin treatment still represses elf18-induced ROS burst in the BRI1 mutant bri1-5 ( we used elf18 in this experiment because bri1-5 is in the Ws-2 background , which is a natural fls2 mutant ) ( Figure 1—figure supplement 3C ) . bri1-5/BZR1Δ plants ( Gampala et al . , 2007 ) still exhibited reduced PAMP-triggered ROS burst ( Figure 1E ) , and treatment with the BR biosynthetic inhibitor brassinazole ( BRZ ) did not affect the BZR1Δ effect ( Figure 1—figure supplement 3D ) . Interestingly , BRZ treatment of wild-type Col-0 plants results in increased ROS production ( Figure 1—figure supplement 3D ) , which is consistent with the fact that BR inhibits PTI responses and suggests that endogenous concentrations of the hormone exert this effect . These results demonstrate that activation of BZR1 affects PTI signaling independently of BR perception or synthesis . 10 . 7554/eLife . 00983 . 007Figure 2 . Activation of BZR1 results in the suppression of specific PTI outputs . ( A ) Co-immunoprecipitation ( Co-IP ) of BAK1 and FLS2 in Col-0 and BZR1Δ seedlings after 10 min mock ( − ) or 1 μM flg22 ( + ) treatment . Proteins were separated in a 10% acrylamide gel and transferred to PVDF membranes . Membranes were blotted with anti-FLS2 or anti-BAK1 antibodies . ( B ) MAPK activation in Col-0 and BZR1Δ seedlings upon 1 μM flg22 treatment . Proteins were separated in a 10% acrylamide gel and transferred to PVDF membranes . Membranes were blotted with phospho-p44/42 MAPK ( Erk1/2; Thr202/Tyr204 ) rabbit monoclonal antibodies . CBB: Coomassie brilliant blue . ( C ) Marker gene ( At2g17700 and NHL10 ) expression in Col-0 and BZR1Δ seedlings after 1 hr mock ( − ) or 1 μM flg22 ( + ) treatment , as determined by qPCR . Bars represent SE of n = 3 . ( D ) and ( E ) Seedling growth inhibition of 10-day-old Col-0 or BZR1Δ seedlings induced by increasing concentrations of flg22 , as indicated . Scale bar ( D ) , 1 cm . Bars ( E ) represent SE of 8 ≤ n ≤ 16 . ( F ) Flg22-induced resistance to P . syringae pv . tomato DC3000 in Col-0 and BZR1Δ plants . Plants were pre-treated with 1 μM flg22 or water 24 hr prior to bacterial infiltration . Bars represent SE of n = 4 . This experiment was repeated seven times with similar results . ( G ) Susceptibility of Col-0 and BZR1Δ plants to P . syringae pv . cilantro 0788-9 . Bars represent SE of n = 4 . Asterisks indicate a statistically significant difference compared to Col-0 according to a Student’s t-test ( p<0 . 05 ) ; ns = not significant . All experiments were repeated at least twice with similar results unless otherwise stated . DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 00710 . 7554/eLife . 00983 . 008Figure 2—figure supplement 1 . Expression of the constitutively active BZR1S173A results in the suppression of specific PTI outputs . ( A ) Co-IP of BAK1 and FLS2 in Col-0 and BZR1S173A seedlings treated with 1 μM flg22 for 10 min . Co-immunoprecipitated proteins were analyzed by using anti-FLS2 or anti-BAK1 antibodies . ( B ) MAPK activation in Col-0 and BZR1S173A seedlings upon treatment with 1 μM flg22 . Membranes were blotted with phospho-p44/42 MAPK ( Erk1/2; Thr202/Tyr204 ) rabbit monoclonal antibodies . ( C ) Flg22-induced resistance to Pto DC3000 in BZR1S173A plants . Plants were pre-treated with 1 μM flg22 or water 24 hr prior to bacterial inoculation . Bars represent SE of n = 4 . Asterisks indicate a statistically significant difference compared to mock-treated plants according to a Student's t-test ( p<0 . 05 ) ; ns = not significant . All experiments were repeated at least twice with similar results . DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 008 To understand how BZR1 mediates the BR-PTI crosstalk , we performed meta-analysis of microarray and ChIP-chip data containing BR-regulated and BZR1 or BES1 target genes ( Sun et al . , 2010; Yu et al . , 2011 ) . Functional enrichment of BR-regulated genes unveiled a statistically significant over-representation of defense-related GO terms of the Biological Process ontology ( Table 1 ) , indicating that BR signaling regulates the expression of defense-related genes . Independent analysis of BR-regulated BZR1 or BES1 targets confirmed BZR1 as the main transcription factor involved in the regulation of defense gene expression ( Table 1 ) . Two out of four over-represented GO terms of the Molecular Function ontology among the BR-regulated BZR1 targets are transcription factor and transcription repressor activity ( Table 2 ) . Interestingly , several defense-related GO terms are also over-represented in the subset of BR-regulated BZR1-targeted transcription factors ( Table 3 ) , pointing at a BZR1-mediated secondary transcriptional wave of defense-related genes . 10 . 7554/eLife . 00983 . 009Table 1 . Defense-related Gene Ontology terms ( Biological Process ontology ) over-represented among all BR-regulated genes , BR-regulated BZR1 targets and BR-regulated BES1 targetsDOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 009Defense-related GO termObserved frequency ( % ) Expected frequency ( % ) p-valueBR-Regulated genes response to bacterium2 . 213 . 31 × 10−08 defense response to bacterium1 . 90 . 83 . 31 × 10−08 response to chitin1 . 40 . 51 . 78 × 10−07 defense response4 . 733 . 32 × 10−07 response to fungus1 . 50 . 73 . 4 × 10−06 response to nematode0 . 70 . 20 . 000532 defense response to fungus10 . 50 . 0035BR-regulated BZR1 targets response to chitin2 . 60 . 59 . 13 × 10−13 response to bacterium2 . 310 . 00112 defense response to bacterium1 . 90 . 80 . 00154 response to fungus1 . 60 . 70 . 00495BR-regulated BES1 targets response to chitin2 . 40 . 50 . 0043910 . 7554/eLife . 00983 . 010Table 2 . Gene Ontology terms ( Molecular Function ontology ) over-represented among all BR-regulated BZR1 targetsDOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 010Over-represented GO termObserved frequency ( % ) Expected frequency ( % ) p valueBR-regulated BZR1 targets nucleic acid binding transcription factor activity14 . 810 . 20 . 000223 transferase activity21 . 616 . 80 . 00333 kinase activity11 . 68 . 10 . 00702 transcription repressor activity1 . 10 . 30 . 0110 . 7554/eLife . 00983 . 011Table 3 . Defense-related Gene Ontology terms ( Biological Process ontology ) over-represented among the BZR1-target BR-regulated transcription factorsDOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 011Defense-related GO TermObserved frequency ( % ) Expected frequency ( % ) p valueBZR1-target BR-regulated TFs response to chitin16 . 60 . 51 . 36 × 10−26 defense response to bacterium7 . 60 . 84 . 71 × 10−07 response to bacterium7 . 614 . 51 × 10−06 regulation of defense response to virus by host1 . 400 . 000964 regulation of immune effector process1 . 400 . 00151 regulation of defense response to virus1 . 400 . 00151 regulation of defense response2 . 80 . 30 . 00484 defense response8 . 330 . 00603 response to fungus3 . 40 . 70 . 01 defense response to fungus2 . 80 . 50 . 02 To identify BZR1-regulated transcription factors with a prominent role in defense , we performed promoter enrichment analysis on the subset of defense-related BR-regulated genes , and found the W-box motif as the only significantly over-represented motif ( Table 4 ) . The W-box motif is the binding site for the WRKY family of transcription factors ( Rushton et al . , 2010 ) , and several members of this family are BR-regulated BZR1-targets ( Table 5 ) . We hypothesized that WRKYs that are BR-induced and BZR1 targets may be involved in PTI signaling . Notably , wrky11 , wrky15 , wrky18 and wrky70 mutants displayed enhanced PAMP-triggered ROS ( Figure 3A ) , suggesting that these transcription factors act as negative regulators of early PTI signaling . This is in accordance with their role as negative regulators of immunity ( Figure 3—figure supplement 1A; Journot-Catalino et al . , 2006 ) . Therefore , the BZR1-mediated inhibition of PTI might be partially explained by the up-regulation of genes encoding WRKY transcription factors that negatively control the expression of genes involved in early PTI signaling . 10 . 7554/eLife . 00983 . 012Table 4 . Over-represented cis-acting promoter elements among the defense-related BR-regulated genes according to Athena ( http://www . bioinformatics2 . wsu . edu/cgi-bin/Athena/cgi/home . pl ) DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 012Enriched TF site% promotersp valueDefense-related BR-regulated genes W-box72 . 4<10−610 . 7554/eLife . 00983 . 013Table 5 . BR-regulated BZR1-target WRKY genesDOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 013AGI numberWRKY TFBR-Induced BZR1 targets AT4G31800WRKY18 AT4G31550WRKY11 AT4G23810WRKY53 AT3G56400WRKY70 AT5G49520WRKY48 AT5G52830WRKY27 AT1G69310WRKY57 AT2G23320WRKY15 ( Yu et al . , 2011 ) BR-repressed BZR1 targets AT4G01250WRKY22 AT1G80840WRKY40 AT2G24570WRKY17 AT2G23320WRKY15 ( Sun et al . , 2010 ) AT2G30590WRKY2110 . 7554/eLife . 00983 . 014Figure 3 . WRKY transcription factors play a dual role on the BR-mediated regulation of PTI signaling . ( A ) Flg22-triggered ROS burst in mutants in each BR-induced BZR1-targeted WRKY . Leaf discs of four- to five-week-old Arabidopsis plants were used in these assays . Flg22 was used at a concentration of 50 nM . Total photon counts were integrated between minutes two and 40 after PAMP treatment . Bars represent SE of n = 28 . Asterisks indicate a statistically significant difference compared to Col-0 according to a Student’s t-test ( p<0 . 05 ) . ( B ) Flg22-triggered ROS burst in epiBL ( BL ) - or mock- pre-treated wrky40 mutant or wild-type plants . Leaf discs of four- to five-week-old plants were pre-treated with a 1 μM BL solution or mock solution for 8 hr . Flg22 was used at a concentration of 50 nM . Total photon counts were integrated between minutes two and 40 after PAMP treatment . Bars represent SE of n = 21 . Asterisks indicate a statistically significant difference compared to Col-0 according to a Student’s t-test ( p<0 . 05 ) . ( C ) Co-IP of BZR1-GFP transiently expressed in N . benthamiana , alone or together with WRKY40-HA or WRKY6-HA . BZR1-GFP was immunoprecipitated with an anti-GFP antibody . Immuniprecipitated or total proteins were separated in a 10% acrylamide gel and transferred to PVDF membranes . Membranes were blotted with anti-HA or anti-GFP antibodies . CBB: Coomassie brilliant blue . ( D ) Co-IP of BZR1-GFP transiently expressed in Arabidopsis protoplasts , alone or together with WRKY40-HA . BZR1-GFP was immunoprecipitated with an anti-GFP antibody . Immuniprecipitated or total proteins were separated in a 10% acrylamide gel and transferred to PVDF membranes . Membranes were blotted with anti-HA or anti-GFP antibodies . CBB: Coomassie brilliant blue . All experiments were repeated at least twice with similar results . DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 01410 . 7554/eLife . 00983 . 015Figure 3—figure supplement 1 . Mutants in WRKY11 , WRKY15 , WRKY18 and WRKY40 are more resistant to Pto DC3000 . ( A ) and ( B ) Pto DC3000 infections in Col-0 , wrky11 , wrky15 , and wrky18 ( A ) and in Col-0 and wrky40 ( B ) plants . Bars represent SE of n = 4 . Asterisks indicate a statistically significant difference compared to Col-0 plants according to a Student's t-test ( p<0 . 05 ) ; ns = not significant . All experiments were repeated at least three times with similar results . DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 015 One of the WRKY genes targeted by BZR1 is WRKY40 ( Sun et al . , 2010 ) . Interestingly , all described targets of WRKY40 ( Pandey et al . , 2010 ) are also targets of BZR1 ( Sun et al . , 2010 ) ( Table 6 ) . The over-representation of the W-box motif among BZR1 targets ( Table 7 ) suggests that BZR1 may interact with WRKY transcription factors ( such as WRKY40 ) to cooperatively regulate transcription . WRKY40 has been described as a negative regulator of defense against biotrophic pathogens and insects ( Xu et al . , 2006; Pandey et al . , 2010; Brotman et al . , 2013; Schon et al . , 2013; Schweizer et al . , 2013 ) . In agreement with this , we found that a null wrky40 mutant is more resistant to Pto DC3000 ( Figure 3—figure supplement 1B ) . Strikingly , wrky40 plants are partially impaired in the BR-mediated suppression of PAMP-triggered ROS ( Figure 3B ) , suggesting that WRKY40 may act coordinately with BZR1 to suppress immunity . Indeed , we found that BZR1 associates with WRKY40 , but not WRKY6 , in co-immunoprecipitation experiments when transiently co-expressed in Nicotiana benthamiana leaves ( Figure 3C ) or Arabidopsis protoplasts ( Figure 3D ) . Collectively , these results indicate that BZR1 and WRKY40 form a protein complex that may participate in the transcriptional inhibition of PTI signaling . 10 . 7554/eLife . 00983 . 016Table 6 . Overlap between the targets of WRKY40 and BZR1DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 016Known targets of WRKY40 ( Pandey et al . , 2010 ) Targets of BZR1 ( Sun et al . , 2010 ) Confirmed by ChIP EDS1Yes RRTF1Yes JAZ8YesPutative ( according to expression analyses ) LOX2Yes AOSYes JAZ7Yes JAZ10Yes10 . 7554/eLife . 00983 . 017Table 7 . Representation of the W-box motif among the BR-regulated BZR1 targets according to Athena ( http://www . bioinformatics2 . wsu . edu/cgi-bin/Athena/cgi/home . pl ) DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 017BZR1 targets% of promoters with W-box motif ( s ) p valueBR-induced66<10−10BR-repressed72<10−4 BZR1 , together with DELLAs and PIF4 , is part of a core transcription module that integrates hormonal ( gibberellin [GA] and BR ) and environmental ( light ) signals ( Gallego-Bartolome et al . , 2012; Li et al . , 2012; Oh et al . , 2012; Bai et al . , 2012b ) . In the dark , BZR1 is activated by endogenous BR and GA to promote growth , partially through the synergistic interaction with PIF4 ( Jaillais and Vert , 2012 ) . Because etiolation requires rapid growth , we hypothesized that plants may prioritize growth over immunity in dark conditions . In fact , we found that PAMP-triggered SGI was partially impaired in dark-grown seedlings ( Figure 4A–D ) . This impairment was abolished in the BR-insensitive mutants bri1-301 and bin2-1 ( Figure 4A , Figure 4—figure supplement 2A ) , indicating that BR signaling is responsible for the dark-induced suppression of this PTI response . Activation of BZR1 in the BZR1Δ line mimicked the dark-induced suppression of SGI in both light and dark ( Figure 4B ) . However , activation of BES1 in the BES1S171A line did not impact SGI ( Figure 4—figure supplement 3A ) . Consistent with the previous results , exogenous BR treatment suppressed SGI in both light and dark ( Figure 4C , Figure 4—figure supplement 2B , C ) . While treatment with GA alone did not have a dramatic effect on SGI , co-treatment with BL and GA resulted in an enhancement of the BR-mediated suppression of SGI ( Figure 4C , Figure 4—figure supplement 2B ) , indicating an additive effect of these two hormones when applied together . Moreover , treatment with the GA synthesis inhibitors paclobutrazol ( PAC ) or uniconazole ( Uni ) abolished the effect of BL on SGI ( Figure 4—figure supplement 1A , B and Figure 4—figure supplement 2B , C ) , and this effect was reduced in the GA biosynthetic mutant ga1-3 ( Figure 4—figure supplement 1C ) . Taken together , these results demonstrate that BR suppress at least one PTI output , SGI , in the dark in a GA-dependent manner , most likely through activation of BZR1 . Notably , although the wrky40 mutant undergoes etiolation normally ( Figure 4—figure supplement 2D ) , it shows a diminished suppression of SGI in the dark ( Figure 4D , Figure 4—figure supplement 2D ) , supporting the idea that WRKY40 is required for the BZR1-mediated inhibition of PTI . 10 . 7554/eLife . 00983 . 018Figure 4 . Activation of BR signaling and BZR1 prioritizes growth over immunity in the dark . ( A ) and ( B ) Relative seedling growth inhibition of 10-day-old ( A ) Col-0 , bri1-301 and bin2-1 or ( B ) Col-0 and BZR1Δ seedlings induced by increasing concentrations of flg22 in either light or dark . ( C ) Relative seedling growth inhibition of 10-day-old Col-0 seedlings grown on medium supplemented or not with BL ( 1 μM ) , GA ( 1 μM ) , BL+GA ( 1 μM + 1 μM ) or mock solution in light or dark . ( D ) Relative seedling growth inhibition of Col-0 or wrky40 seedlings induced by increasing concentrations of flg22 in either light or dark . Bars represent SE of n = 16 ( A , B and D ) or n = 8 ( C ) Asterisks indicate a statistically significant difference compared to Col-0 in the same condition ( light or dark and same concentration of flg22 ) , according to a Student’s t-test ( p<0 . 05 ) ; ‘a’ indicates a statistically significant difference compared to the same genotype/treatment and concentration of flg22 in light , according to a Student’s t-test ( p<0 . 05 ) . All experiments were repeated at least three times with similar results . Values are relative to Col-0 ( A , B and D ) or mock-treated seedlings ( C ) ( set to 100 ) . Absolute values of these experiments are shown in Figure 4—figure supplement 3 . ( E ) Schematic model depicting the BZR1-mediated inhibition of PTI . Upon BR- and DELLA-dependent activation , BZR1 induces the expression of negative regulators of PTI , such as WRKY11 , WRKY15 , WRKY18 , or HBI1 . In addition , BZR1 also inhibits the expression of immune genes , acting cooperatively with WRKY40 and possibly other WRKYs . Ultimately , the BZR1-mediated changes in transcription would lead to the suppression of PTI signaling . The PTI signaling pathway is shadowed in violet; the BR signaling pathway is shadowed in green . DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 01810 . 7554/eLife . 00983 . 019Figure 4—figure supplement 1 . The BR-mediated suppression of seedling growth inhibition in the dark requires GA synthesis . Seedling growth inhibition of 10-day-old Col-0 seedlings grown on medium supplemented or not with ( A ) BL ( 1 μM ) , paclobutrazol ( PAC; 1 μM ) , BL+PAC ( 1 μM + 1 μM ) and PAC+GA ( 1 μM + 1 μM ) , or ( B ) uniconazole ( Uni; 100 μM ) , BL ( 1 μM ) and Uni+BL ( 100 μM + 1 μM ) induced by increasing concentrations of flg22 in light or dark . ( C ) Seedling growth inhibition of 10-day-old Ler and ga1-3 seedlings grown on medium supplemented or not with BL ( 1 μM ) . Bars represent SE of n = 8 . Asterisks indicate a statistically significant difference compared to Col-0 in the same condition ( light or dark and same concentration of flg22 ) , according to a Student's t-test ( p<0 . 05 ) ; ‘a’ indicates a statistically significant difference compared to the same genotype/treatment and concentration of flg22 in light , according to a Student's t-test ( p<0 . 05 ) . Values are relative to mock-treated seedlings ( set to 100 ) . All experiments were repeated at least twice with similar results . DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 01910 . 7554/eLife . 00983 . 020Figure 4—figure supplement 2 . Phenotype of the light- or dark-grown seedlings used in the seedling growth inhibition assays ( Figure 4 and Figure 4—figure supplement 1 ) . Representative seedlings of the seedling growth inhibition experiments depicted in: ( A ) Figure 4A; ( B ) Figure 4—figure supplement 1A; ( C ) Figure 4–figure supplement 1B; ( D ) Figure 4D . Scale bar , 1 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 02010 . 7554/eLife . 00983 . 021Figure 4—figure supplement 3 . Absolute fresh weight values of seedling growth inhibition assays . ( A ) Seedling growth inhibition of 10-day-old Col-0 or BES1S171A seedlings induced by increasing concentrations of flg22 . ( B ) Absolute fresh weight values of the seedling growth inhibition assay depicted in Figure 4B , dark . ( C ) Absolute fresh weight values of the seedling growth inhibition assay depicted in Figure 4A . ( D ) Absolute fresh weight values of the seedling growth inhibition assay depicted in Figure 4C . ( E ) Absolute fresh weight values of the seedling growth inhibition assay depicted in Figure 4—figure supplement 1A . ( F ) Absolute fresh weight values of the seedling growth inhibition assay depicted in Figure 4—figure supplement 1B . ( G ) Absolute fresh weight values of the seedling growth inhibition assay depicted in Figure 4D . Error bars represent SE as indicated in Figure 4 , Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00983 . 021 Previously , a unidirectional negative crosstalk between the growth-promoting hormone BR and PTI had been described ( Albrecht et al . , 2012; Belkhadir et al . , 2012 ) . In this work , we show that activation of one of two major BR-activated transcription factors , BZR1 , is sufficient to suppress PTI , measured as PAMP-triggered ROS production , PAMP-triggered gene expression , SGI or induced resistance ( Figures 1 and 2 , Figure 1—figure supplement 3 , Figure 2—figure supplement 1 ) . Of note , another PTI output , MAPK activation , is not affected by activation of the BR pathway ( Figure 2B , Figure 1—figure supplement 2 , Figure 2—figure supplement 1B ) . BR treatment results in BZR1-dependent changes in the expression of defense-related genes , among which several members of the WRKY family of transcription factors can be found . Because the promoters of BR-regulated defense-related genes are enriched in the W-box motif ( Table 4 ) , BZR1-targeted WRKY transcription factors could be responsible for a secondary wave of transcription , ultimately leading to the suppression of PTI . In agreement with this idea , a subset of WRKYs induced by BR ( WRKY11 , WRKY15 and WRKY18 ) ( Figure 3A ) act as negative regulators of PAMP-triggered ROS , potentially by controlling the steady-state expression of genes encoding components required for this response . The over-representation of the W-box motif among the BZR1 targets ( Table 7 ) raises the possibility that , additionally , WRKY transcription factors could also act together with BZR1 to cooperatively regulate gene expression . We found that WRKY40 associates with BZR1 directly or indirectly in planta ( Figure 3C , D ) ; in the absence of WRKY40 , the BR-mediated suppression of PAMP-triggered ROS burst is partially impaired ( Figure 3B ) . Therefore , WRKYs may play a dual role in the BZR1-mediated suppression of defense , as both co- and secondary regulators of defense gene expression . Given that the loss of BR-mediated suppression of PAMP-triggered ROS burst in the wrky40 mutant is only partial , BZR1 may interact with other members of the WRKY family , such as WRKY18 or WRKY60 , to repress immunity . Furthermore , we recently described that the bHLH transcription factor HBI1 , which is a BRZ1 target ( Sun et al . , 2010; Bai et al . , 2012a ) , negatively regulates PTI ( Malinovsky et al . , under revision ) . All together , these results illustrate that BZR1 controls the expression of transcription factors ( e . g . WRKY11 , WRKY15 , WRKY18 and HBI1 ) , which themselves might control the expression of PTI components ( see model in Figure 4E ) whose identities remain to be identified . Plants need to finely regulate allocation of resources upon integration of environmental cues , both biotic and abiotic , in order to rapidly and readily adapt to changing conditions and ensure survival in a cost-efficient manner . Dark conditions impose an energetic limitation due to lack of photo-assimilates; in this situation , the restoration of normal photosynthesis by reaching light is an essential requirement to guarantee perpetuation , and as such must be given priority ( Casal , 2013 ) . We hypothesize that when plants face conditions that require rapid growth , such as when germinating in soil or when under a canopy , limited resources are invested in this developmental process at the expense of immunity in what must be a quantitative choice . Indeed , we show that etiolated seedlings do not arrest their growth in response to PAMPs as light-grown seedlings do , as measured by total fresh weight ( Figure 4A–D ) . In addition , BR signaling , acting cooperatively with GA signaling , is required for the dark-induced suppression of this PTI response ( Figure 4C , Figure 4—figure supplement 1 ) , and activation of BZR1 is sufficient to exert this effect regardless of light conditions ( Figure 4B ) . Although seedlings were used in these experiments due to technical reasons , BR also regulate growth at later developmental stages , so this phenomenon may be more general . Based on these findings , we propose a model in which BZR1 regulates the expression of defense genes , assisted by WRKY40 ( and potentially other WRKYs ) , ultimately resulting in a quantitative suppression of immunity ( Figure 4E ) . Because the activation status of BZR1 depends on BR , GA and light signaling , BZR1 would act as a molecular integrator of these inputs to effectively regulate the trade-off between growth and immunity .
Col-0 plants were used as control . The transgenic lines BZR1Δ , bri1-5/BZR1Δ and BES1S171A ( Gampala et al . , 2007 ) , BZR1S173A and BZR1-CFP ( Ryu et al . , 2007 ) , 35S:BRI1-cit , BRI1p:BRI1sud-cit , 35S:DWF4 and BAK1-HA ( Belkhadir et al . , 2012 ) are published . The mutant lines Triple GSK3 mutant ( Vert and Chory , 2006 ) , bri1-5 ( Noguchi et al . , 1999 ) , bri1-301 ( Xu et al . , 2008 ) , bin2-1 ( Peng et al . , 2008 ) , wrky11 ( Journot-Catalino et al . , 2006 ) , wrky18 , wrky53 and wrky70 ( Wang et al . , 2006 ) wrky27 ( Mukhtar et al . , 2008 ) , wrky40 ( Pandey et al . , 2010 ) and ga1-3 ( Navarro et al . , 2008 ) are published . wrky15 mutant was identified in the ZIGIA population ( Wisman et al . , 1998a , 1998b ) ; wrky48 and wrky57 are from the SALK collection ( Alonso et al . , 2003 ) . Arabidopsis plants and seedlings were grown as described in Albrecht et al . ( 2012 ) . Flg22 and elf18 peptides were purchased from Peptron , and chitin oligosaccharide from Yaizu Suisankagaku . epiBL was purchased from Xiamen Topusing Chemical . LiCl , bikinin , brassinazole and GA were purchased from Sigma ( St Louis , MO , USA ) . Paclobutrazol was purchased from Duchefa ( Haarlem , NL ) . Uniconazole was purchased from Sigma . The measurement of ROS generation was performed as described in Albrecht et al . ( 2012 ) . Leaf discs from five-week-old Arabidopsis plants were used in each experiment , as indicated in the figure legends . Total photon counts were measured over 40 min by using a high-resolution photon counting system ( HRPCS218 ) ( Photek , St Leonards on Sea , UK ) coupled to an aspherical wide lens ( Sigma Imaging , Welwyn Garden City , UK ) . Protein extraction and immunoprecipitation of Arabidopsis was performed as described in Schwessinger et al . ( 2011 ) . Arabidopsis mesophyll protoplasts were prepared from 4 to 5-week-old plants , transfected with the indicated constructs and incubated for 16 hr prior to BL treatment . Protein extraction of N . benthamiana was performed as described in Schwessinger et al . ( 2011 ) ; immunoprecipitations were performed using the μMACS GFP Isolation Kit ( Miltenyi Biotec , Church Lane Bisley , UK ) , following the manufacturer’s instructions . In N . benthamiana , BZR1-GFP was expressed from the pUb-cYFP-Dest vector ( Grefen et al . , 2010 ) ; WRKY40-HA and WRKY6-HA were expressed from the pAM-PAT vector ( AY436765; GeneBank ) . In protoplasts , WRKY40-HA was expressed from the pGWB414 vector ( Nakagawa et al . , 2007 ) ; the construct to express BZR1-GFP has been described elsewhere ( Ryu et al . , 2007 ) . In both cases , samples were treated with 1 μM epiBL solution for 1 hr prior to protein extraction . MAP kinase activation assays were performed as described in Schwessinger et al . ( 2011 ) . Phospho-p44/42 MAPK ( Erk1/2; Thr202/Tyr204 ) rabbit monoclonal antibodies ( Cell Signaling Technologies , Hitchin , UK ) were used according to the manufacturer’s protocol . RNA isolation was performed from ten-day-old seedling following the protocol described in Onate-Sanchez and Vicente-Carbajosa ( 2008 ) . First-strand cDNA synthesis was performed with the SuperScript III RNA transcriptase ( Invitrogen , Paisley , UK ) and oligo ( dT ) primer , according to the manufacturer’s instructions . For qPCR reactions , the reaction mixture consisted of cDNA first-strand template , primers ( 5 nmol each ) and SYBR Green JumpStart Taq ReadyMix ( Sigma ) . qPCR was performed in a BioRad CFX96 real-time system . UBQ10 was used as the internal control; expression in mock-treated Col-0 seedlings was used as the calibrator , with the expression level set to one . Relative expression was determined using the comparative Ct method ( 2-ΔΔCt ) . Each data point is the mean value of three experimental replicate determinations . Primers for At2g17740 are described in Albrecht et al . ( 2012 ) ; for NHL10 are described in Boudsocq et al . ( 2010 ) ; for LOX2 are described in Pandey et al . ( 2010 ) ; for UBQ10 ( U-box ) are described in Albrecht et al . ( 2012 ) . Seedling growth inhibition assays were performed as described in Nekrasov et al . ( 2009 ) . In brief , four-day-old Arabidopsis seedlings were grown in liquid Murashige–Skoog medium containing 1% sucrose supplemented with flg22 and the appropriate chemicals . Seedlings were weighed between 6 and 10 days after treatment . Induced resistance assays were performed as described previously ( Zipfel et al . , 2004 ) . In brief , water or a 1 μM flg22 solution were infiltrated with a needleless syringe into leaves of four-week-old Arabidopsis plants 24 hr prior to bacterial inoculation ( Pto DC3000 , 105 cfu/ml ) . Bacterial growth was determined 2 days after inoculation by plating serial dilutions of leaf extracts . Spray inoculation of P . syringae pv . cilantro ( Pci ) 0788-9 was performed as described in Schwessinger et al . ( 2011 ) . In brief , bacteria were grown in an overnight culture in LB medium , cells were harvested by centrifugation , and pellets were re-suspended to OD600 = 0 . 02 in 10 mM MgCl2 with 0 . 04% Silwet L-77 . Bacterial suspensions were sprayed onto leaf surfaces and plants were kept uncovered . Bacterial growth was determined 3 days after inoculation by plating serial dilutions of leaf extracts . Functional enrichment analyses of the Biological Process ontology were performed using VirtualPlant ( Katari et al . , 2010 ) . Functional enrichment analysis of the Molecular Function ontology was performed using the Classification SuperViewer tool of the Bio-Array Resource for Arabidopsis Functional Genomics , BAR ( Toufighi et al . , 2005 ) . Promoter analyses were performed using Athena ( O’Connor et al . , 2005 ) . | Like all organisms , plants must perform a careful balancing act with their resources . Investing in the growth of new roots or leaves can allow a plant to better exploit its environment—but it must not be at the expense of leaving the plant vulnerable to attack by pests and pathogens . As such , there is an obvious trade-off between allocating resources to growth or defense against disease . This trade-off must be finely balanced , and must also be responsive to different cues in the environment that would favor either growth or defense . The plant’s immune system is able to detect invading microbes , and trigger a defensive response against them . At the surface of plant cells , proteins called pattern recognition receptors are able to recognize specific molecules that are the tell-tale signs of microbes and pathogens—such as the proteins in the molecular tails that bacteria use to move around . For many pattern recognition receptors , signaling that they have recognized a potential invading microbe requires the actions of a co-receptor called BAK1 . Interestingly , BAK1 also interacts with the receptor that identifies brassinosteroids—hormones that stimulate plant growth . Since growth and a functioning immune system are both reliant on BAK1 , it was hypothesized that competition for this co-receptor could have a role in the trade-off between the two processes in plants . However , this explanation was controversial and the mechanisms underlying the trade-off still required clarification . Now , Lozano-Durán et al . have debunked the idea that competition for BAK1 is directly responsible for the trade-off between growth and immunity . By examining how BAK1 interacts with immune receptors in the plant model species Arabidopsis thaliana , the trade-off was actually shown to be independent of BAK1 . Instead , it was discovered that activation of a protein , called BZR1 , reprogramed gene expression to ‘switch off’ immune signaling in response to brassinosteroids . Lozano-Durán et al . also show that BZR1 allows the balance of the trade-off between growth and immunity to be shifted in response to cues from the environment . The suppression of the immune system by BZR1 was particularly pronounced when the conditions required fast plant growth—for example , when they mimicked the conditions experienced by seedlings before they emerge from the soil , and must grow swiftly to reach the light before they starve . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology"
] | 2013 | The transcriptional regulator BZR1 mediates trade-off between plant innate immunity and growth |
The key component of the nuclear pore complex ( NPC ) controlling permeability , selectivity , and the speed of nucleocytoplasmic transport is an assembly of natively unfolded polypeptides , which contain phenylalanine-glycine ( FG ) binding sites for nuclear transport receptors . The architecture and dynamics of the FG-network have been refractory to characterization due to the paucity of experimental methods able to probe the mobility and density of the FG-polypeptides and embedded macromolecules within intact NPCs . Combining fluorescence polarization , super-resolution microscopy , and mathematical analyses , we examined the rotational mobility of fluorescent probes at various locations within the FG-network under different conditions . We demonstrate that polarization PALM ( p-PALM ) provides a rich source of information about low rotational mobilities that are inaccessible with bulk fluorescence anisotropy approaches , and anticipate that p-PALM is well-suited to explore numerous crowded cellular environments . In total , our findings indicate that the NPC’s internal organization consists of multiple dynamic environments with different local properties .
Intracellular environments are highly crowded , with typical local macromolecular concentrations of ~80–400 mg/mL , and some cellular environments contain only ~50% water ( Kuznetsova et al . , 2014 ) . Under crowded conditions , excluded volume effects and local interactions can change protein activities by over an order of magnitude compared with the 'dilute' solutions typically used for most in vitro studies ( Aumiller et al . , 2014 ) . Crowded conditions can affect protein folding , structure , shape , conformational stability and dynamics , binding interactions , and enzymatic activity ( Kuznetsova et al . , 2014; Zhou et al . , 2008 ) . Biological polymers play central roles in generating a variety of crowded environments . For example , the polymers in mucus , the extracellular matrix , the cytoskeleton , the vitreous humor of the eye , and the Nuclear Pore Complex ( NPC ) produce complex environments that restrict diffusion and trap molecules ( Leterrier , 2001; Lieleg and Ribbeck , 2011 ) . In addition , the numerous distinct bodies/granules within the nucleus and the cytoplasm have been interpreted to form via a phase separation-like mechanism due to high local concentrations of self-cohesive nucleic acid and/or intrinsically disordered protein polymers ( Aumiller et al . , 2014; Toretsky and Wright , 2014 ) . Characterization of the physical , structural , dynamical , and functional properties of these crowded environments remains challenging due to the dearth of appropriate tools that are needed to investigate the complexity and heterogeneity of these environments on the nanoscale . One example of a crowded environment is the pore of the NPC , which mediates bidirectional traffic between the cytoplasm and the nucleoplasm of eukaryotic cells . The translocation passageway of the NPC is occupied by hundreds of intrinsically disordered polypeptides ( Lim et al . , 2008; Peleg and Lim , 2010; Suntharalingam and Wente , 2003 ) , 50–100 nuclear transport receptors ( NTRs ) ( Lowe et al . , 2015; Tokunaga et al . , 2008 ) , and protein and nucleic acid cargo complexes moving in opposite directions . NTRs are classified as importins or exportins , reflecting their ability to carry cargos into or out of the nucleus , respectively ( for reviews , see [Chook and Süel , 2011; Güttler and Görlich , 2011; Jamali et al . , 2011; Stewart , 2007; Wente and Rout , 2010] ) . On the nuclear side , RanGTP promotes disassembly of NTR/cargo import complexes , freeing the cargo and allowing NTRs to diffuse back to the cytoplasm ( Chook and Blobel , 2001; Izaurralde et al . , 1997; Rexach and Blobel , 1995; Siomi et al . , 1997 ) . NTR/cargo/RanGTP export complexes are disassembled on the cytoplasmic side after GTP hydrolysis , which results from interactions with RanGAP and a Ran-binding protein ( RanBP ) ( Bischoff and Görlich , 1997; Bischoff et al . , 1994; Güttler and Görlich , 2011; Kutay et al . , 1997a; Okamura et al . , 2015 ) . Many of these assembly and disassembly reactions are coordinated to occur at the cytoplasmic and nucleoplasmic exits of the NPC’s central pore ( Sun et al . , 2013; Sun et al . , 2008 ) . Exactly how cargo complexes are specifically recognized and yet rapidly migrate in milliseconds ( Dange et al . , 2008; Grünwald and Singer , 2010; Kubitscheck et al . , 2005; Tu et al . , 2013; Yang et al . , 2004; Yang and Musser , 2006a ) through the NPC’s crowded environment remains enigmatic . NPCs are large ( ~60–120 MDa ) structures with octagonal rotational symmetry . They are comprised of ~30 different nuclear pore proteins ( nucleoporins , or Nups ) , each of which are thought to be present in an integer multiple of eight copies ( Cronshaw et al . , 2002; Fahrenkrog and Aebi , 2003; Mi et al . , 2015; Ori et al . , 2013; Rout and Aitchison , 2001 ) . The vertebrate NPC has an outer diameter of ~120 nm , and extends ~200 nm along the transport axis ( Fahrenkrog and Aebi , 2003; Stoffler et al . , 1999 ) . Eight flexible filaments extend ~50 nm into the cytoplasm , and an additional eight filaments extend ~75 nm into the nucleoplasm and terminate in a ring to form the nuclear basket ( Fahrenkrog and Aebi , 2003; Stoffler et al . , 2003 ) . In humans , the hourglass-shaped central pore has a minimum diameter of ~50 nm and a length of ~85 nm ( Maimon et al . , 2012 ) . Within this large pore and decorating its openings is a network of ~200–250 intrinsically disordered polypeptides , which generates a permeability barrier impeding macromolecular transport ( Lim et al . , 2008; Ori et al . , 2013; Peleg and Lim , 2010; Suntharalingam and Wente , 2003 ) and which is particularly selective against larger cargos ( Mohr et al . , 2009; Popken et al . , 2015; Ribbeck and Görlich , 2001; Timney et al . , 2016 ) . These disordered polypeptides contain , in total , 3000–4000 phenylalanine-glycine ( FG ) repeats to which NTRs transiently bind as they carry cargos through NPCs ( Cronshaw et al . , 2002; Denning et al . , 2003; Rout et al . , 2000; Strawn et al . , 2004; Tran and Wente , 2006 ) . We term this assembly of intrinsically disordered FG-containing polypeptides the FG-network . Each FG-containing nucleoporin ( FG-Nup ) has a globular anchor domain that is embedded in or attached to the NPC scaffold , and thus , it acts as an anchor point for the flexible and mobile FG-domain . The FG-repeat motifs are separated by short ( ~10–20 amino acid residues ) , largely hydrophilic segments ( Denning and Rexach , 2007; Yamada et al . , 2010 ) . The FG-domains do not form readily recognizable secondary structures , but rather are more appropriately described as flexible polymers with alternating hydrophobic and hydrophilic domains ( Lim et al . , 2006; Yamada et al . , 2010 ) . The FG-network is sufficiently fluid and mobile that it is rapidly displaced by transporting cargos , which can be up to ~40 nm in diameter ( Frey and Görlich , 2009; Hough et al . , 2015; Lim et al . , 2007; Milles et al . , 2015; Panté and Kann , 2002 ) . The ‘polymer brush’ ( Lim et al . , 2006; Peleg and Lim , 2010 ) and ‘hydrogel’ ( Frey and Görlich , 2007; Frey et al . , 2006 ) models are the most widely cited descriptions of the biophysical nature of FG-polypeptide assemblies . These models are two extremes in the model space describing the potential morphologies and properties of the FG-Nup assemblies within the NPC ( Eisele et al . , 2013; Vovk et al . , 2016 ) . The polymer brush model postulates that the FG-polypeptides are largely non-interacting ( beyond steric repulsion ) , relatively extended and minimally entangled ( Lim et al . , 2006; Peleg and Lim , 2010 ) , and their spatial assemblies are stabilized mostly by entropic forces ( Vovk et al . , 2016 ) . The hydrogel model posits that the FG-polypeptides exhibit significant inter- and intra-strand cohesiveness via FG-FG interactions , which results in a connected dense network ( Frey and Görlich , 2007; Frey and Görlich , 2009; Frey et al . , 2006; Hülsmann et al . , 2012 ) . A hybrid , two-gate model postulates brush-like structures on both cytoplasmic and nuclear sides of the NPC , suitable for binding and ( dis ) assembly reactions , and a central cohesive structure in the center of the pore that provides the permeability barrier ( Patel et al . , 2007 ) . The spatial distribution of functional activities in this two-gate model is supported by single molecule transport results ( Sun et al . , 2013; Tu et al . , 2013 ) . Quantitative modeling of FG-polypeptide behavior predicts a smooth transition between brush-like and gel-like behaviors in response to relatively small changes in physical properties and favors a picture intermediate between a brush and a gel ( Vovk et al . , 2016 ) . The magnitude of the inter- and intra-chain cohesiveness that differentiates these two descriptions could be different for different FG-polypeptides , or different segments of the same FG-polypeptides , in distinct spatial locations within the NPC ( Vovk et al . , 2016 ) . Avidity calculations indicate that the multivalent affinities of NTRs depend critically upon the local free FG-repeat concentration ( Tu et al . , 2013 ) . In agreement with these predictions , experimental results indicate that some sub-populations of NTRs have very long dissociation times , and therefore , they potentially can form an integral part of the permeability barrier ( Kapinos et al . , 2014; Lowe et al . , 2015; Schleicher et al . , 2014 ) . Taken together , these findings suggest that the FG-polypeptides and NTRs act together to form different local environments with different properties within the NPC ( Coalson et al . , 2015; Ghavami et al . , 2014; Lowe et al . , 2015; Eskandari Nasrabad et al . , 2016; Osmanović et al . , 2013; Tagliazucchi et al . , 2013; Yamada et al . , 2010 ) . Considering the uncertainty in the structural arrangement of and interactions between FG-polypeptides , and knowing that many tens to over a hundred macromolecules ( including NTRs , Ran , and cargos ) interact with the FG-network during steady-state transport ( Abu-Arish et al . , 2009; Lowe et al . , 2015; Tokunaga et al . , 2008 ) , developing a general picture of FG-polypeptide distributions and local crowding conditions , and discerning their functional effects on cargo transport , is a challenging problem , but nevertheless essential for establishing the mechanism of nucleocytoplasmic transport and its implications . Here , we used the super-resolution approach photoactivated localization microscopy ( PALM ) ( Betzig et al . , 2006 ) to probe the locations of a number of FG-polypeptides and transport-related proteins within the NPC . Our main focus , however , was on using polarization PALM ( p-PALM ) ( Gould et al . , 2008 ) to measure rotational mobility , which is sensitive to local crowding conditions , and which enables probing of the local properties of crowded macromolecular assemblies that are currently inaccessible by other means . Crucially , we developed a theoretical model that enables detailed analysis of the experimental p-PALM data in terms of rotational diffusion constants . While numerous previous super-resolution approaches on NPCs utilized fixed samples , and most concentrated on scaffold structural questions ( Löschberger et al . , 2014 , 2012; Lowe et al . , 2015; Otsuka et al . , 2016; Pleiner et al . , 2016; Szymborska et al . , 2013; Winterflood and Ewers , 2014 ) , the NPCs in our samples were fully functional since our goal was to probe the properties of the FG-network , which is intrinsically dynamic . The results of our analysis of protein localization and local mobility within the NPC demonstrate that the FG-network is heterogeneous with regard to molecular crowding and that this can be influenced by the presence of embedded proteins , which argues for a remarkable complexity in nucleocytoplasmic trafficking pathways and their regulation .
The relatively wide particle distribution maps observed in PALM experiments ( Figure 7 ) suggested that regions of different rotational mobilities could potentially be resolved by combining the 2D localization maps generated via PALM with p-PALM rotational mobility information . This combined approach is challenging for the following reasons , which significantly reduced the size of current datasets . First , all p-PALM experiments reported thus far were performed by imaging the bottom of the nucleus , yielding spots from an approximately planar ( 2D ) distribution of NPCs . In contrast , in order to obtain the position of the NE , PALM experiments were performed at the nuclear equator , yielding spots from a pseudo-linear ( 1D ) distribution of NPCs . Thus , in combined PALM/p-PALM experiments , the NE was imaged at the nuclear equator , limiting the number of NPCs that could be simultaneously examined . Second , whereas entire trajectories were used for PALM , a single image per molecule was used for p-PALM to avoid biasing the data ( see Materials and methods ) . This p-PALM constraint was retained in combined PALM/p-PALM experiments . And third , while xy spatial information was readily obtained from p-PALM fluorescent spots , localization precision was reduced in combined PALM/p-PALM experiments compared with typical PALM data since the emission intensity was distributed over two images ( partially compensated by increasing the excitation intensity ) . Nonetheless , we demonstrate here that PALM localizations can be combined with p-PALM measurements . In order to explore the distribution of WGA-binding sites , we focused on the following question: did WGA reduce rotational mobility throughout the FG-network ? We examined mEos3-Nup98 and RanGAP-mEos3 , both of which yielded probe localizations widely distributed throughout the FG-network ( Figure 7 ) . The data were divided into those with |pcir| > 0 . 3 , which is only expected for molecules with Dr < ~103 rad2/s , and those with |pcir| < 0 . 3 , which could be observed for molecules with any Dr value ( Figure 2C and D ) . Thus , if WGA reduced rotational mobility in a specific region of the FG-network , the two datasets should yield spatially distinguishable distributions . This was not observed ( Figure 8 ) . Therefore , the data support the hypothesis that WGA inhibits rotational mobility throughout most , if not all , of the FG-network .
In this study , we have developed a combined experimental and theoretical framework for inferring the local rotational mobility of macromolecules in crowded environments using p-PALM . The p-PALM method was used to examine the macromolecular crowding in the vicinity of mEos3 probes positioned at different locations within the FG-network of NPCs , and PALM was used to determine the spatial distributions of these probes . Our major findings are: ( 1 ) different FG-polypeptides and different domains within the same FG-polypeptide experience different environments that are distinguishable by a probe’s rotational mobility; ( 2 ) in some cases , the binding of NTRs can increase crowding , thus producing significant differences in the properties of the local environments; and ( 3 ) WGA strongly influences rotational mobility throughout the FG-network , demonstrating that the local properties of the FG network can be modulated by embedded macromolecules . The implications of these findings provide a substantially improved understanding of the complexities of the FG-network , which we now discuss . The average positions vis-à-vis the transport axis for the mEos3 probe attached to FG-polypeptides are largely consistent with previous results , as indicted in the Results section . Thus , the PALM data support the hypothesis that these mEos3-tagged proteins behave similar to their wild-type counterparts , and therefore , they enable probing of the FG-network . The mEos3 probe on the N-terminus of Nup98 was widely distributed along the transport axis with localizations within the central pore region as well as relatively far from the NPC center on both the nuclear and cytoplasmic sides ( Figure 7E ) . This broad distribution pattern is consistent with Nup98 anchoring sites on the inner and outer ring structures of the NPC scaffold ( Kosinski et al . , 2016; Lin et al . , 2016 ) and on the cytoplasmic filaments and nuclear basket ( Frosst et al . , 2002; Stuwe et al . , 2012 ) , which together agree with the high copy number ( 48 ) for Nup98 ( Lin et al . , 2016; Ori et al . , 2013 ) . While the probe on the C-terminus of Pom121 also yielded a broad spatial distribution pattern , consistent with the long C-terminal FG-domain , labeling of the Pom121 N-terminus yielded a narrower distribution , consistent with anchoring of this part of the protein at the NE ( Figure 7C and D ) . Two anchoring sites for Pom121 via its N-terminal domain to Nup155 and/or Nup160 on both the inner and outer ring complexes ( Kosinski et al . , 2016; Lin et al . , 2016 ) is consistent with a stoichiometry of 16 copies/NPC ( Ori et al . , 2013 ) , although such dual anchoring is unresolvable at our current resolution . For RanGAP , there were a surprising number of localizations within the central pore and basket regions ( Figure 7B ) , seemingly inconsistent with the cytoplasmic distribution expected considering the known binding site for SUMOlyated RanGAP on the cytoplasmic filaments ( Hutten et al . , 2008; Mahajan et al . , 1997; Matunis et al . , 1998; Reverter and Lima , 2005; Wälde et al . , 2012 ) . RanGAP has a role in heterochromatin assembly ( Nishijima et al . , 2006 ) , it has both nuclear localization and nuclear export signals ( Feng et al . , 1999 ) , and , although found predominantly at the NE , it is also found in both the cytoplasmic and nucleoplasmic compartments ( Mahajan et al . , 1997; Matunis et al . , 1998 ) . These data suggest that RanGAP trafficks through the NPC , which could result in trapping within the FG-network during cell permeabilization . Alternatively , RanGAP could have additional roles within the FG-network other than export complex disassembly on the cytoplasmic filaments . Notably , RanGAP is not expected to catalyze disassembly of RanGTP-containing export complexes without a RanBP ( Bischoff and Görlich , 1997; Bischoff et al . , 1994; Güttler and Görlich , 2011; Kutay et al . , 1997a; Okamura et al . , 2015 ) , thus suggesting that only the portion of RanGAP attached to the cytoplasmic filaments ( RanBP2 ) may be active ( Mahajan et al . , 1997; Matunis et al . , 1998; Wu et al . , 1995; Reverter and Lima , 2005 #688; Yokoyama et al . , 1995 ) . The goal of our p-PALM approach was to identify regions of increased crowding within the FG-network and thus map the protein density distribution within the pore . While attaching the mEos3 probe to the end or to the middle of an FG-polypeptide could potentially introduce severe anisotropy in the rotational diffusion constants , the entirety of our results and simulations suggest that the probe’s rotation mobility behavior is at most only slightly anisotropic or the angle between the dominant rotational axis and the transition dipole is near the magic angle . In either case , the probe’s behavior largely resembled that of an untethered spherical particle . This was a somewhat surprising finding . However , this conclusion greatly simplifies the interpretation of p-PALM data since it substantially limits the parameter space that needs to be considered . Since the mEos3 probe’s rotational mobility behavior resembled that of an isotropic particle , comparison of the experimental <p>lin and Var ( p ) cir values with the simulation results for a spherical particle ( Figure 2 ) enables the rotational mobility of the mEos3 probe under different conditions to be interpreted in terms of the approximate values of the average rotational diffusion coefficient ( Dr ) . Assuming an mEos3 fluorescence lifetime > ~3 ns , which is true for most fluorescent proteins ( Bajar et al . , 2016; Moeyaert et al . , 2014 ) , the <p>lin values for all the conditions tested ( Supplementary file 1 ) indicate that Dr ≤ ~106 rad2/s ( Figure 2 and Supplements ) . Since Dr for mEos3 free in solution is ~107 rad2/s ( calculated for a sphere [Loman et al . , 2010] ) , the experimental <p>lin values suggest that rotational mobility was reduced by at least an order of magnitude by crowding within the FG-network . Var ( p ) cir values of ~0 . 2–0 . 3 ( Supplementary file 1 ) for the mEos3 probe under some conditions , in particular in the presence of WGA , indicate that the Dr was reduced to <~100 rad2/s ( Figure 2 and Supplements ) , that is , at least a 5 orders of magnitude change in rotational mobility from the free particle ( for at least a fraction of the particles in the population , considering that most populations likely consisted of particle distributions with multiple Dr values – see Appendix 1 ) . The p-PALM method therefore enables detection of large changes in rotational mobility . Moreover , since the p-PALM technique measures the polarization of individual molecules , which allows calculation of the variance of the polarization , it permits discrimination between rotational diffusion behaviors at much lower Dr values than traditional anisotropy approaches , in which fluorescence depolarization is governed by the fluorescence lifetime ( Lakowicz , 2006 ) . Notably , using a bulk fluorescence anisotropy approach on yeast NPCs ( Atkinson et al . , 2013; Mattheyses et al . , 2010; Kampmann et al . , 2011 ) , it was reported that some GFP probes were oriented when attached to some FG-polypeptides , particularly when they were near to the NPC scaffold . As the anisotropy signals were weak relative to homogeneous models , it appears that either the percentage of oriented molecules was low , or the orientation bias was weak . For either explanation , the assumption that probes were initially isotropically oriented in our random walk simulations is valid in most cases and leads to only minor errors in other cases . The fact that WGA had substantial effects on rotational mobility , as we observed here , and yet had very little , if any , effect on probe orientation ( Atkinson et al . , 2013 ) emphasizes the different physical parameters measured by the p-PALM and bulk fluorescence anisotropy approaches . Under wild-type conditions , <p>lin and Var ( p ) cir values suggest that the Dr was 103–106 rad2/s for the large majority of probe molecules , consistent with the high mobility expected for the FG-polypeptides and a dynamically flexible FG-network . Conditions that decreased the rotational mobility of the mEos3 probe have been interpreted to result from an increase in macromolecular crowding . A high density of macromolecules reduces molecular motion ( Dix and Verkman , 2008; McGuffee and Elcock , 2010 ) , presumably through an increased number of contacts with surrounding macromolecules . An mEos3 probe molecule within the FG-network can interact with FG-polypeptides , embedded macromolecules , or both . While the parameter Var ( p ) cir provided an initial indication of the reduction of rotational mobility due to crowding , it is still a population average , and a more refined picture was obtained by examining the full distribution of polarization values via polarization histograms and photon scatterplots . In multiple instances , two distinct rotational mobilities were necessary to explain the data , indicating heterogeneity in the environment around the different probe molecules . We consider it likely that most , if not all , of the high Var ( p ) cir values arose from mixed populations ( Figure 6 and discussion in Appendix 1 ) , one sub-population of which had a relatively low rotational mobility ( <103 rad2/s ) . Therefore , despite being fused to a single location in a given protein , mEos3 probes often resided in multiple distinct environments , and variations in Var ( p ) cir values likely arose from both differences in local protein densities as well as differential partitioning between environments . In all cases that we examined , WGA had a significant effect on the probe’s rotational mobility ( Figures 4B and 5B ) . WGA binds to O-GlcNAc-modified Nups , of which there are at least five in humans ( Finlay and Forbes , 1990; Hülsmann et al . , 2012 ) . The eight GlcNAc binding sites on the WGA dimer ( Schwefel et al . , 2010 ) suggest that it likely inhibited rotational mobility by non-covalently ‘crosslinking’ the FG-network . Considering that WGA affected the rotational mobility of mEos3 probes that were located both in the central pore and on the cytoplasmic and nucleoplasmic sides ( Figure 7—figure supplement 1 ) , the most parsimonious conclusion is that WGA binding sites are found throughout the FG-network . This conclusion is also supported by the combined PALM and p-PALM data ( Figure 8 ) . However , this conclusion that WGA binding sites are located throughout the FG-network is inconsistent with previous dSTORM microscopy studies that localized WGA to an ~40 nm diameter ring near the scaffold of the central pore ( Löschberger et al . , 2012 ) . Electron microscopy using WGA-gold revealed a similar picture to the dSTORM study , although more central localizations were also revealed ( Akey and Goldfarb , 1989 ) . It is possible that freezing or fixation influences the distribution of WGA-binding sites in these previous studies , which could explain the apparent conflict with the p-PALM data collected here on functional pores . However , there is an alternate interpretation . WGA could significantly increase Var ( p ) cir by binding to a distinct region of the FG-network , and strongly influencing the rotational mobility of the sub-population of probes in the neighborhood of these binding sites . In a highly interconnected network , such as a hydrogel or NTR/FG-polypeptide mixed network , WGA binding in one localized spatial region could influence more distant regions of the FG network via long-range allosteric-type effects . In this way , binding of WGA to one or more discrete spatial locations could influence the rotational motion of a probe throughout the FG-network . This interpretation is consistent with the WGA localization data obtained via dSTORM and electron microscopy ( Akey and Goldfarb , 1989; Löschberger et al . , 2012 ) . In the case of the mEos3 probe on the N-terminus of Pom121 , whose rotational mobility was also significantly reduced by the WGA , it seems unlikely that WGA-binding interactions within the central pore could be transmitted across the NE membrane and influence the rotational mobility of a probe within the perinuclear space . For this reason , we have concluded that the N-terminus of Pom121 is likely to reside within the central pore ( Figure 4 ) . Our results also shed light on the NTR distribution within the NPC . Nearly a hundred molecules of Imp β1 are bound within each NPC during steady-state ( Lowe et al . , 2015; Paradise et al . , 2007; Tokunaga et al . , 2008 ) , consistent with the finding that NTRs have high affinities for FG-polypeptides ( summarized in [Tetenbaum-Novatt et al . , 2012] ) . A high number of NTRs within the FG-network increases macromolecular crowding , which is expected to influence the structural and functional properties of the FG-network ( Kapinos et al . , 2014; Lowe et al . , 2015; Schleicher et al . , 2014; Vovk et al . , 2016; Wagner et al . , 2015 ) . In particular , the NTR-centric model postulates that NTR/FG-polypeptide effective affinities are higher nearest the NPC scaffold , and significantly weaker in the center of the pore , thus enabling rapid transport only through a narrow channel ( ~10–20 nm diameter ) in the center of the ~50 diameter pore ( Kapinos et al . , 2014; Schleicher et al . , 2014; Wagner et al . , 2015 ) . This model therefore predicts significantly higher macromolecular crowding near the scaffold anchor domain of FG-polypeptides . This hypothesis was directly tested via the p-PALM measurements on the mEos3 probe placed at different locations within the Nup98 FG-polypeptide , which support the hypothesis that crowding is indeed higher near the Nup98 anchor domain ( Figure 5 ) . Considering mixed populations , the average weighted rotational diffusion constant ( Dr , ave ) for the probes on mEos3-Nup98 and mEos3-700midNup98 were ~2800 and ~910 rad2/s , respectively ( Figure 6—figure supplement 1A C ) . Similarly , the Dr , ave for the probes on Pom121-mEos3 and mEos3-Pom121 were ~2500 and ~850 rad2/s , respectively ( Figure 6—figure supplement 2 ) . Therefore , the rotational mobility data for both Pom121 and Nup98 suggest higher macromolecular crowding near the NPC scaffold than at the tips of the FG-polypeptides , consistent with the NTR-centric model . For comparison , in the absence of molecular crowding effects , Dr ≈ 1000 rad2/s for the mEos3 probe would correspond to a viscosity of ~104 cP . Notably , the probes on both mEos3-700midNup98 and mEos3-Pom121 experience multiple environments distinguished by at least two distinct rotational mobilities ( Figure 6—figure supplements 1 , 2 ) . Since the mEos3 probes in both of these constructs are attached near their respective anchor domains , and thus cannot migrate to spatially distinct sites within the FG-network , the local environment must be heterogeneous within an individual NPC , or with respect to different NPCs . Consequently , crowding near the NPC scaffold is somewhat heterogeneous . High time-resolution super-resolution methods on functional NPCs in unfixed cells will continue to be instrumental in deciphering the complex , amorphous biomaterial that is the FG-network . We demonstrated here that the p-PALM method allows examination of rotational mobility over a range of at least 6 orders of magnitude . This range can be tuned by both acquisition conditions and experimental design , offering significant advantages over bulk measurements of fluorescence anisotropy . The results have allowed us to infer the local binding interactions and molecular crowding within NPCs , and have elucidated multiple aspects of the structural and dynamic complexity of the FG-network . While dynamics are an essential feature of the FG-network , enabling both rapid transport and dynamic maintenance of the permeability barrier , the extent to which newly identified heterogeneities play a role in functional properties of the NPC remains to be explored . While we expect that p-PALM will enable further dissection of the intricacies of the FG-network , it is also well-suited for probing the nanoscale structure of other dense molecular aggregates , such as the poorly understood organization of numerous nucleoplasmic and cytoplasmic membrane-less compartments ( ‘bodies’ ) , for example , nucleoli , stress granules , and RNA and protein processing bodies ( Aumiller et al . , 2014; Mitrea and Kriwacki , 2016 ) . These bodies typically contain high concentrations of proteins , and often nucleic acids , and their high densities promote phase separation . These highly crowded environments are difficult to probe because of their rapid dynamics , and often , their small size ( <1 µm ) ( Aumiller et al . , 2014; Mitrea and Kriwacki , 2016 ) , and thus , the high time- and super-resolution capability of PALM and the molecular crowding sensitivity of p-PALM provide an important novel tool . | Most of the genetic material inside an animal cell is enclosed within a compartment called the nucleus . This compartment is separated from the rest of the cell by the nuclear envelope , a double-membrane structure containing thousands of pores that selectively allow certain molecules ( collectively referred to as cargo ) to enter and exit the nucleus . The movement of cargo through the pores is controlled by large groups of proteins called nuclear pore complexes . The pore is at the center of the complex and is filled by a selective barrier made of an extensive network of flexible proteins known as the FG-network . Other proteins known as nuclear transport receptors bind to the proteins in the FG-network and carry cargos through the barrier . The properties of the nuclear pore barrier and how it rapidly selects the right cargos have been difficult to study , in part , because the barrier network is constantly changing and is crowded with hundreds of transport receptors . New techniques are needed to investigate such highly crowded environments inside cells . Now , Fu et al . use a technique called polarization photoactivated localization microscopy ( p-PALM ) to explore the molecular crowding within the nuclear pore barrier in human cells . This technique measures the freedom with which a single molecule embedded in the network can rotate , providing information about the local environment . In a crowded environment , it is harder for the probe molecule to rotate as it is more likely to bump into other molecules . Fu et al . found that there are different levels of crowding within the barrier . This is consistent with previous ideas of how the pore barrier could work , which propose that the nuclear transport receptors are less tightly packed in the center of the FG-network . This enables transport receptor and cargo complexes to move more rapidly through the center of the pore . The molecular crowding in the barrier of nuclear pores parallels that observed in other cellular compartments that also rely on assemblies of proteins with flexible structures . Thus , future work using p-PALM is expected to reveal more details about the biophysical properties of nuclear pores as well as those of other structures inside cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2017 | Investigating molecular crowding within nuclear pores using polarization-PALM |
In Arabidopsis , root hair and non-hair cell fates are determined by a MYB-bHLH-WD40 transcriptional complex and are regulated by many internal and environmental cues . Brassinosteroids play important roles in regulating root hair specification by unknown mechanisms . Here , we systematically examined root hair phenotypes in brassinosteroid-related mutants , and found that brassinosteroid signaling inhibits root hair formation through GSK3-like kinases or upstream components . We found that with enhanced brassinosteroid signaling , GL2 , a cell fate marker for non-hair cells , is ectopically expressed in hair cells , while its expression in non-hair cells is suppressed when brassinosteroid signaling is reduced . Genetic analysis demonstrated that brassinosteroid-regulated root epidermal cell patterning is dependent on the WER-GL3/EGL3-TTG1 transcriptional complex . One of the GSK3-like kinases , BIN2 , interacted with and phosphorylated EGL3 , and EGL3s mutated at phosphorylation sites were retained in hair cell nuclei . BIN2 phosphorylated TTG1 to inhibit the activity of the WER-GL3/EGL3-TTG1 complex . Thus , our study provides insights into the mechanism of brassinosteroid regulation of root hair patterning .
The Arabidopsis root epidermal cell types are defined by position in a predictable manner ( Ishida et al . , 2008 ) . Hair ( H ) cells , or trichoblasts , are specified early from root epidermal cells that lie over clefts between two underlying cortical cells , whereas the root epidermal cells that lie over a single cortical cell develop as non-hair ( N ) cells , or atrichoblasts ( Ishida et al . , 2008 ) . Hair cell and non-hair cell files are patterned alternately in rows within the Arabidopsis root epidermis , with columns of hair cells interspersed with columns of non-hair cells ( Schiefelbein et al . , 2009 ) . Prior to root hair outgrowth , root epidermal cells in the H position can be distinguished from those in the N position by many visible cellular features , including a greater rate of cell division ( Berger et al . , 1998 ) , reduced cell length and vacuolation ( Dolan et al . , 1994; Galway et al . , 1994 ) , and enhanced cytoplasmic density ( Dolan et al . , 1994 ) . It is proposed that positional signals and a putative receptor-like kinase SCRAMBLED ( SCM ) ( Kwak et al . , 2005 ) function through a MYB-bHLH-WD40 repeat transcriptional complex to determine root epidermal cell fate ( Schiefelbein et al . , 2009 ) . Based on this model , in N cells , WEREWOLF ( WER ) ( Lee and Schiefelbein , 1999 ) , a R2R3 MYB-domain transcription factor , forms a complex with basic helix-loop-helix transcription factors , GLABRA3 ( GL3 ) /ENHANCER OF GLABRA3 ( EGL3 ) ( Bernhardt et al . , 2003; Zhang et al . , 2003 ) , and a WD40 repeat protein , TRANSPARENT TESTA GLABRA1 ( TTG1 ) ( Galway et al . , 1994 ) , to promote expression of GLABRA2 ( GL2 ) and CAPRICE ( CPC ) ( Ryu et al . , 2005; Song et al . , 2011 ) . GL2 , a homeodomain/leucine zipper transcription factor , negatively regulates H cell fate and positively regulates N cell fate ( Masucci et al . , 1996 ) . CPC ( Wada et al . , 1997 ) , a MYB-type transcription factor , moves from N cells to H cells ( Kurata et al . , 2005a ) to compete with WER for binding to GL3/EGL3 to form a CPC-GL3/EGL3-TTG1 complex , which is unable to induce GL2 expression ( Song et al . , 2011 ) . In addition to CPC , the bHLH transcription factor GL3 is also a mobile protein ( Bernhardt et al . , 2005 ) . GL3 and its homologue EGL3 are both expressed in H cells , but GL3 protein is only localized in the N cell nucleus , indicating that GL3 protein moves into the adjoining N cell nucleus to determine N cell fate ( Bernhardt et al . , 2003 , 2005 ) . Integration of existing genetic and biochemical data also supports an alternative mechanism centered on the movement of transcriptional factors between epidermal cells rather than a putative local activation of the WER gene function to determine root epidermis pattern formation ( Savage et al . , 2008 ) . In addition , root hair development is highly regulated by many external and internal cues , including phytohormones . For instance , abscisic acid ( ABA ) plays a role in the early stage of root epidermal cell specification ( Van Hengel et al . , 2004 ) and in inhibiting root hair tip growth in Arabidopsis ( Schnall and Quatrano , 1992 ) , while both ethylene and auxin may act downstream of TTG1 and GL2 to promote root hair formation and elongation ( Masucci and Schiefelbein , 1994 , 1996 ) . Moreover , jasmonic acids ( JAs ) promote root hair formation through their interaction with ethylene ( Zhu et al . , 2006 ) . However , the underlying cellular and molecular mechanisms of how these internal hormones integrate with environmental cues to regulate root hair cell fate determination are still poorly understood . The plant steroid hormones , brassinosteroids ( BRs ) , play essential roles in regulating many developmental processes , including shoot , root , and reproductive development ( Savaldi-Goldstein et al . , 2007; Ye et al . , 2010; Hacham et al . , 2011; Yang et al . , 2011 ) . BRs are perceived by the receptor kinase BRASSINOSTEROID INSENSITIVE 1 ( BRI1 ) ( Li and Chory , 1997; Hothorn et al . , 2011; She et al . , 2011 ) . The BR-activated BRI1 phosphorylates BRI1 KINASE INHIBITOR 1 ( BKI1 ) to release its inhibition ( Wang and Chory , 2006 ) , and then BKI1 acts as a positive regulator by binding to a subset of 14-3-3 proteins ( Wang et al . , 2011 ) . Another BRI1 substrate , BR-SIGNALING KINASE ( BSK ) , transduces the BR signaling through bri1 SUPPRESSORS 1 ( BSU1 ) to inactivate a GSK3-like kinase BRASSINOSTEROID INSENSITIVE 2 ( BIN2 ) , which leads to accumulation of the dephosphorylated form of transcriptional factors BRI1 EMS SUPPRESSOR 1 ( BES1 ) /BRASSINAZOLE RESISTANT 1 ( BZR1 ) in the nucleus to regulate gene expression ( Yang et al . , 2011 ) . A previous study suggests that BRs play an important role in determining root epidermal cell fate by regulating WER and GL2 expression ( Kuppusamy et al . , 2009 ) . However , the elaborate molecular mechanism by which BRs regulate root epidermal cell fate and development is still unknown . Here , we first systematically examined root epidermal cell patterning and PGL2::GUS expression in a series of BR-deficient and signaling mutants . We found that BRs regulate root epidermal cell fate through promoting GL2 expression in both H and N cells , which is mediated by GSK3-like kinases and the WER-GL3/EGL3-TTG1 complex as indicated by genetic analysis and biochemical studies . Our study further demonstrated that BIN2 , one of the GSK3-like kinases , interacted with and phosphorylated EGL3 on T399 and T209/T213 , leading to its trafficking from nucleus to cytosol in H cells , which may facilitate its movement from H cells to N cells . BIN2 also phosphorylated TTG1 to inhibit the activity of the WER-GL3/EGL3-TTG1 transcriptional complex . These results explain how BR signaling regulates both the formation and activity of the WER-GL3/EGL3-TTG1 complex through GSK3-like kinases to coordinate root epidermal cell fate specification .
To broadly explore the role of BRs in root hair formation , we systematically examined the root hair phenotype of BR-biosynthetic mutants , det2-1 and cpd , and BR-responsive mutants , including bri1-116 , BRI1-OX ( a BRI1-overexpression line ) , bin2-3 bil1 bil2 ( a triple knockout mutant of BIN2 and its two closest homologues ) , and a BES1-RNAi line . We found that the relative hair number ( =root hair density × root hair cell length ) was higher in bri1-116 ( 4 . 67 ± 0 . 47 ) , det2-1 ( 4 . 75 ± 0 . 52 ) , and cpd ( 4 . 65 ± 0 . 54 ) , and significantly lower in BRI1-OX ( 3 . 40 ± 0 . 46 ) and bin2-3 bil1 bil2 ( 2 . 86 ± 0 . 39 ) than in their corresponding wild types Col-0 ( 3 . 89 ± 0 . 43 ) and WS-2 ( 3 . 85 ± 0 . 41 ) ( Figure 1—source data 1 ) . However , there was no significant difference between BES1-RNAi and Col-0 . Images at the highest magnification ( ×100; Figure 1 ) showed that in the wild type root ( Figure 1A ) , the H cells and N cells were arranged in alternating files with the H cell columns regularly interspaced with the N cell columns; no adjacent H cell columns were found . However , in the BR signaling-inhibited mutants , including bri1-116 , det2-1 , and cpd ( Figure 1B–D ) , many root hair columns were next to each other , leading to more root hairs , suggesting that some N cell fate might be changed into H cell fate . In contrast , the BR signaling-enhanced plants , including BRI1-OX and bin2-3 bil1 bil2 , grew fewer root hairs than the wild type , due to the fact that they lacked root hairs in many H cell positions ( Figure 1A , E–G ) . Interestingly , the BES1-RNAi line showed a similar root hair pattern as the wild type ( Figure 1A , H ) . 10 . 7554/eLife . 02525 . 003Figure 1 . Root epidermal cell patterning is altered in the BR-related mutants . ( A–H ) Root hair patterning of the BR-related mutants and their wild type counterparts . bin2-3 bil1 bil2 is in the WS-2 background , and all of the others are in the Col-0 background . ( I–L ) Root hair phenotype of the wild type plants grown on 1/2 MS ( Murashige and Skoog ) medium with DMSO ( mock ) ( I ) , 100 nM epibrassinolide ( eBL ) ( J ) , 30 μM bikinin ( Bik ) ( K ) , or 1 μM brassinazole ( Brz ) ( L ) . Right images are the outlined areas of left images with higher magnification . Arrows indicate ectopic root hair cells , and arrowheads indicate ectopic non-hair cells . Areas outlined with blue lines indicate the ectopic non-hair cells . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 00310 . 7554/eLife . 02525 . 004Figure 1—source data 1 . Root hair density , cell length , and relative hair number of the BR-related mutants and wild type plants treated with eBL , bikinin , Brz , or DMSODOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 00410 . 7554/eLife . 02525 . 005Figure 1—figure supplement 1 . Bikinin treatment inhibited H cell fate in bin2-3 bil1 bil2 mutants . Root hair phenotype of the bin2-3 bil1 bil2 mutants grown on medium with DMSO ( mock ) ( A ) or 30 μM bikinin ( B ) ; right images are the outlined areas of left images with higher magnification . ( C ) The root hair density , cell length , and relative hair number of the bin2-3 bil1 bil2 mutants treated with bikinin . Values are means ± SD . The two-tailed t test with equal variance or unequal variance was used to determine the significance level of the difference between the bin2-3 bil1 bil2 mutants treated with 30 μM bikinin and mock medium . *p<0 . 05; **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 005 To further test whether exogenously applied eBL ( epibrassinolide ) , bikinin ( a specific GSK3 kinase inhibitor ) ( De Rybel et al . , 2009 ) , or Brz ( brassinazole , an inhibitor of BR biosynthesis ) regulate root hair specification , we planted seeds on 1/2 MS ( Murashige and Skoog ) medium containing each of these chemicals or DMSO ( as the mock treatment ) , and carefully observed their root hair phenotypes . We found that compared with plants grown on the mock medium ( 3 . 80 ± 0 . 45 ) , the relative root hair number of plants grown on medium containing 100 nM eBL ( 3 . 22 ± 0 . 42 ) or 30 μM bikinin ( 2 . 71 ± 0 . 50 ) was significantly reduced , while it was greatly increased in plants grown on medium containing 1 μM Brz ( 4 . 17 ± 0 . 43 ) ( Figure 1—source data 1 ) . Images at higher magnification ( ×100 ) show that , compared to seedlings grown on the mock medium ( Figure 1I ) , those grown on medium containing eBL or bikinin produced fewer root hairs in the H position of epidermal cells ( Figure 1J , K ) , while those grown on medium containing Brz grew more root hairs in the N position ( Figure 1L ) . We also found that the bin2-3 bil1 bil2 seedlings grown on medium containing 30 μM bikinin produced very few root hairs ( Figure 1—figure supplement 1 ) , suggesting that besides BIN2 , BIL1 , and BIL2 , other GSK3-like kinases may also be involved in root hair specification in Arabidopsis . Taken together , these findings indicate that the BR-mediated root epidermal cell pattern formation largely relies on GSK3-like kinases and/or their upstream components . GL2 has been widely used as a molecular maker of N cell fate determination ( Masucci et al . , 1996; Kuppusamy et al . , 2009 ) . In order to test whether the disordered root hair patterning in the BR-related mutants resulted from an altered root epidermal cell fate , we analyzed the GL2 expression pattern in these mutants using PGL2::GUS as a reporter . We found that in the wild type , 1 . 3% of epidermal cells in the N position lacked GL2 expression in cross-sections ( Figure 2A , B ) , and in the longitudinal view of root epidermis , the root epidermal cell files were arranged regularly , with GL2-expressing columns ( N cell columns ) interspaced with columns without GL2 expression ( H cell columns ) ( Figure 2C ) . However , about 15 . 3% of bri1-116 and 18 . 8% of det2-1 cells showed suppressed PGL2::GUS expression , and there were adjacent root epidermal cells without GL2 expression in both bri1-116 and det2-1 ( Figure 2D–I ) , which supports a previous finding with bri1-116 and the Brz-treated wild type ( Kuppusamy et al . , 2009 ) . These results indicated that adjacent root hairs in bri1-116 and det2-1 were caused by some N cell fate changing to H cell fate . In contrast , GL2 was ectopically expressed in about 13 . 5% of H cells in BRI1-OX plants ( Figure 2J–L ) and in about 19 . 6% in bin2-3 bil1 bil2 ( Figure 2M–O ) , as compared with only 3 . 2% in the wild type ( Figure 2A–C ) , indicating that lack of GL2 expression in N cells may correspond to the ectopic root hairs observed in bri1-116 and det2-1 , and that the ectopically expressed GL2 in H cells partially inhibits H cell fate in BRI1-OX and bin2-3 bil1 bil2 plants . Taken together , the above results suggested that BR signaling has an important role in suppressing H cell fate and promoting N cell fate in both the N and the H positions , and BR signaling regulates root epidermal cell fate by controlling GL2 expression through GSK3-like kinases , or their upstream components , but not through downstream transcription factors . 10 . 7554/eLife . 02525 . 006Figure 2 . Expression pattern of PGL2::GUS is altered in the BR-related mutants . Transverse sections from root meristem of Col-0 ( A ) , bri1-116 ( D ) , det2-1 ( G ) , BRI1-OX ( J ) , and bin2-3 bil1 bil2 ( M ) . Frequency of cells without PGL2::GUS expression in the N cell position ( open bars ) and cells with ectopically expressed PGL2::GUS in the H cell position ( solid bars ) of Col-0 ( B ) , bri1-116 ( E ) , det2-1 ( H ) , BRI1-OX ( K ) , and bin2-3 bil1 bil2 ( N ) . Longitudinal images of the root epidermal cells in Col-0 ( C ) , bri1-116 ( F ) , det2-1 ( I ) , BRI1-OX ( L ) , and bin2-3 bil1 bil2 ( O ) . Scale bars , 25 μm . Red arrows indicate N cells without PGL2::GUS expression , and red arrowheads indicate H cells ectopically expressing PGL2::GUS . For each genotype , n = 8 . Error bars indicate standard deviation ( SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 006 It was known that GL2 expression is directly regulated by the WER-GL3/EGL3-TTG1 but not the CPC-GL3/EGL3-TTG1 transcriptional complex ( Schiefelbein et al . , 2009 ) . To explore whether the BR-regulated GL2 expression and root epidermal cell fate determination are dependent on these complexes , we first created a set of double mutants of cpc-1 , a mutant with fewer root hairs than its counterpart ( Figure 3A , B ) , with bri1-116 or cpd ( Figure 3C , D ) . Similar to cpc-1 , both double mutants bri1-116 cpc-1 ( Figure 3E ) and cpd cpc-1 ( Figure 3F ) produced few root hairs . We also generated double or multiple mutants of wer-1 , a mutant with more root hairs than Col-0 ( Figure 3G , H ) , with BRI1-OX or bin2-3 bil1 bil2 ( Figure 3I , J ) , and found that both BRI1-OX wer-1 and bin2-3 bil1 bil2 wer-1 ( Figure 3K , L ) were similar to wer-1 , with many ectopic root hairs formed at the N cell position . These genetic analyses indicated that the WER-GL3/EGL3-TTG1 and CPC-GL3/EGL3-TTG1 transcriptional complexes act downstream of BR early signaling . 10 . 7554/eLife . 02525 . 007Figure 3 . BR signaling acts upstream of CPC and WER to regulate root epidermal cell fate . Root hair phenotype of the wild type WS-2 ( A ) and double mutants of cpc-1 ( B ) with bri1-116 ( C ) or cpd ( D ) , including bri1-116 cpc-1 ( E ) and cpd cpc-1 ( F ) . Root hair phenotype of the wild type Col-0 ( G ) and the double/multiple mutants of wer-1 ( H ) with BRI1-OX ( I ) or bin2-3 bil1 bil2 ( J ) , including BRI1-OX wer-1 ( K ) and bin2-3 bil1 bil2 wer-1 ( L ) . cpc-1 is in the WS-2 background . BR: brassinosteroid . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 007 Therefore , we inferred that BR-mediated root epidermal cell fate may be dependent on GSK3-like kinases , the key negative components in the BR signaling pathway , acting upstream of the WER-EGL3/GL3-TTG1 or CPC-EGL3/GL3-TTG1 transcriptional complex . We then conducted yeast two-hybrid assays to test whether any components in the WER-EGL3/GL3-TTG1 or CPC-EGL3/GL3-TTG1 complex interact with BIN2 , a well-studied GSK3-like kinase , and found that BIN2 can interact with EGL3 ( Figure 4A ) but not with CPC in yeast ( Figure 4—figure supplement 1 ) . However , due to strong auto-activation of WER and TTG1 fused with GAL4-DNA binding domain ( DB ) in yeast two-hybrid assays ( Figure 4—figure supplement 1 ) , we conducted GST pull-down and BiFC ( biomolecular fluorescence complementation ) assays to test their interactions , and found that BIN2 can interact with WER , EGL3 , and TTG1 ( Figure 4B , C ) . Furthermore , because BIN2 can regulate many transcription factors by phosphorylation ( Saidi et al . , 2012 ) , and WER or CPC can interact with EGL3/GL3-TTG1 in vivo to form WER-EGL3/GL3-TTG1 or CPC-EGL3/GL3-TTG1 complexes , respectively ( Zhao et al . , 2008; Song et al . , 2011 ) , we then conducted in vitro kinase assays to test whether BIN2 can phosphorylate any of these components . We found that BIN2 did not phosphorylate WER and CPC , but was able to phosphorylate EGL3 and TTG1 ( Figure 5 ) . 10 . 7554/eLife . 02525 . 008Figure 4 . BIN2 interacts with EGL3 , TTG1 , and WER . ( A ) BIN2 interacts with EGL3 in yeast two-hybrid assays . ( B ) The interaction of BIN2-His with CPC-GST , EGL3-GST , TTG1-GST , and WER-GST in vitro . The BIN2-His pulled-down by CPC-GST , EGL3-GST , TTG1-GST , and WER-GST , or GST was detected by western blotting with anti-His antibody ( top ) . The purified BIN2-His protein was used as inputs . An equal loading of recombinant proteins was indicated by Coomassie brilliant blue ( CBB ) staining ( bottom ) . ( C ) BiFC assays of the interaction between BIN2 with EGL3 , TTG1 , and WER . Scale bars , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 00810 . 7554/eLife . 02525 . 009Figure 4—figure supplement 1 . Yeast two-hybrid assays to test interactions of BIN2 with WER , TTG1 , or CPC . The full-length cDNA of each corresponding gene was fused with the GAL4-BD or AD domain . Yeast cells harboring the indicated constructs were grown on the synthetic media lacking Trp and Leu , or Trp , Leu , and His with an additional 1 mM 3AT . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 00910 . 7554/eLife . 02525 . 010Figure 5 . BIN2 phosphorylates EGL3 and TTG1 , but not WER and CPC in vitro . An equal amount of recombinant BIN2 kinase indicated by Coomassie brilliant blue ( CBB ) staining ( bottom ) was incubated with recombinant MBP , WER-MBP , CPC-MBP , EGL3-MBP , or TTG1-MBP , separated by SDS–PAGE , and followed by autoradiography ( top ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 01010 . 7554/eLife . 02525 . 011Figure 5—figure supplement 1 . Mass spectrometry analysis of EGL3 phosphorylation sites . Putative phosphorylation sites are in T399PEET403 ( A ) and T209TIST213 fragments ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 011 To investigate the biological relevance of EGL3 phosphorylation by a GSK3-like kinase BIN2 , we used mass spectrometry and identified four potential phosphorylation sites ( T209 , T213 , T399 , and T403 ) of EGL3 by BIN2 ( Figure 5—figure supplement 1 ) , which are located in two regions that contain typical recognition sites of GSK3 kinases ( Cohen and Frame , 2001 ) . We then mutated threonine residues into alanine to make single- or double-site mutated forms of EGL3 . In vitro phosphorylation assays showed that phosphorylation levels of EGL3T399A and EGL3T209A/T213A by BIN2 were significantly reduced ( Figure 6A ) , indicating that T399 and T209/T213 are the main phosphorylated residues . To investigate the biological function of EGL3 phosphorylation , we transformed EGL3 and its mutated forms driven by its own promoter into Col-0 to examine their subcellular localization . We found that the EGL3-GFP signal in N cells was apparently higher than that in H cells , and that it was mainly localized to cytosol in H cells , but to both cytosol and nucleus in N cells ( Figure 6B ) . As EGL3 mRNA is only expressed in H cells ( Bernhardt et al . , 2005 ) , this result indicated that , like its homologue GL3 , EGL3 is also a mobile protein that moves from H cells to N cells . However , the EGL3T399A-GFP and EGL3T209A/T213A-GFP were solely localized to the nuclei of H cells ( Figure 6C , D ) , indicating that EGL3 phosphorylation is required not only for its cytoplasmic accumulation , but also for its movement from H cells to N cells . In addition , we found that the root hair patterning of EGL3-GFP transgenic plants was not altered ( Figure 6E; Table 1 ) , indicating that correctly localized EGL3 , mainly in N cell nuclei to promote N cell fate and less in H cell nuclei not to promote N cell fate , has no influence on root epidermal fate . Moreover , although root epidermal patterning in the EGL3T399A-GFP transgenic plants was normal ( Figure 6F; Table 1 ) , the number of root hairs in EGL3T209A/T213A-GFP plants was significantly reduced , likely due to a misspecification of non-hair cells in the H position ( Figure 6G; Table 1 ) , suggesting that the nucleus-localized EGL3 in H cells may determine N cell fate specification . Taken together , our results indicate that BIN2 phosphorylation on T399 , T209 , and/or T213 of EGL3 in H cells promotes EGL3 cytoplasmic localization , which likely helps its movement from H to N cells to regulate root epidermal cell fate . 10 . 7554/eLife . 02525 . 012Figure 6 . BIN2 phosphorylates EGL3 to regulate its subcellular localization and root epidermal cell fate . ( A ) BIN2 phosphorylates EGL3 on T399 and T209/T213 . An equal amount of recombinant protein , as indicated by Coomassie brilliant blue ( CBB ) staining ( bottom panel ) , was incubated in phosphorylation buffer , separated by SDS–PAGE , and followed by autoradiography ( top panel ) . ( B ) EGL3-GFP is predominantly localized in N cell nuclei . Both EGL3T399A-GFP ( C ) and EGL3T209A/T213A-GFP ( D ) are solely localized in H cell nuclei . For ( B–D ) , the 5-day-old roots were stained with propidium iodide ( red ) for 10 s for visualizing the cell wall . The top panels show the underlying cortex . The stars indicate H cells . Scale bars , 20 μm . ( E–G ) Root hair patterns of EGL3-GFP ( E ) , EGL3T399A-GFP ( F ) , and EGL3T209A/T213A-GFP ( G ) transgenic plants . Outlined areas in the left images are magnified in the right images . Red arrowheads and areas outlined with blue lines indicate ectopic non-root hair cells in the H position . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 01210 . 7554/eLife . 02525 . 013Figure 6—source data 1 . EGL3 amino acid sequence analysisDOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 01310 . 7554/eLife . 02525 . 014Figure 6—figure supplement 1 . Alignment of EGL3 amino acid sequence with other bHLH homologues in Arabidopsis . The sequences underlined in red indicate the 209TTIST213 and 399TPEET403 regions of EGL3 . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 01410 . 7554/eLife . 02525 . 015Table 1 . The effect of EGL3 and its two mutated forms on root epidermal cell pattern formationDOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 015GenotypeCells in the H positionCells in the N positionHair cells ( % ) Non-hair cells ( % ) Hair cells ( % ) Non-hair cells ( % ) Col-098 . 9 ± 3 . 31 . 1 ± 3 . 32 . 0 ± 4 . 598 . 0 ± 4 . 5PEGL3::EGL3-GFP95 . 8 ± 6 . 14 . 2 ± 6 . 11 . 9 ± 4 . 298 . 1 ± 4 . 2PEGL3::EGL3T399A-GFP93 . 1 ± 6 . 46 . 9 ± 6 . 42 . 8 ± 6 . 097 . 2 ± 6 . 0PEGL3::EGL3T209A/T213A-GFP85 . 5 ± 4 . 4*14 . 5 ± 4 . 4*1 . 0 ± 3 . 299 . 0 ± 3 . 2At least 10 different 5-day-old roots were examined for each strain . Values represent means ± SD . For statistical analysis , the F test was used to determine the variance , and the two-tailed t test with equal variance or unequal variance was used to determine the significance level of the difference among the transgenic plants . *p<0 . 05 . TTG1 is required for a normal expression level of GL2 ( Di Cristina et al . , 1996 ) but not for its expression pattern ( Hung et al . , 1998 ) . Transgenic TTG1-GFP plants driven by its own promoter indicated that TTG1 was preferentially localized in the cytoplasm and slightly in the nucleus of both N and H cells ( Figure 7—figure supplement 1 ) , which was consistent with the subcellular localization of its petunia homologue AN11 ( De Vetten et al . , 1997 ) . To understand the biological relevance of TTG1 phosphorylation by a GSK3-like kinase , we conducted transient transcription assays in Nicotiana benthamiana leaves to examine whether BIN2 regulated the complex's activity in a TTG1-dependent manner . We constructed a dual-luciferase reporter system using PGL2::LUC as a reporter gene and 35S::REN as an internal control ( Figure 7A ) . Because the protein of the gain-of-function mutation bin2-1 ( E263K ) is more stable and has higher activity than the wild type BIN2 ( Peng et al . , 2008 ) , and GSK3-like kinases are quite conserved among different species ( Saidi et al . , 2012 ) , we used bin2-1 to conduct this study . As shown in Figure 7B , transient expression of WER alone was able to slightly induce PGL2::LUC gene expression . In contrast , transient expression of EGL3 alone was unable to induce reporter gene expression . Co-expression of WER and EGL3 can dramatically promote LUC expression , which is consistent with a previous study ( Song et al . , 2011 ) . Additional bin2-1 did not alter the effect of WER , EGL3 , or both WER and EGL3 on PGL2::LUC expression . When WER , EGL3 , and TTG1 were used together , the expression of PGL2::LUC was further enhanced , indicating that TTG1 can promote the activity of the WER-EGL3 complex . Interestingly , additional bin2-1 significantly inhibited reporter gene expression regulated by the WER-EGL3-TTG1 complex ( Figure 7B ) , indicating that TTG1 is mediating the negative effect of BIN2 on this transcriptional complex . 10 . 7554/eLife . 02525 . 016Figure 7 . BIN2 inhibits the transcription activity of the WER-EGL3-TTG1 complex through TTG1 . ( A ) Schematic diagram of the dual-luciferase reporter construct . The firefly luciferase ( LUC ) reporter gene was driven by GL2 promoter . The Renillia luciferase ( REN ) reporter gene was controlled by Cauliflower mosaic virus promoter ( 35S ) and terminator ( Ter ) . ( B ) bin2-1 inhibits PGL2::LUC expression only when TTG1 is co-expressed with WER and EGL3 . Relative reporter activity in Nicotiana benthamiana leaf cells transiently transformed with the indicated effector , reporter , and regulatory constructs . G , W , E , and T indicate GL2 , WER , EGL3 , and TTG1 , respectively . Error bars indicate SD . **p<0 . 01 determined by the two-tailed Student's t test . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 01610 . 7554/eLife . 02525 . 017Figure 7—figure supplement 1 . Subcellular localization of TTG1-GFP in Col-0 root epidermal cells . Stars indicate H cells . Scale bars , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 017
We provide several lines of evidence to strongly support an important role for BR signaling in directly regulating root hair cell fate . First , root hair patterning in the BR-biosynthetic and responsive mutants or in the wild type grown on eBL , bikinin , or BRZ , was dramatically altered , demonstrating that GSK3 kinases and/or their upstream components are mediating this regulation . Second , the expression pattern of the non-hair cell fate marker PGL2::GUS indicates that BR early signaling promotes N cell fate in the whole root epidermis , which is a reasonable explanation for the abnormal root hair patterning in the BR-related mutants: when BR signaling is enhanced , fewer root hairs are formed in the H position; when BR signaling is inhibited , more ectopic root hairs are produced in the N position . This finding supports the previous report that BRs positively regulate the expression of WER and GL2 ( Kuppusamy et al . , 2009 ) . Third , genetic analysis revealed that major components of the WER-EGL3-TTG1 or CPC-EGL3-TTG1 complex act downstream of BR signaling-mediated root epidermis patterning . Finally , BIN2 phosphorylation on EGL3 and TTG1 suggested that GSK3-like kinases directly regulate EGL3 movement and the transcription activity of the WER-EGL3-TTG1 complex to mediate root hair development . This study revealed a key mechanism in the regulation of intercellular communication of transcription factors by a hormonal signal to determine epidermal cell specification . Non-cell autonomous movement of some transcriptional factors is an important mechanism to regulate some developmental processes ( Kurata et al . , 2005b ) , and cytoplasmic localization of these mobile proteins may be required for their intercellular movement . For example , in maize shoot apical meristem , a mutation in its potential nuclear localization signal ( NLS ) of KNOTTED 1 abolished its intercellular movement ( Vollbrecht et al . , 1991; Lucas et al . , 1995 ) , and cytoplasmic localization of LEAFY ( Schultz and Haughn , 1991 ) , a transcriptional factor in floral identity , is also strongly correlated with its adjacent cell movement ( Wu et al . , 2003 ) . Moreover , the movement of SHORT-ROOT , another mobile transcriptional factor in root radical patterning ( Helariutta et al . , 2000 ) , was abolished when it was fused to a NLS , leading to its diminished cytoplasmic localization ( Gallagher et al . , 2004 ) . However , it is largely unknown how the nuclear-cytoplasmic trafficking of transcription factors is regulated by internal cues to influence their intercellular movement . Although the intercellular movement of mobile factors in the WER/CPC-EGL3/GL3-TTG1 complex may determine root epidermis patterning ( Savage et al . , 2008 ) , it is not clear how their movement is regulated by internal cues . Because EGL3 mRNA was expressed only in H cells , the major nuclear localization of EGL3 protein in N cells indicated that , like GL3 and CPC , EGL3 can also move from H to N cells . In addition , we found that EGL3T399A and EGL3T209A/T213A with abolished phosphorylation sites were solely localized in the nucleus of H cells , indicating that the unphosphorylated EGL3 may not move between cells . Although we do not know how T209/T213 phosphorylation regulated EGL3 subcellular localization , the T399 phosphorylation likely affected a NLS , because T399 is located in a predicted non-canonical NLS ( Figure 6—source data 1 ) , which is found in many other plant bHLHs ( Galstyan et al . , 2012 ) . Although the EGL3T399A-GFP plants showed normal root hair patterning , this can be explained by the close proximity of T399 to its bHLH domain , which may have affected its interaction with TTG1 and MYB ( Zhang et al . , 2003 ) , leading to the nucleus-localized EGL3T399A in H cells unable to induce GL2 expression . However , EGL3T209A/T213A-GFP may still interact with TTG1 and have DNA-binding activity , because T209/T213 was located in the N-terminal region far away from the bHLH domain ( Figure 6—figure supplement 1 ) , which led to GL2 expression in H cells and EGL3T209A/T213A-GFP plants growing fewer root hairs ( Figure 6G; Table 1 ) . Besides the regulation of EGL3 nuclear-cytoplasmic trafficking , we also provided strong evidence to support GSK3-like kinases’ inhibition on the transcriptional activity of the WER-EGL3-TTG1 complex through TTG1 . In Nicotiana benthamiana pavement cells , it was demonstrated that BIN2 has a negative role in WER-EGL3-TTG1 transcriptional activity , but has no effect on the activity of WER , EGL3 , or both WER and EGL3 . Although BIN2 phosphorylates EGL3 , its failure to regulate WER-EGL3 transcriptional activity can be explained by a possible ubiquitous expression of WER , EGL3 , and GSK3-like kinases in Nicotiana benthamiana leaves . Furthermore , it was reported that TTG1 interacts with EGL3 ( Zhang et al . , 2003 ) , and TTG1 is necessary for the full functioning of other bHLH partners , such as GL3 and TRANSPARENT TESTA8 ( Baudry et al . , 2004; Zhao et al . , 2008 ) . Therefore , it is very likely that TTG1 phosphorylation by GSK3-like kinases may affect its regulation of EGL3 and the activity of the WER-EGL3-TTG1 complex . Our data also support the suggestion that the N cell is a default cell type in root epidermis , and that H cell fate is produced due to inhibition of N cell fate by internal or external cues . First , we observed that WER , a positive regulator for GL2 expression , is expressed in both N and H cells in the early root meristem ( Figure 8—figure supplement 1 ) , which is also supported by a previous report that WER exhibits uniform promoter activity in both N and H cells proximal to the initial cells ( Savage et al . , 2008 ) . Second , in Arabidopsis , both EGL3 and GL3 are expressed in H cells , but their proteins move to adjacent cells to promote N cell fate ( Bernhardt et al . , 2005 ) . If they stay in H cells with the ability to interact with WER and TTG1 , the H cells may develop into N cells as shown in the EGL3T209A/T213A-GFP transgenic plants . Moreover , over-expression of GL3 and EGL3 promoted non-hair cell fate ( Bernhardt et al . , 2003 ) . Third , TTG1 is localized in both N and H cells , and TTG1 and EGL3 may synergistically promote WER-EGL3 transcriptional activity and enhance N cell fate . Apparently , BR signaling can promote N cell fate in several ways . Besides the inhibition of BR signaling on EGL3 cell–cell movement and the promotion of TTG1 activity , BR signaling also promotes WER expression as WER up-regulation in bin2-3 bil1 bil2 ( Figure 8—figure supplement 1 ) , which is consistent with the positive role of BRs in WER expression ( Kuppusamy et al . , 2009 ) . Thus , we proposed a model to illustrate how BR signaling regulates WER-EGL3-TTG1 complex formation and activity to control root epidermal cell fate . As shown in Figure 8 , without BRs , WER-GL3/EGl3-TTG1 complex formation and activity is inhibited in both N and H cells , as WER expression is reduced in both H and N cells , and the activated GSK3-like kinases phosphorylate EGL3 in H cells to promote its cytoplasmic localization in both H cells and N cells , both of which lead to less WER-GL3/EGL3-TTG1 complex formation in nuclei and suppression of GL2 expression . The activity of some formed WER-bHLH-TTG1 complexes may be further inhibited by GSK3-like kinases phosphorylating TTG1 . In contrast , enhanced BR early signaling inhibits GSK3-like kinases , leading to nuclear accumulation of the unphosphorylated EGL3 in H cells and normal function of unphosphorylated TTG1 in both cell types . Although CPC can move into H cells , due to enhanced WER expression and more efficient interaction of EGL3 with WER than with CPC ( Song et al . , 2011 ) , more WER-EGL3-TTG1 complex is formed in H cells to promote GL2 expression and determine N cell fate . In N cells , the nucleus-localized GL3 can interact with WER and TTG1 to promote GL2 expression and maintain N cell fate . However , it remains to be investigated how TTG1 and WER expression is regulated by BR signaling . It is also not clear how BR signaling coordinates with positional signals and other phytohormones to regulate root hair patterning . 10 . 7554/eLife . 02525 . 018Figure 8 . A proposed model to illustrate how BR signaling regulates root epidermal cell fate . Without BR early signaling , WER expression is reduced , and the activated GSK3-like kinases phosphorylate EGL3 and TTG1 in both H and N cells , leading to reduced formation and/or activity of the WER-EGL3/GL3-TTG1 complex , which inhibits GL2 expression in some N cells . With enhanced BR early signaling , WER expression is enhanced in both H and N cells , and the GSK3-like kinases activity is inhibited , leading to reduced phosphorylation of EGL3 and TTG1 in both cell types . Thus , WER-EGL3-TTG1 and WER-GL3-TTG1 complexes with transcriptional activity are formed in H and N cells , respectively , to promote GL2 expression and non-root hair cell fate . BR: brassinosteroid . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 01810 . 7554/eLife . 02525 . 019Figure 8—figure supplement 1 . WER expression pattern in the root early meristem and its expression level in the bin2-3 bil1 bil2 and wild type Col-0 . ( A ) Transverse section from the root meristem of the PWER::GUS transgenic plant . Scale bar , 25 μm . ( B ) The WER expression level was enhanced in the bin2-3 bil1 bil2 mutant . CPD ( CONSTITUTIVE PHOTOMORPHOGENIC DWARF ) , a BR biosynthetic gene feedback inhibited by BR signaling , was used as a control . The expression level of CPD and WER in WS-2 was normalized to ‘1’ , and a U-BOX gene ( At5g15400 ) was used as an internal control . Error bars indicate SD . **p<0 . 01 with a two-tailed Student's t test . BR: brassinosteroid . DOI: http://dx . doi . org/10 . 7554/eLife . 02525 . 019
The seeds of the wer-1 and PGL2::GUS lines were obtained from Dr John Schiefelbein ( University of Michigan ) , the bin2-3 bil1 bil2 seeds were obtained from Jianming Li ( University of Michigan ) , and the cpc-1 seeds ( CS6399 ) were obtained from the Arabidopsis Biological Resource Center ( Ohio State University ) . Combinations of the BR-related mutants with the root hair mutants or the PGL2::GUS line were generated by crossing and selected by GUS staining based on the mutant phenotype , or antibiotic selection marker analysis . For root hair observation , seeds were grown on 1/2 MS medium ( pH 5 . 8 ) with 1% sucrose , chilled for 3 d at 4°C , and grown for 5 d at 23°C under long-day conditions ( 16 hr light/8 hr dark ) . Nicotiana benthamiana plants were grown at 28°C under long-day conditions ( 16 hr light/8 hr dark ) . The double mutants or multiple mutants were derived from genetic crosses of the parental mutants ( or transgenic lines ) . For generation of the BRI1-OX wer-1 and bil1 bil2 bil3 wer-1 double/multiple mutants , the wer-1 was genotyped with its point mutation-derived cleaved amplified polymorphic sequence ( CAPS ) marker ( Lee and Schiefelbein , 1999 ) ( Supplementary file 1 ) , the bin2-3 bil1 bil2 was genotyped as described ( Yan et al . , 2009 ) , and the BRI1-OX was selected by the antibiotic selection markers . For generation of bri1-116 cpc-1 and cpd cpc-1 double mutants , the cpc-1 was identified by PCR and phenotypic analysis , and the bri1-116 and the cpd were isolated by phenotype . For GST pull-down assays , BIN2 was cloned into the PET28a vector , and EGL3 was cloned into the pGEX4T-1 vector . For in vitro kinase assays , WER , TTG1 , CPC , and EGL3 were cloned into the pMAL-C2X vector . His-fused BIN2 ( BIN2-His ) , GST-fused EGL3 ( EGL3-GST ) , and MBP-fused WER ( WER-MBP ) , TTG1 ( TTG1-MBP ) , CPC ( CPC-MBP ) , and EGL3 ( EGL3-MBP ) were expressed in BL21 ( DE3 ) pLySs strain and purified with either Ni-NTA agarose ( Clontech , Mountain View , CA ) , glutathione resin ( Genescript , Piscataway , NJ ) , or amylase resin ( NEB , Ipswich , MA ) , respectively . To generate plants expressing GFP-tagged EGL3 or mutated EGL3T399A and EGL3T209A/T213A , the various EGL3 cDNAs were cloned in-frame with GFP into the pCAMBIA2302 vector and driven by the EGL3 promoter ( 2 kb upstream of the start codon ) , and the plants expressing GFP-tagged TTG1 were generated by cloning the TTG1 cDNA in-frame with GFP into the pCAMBIA2302 vector and driven by the TTG1 promoter ( 2 kb upstream of the start codon ) . The constructs were transformed into Agrobacterium tumefaciens GV3101 strains . All transgenic plants were generated by floral dip transformation . T0 seeds were harvested and screened by germinating on MS solid medium with antibiotic selection . For each transformation , at least five individual T1 transgenic lines were selected . Transgenic lines of PEGL3::EGL3-GFP , PEGL3::EGL3T399A-GFP , PEGL3::EGL3T209A/T213A-GFP , and PTTG1::TTG1-GFP with T2 or higher generations were used for further analysis . The root hair pattern of the 5-day-old seedlings was observed , and images at ×100 magnification were taken with a Leica MZ FLIII stereomicroscope ( Leica Microsystems ) . The root hair density was counted as described ( Galway et al . , 1994; Jones et al . , 2002 ) with some modifications . Any visible protrusion from the epidermal cell was regarded as a root hair , regardless of length . The number of root hairs was counted from one side of a 1 mm segment from the imitated differentiation region of the 5-day-old roots , and at least eight roots were measured for each stain . The hair cell length was measured along the longitudinal plane at ×100 magnification using the software Scion Image , and at least 10 root hair cells were measured for each root . The relative root number was calculated as root hair density × root hair cell length for each root as described ( Wada et al . , 1997 ) . The histochemical staining of 5-day-old roots harboring the GUS reporter was performed as described ( Masucci et al . , 1996 ) . Transverse sections of root meristem were prepared as described ( Ye et al . , 2010 ) with modifications . The proportion of cells expressing or not expressing PGL2::GUS reporter in H cells or N cells was measured by examining sections at least from eight seedlings in each strain . For protein localization of EGL3-GFP and its mutated forms , the 5-day-old transgenic plants were examined by confocal microscope ( Zeiss ) after staining with 5 μg/ml propidium iodide ( PI ) ( Sigma , St . Louis , MO ) for 10 s at room temperature , and images were captured at 489 nm and 538 nm laser excitation and at 509 nm and 617 nm emission for GFP and PI staining . The pattern of epidermal cell types was determined as described ( Lee and Schiefelbein , 2002 ) . For yeast two-hybrid assays , the full length cDNA of BIN2 was cloned into vector pEXP-AD502 ( BIN2-AD ) and used as a prey , and the full length cDNAs of EGL3 , WER , TTG1 , and CPC were cloned into the pDBLeu vector ( EGL3-DB , WER-DB , TTG1-DB , CPC-DB ) , respectively , and used as a bait . The prey and bait plasmids were transformed into the yeast strains AH109 and Y187 , respectively . After yeast mating , the protein–protein interactions were tested on SD medium minus Leu , Trp , and His , and containing 2 mM 3-amino-1 , 2 , 4-triazole ( 3AT ) ( Sigma , St . Louis , MO ) . To generate the vector system for BiFC analysis , the full length cDNAs of EGL3 , WER , TTG1 , and CPC were cloned into the pXY104 vector ( cYFP ) , respectively , to generate EGL3-cYFP , WER-cYFP , TTG1-cYFP , and CPC-cYFP constructs , and BIN2 cDNA was cloned into the pXY106 ( nYFP ) vector to generate BIN2-nYFP construct . For transient expression , Agrobacterium strains ( GV3101 ) carrying the constructs for testing the specific interaction were transformed into 4–5-week-old Nicotiana benthamiana leaves as described previously ( Walter et al . , 2004 ) . After infiltration for 4 d , the lower leaf epidermis cells were used for analyzing the fluorescence signal by confocal microscopy ( Zeiss ) . For dual-luciferase assays , cDNAs of the effectors WER , EGL3 , and TTG1 , and the regulator bin2-1 were cloned with Flag tag into pCAMBIA2302 driven by a 35S promoter . GL2 promoter ( 2 kb upstream of the start codon ) was cloned into the pGreenII 0800-LUC vector to be used as the reporter . The method of transient expression used was as previously described ( Hellens et al . , 2005 ) . The purified proteins , EGL3-GST , WER-GST , TTG1-GST , CPC-GST , and GST , were bound with 25 μl GST resin in binding buffer ( 10 mM phosphate buffer , pH 7 . 4 , 140 mM NaCl , 3 mM KCl , 0 . 1% Triton X-100 ) for 2 hr at 4°C . After washing three times with the binding buffer , an equal amount of BIN2-His was added and rebound for 2 hr at 4°C . After boiling in SDS loading buffer for 5 min , the pull-down proteins were separated on 10% SDS–PAGE gels and detected by immunoblotting with anti-His antibody ( Abmart , Shanghai , China ) . In vitro kinase assays were performed in 24 μl reaction buffer ( 20 mM Tris , pH 7 . 5 , 10 mM MgCl2 , 5 mM DTT ) containing 20 μM ATP and 1 μl of 10 μCi [32P] γATP ( PerkinElmer , Waltham , Massachusetts ) and purified proteins . The reaction was carried out at 30°C for 1 hr and terminated by adding 6 μl of 5 × SDS loading buffer . After boiling for 3 min , proteins were separated on 10% SDS–PAGE . Gels were stained with Coomassie brilliant blue , and then dried and autoradiographed . For phosphorylation site identification , in vitro kinase assays were performed , and protein bands were excised to be used for mass spectrometry analysis . The mutated forms of EGL3 were generated by a PCR-based site-directed mutagenesis ( Supplementary file 1 ) . | Roots anchor a plant into the ground , and allow the plant to absorb water and mineral nutrients from the soil . As roots grow and branch , they increase the surface area of root exposed to the soil—and many plant cells in the root's outer layer have a hair-like projection to further increase this surface area . Thus , root hairs are where most water and mineral nutrients are absorbed . Many factors affect whether , or not , a plant cell will develop into a root hair . These factors include both external cues ( such as the mineral content of the soil ) and signals from the plant itself ( such as hormones ) . Brassinosteroids are plant hormones that regulate the development of shoots and roots , as well as the timing of when flowers begin to develop . These hormones are detected on the outside of plant cells , and activate a signaling pathway within the cell that causes changes in gene expression . Brassinosteroids also control if a root cell will become a hair cell or not , although the mechanism behind this activity is unclear . Here , Cheng et al . have looked at the root hairs of mutant Arabidopsis thaliana plants that have had individual genes involved in brassinosteroid signaling knocked-out . Plant biologists commonly study this plant species because it is small and grows quickly—and Arabidopsis has regular stripes of root hair cells and ‘non-hair cells’ in the outer layer of its roots . Cheng et al . reveal that brassinosteroids prevent the formation of root hairs via signaling pathways that involve proteins called GSK3-like kinases . These hormones ‘switch off’ these kinases’ activity , so knocking-out the genes that code for these kinases has the same effect as adding extra brassinosteroids to the plant roots: fewer root hair cells . Cheng et al . show that one of the GSK3-like kinases binds and adds phosphate groups to protein complexes that control gene expression—and this causes these protein complexes to be less active . When GSK3-like kinase activity is switched off by brassinosteroids , these complexes instead become more active and trigger the expression of genes that direct a plant cell to become a non-hair cell . The findings of Cheng et al . reveal the pathways that allow brassinosteroids to stop plant cells in roots from becoming hair cells , and that instead encourage these cells to become non-hair cells . However , further work is needed to uncover how the striped pattern of hair cells and non-hair cells on Arabidopsis roots is established , and how brassinosteroids work with other plant hormones to control this pattern . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology"
] | 2014 | Brassinosteroids control root epidermal cell fate via direct regulation of a MYB-bHLH-WD40 complex by GSK3-like kinases |
The universally conserved signal recognition particle ( SRP ) is essential for the biogenesis of most integral membrane proteins . SRP scans the nascent chains of translating ribosomes , preferentially engaging those with hydrophobic targeting signals , and delivers these ribosome-nascent chain complexes to the membrane . Here , we present structures of native mammalian SRP-ribosome complexes in the scanning and engaged states . These structures reveal the near-identical SRP architecture of these two states , show many of the SRP-ribosome interactions at atomic resolution , and suggest how the polypeptide-binding M domain selectively engages hydrophobic signals . The scanning M domain , pre-positioned at the ribosomal exit tunnel , is auto-inhibited by a C-terminal amphipathic helix occluding its hydrophobic binding groove . Upon engagement , the hydrophobic targeting signal displaces this amphipathic helix , which then acts as a protective lid over the signal . Biochemical experiments suggest how scanning and engagement are coordinated with translation elongation to minimize exposure of hydrophobic signals during membrane targeting .
Roughly 20% of eukaryotic genes encode membrane proteins that must be inserted into the endoplasmic reticulum ( ER ) early in their biogenesis ( Krogh et al . , 2001; Shao and Hegde , 2011b ) . The majority of these proteins are recognized co-translationally at the ribosome by the signal recognition particle ( SRP ) and targeted via the SRP receptor to the Sec61 translocon for insertion ( Halic and Beckmann , 2005; Shan and Walter , 2005 ) . SRP has several critical functions in coordinating this targeting reaction . First , SRP must selectively recognize its cargo via a hydrophobic targeting signal . Second , it must shield this hydrophobic domain to preclude inappropriate interactions . Third , cargo-loaded SRP must interact with its receptor and release the cargo to the translocon . All of these events must occur before substantial synthesis beyond the targeting signal precludes successful engagement of the translocon ( Siegel and Walter , 1988 ) . In eukaryotes , SRP is composed of six protein subunits assembled on a ∼300 nucleotide RNA ( Keenan et al . , 2001 ) . The elongated assembly is operationally divided into an S domain and an Alu domain ( Siegel and Walter , 1986 ) . The S domain is involved in recognizing the targeting signal and interacting with the SRP receptor . Its functionally essential SRP54 subunit contains three modules: an N domain that interacts with the ribosome near its exit tunnel ( Halic et al . , 2004 , 2006; Schaffitzel et al . , 2006 ) , an M domain responsible for signal recognition ( Zopf et al . , 1990; Halic et al . , 2006; Janda et al . , 2010 ) , and a G domain that ( together with the N domain ) coordinates receptor targeting ( Miller et al . , 1993; Egea et al . , 2004; Focia et al . , 2004 ) . The Alu domain interacts near the GTPase center at the interface between the two ribosomal subunits ( Halic et al . , 2004 ) . This interaction is thought to compete with translation factors to slow nascent polypeptide elongation and provide additional time for successful targeting . Many prokaryotes lack an Alu domain and instead contain a simplified SRP composed solely of an SRP54 homolog and a short ∼110 nucleotide RNA . Despite extensive biochemical and structural studies , several basic aspects of SRP biology remain unresolved . First , the interactions between SRP and the ribosome are understood only in general terms , with regions of close proximity deduced from moderate resolution cryo-EM structures ( Halic et al . , 2004 , 2006; Schaffitzel et al . , 2006 ) . The specific residues and chemical details of the binding surfaces are not known . Furthermore , the eukaryotic SRP-ribosome interaction has only been visualized for mammalian SRP bound to the plant ribosome ( Halic et al . , 2004 , 2006 ) . Although functional for translocation , this heterologous complex strongly arrests translation ( Walter and Blobel , 1981 ) . By contrast , endogenous SRP in a homologous mammalian system has a remarkably subtle effect on translation elongation ( Wolin and Walter , 1989 ) . Hence , the nature of the Alu domain-ribosome interaction and its relationship to translation factor binding in a native homologous system is unknown . Second , the molecular events leading to signal recognition by the SRP M-domain are also incompletely resolved . Structural analyses of the M domain show that it contains a large hydrophobic groove ( Keenan et al . , 1998 ) capable of housing a hydrophobic signal ( Janda et al . , 2010 ) . While this has convincingly established the binding site , the configuration of the pre-engaged state of the M domain and how it transitions to the engaged state are less clear . An exposed hydrophobic groove , as seen in isolated M domain crystal structures ( e . g . , Keenan et al . , 1998 ) , would seem energetically unfavorable in an aqueous environment and may exhibit promiscuous interactions . Indeed , other factors that bind hydrophobic clients contain hydrophobic grooves that are occluded in the unengaged state ( Pellecchia et al . , 2000; Mateja et al . , 2009 ) . Thus , it is not known how SRP's substrate binding groove is kept shielded until it encounters its hydrophobic clients specifically at the ribosomal exit tunnel . Third , the timing of SRP recruitment to the ribosome relative to its engagement of the targeting signal remains a source of considerable debate . The simplest and earliest idea was that SRP has high affinity and selectivity for a ribosome exposing a hydrophobic signal ( Walter et al . , 1981 ) . However , such a model would not easily explain why far more abundant hydrophobic binding proteins do not outcompete SRP , why SRP cannot bind those same sequences post-translationally , and how SRP can rapidly find these transient and rare species to prevent aggregation . Thus , it was suggested that SRP may repeatedly scan all translating ribosomes until the appropriate cargo is encountered ( Ogg and Walter , 1995 ) . Studies in the simplified bacterial system have extensively debated both the basic premise and mechanistic details of this scanning model ( Bornemann et al . , 2008; Zhang et al . , 2010; Holtkamp et al . , 2012; Zhang and Shan , 2012; Saraogi et al . , 2014; Noriega et al . , 2014a , 2014b ) . The extent to which these prokaryotic studies apply to eukaryotic SRP is unclear and has been minimally studied . Not only is its ribosome interaction more complex , but the Alu domain merits consideration of how SRP recruitment is integrated into the translation cycle . Furthermore , Escherichia coli SRP seems to scan any translating ribosome ( Bornemann et al . , 2008; Holtkamp et al . , 2012 ) , while eukaryotic SRP may display a further preference for ribosomes containing a hydrophobic signal inside the exit tunnel ( Flanagan et al . , 2003; Berndt et al . , 2009; Mariappan et al . , 2010; Zhang et al . , 2012 ) . At present , the existence and properties of this scanning mode remain a point of uncertainty , but is crucial for understanding how hydrophobic signals are promptly found by SRP before their cytosolic exposure leads to off-pathway fates . Here we demonstrate that endogenous SRP is efficiently recruited to mammalian ribosomes containing a transmembrane domain ( TMD ) inside the ribosomal exit tunnel . The cryo-EM reconstruction of this native complex provided the first structure of SRP bound to the ribosome in a ‘scanning’ mode . A parallel structure of SRP bound after the TMD has emerged from the ribosome revealed the conformational changes that accompany TMD engagement by SRP . The structures show how the hydrophobic cavity of the SRP M domain is occluded by an amphipathic C-terminal ‘placeholder’ helix until its displacement by a bona fide TMD . Competition assays between SRP and a translational GTPase provide insight into how scanning and engagement are integrated during polypeptide elongation to maximize targeting efficiency while limiting TMD exposure to the cytosol . These findings have functional implications for the mechanism of membrane protein recognition and targeting .
To examine the timing and specificity of SRP recruitment , a variety of stalled ribosome-nascent chain complexes ( RNCs ) were prepared by in vitro translation of truncated mRNAs in rabbit reticulocyte lysate , affinity purified via the nascent chain , and analyzed biochemically . Constructs and truncation points were designed to position a 20 residue TMD either within the exit tunnel or just emerged from the ribosome ( Figure 1A ) . These two states were chosen to represent moments when SRP would be either scanning in anticipation of , or successfully engaged with , the TMD . Specificity was controlled by using a mutant ( hereafter termed 3R ) in which three hydrophobic residues in the TMD were changed to arginine . 10 . 7554/eLife . 07975 . 003Figure 1 . The timing of SRP recruitment to the mammalian ribosome . ( A ) Schematic of the constructs used for production and purification of stalled RNCs . Key domains and their positions ( in amino acids ) are indicated , along with the mutation ( 3R ) used to disrupt the TMD . The diagrams to the right depict RNCs with the TMD inside or outside the tunnel used for biochemical and structural analyses . ( B ) Anti-FLAG affinity purifications were performed on translation reactions programmed with no RNA ( mock ) , TMD-containing transcripts , or 3R mutant transcripts . Truncation was at position 95 ( tunnel ) or 128 ( exposed ) . The samples were analyzed by SDS-PAGE and visualized for total protein ( top panel ) or immunoblotted for SRP54 , the large ribosomal protein uL6 , or the small subunit protein uS9 . ( C ) The unbound fractions of translation reactions following affinity purification of the TMD and 3R constructs ( as in panel B ) were analyzed relative to serial dilutions of total lysate . RNCs containing an intact TMD , whether in the tunnel or exposed , selectively deplete more than 75% of SRP from the lysate . ( D ) RNCs truncated at different positions relative to the TMD were affinity purified and probed for SRP and the ribosome as in panel B . Maximal recruitment is observed at all points after truncation at residue 89 , when 14 TMD residues have entered the tunnel . ( E ) Analysis of SRP recruitment as in panel B using mutant TMDs in which 4 to 10 residues have been deleted . The truncation point was at residue 95 . ( F ) Experiment as in panel D for the indicated truncation points using the TMD or 3R construct . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 003 RNCs with an intact TMD , either exposed or buried in the tunnel , efficiently recovered SRP as judged by immunoblotting for SRP54 ( Figure 1B ) . Furthermore , the SRP68 and SRP72 subunits were recovered at sufficient levels to be clearly visible on the stained gel of the purified complexes ( Figure 1B ) . Indeed , immunoblotting of the unbound fraction after affinity purification revealed that ∼75% and 90% of SRP in the lysate was depleted by RNCs with the TMD in the tunnel or exposed , respectively ( Figure 1C ) , despite these representing only ∼5% of total ribosomes . Appreciable SRP depletion was not observed with either of the 3R RNCs , and post-translational TMD-binding proteins such as TRC40 ( Stefanovic and Hegde , 2007 ) were not depleted by any of the constructs . These results illustrate that RNCs containing a TMD inside the ribosomal tunnel recruit SRP nearly as efficiently as RNCs containing an exposed TMD . The observation that the 3R mutant largely abolishes SRP recruitment suggests that shorter hydrophobic regions , such as those remaining on either side of the mutation , are insufficient for robust recruitment . Serial truncations verified that maximal recruitment required the presence of 14 hydrophobic residues in the tunnel , with lower ( but detectable ) recruitment once 10 or 11 residues has been synthesized ( Figure 1D ) . Analysis of shortened TMDs were consistent with this result , showing sharply diminished recruitment with deletion of 4–8 residues , and no detectable recruitment after deletion of 10 residues ( Figure 1E ) . Systematically varying the position of the full TMD inside the tunnel showed efficient SRP recruitment at all positions ( Figure 1F ) , while the shortened ∆8 mutant showed low recruitment at all positions ( data not shown ) . These data collectively suggest that SRP binds translating ribosomes in two distinct states: a ‘scanning’ mode when the TMD is inside the exit tunnel , and an ‘engaged’ mode where SRP interacts directly with the hydrophobic nascent chain . These scanning and engaged SRP-RNCs were assembled in total cytosol from endogenous mammalian components , thereby representing native complexes . Importantly , both complexes could be purified in sufficient amounts for structural analysis by electron cryomicroscopy ( cryo-EM ) . Ribosomal particles from the scanning and engaged SRP complexes were visualized by cryo-EM and subjected to iterative in silico classification ( Figure 2—figure supplement 1 ) to select RNCs containing both P-site tRNA ( as a surrogate for the nascent chain ) and SRP . Refinement of the resulting RNC populations resulted in final reconstructions at an overall resolution of 3 . 9 Å and 3 . 8 Å for the 80S ribosome bound to SRP in its scanning and engaged modes , respectively ( Figure 2A , B , Table 1 , Figure 2—figure supplement 2 ) . The inherent flexibility of regions of SRP was reflected by the large variation in local resolution within the particle , which spans from ∼3 . 5 Å to greater than 7 . 5 Å resolution ( Figure 2—figure supplement 3 ) . 10 . 7554/eLife . 07975 . 004Figure 2 . Overview of scanning and engaged SRP-RNC cryo-EM reconstructions . ( A , B ) Two views depicting the cryo-EM density of the scanning and engaged complex , in which the density for SRP is green , 40S is yellow , 60S is blue , and P-site tRNA is purple . The top and bottom panels show views of the GTPase center ( with bound Alu domain ) and the exit tunnel ( with bound S domain ) , respectively . ( C , D ) Models for the Alu and S domains of SRP are overlayed on the appropriately colored density map for the engaged complex ( yellow: 40S , blue: 60S , purple:tRNA , green: SRP ) . See also Figure 2—figure supplements 1–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 00410 . 7554/eLife . 07975 . 005Figure 2—figure supplement 1 . Schematic of computational classification . Displayed is the classification scheme used for both the engaged and scanning EM datasets in order to isolate the population of particles used to reconstruct the three maps displayed in the manuscript , highlighted in red . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 00510 . 7554/eLife . 07975 . 006Figure 2—figure supplement 2 . Map and model quality . ( A–C ) Gold-standard Fourier Shell Correlation ( FSC ) curves for the density maps for the scanning , engaged , and ternary complex reconstructions . Resolution is demarcated using the FSC = 0 . 143 criterion . ( D ) FSC curves of the final model vs the complete engaged map ( black ) ; of a model refined in the first of two independent halves of the map ( red ) ; and of that model vs the independent second map ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 00610 . 7554/eLife . 07975 . 007Figure 2—figure supplement 3 . Local resolution of cryo-EM maps . Density maps for the scanning and engaged samples colored by local resolution in Å as calculated using ResMap . Regions of functional importance are labelled . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 00710 . 7554/eLife . 07975 . 008Figure 2—figure supplement 4 . Nascent polypeptide inside the exit tunnel . Density for the nascent chain within the ribosomal exit tunnel in the scanning and engaged complexes . The weaker density for the nascent peptide in the scanning complex may be due to increased flexibility of the chain in this sample , as it is not anchored at the N-terminus through interaction with the M-domain . Alternatively it is possible the nascent chain is present at lower occupancy . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 00810 . 7554/eLife . 07975 . 009Table 1 . Refinement and model statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 009Data collectionScanningEngagedParticles27 , 41552 , 061Pixel size ( Å ) 1 . 341 . 34Defocus range ( μm ) 2 . 0–3 . 52 . 0–3 . 5Voltage ( kV ) 300300Electron dose ( e/Å2 ) 2727Map sharpening B-factor ( Å2 ) −77 . 0−82 . 860S-S domain40S-Alu domainModel composition Non-hydrogen atoms149 , 73378 , 464 Protein residues75715021 RNA residues42571824 Ions13336Refinement Resolution used ( Å ) 3 . 83 . 8 Average B factor ( Å2 ) 89 . 1109 . 6 R factor*0 . 300 . 32 Fourier Shell Correlation ( FSC ) †0 . 850 . 79Rms deviations Bonds ( Å ) 0 . 0110 . 011 Angles ( ° ) 1 . 91 . 9Ramachandran plot Favored ( % ) 9391 Outliers ( % ) 79*R factor = Σ||Fobs| − ||Fcalc|/Σ|Fobs| . †FSC = Σ ( FobsF*calc ) /√ ( Σ|Fobs|2 Σ|Fcalc|2 ) . In both structures , the ribosome is unratcheted and contains a canonical P-site tRNA and density for the nascent chain in the exit tunnel ( Figure 2—figure supplement 4 ) . Density for the mRNA codon in the P site is also present , but is poorly defined , likely reflecting increased flexibility due to truncation of the mRNA . Additional density representing the E-site tRNA is also observed , though its anticodon is disordered suggesting it is not in a single orientation . We do not observe any conformational changes in the ribosome exit tunnel in the scanning structure relative to either the engaged structure or earlier mammalian ribosome structures ( Voorhees et al . , 2014 ) . Thus , SRP recruitment by a TMD inside the exit tunnel cannot be explained purely based on structural changes observed at this resolution , a point we address further in the ‘Discussion’ . The overall architecture of SRP in the engaged binding mode ( Figure 2B ) is very similar to earlier reconstructions of this state visualized using heterologous plant–mammalian complexes ( Halic et al . , 2004 , 2006 ) . As expected , the Alu domain is localized to the GTPase center between the 40S and 60S subunits ( Figure 2C ) and is connected by a flexible RNA linker to the S domain positioned at the exit tunnel ( Figure 2D ) . At low-resolution , the observed density accounts for the majority of the SRP molecule , including most of the RNA , the Alu domain proteins ( SRP9 and SRP14 ) , and most of the S domain proteins ( SRP54 , SRP19 , and RNA binding regions of SRP68 ) . However , only the Alu domain and SRP54 were sufficiently well ordered to allow direct modelling into the density maps , and thus form the focus of our interpretations in the sections below . The others areas , whose functional relevance remain poorly understood , were rigid-body fit with minor adjustments from available crystal structures ( Grotwinkel et al . , 2014 ) , and were not interpreted further here . The SRP density in the matched scanning mode structure was found to be remarkably similar in overall conformation , position , and relative occupancy of all domains ( Figure 2A ) . In particular , the Alu domain , whose binding has been implicated in slowing translational elongation after SRP engages with a signal sequence ( Wolin and Walter , 1989 ) , is positioned in the GTPase center . Similarly , the entire S domain is grossly indistinguishable , indicating that the TMD-interacting M domain of SRP54 is positioned at the exit tunnel even prior to engagement . This suggests that in mammals , SRP has a single stable binding site on the ribosome , and carries out its biological functions during both scanning and engagement from this position . The considerably higher resolution of these structures relative to earlier reconstructions permitted interpretation of several SRP-RNC interactions in atomic terms . SRP is anchored on the ribosome primarily by its SRP54 subunit near the exit tunnel ( Figure 3A ) and the Alu domain at the ribosomal GTPase center . Two loops in the SRP54 N-domain previously implicated in ribosome binding ( Halic et al . , 2004 , 2006; Schaffitzel et al . , 2006 ) can now be sufficiently resolved for unambiguous placement of many protein side chains and analysis of the stabilizing chemical interactions ( Figure 3B ) . For example , Thr21 and Asp19 of SRP54 are within hydrogen bonding distance of Lys46 in uL29 and the phosphate oxygen of A80 in 5 . 8S rRNA ( Figure 3C ) . Similarly , the backbone of residue 67 is within hydrogen bonding distance of the carbonyl oxygen of residue 116 of uL23 ( Figure 3D ) . These specific contacts , along with other electrostatic interactions involving the conserved basic residues in this region , provide the primary stabilization of SRP54 and serve to orient its signal-binding M domain at the ribosomal exit tunnel . 10 . 7554/eLife . 07975 . 010Figure 3 . SRP-ribosome interactions near the exit tunnel . ( A ) Overview of the N-domain of SRP54 ( green ) , which is positioned near the exit tunnel by its interactions with the 5 . 8S rRNA and ribosomal proteins uL23 and uL29 . ( B ) Representative density for one of the loops of SRP54 , which anchors SRP to the ribosome , and its adjoining helices . ( C , D ) Specific hydrogen bonding interactions of SRP54 with the 5 . 8S rRNA ( cyan ) and ribosomal proteins uL29 and uL23 ( dark blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 010 Within the ribosomal GTPase center , the RNA portion of the well resolved Alu domain approaches the 60S subunit , interacting with the ribosomal stalk protein uL10 , while the proteins SRP14 and SRP9 approach the 40S subunit , interacting with the backbone of the 18S rRNA ( Figure 4A ) . The structure reveals that Asn77 and Lys95 of SRP14 are positioned within hydrogen bonding distance of the phosphate oxygens of U488 and A464 of the 18S rRNA ( Figure 4B–D ) . Of note , alignment of our SRP structure bound to the mammalian ribosome with an earlier structure of mammalian SRP bound to the plant ribosome ( Halic et al . , 2004 ) suggests changes in relative position of the S and Alu domains . Such changes , together with sequence divergence of uL10 and uL11 , may explain why mammalian SRP arrests translation in a plant system while only subtly slowing translation in mammals ( Walter and Blobel , 1981; Wolin and Walter , 1989 ) . 10 . 7554/eLife . 07975 . 011Figure 4 . Interactions of the SRP Alu domain at the GTPase center . ( A ) Overview of the Alu domain ( green ) contacting the 60S subunit via the ribosomal stalk ( dark blue ) and the 40S subunit through the 18S rRNA ( orange ) . The engaged structure is shown; the scanning structure was essentially indistinguishable . ( B ) Representative density for the region of SRP14 that interacts with the 40S subunit . ( C , D ) Specific hydrogen bonding interactions of SRP14 with the backbone of the 18S rRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 011 The main functional transition between a scanning and engaged SRP is substrate binding by the M domain . Our structures afforded the first opportunity to visualize the conformation of the scanning M domain and deduce changes that accompany TMD engagement . We began by modelling the engaged M domain ( Figure 5A ) , guided by earlier crystal structures involving this region ( Keenan et al . , 1998; Clemons et al . , 1999; Batey et al . , 2000; Rosendal et al . , 2003; Egea et al . , 2008 ) . The hydrophobic groove of this domain is occupied by density consistent with an alpha helix of ∼12 residues , accounting for roughly half of the nascent chain's TMD . This assignment is consistent with earlier structures ( Janda et al . , 2010; Hainzl et al . , 2011 ) and crosslinking analyses ( Zopf et al . , 1990; Lutcke et al . , 1992 ) . 10 . 7554/eLife . 07975 . 012Figure 5 . The engaged SRP54 M domain . ( A ) The final model of the SRP54 M-domain ( green ) and bound TMD ( cyan ) is displayed in the density for the engaged complex . The C-terminus of this domain is indicated with a red asterisk . Density for two additional helices ( termed αC1 and αC2 ) , which were not part of the earlier crystal structures , represent the C-terminus of SRP54 and would be connected via flexible linkers ( not modelled ) to the remainder of the M-domain . These helices enclose the TMD substrate , minimizing the exposed hydrophobic surface area . ( B ) Displayed is a surface representation of the M-domain , colored by hydrophobicity , indicating the hydrophobic binding pocket for the TMD enclosed by the C-terminal helices of SRP54 . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 01210 . 7554/eLife . 07975 . 013Figure 5—figure supplement 1 . Sequence conservation of the amphipathic helices at the C-terminus of SRP54 . ( A ) Sequence of the ∼70 C-terminal amino acids of human SRP54 , which were truncated in previous crystal structures . Regions predicted to be helical are indicated in red . ( B ) Helical wheel depictions of the two predicted C-terminal helices from diverse eukaryotic species ( each represented in concentric rings ) . Note that the amphipathic nature of both helices is highly conserved , with the hydrophobic face strongly enriched in methionines . This is consistent with their observed positioning surrounding the TMD substrate in the M-domain within the engaged structure . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 013 After fitting the density corresponding to earlier crystal structures and assigning the substrate , two additional regions of helical density were observed abutting the TMD ( Figure 5A ) . Several arguments suggest that these represent the C-terminus of SRP54 . First , the C-terminal ∼70 residues , truncated from all earlier SRP54 crystal structures , are the only unaccounted amino acids in this region . The only other nearby polypeptide would be the hydrophilic nascent chain , which is unlikely to be helical or rigid . Second , this 70 residue stretch of mammalian SRP54 contains two conserved predicted helices of appropriate length ( Figure . 5—figure supplement 1A ) . Third , these helices are amphipathic , making them ideal for surrounding the hydrophobic TMD ( Figure 5—figure supplement 1B ) . We therefore posit that these two additional helices , provisionally termed αC1 and αC2 of SRP54 , serve as an intramolecular ‘lid’ that surrounds the hydrophobic substrate to shield it from aqueous solvent ( Figure 5B ) . Although we cannot unambiguously assign these helical regions to a particular sequence , we have provisionally named them based on the relationship between their apparent length in the structure and predicted size . Relative to the engaged state , the scanning structure showed markedly underrepresented density in the region of αC2 , but clear helical density within the SRP hydrophobic groove ( Figure 6A , compare top and bottom panels ) . This suggests that in the absence of a TMD substrate , αC2 shifts to occupy the hydrophobic groove . This helix may be somewhat dynamic since a small amount of density is still observed in the alternate position of αC2 ( Figure 6A , bottom panel ) . Although this is our favored hypothesis , we cannot formally exclude the possibility that the density in the hydrophobic groove represents the nascent chain . However , this seems unlikely because a hydrophilic sequence would be disfavored in the hydrophobic groove , this region of polypeptide is unlikely to be helical , earlier crosslinking data suggest SRP does not interact with hydrophilic segments of nascent chain ( Berndt et al . , 2009 ) , and disappearance of the αC2 density would remain unexplained . Thus , we propose that the amphipathic αC2 normally resides in the hydrophobic groove , but is displaced by TMD engagement to its ‘lid’ position between the M domain and exit tunnel ( Figure 6B ) . 10 . 7554/eLife . 07975 . 014Figure 6 . Comparison of the engaged and scanning M-domain . ( A ) Density and associated model of the engaged M-domain ( top panel ) and rigid-body fitting of the engaged model into the density for the scanning structure ( bottom panel ) . Note that density in the engaged position of αC2 is nearly absent in the scanning structure . The C-terminus of the M-domain preceding αC1 and αC2 is indicated by a red asterisk . ( B ) Superposition of the scanning and engaged M-domain models suggests that αC2 is repositioned ( arrow ) upon binding to a TMD ( not depicted for clarity ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 014 Our structures indicate that Alu domain occupancy at the GTPase center is mutually exclusive with translational GTPase interactions that occur with each cycle of nascent chain elongation . Yet , 35 cycles of translational elongation separate our two structures , and SRP does not arrest translation even after successful engagement ( Wolin and Walter , 1989 ) . To investigate this apparent incompatibility , we sought to examine the relationship between translational GTPase binding and the SRP-ribosome interaction in the scanning vs engaged modes . Such an experiment is complicated because neither dominant negative eEF1 nor eEF2 would bind our staged SRP-RNC complexes: eEF1 ternary complex binding requires an appropriate codon in the ribosomal A site , while eEF2 , when in its GTP-bound state , preferentially acts on pre-translocation RNCs in the hybrid state . Furthermore , adding mutant versions of these factors to our reactions would inhibit translation . To bypass these technical obstacles , we utilized the GTPase Hbs1 as a surrogate for eEF1 and eEF2 . The G-domain of all three factors share a high degree of sequence and structural homology , and the binding of Hbs1 to the ribosomal GTPase center is well characterized ( Shoemaker et al . , 2010; Becker et al . , 2011; Pisareva et al . , 2011 ) . Importantly , a GTPase deficient dominant negative mutant ( Hbs1-DN ) added to our translation reactions will stably bind to the translation factor binding site ( Becker et al . , 2011 ) only when ribosomes reach the end of the truncated message ( Shao and Hegde , 2014 ) , allowing us to query the consequence of GTPase center occupancy for the SRP-RNC interaction without inhibiting translation . Translation extracts were supplemented with increasing amounts of Hbs1-DN , and used for the synthesis and purification of scanning and engaged RNC complexes via the nascent chain . Hbs1-DN binding , observed in both cases , reduced SRP recovery by up to ∼70% with the scanning RNCs , but had no effect on engaged SRP-RNC complexes ( Figure 7A ) . Because a substantial excess of Hbs1-DN relative to ribosomes was used , its failure to displace engaged SRP suggested the existence of a ternary complex between the ribosome , Hbs1-DN , and SRP . The relatively modest displacement of scanning SRP further hinted that this too may permit ternary complex formation . 10 . 7554/eLife . 07975 . 015Figure 7 . Integration of SRP recruitment with the translational elongation cycle . ( A ) Stalled RNCs were produced and affinity purified via the nascent chain from translation reactions supplemented with increasing amounts of a GTPase deficient Hbs1 ( Hbs1-DN ) . SRP recovery is reduced by Hbs1-DN for scanning mode RNCs , but unaffected for engaged RNCs . ( B ) Following translation in the presence of 3X-FLAG tagged Hbs1-DN , HA-tagged RNCs were purified via the FLAG tag and analyzed for the indicated components . Recovery of SRP in a ternary complex with Hbs1 and the ribosome is observed for both the scanning and engaged mode RNCs , but SRP recovery is substantially reduced in the scanning mode . ( C ) Two views depicting the cryo-EM density for a ternary complex of the ribosome bound to eEF2 ( blue ) in the GTPase center and the S domain of SRP ( green ) at the exit tunnel . ( D ) Left panel: diagram illustrating that translation of a construct containing a stop codon 35 amino acids from the beginning of the TMD will only expose the TMD after termination . Right panel: the construct from the diagram ( or a 3R control ) was translated for a short 3 min ‘pulse’ followed by either immediate cooling , or addition of an initiation inhibitor and a 30 min ‘chase’ translation . Affinity purification of the nascent chain from the pulse and chase samples reveals that actively translating RNCs can recruit SRP in a TMD-dependent manner . See also Figure 7—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 01510 . 7554/eLife . 07975 . 016Figure 7—figure supplement 1 . Simultaneous binding of eEF2 and SRP . View of the GTPase center of the cryo-EM map of the ternary complex containing the ribosome , eEF2 , and SRP . Note that at the lower threshold that is depicted in this figure , density for the Alu domain and linker regions of SRP are observed . The comparatively poor density of these areas of SRP relative to the S-domain suggests that they are probably dynamic . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 016 To test this idea , scanning and engaged HA-tagged RNCs were assembled in the presence of FLAG-tagged Hbs1-DN , affinity purified via the FLAG tag , and tested for SRP recovery ( Figure 7B ) . Consistent with the competition experiment , equal amounts of ribosomes were recovered from both reactions in an Hbs1-dependent manner , but substantially less SRP was found with scanning vs engaged RNCs . Thus , when SRP is in its scanning mode , binding of Hbs1-DN can partially ( but not completely ) displace SRP . However , once the TMD is exposed , both Hbs1-DN and engaged SRP are able to stably associate with the ribosome . Such an observation can be explained by the available structural information . Due to the flexible RNA linker that connects the two halves of SRP , Alu domain dissociation need not lead to immediate dissociation of SRP . Rather , the SRP S domain interactions could retain SRP on the ribosome , with dissociation governed by the off rate of this complex . For E . coli SRP , which lacks an Alu domain , this off rate is ∼0 . 1–0 . 3 s−1 ( Saraogi et al . , 2014 ) , probably explaining why scanning-mode SRP is lost during the 2–3 hr needed for RNC affinity purification . By contrast , the additional interaction between the TMD and the M domain reduces this off rate effectively to zero relative to the timeframe of this experiment , explaining why Hbs1-DN cannot displace engaged SRP . Other translational GTPases ( eEF1 , eEF2 , and the release factor complexes ) would likely share the same qualitative relationship with the different states of ribosome-bound SRP , explaining why SRP , even after RNC engagement , is compatible with active elongation . This logic suggests that a population of ribosomes bound to both a translation factor and SRP might be present in our EM samples . The use of a truncated mRNA for sample preparation precludes eEF1 binding , but eEF2-ribosome-SRP complexes were feasible . Such a population was not identified in our scanning dataset , likely due to the lower stability of this ternary complex when SRP is bound in its scanning mode ( Figure 7B ) . By contrast , the engaged dataset contained a small subset of particles in which density was observed for both eEF2 and the S domain of SRP bound in its canonical location at the exit tunnel ( Figure 7C ) . At a lower threshold , density for the flexible RNA linker and the Alu domain could also be observed , with the Alu domain at the so-called right foot of the 40S subunit ( Figure 7 , Figure 7—figure supplement 1 ) . The relatively poor density for this half of SRP suggests it is probably dynamic when the Alu domain cannot bind its canonical site at the GTPase center . Whether the weakly observed alternate Alu domain position on the 40S represents a physiologically relevant binding site remains to be determined . Nevertheless , this structure supports our conclusion that the Alu domain can be displaced from the GTPase center by translation factors without dissociating SRP from the ribosome . The compatibility of SRP and translation factor binding to the same ribosome suggested that scanning SRP need not dissociate between rounds of polypeptide elongation . To investigate this idea , we designed an experiment to determine if scanning SRP can be observed on actively translating ribosomes . A stop codon was positioned 35 residues from the beginning of the TMD , ensuring that translation would terminate before emergence of the TMD from the exit tunnel ( Figure 7D , diagram ) . During translation , a sub-population of ribosomes containing nascent peptides long enough to expose the N-terminal epitope tag would also contain the TMD inside the tunnel . As SRP cannot interact with its clients post-translationally , its recovery with the nascent chain would necessarily be via the ribosomal scanning mode . Affinity purification of RNCs after 3 min of translation recovered substantially more SRP with RNCs containing a TMD than the 3R mutant ( Figure 7D , right panel ) . Parallel reactions in which translation initiation was inhibited at 3 min and the ribosomes were allowed to complete translation produced full-length protein that recovered neither ribosomes nor SRP . These observations suggest that SRP is recruited in a TMD- and nascent chain-dependent manner to actively translating ribosomes despite never exposing the TMD outside the exit tunnel . Thus , SRP might be able to continuously scan translating ribosomes as a TMD elongates through the exit tunnel , thereby permitting engagement with essentially no opportunity for TMD exposure to the cytosol .
In this study , we have characterized an anticipatory scanning mode of SRP-ribosome interaction , validated its existence during ongoing translation , and provided the first structure of this complex . Together with our engaged SRP complex , these structures represent markedly improved views of many functional regions of mammalian SRP and provide the first entirely homologous SRP-RNC structures assembled with native endogenous factors . Additional experiments analysing the relationship between SRP and a translational GTPase permit us to consolidate our findings into a dynamic model of nascent chain scanning and TMD capture by SRP during the earliest stages of membrane protein biosynthesis ( Figure 8 ) . 10 . 7554/eLife . 07975 . 017Figure 8 . Working model for SRP scanning and engagement of a nascent membrane protein . ( A ) Model for SRP dynamics and conformation during translational elongation in its scanning and engaged states . The Alu domain swings away from the ribosome to accommodate translation factor binding ( eEF2 is shown ) . Scanning SRP is more prone to displacement by translation factors , while engaged SRP remains stably bound . ( B ) Close-up views of the relative positions and conformations of the SRP54 M-domain ( green ) and hypothetical nascent chain ( blue ) during scanning ( left ) and engagement ( right ) . The arrows in the scanning diagram depict hypothesized dynamics of the αC2 helix to permit nascent chain sampling . A hydrophobic domain ( blue cylinder ) displaces the αC2 helix to its lid position . See also Figure 8—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 01710 . 7554/eLife . 07975 . 018Figure 8—figure supplement 1 . Superposition of various M-domain structures . ( A ) Comparison of the positioning of the hydrophobic substrate bound to the SRP M-domain from crystal structures of SRP54 and an earlier cryo-EM reconstruction of SRP bound to the plant ribosome . ( B ) Comparison of localization of the C-terminus of SRP54 between this study and the earlier reconstruction of SRP bound to the plant ribosome . DOI: http://dx . doi . org/10 . 7554/eLife . 07975 . 018 Once as few as 12–14 residues of a TMD enter the ribosomal exit tunnel , an RNC becomes competent to recruit SRP in its scanning mode . SRP binding appears to favor an unratcheted ribosome and requires an unoccupied GTPase center . An A site tRNA , while not observed in our structure due to the use of a truncated mRNA , is sterically compatible with SRP binding . Thus , Alu domain recruitment could potentially occur immediately after dissociation of eEF1 or eEF2 , from the ribosome . The S domain could presumably bind transiently anytime , and may increase local SRP concentration to facilitate the Alu domain interaction . The proportion of each ∼200 ms translation cycle during which RNCs are compatible for Alu domain binding remains to be determined . Regardless , a twenty residue TMD would have approximately 21 translation cycles to recruit SRP while it is still inside the tunnel . Once recruited , SRP would be ribosome-associated via both the S and Alu domains . Arrival of the next translation factor would displace the Alu domain , while the S domain can remain fixed due to the flexible RNA linker between these two domains ( Figure 8A ) . Whether SRP completely dissociates from the ribosome would depend on the off-rate of the S domain relative to the next Alu domain binding opportunity within the translation cycle . Estimates from the E . coli system suggest that the S domain dissociation would be far slower than an elongation cycle , and thus the Alu domain could re-bind and detach repeatedly while SRP nevertheless maintains ribosome association during ongoing translation . This would explain how we are able to isolate scanning mode SRP-RNC complexes from an active translation reaction ( Figure 7D ) and recover at least some GTPase-SRP-RNC ternary complex in the scanning state ( Figure 7B ) . Early recruitment followed by maintenance of the M domain at the exit tunnel would allow for continuous sampling of the nascent chain until the TMD emerges . The M domain is optimally pre-positioned for this function , with the hydrophobic groove located ∼20 Å from the exit site . Although the precise path of the nascent chain is not known , the limited space in this region means it necessarily passes the M domain ( Figure 8B , left panel ) . We envisage a dynamic αC2 helix that transiently exposes the hydrophobic groove to provide opportunities for nascent chain sampling . If a segment of the nascent polypeptide is sufficiently hydrophobic to replace the αC2 helix , it would successfully engage SRP54 ( Figure 8B , right panel ) . In this model , the biophysical properties of αC2 effectively set the hydrophobicity threshold for client engagement . After nascent chain binding to the M domain , αC2 would occupy a position on the opposite side of the hydrophobic groove , oriented with its hydrophobic face toward the substrate , forming part of a protective lid . The conformation of this engaged state may be flexible to accommodate different substrates , as suggested by other variant configurations ( Figure 8—figure supplement 1 ) . Regulated access to a hydrophobic binding groove and substrate shielding by an amphipathic helix appears to be a general principle that is shared by the Get3 targeting factor ( Mateja et al . , 2015 ) . In the engaged state , the SRP S domain is able to remain stably bound to the ribosome via its dual interactions with the exit tunnel and nascent chain , ensuring shielding of the hydrophobic substrate . Elongation factors could still access the GTPase center by displacing the Alu domain , which must swing away from the ribosome . Its local concentration would necessarily remain high due to S domain tethering , presumably improving its ability to compete with eEF1 and eEF2 . This is the likely cause of the reported slowing of translation by SRP ( Wolin and Walter , 1989 ) that provides a subtly increased time for successful delivery to the translocon ( Mason et al . , 2000; Lakkaraju et al . , 2008 ) . It is noteworthy that the minimum hydrophobic element for robust early recruitment of SRP is longer than the 7 to 9 hydrophobic residues within most cleavable N-terminal signal peptides ( von Heijne , 1990 ) . Early recruitment may not be critical in this case because signal peptides are more soluble and less aggregation-prone than TMDs , and sufficiently short to be fully shielded by the hydrophobic groove of the M domain . By contrast , TMDs responsible for targeting are markedly more hydrophobic and between 17 to 24 residues long . They are effectively insoluble in aqueous solvent and too long to be completely shielded by the M domain , making their efficient and early targeting more important . Moreover , early recruitment would provide SRP the opportunity for first refusal of an emerging TMD , which could allow SRP to out-compete other more abundant binding factors such as TRC40 ( Stefanovic and Hegde , 2007 ) , SGTA ( Wang et al . , 2010 ) , Hsp70 ( Rabu et al . , 2008 ) , and calmodulin ( Shao and Hegde , 2011a ) . Another important functional implication of early recruitment is increased time for targeting the RNC to the translocon . We cannot , at this level of resolution , see any differences in the configuration of the SRP54 NG domain between the scanning and engaged states . This suggests that the scanning state should be competent for interaction with the SRP receptor , potentially permitting targeting while the TMD is still inside the tunnel . Based on the ∼35 residue length of the exit tunnel , recruitment before the TMD is synthesized provides an additional ∼7 s for targeting relative to a model where targeting only occurs after TMD emergence . This extra time would be sufficient for delivery to the translocon before or as the TMD emerges , essentially eliminating its cytosolic exposure and allowing its synthesis to occur directly at Sec61 . Indeed , recent ribosome profiling experiments in yeast suggest that many proteins arrive at the Sec61 translocon before the TMD is exposed ( Jan et al . , 2014 ) . Such early delivery may be particularly important for proteins whose downstream domains may interfere with correct translocation such as downstream TMDs of multispanning membrane proteins or rapidly folding domains ( Conti et al . , 2014 ) . Our findings suggest that early SRP recruitment is another mechanism for lengthening the targeting window , and may operate in conjunction with translational slowing ( Wolin and Walter , 1989 ) and strategic positioning of slowly decoded rare codons ( Pechmann et al . , 2014 ) . A major unresolved question is the mechanistic basis for how a TMD can influence SRP binding from inside the exit tunnel . One possibility involves conformational changes on the ribosome surface triggered by TMD interactions with the proteins or RNA lining the tunnel . However , we have been unable to detect any differences in the tunnel or surface regions for SRP binding by comparing our scanning RNCs with those containing hydrophilic sequences . Instead , we currently favor a kinetic model in which a hydrophobic sequence in the tunnel biases ribosomes into a conformation favorable for Alu domain binding . If the presence of a hydrophobic sequence in the ribosomal tunnel would preferentially extend those stages of the translation cycle compatible with SRP binding , this could provide a mechanism for tunnel-initiated SRP recruitment . Evidence for preferential SRP binding to some but not other states of the ribosome comes from experiments using different translation elongation inhibitors ( Ogg and Walter , 1995 ) . Cycloheximide , an elongation inhibitor that causes ribosomes to pause in a presumably unratcheted state , can rescue translocation defects caused by limiting SRP availability . By contrast , another elongation inhibitor anisomycin , which pauses ribosomes at the ratcheted state , does not rescue under similar conditions . This implies that situations preferentially favoring the cycloheximide state may permit SRP recruitment . Thus , it is plausible that a TMD inside the tunnel , or elongation of hydrophobic amino acids , would subtly influence the translation cycle to favor this state . Indeed , sequences inside the ribosomal tunnel are known to influence the translation cycle ( Seidelt et al . , 2009; Bhushan et al . , 2011; Wilson and Beckmann , 2011 ) , while the biophysical characteristic of the aminoacyl-tRNA in the A site influences the time spent in different ribosome states ( Lareau et al . , 2014 ) . A model invoking the Alu domain would also explain why sequence-independent scanning in the E . coli system appears to operate differently than the TMD-dependence observed in eukaryotes . Finally , it is worth noting that tools are now readily available to analyze early recruitment of SRP in vivo . The combination of ribosome profiling ( Ingolia et al . , 2009 ) together with selective retrieval of SRP complexes ( del Alamo et al . , 2011 ) should permit the mapping of all SRP-containing RNCs . Such analyses can provide a global view of not only SRP substrates in vivo , but also the precise timing of its recruitment for different types of cargos . We anticipate that , based on our analyses in the mammalian system , early recruitment should be seen for proteins with the most hydrophobic targeting sequences . Such proteins are also the most aggregation-prone when targeting fails , perhaps providing a strong selection pressure for evolution of an early targeting mechanism .
An SP64 vector-based construct encoding the transferrin receptor TMD ( AIAVIVFFLIGFMIGYLGYA ) inserted into Sec61β ( Hessa et al . , 2011 ) was modified to contain an N-terminal affinity tag ( 3xFLAG or 3xHA ) . Phusion mutagenesis was used to generate the following TMD mutants: 3R ( AIAVIRRRLIGFMIGYLGYA ) , ∆4 ( AVIVFFLIGFGYLGYA ) , ∆6 ( AIVFFLIGGYLGYA ) , ∆8 ( AVFFLIGYLGYA ) , and ∆10 ( AFFLGYLGYA ) . A stop codon was introduced by Phusion mutagenesis at position 110 for the experiment in Figure 7D . The mammalian expression construct for Hbs1-DN has been described ( Shao et al . , 2013 ) , and was purified as before ( Shao and Hegde , 2014 ) . Antibodies against uL6 ( anti-L9 ) and uS9 ( anti-S16 ) were from Santa Cruz Biotechnology ( Dallas , TX ) . Anti-SRP54 was from BD Biosciences ( San Jose , CA ) . The antibody against TRC40 has been described ( Stefanovic and Hegde , 2007 ) . Anti-Flag and HA resin and 3X Flag and HA peptides were obtained from Sigma ( St . Louis , MO ) . Preparation and purification of stalled RNCs was performed as previously described ( Shao et al . , 2013 ) . Briefly , the template for in vitro transcription was prepared by PCR from the constructs above using a 5′ primer just preceding the SP6 promoter , and a 3′ primer at the desired site of truncation ( or downstream of the stop codon for full length products ) . All 3′ primers for truncations encoded a final valine codon , whose peptidyl-tRNA is least labile to hydrolysis ( Shao et al . , 2013 ) . PCR products were purified and used for in vitro transcription and translation in rabbit reticuloclyte lysate as described ( Sharma et al . , 2010 ) . Translations were for were for 20–25 min at 32°C . For Figure 7A , Hbs1-DN was included at 3 . 3 nM , 17 nM , 33 nM and 133 nM . For Figure 7B , Hbs1-DN was included at 133 nM . For Figure 7D , actively translating , unstalled complexes were produced by allowing 3 min of translation at 32°C followed by either a rapid fivefold dilution in chilled buffer 1 ( 50 mM HEPES pH 7 . 5 , 200 mM KAc , 15 mM MgAc2 , and 1 mM DTT ) , or addition of 70 μM aurin tricarboxylic acid ( ATA ) for the remainder of a 30 min translation before dilution . All samples were then affinity purified via either the nascent chain or the tag on Hbs1-DN using anti-Flag or HA resin . The Flag tag was used on the nascent chain in all experiments except those that included Hbs1-DN , which instead used HA-tagged nascent chains . For biochemical analysis , the affinity resin was either added directly to undiluted translation reactions , or after dilution in buffer 1 . The ratio of affinity resin to translation reaction was typically 1:50 . Binding was performed in batch at 4°C for ∼1–2 hr with gentle mixing , transferred to a micro-spin column , washed with ∼25 vol ( relative to resin ) of buffer 1 , and eluted for 30 min at 22°C in the same buffer supplemented with 0 . 2 mg/ml of the appropriate peptide . Samples for structural analysis were washed as above in buffer 1 containing an additional 200 mM or 400 mM KAc for the scanning and engaged complexes , respectively . Samples were eluted in buffer 1 supplemented with 3 mM GDPCP , concentrated by centrifugation ( 50 , 000 rpm for 75 min in a TLA55 rotor ) , and resuspended in a volume of buffer 1 to achieve ∼125 nM ribosomes in the presence of 100 μM GDPCP . Purified samples were applied to glow-discharged holey carbon grids ( Quantifoil R2/2 ) , which had been coated with a ∼70 Å thick layer of amorphous carbon . Using an FEI Vitrobot , 3 μl of sample was applied to the grid , followed by a 30 s incubation at 4°C , 3 s of blotting , and flash-cooling in liquid ethane . Data were collected on an FEI Titan Krios at 300 KV using FEI's automated single particle acquisition software and defocus values of 2–3 . 5 μm . Images were recorded using a back-thinned FEI Falcon II detector at a calibrated magnification of 104 , 478 ( pixel size of 1 . 34 Å ) . Individual frames from the detector were recorded as previously described ( Bai et al . , 2013 ) . Contrast transfer function parameters were estimated using CTFFIND3 ( Mindell and Grigorieff , 2003 ) , and micrographs that had evidence of astigmatism or drift were discarded . All automated particle picking , 2D and 3D classifications , and refinements were performed using RELION as described below ( Scheres , 2012 ) . Unsupervised 2D class averaging was used to discard any non-ribosome particles , resulting in a combined 326 , 981 and 639 , 184 particles for the engaged and scanning samples , respectively . Iterative rounds of 3D classification were then utilized to identify the population of ribosomes bound to SRP ( Figure 2—figure supplement 1 ) . For the scanning sample , 27% of selected 80S particles ( 171 , 143 ) were computationally classified as being in an unratcheted conformation and bound to a P-site tRNA . As weak density for SRP in this population could be observed , a further focused classification was performed utilizing a mask around the observed extra-ribosomal density to identify 16% of this unratcheted population ( 27 , 627 particles ) that contained SRP . An additional focussed classification using a mask around the P-Site tRNA identified a final 27 , 415 particles that are unratcheted , bound to SRP , and contain improved density for the nascent chain . The somewhat weaker density observed for the nascent chain in the scanning complex ( Figure 2—figure supplement 4 ) might be due to either increased flexibility of the nascent chain when it is not anchored at the N-terminus by an interaction with SRP , or decreased occupancy resulting from hydrolysis during sample freezing that was not resolved by computational sorting . Similarly , in the engaged sample , 58% of ribosomes ( 189 , 099 particles ) were unratcheted and contained tRNA , while 28% of these particles ( 52 , 061 ) were bound by SRP . In order to isolate the population of ribosomes bound to both SRP and eEF2 in the engaged sample , a mask for SRP was used to further sub-classify the population of ribosomes bound to eEF2 ( 20%: 64 , 416 ) resulting in a final population of 8418 particles . Final 3D refinements of the resulting populations were performed without external masking , utilizing statistical movie processing ( Bai et al . , 2013 ) , and particle polishing ( Scheres , 2014 ) . This resulted in final reconstructions at overall resolution of 3 . 9 Å , 3 . 75 Å , and 5 . 0 Å for the scanning , engaged , and ternary complex structures , respectively , using the gold-standard FSC = 0 . 143 criteria ( Scheres and Chen , 2012 ) . All models were initially built and refined in the density for the higher resolution engaged complex , and then rigid body fit into the density for the scanning sample . The resulting scanning model required only minor adjustments to the M domain to fit the respective density ( e . g . , Figure 6 ) . The 60S subunit , 40S body , and 40S head were individually placed using models of the porcine ribosome ( Voorhees et al . , 2014 ) , and the P-site tRNA was homology modelled using the bacterial structure ( Voorhees et al . , 2009 ) . SRP was built using a combination of crystal structures of its individual domains , which were modified to fit the observed density as described below . Within the S domain , the N-domain of SRP54 was built primarily using the solution structure of the human domain ( PDB ID: 1WGW ) with minor modification to the loops that interact with the ribosome . The G-domain was modelled using the structure from ( Janda et al . , 2010 ) ( PDB ID: 3KL4 ) . The M-domain itself was modelled using the crystal structure of the human M-domain ( Clemons et al . , 1999 ) ( PDB ID: 1QB2 ) , as well as models of the occupied M-domain from homologous species ( Janda et al . , 2010; Hainzl et al . , 2011 ) , with modifications to account for the interface with the ribosome and other SRP components . Due to lower resolution in this region , the S-domain RNA , SRP19 , and a portion of SRP68 were simply rigid-body fit using the model from Grotwinkel et al . ( 2014 ) ( PDB ID: 4P3E ) . The Alu domain was built based on models from ( Weichenrieder et al . , 2000 ) , including the proteins SRP9 and SRP14 ( PDB ID: 1E8O ) , and the Alu RNA ( assembled from PDB IDs: 1E8O and 1E8S ) . All models were built in COOT ( Emsley et al . , 2010 ) . Refinement of the 40S subunit plus Alu domain , and 60S subunit plus SRP54 , were carried out individually using using REFMAC v5 . 8 ( Murshudov et al . , 2011 ) as previously described ( Amunts et al . , 2014; Brown et al . , 2015 ) . Secondary structure restraints were generated in ProSMART ( Nicholls et al . , 2012 ) , and nucleic acid base-pairing and stacking restraints were generated as before ( Amunts et al . , 2014 ) and were maintained throughout refinement to prevent over-fitting . Local resolution was calculated using ResMap ( Kucukelbir et al . , 2014 ) and all figures were generated using Pymol ( DeLano , 2006 ) and Chimera ( Goddard et al . , 2007 ) . | Proteins are long chain-like molecules built from smaller building blocks , called amino acids , by a large molecular machine known as a ribosome . Although all proteins are assembled inside cells , some of them must be delivered to the outside or inserted into cell membranes . It is important to understand how this selective delivery system works because secreted proteins ( i . e . , those delivered outside ) and membrane-embedded proteins are essential for cells to communicate with their surroundings . Proteins destined for secretion or membrane insertion contain characteristic stretches of amino acids that act as a targeting signal for delivery to the membrane . These targeting signals are recognized by the ‘signal recognition particle’ ( or SRP for short ) , a large complex found in all living organisms . The SRP has the task of finding ribosomes that are assembling proteins with a targeting signal , and then taking them to the membrane . The protein being assembled can then either cross the membrane for secretion by the cell , or get embedded within the membrane . So , how can the SRP scan the broad range of proteins that are made by the ribosome and engage with only those containing targeting signals ? Voorhees and Hegde investigated this question by analyzing SRPs bound to ribosomes that were at different stages of building a membrane protein . The experiment was devised so that SRP would be in two different states: in the first state , the SRP was scanning for its targeting signal and , in the second , it was engaged with the targeting signal . Voorhees and Hegde took many thousands of pictures of these samples using a technique called cryo-electron microscopy , and reconstructed the three-dimensional structures of both states . This revealed fine details of how SRP positions itself immediately next to the part of the ribosome where newly formed protein chains emerge . From here , the SRP scans the protein until the targeting signal emerges and then it engages with the protein . Engaging the targeting signal just as it emerges from the ribosome is probably important because targeting signals tend to aggregate if they are exposed to the contents of a cell . The new structures show how SRP cradles the targeting signal inside a binding groove and covers it with a protective lid to minimize its risk of aggregation . The next challenges are to figure out how SRP chooses which ribosomes to scan , and how it releases the targeting signal when it has delivered it to the membrane . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2015 | Structures of the scanning and engaged states of the mammalian SRP-ribosome complex |
How does attentional modulation of neural activity enhance performance ? Here we use a deep convolutional neural network as a large-scale model of the visual system to address this question . We model the feature similarity gain model of attention , in which attentional modulation is applied according to neural stimulus tuning . Using a variety of visual tasks , we show that neural modulations of the kind and magnitude observed experimentally lead to performance changes of the kind and magnitude observed experimentally . We find that , at earlier layers , attention applied according to tuning does not successfully propagate through the network , and has a weaker impact on performance than attention applied according to values computed for optimally modulating higher areas . This raises the question of whether biological attention might be applied at least in part to optimize function rather than strictly according to tuning . We suggest a simple experiment to distinguish these alternatives .
Covert visual attention—applied according to spatial location or visual features—has been shown repeatedly to enhance performance on challenging visual tasks ( Carrasco , 2011 ) . To explore the neural mechanisms behind this enhancement , neural responses to the same visual input are compared under different task conditions . Such experiments have identified numerous neural modulations associated with attention , including changes in firing rates , noise levels , and correlated activity ( Treue , 2001; Cohen and Maunsell , 2009; Fries et al . , 2001; Maunsell and Cook , 2002 ) . But how do these neural activity changes impact performance ? Previous theoretical studies have offered helpful insights on how attention may work to enhance performance ( Navalpakkam and Itti , 2007; Rolls and Deco , 2006; Tsotsos et al . , 1995; Cave , 1999; Hamker and Worcester , 2002; Wolfe , 1994; Hamker , 1999; Eckstein et al . , 2009; Borji and Itti , 2014; Whiteley and Sahani , 2012; Bundesen , 1990; Treisman and Gelade , 1980; Verghese , 2001; Chikkerur et al . , 2010 ) . However , much of this work is either based on small , hand-designed models or lacks direct mechanistic interpretability . Here , we utilize a large-scale model of the ventral visual stream to explore the extent to which neural changes like those observed experimentally can lead to performance enhancements on realistic visual tasks . Specifically , we use a deep convolutional neural network trained to perform object classification to test effects of the feature similarity gain model of attention ( Treue and Martínez Trujillo , 1999 ) . Deep convolutional neural networks ( CNNs ) are popular tools in the machine learning and computer vision communities for performing challenging visual tasks ( Rawat and Wang , 2017 ) . Their architecture—comprised of layers of convolutions , nonlinearities , and response pooling—was designed to mimic the retinotopic and hierarchical nature of the mammalian visual system ( Rawat and Wang , 2017 ) . Models of a similar form have been used to study the biological underpinnings of object recognition for decades ( Fukushima , 1988; Riesenhuber and Poggio , 1999; Serre et al . , 2007 ) . Recently it has been shown that when these networks are trained to successfully perform object classification on real-world images , the intermediate representations learned are remarkably similar to those of the primate visual system , making CNNs state-of-the-art models of the ventral stream ( Yamins et al . , 2014; Khaligh-Razavi et al . , 2017; Khaligh-Razavi and Kriegeskorte , 2014; Kheradpisheh et al . , 2016; Kar et al . , 2017; Cadena et al . , 2017; Tripp , 2017; Love et al . , 2017; Kubilius et al . , 2016 ) . A key finding has been the correspondence between different areas in the ventral stream and layers in the deep CNNs , with early convolutional layers best able to capture the representation of V1 and middle and higher layers best able to capture V4 and IT , respectively ( Güçlü and van Gerven , 2015; Eickenberg et al . , 2017; Yamins et al . , 2014 ) . The generalizability of these networks is limited , however , and the models are not able to match all elements of visual behaviour ( Ullman et al . , 2016; Azulay and Weiss , 2018; Baker et al . , 2018 ) . But given that CNNs can reach near-human performance on some visual tasks and have architectural and representational similarities to the visual system , they are well-positioned for exploring how neural correlates of attention can impact behaviour . One popular framework to describe attention’s effects on firing rates is the feature similarity gain model ( FSGM ) . This model , introduced by Treue and Martinez-Trujillo , claims that a neuron’s activity is multiplicatively scaled up ( or down ) according to how much it prefers ( or doesn’t prefer ) the properties of the attended stimulus ( Treue and Martínez Trujillo , 1999; Martinez-Trujillo and Treue , 2004 ) . Attention to a certain visual attribute , such as a specific orientation or color , is generally referred to as feature-based attention ( FBA ) . FBA effects are spatially global: if a task performed at one location in the visual field activates attention to a particular feature , neurons that represent that feature across the visual field will be affected ( Zhang and Luck , 2009; Saenz et al . , 2002 ) . Overall , this leads to a general shift in the representation of the neural population towards that of the attended stimulus ( Çukur et al . , 2013; Kaiser et al . , 2016; Peelen and Kastner , 2011 ) . Spatial attention implies that a particular portion of the visual field is being attended . According to the FSGM , spatial location is treated as an attribute like any other . Therefore , a neuron’s modulation due to attention can be predicted by how well it’s preferred features and spatial receptive field align with the features or location of the attended stimulus . The effects of combined feature and spatial attention have been found to be additive ( Hayden and Gallant , 2009 ) . A debated issue in the attention literature is where in the visual stream attention effects can be seen . Many studies of attention focus on V4 and MT/MST ( Treue , 2001 ) , as these areas have reliable attentional effects . Some studies do find effects at earlier areas ( Moro et al . , 2010 ) , though they tend to be weaker and occur later in the visual response ( Kastner and Pinsk , 2004 ) . Therefore , a leading hypothesis is that attention signals , coming from prefrontal areas ( Moore and Armstrong , 2003; Monosov et al . , 2011; Bichot et al . , 2015; Kornblith and Tsao , 2017 ) , target later visual areas , and the feedback connections that those areas send to earlier ones cause the weaker effects seen there later ( Buffalo et al . , 2010; Luck et al . , 1997 ) . In this study , we define the FSGM of attention mathematically and implement it in a deep CNN . By applying attention at different layers in the network and for different tasks , we see how neural changes at one area propagate through the network and change performance .
The feature similarity gain model of attention posits that neural activity is modulated by attention in proportion to how strongly a neuron prefers the attended features , as assessed by its tuning . However , the relationship between a neuron’s tuning and its ability to influence downstream readouts remains a difficult one to investigate biologically . We use our hierarchical model to explore this question . We do so by using back propagation to calculate 'gradient values' , which we compare to tuning curves ( see Materials and methods , 'Object category gradient calculations' and 'Tuning values' for details ) . Gradient values indicate the ways in which feature map activities should change in order to make the network more likely to classify an image as being of a certain object category . Tuning values represent the degree to which the feature map responds preferentially to images of a given category . If there is a correspondence between tuning and classification , a feature map that prefers a given object category ( that is , responds strongly to it ) should also have a high positive gradient value for that category . In Figure 2A we show gradient values and tuning curves for three example feature maps . In Figure 2C , we show the average correlation coefficients between tuning values and gradient values for all feature maps at each of the 13 convolutional layers . As can be seen , tuning curves in all layers show higher correlation with gradient values than expected by chance ( as assayed by shuffled controls ) , but this correlation is relatively low , increasing across layers from about . 2 to . 5 . Overall tuning quality also increases with layer depth ( Figure 2B ) , but less strongly . Even at the highest layers , there can be serious discrepancies between tuning and gradient values . In Figure 2D , we show the gradient values of feature maps at the final four convolutional layers , segregated according to tuning value . In red are gradient values that correspond to tuning values greater than one ( for example , category 12 for the feature map in the middle pane of Figure 2A ) . As these distributions show , strong tuning values can be associated with weak or even negative gradient values . Negative gradient values indicate that increasing the activity of that feature map makes the network less likely to categorize the image as the given category . Therefore , even feature maps that strongly prefer a category ( and are only a few layers from the classifier ) still may not be involved in its classification , or even be inversely related to it . This is aligned with a recent neural network ablation study that shows category selectivity does not predict impact on classification ( Morcos et al . , 2018 ) . To determine if manipulation according to tuning values can enhance performance , we created challenging visual images composed of multiple objects for the network to classify . These test images are of two types: merged ( two object images transparently overlaid , such as in Serences et al . , 2004 ) or array ( four object images arranged on a grid ) ( see Figure 1C examples ) . The task for the network is to detect the presence of a given object category in these images . It does so using a series of binary classifiers trained on standard images of these objects , which replace the last layer of the network ( Figure 1B ) . The performance of these classifiers on the test images indicates that this is a challenging task for the network ( 64 . 4% on merged images and 55 . 6% on array , Figure 1D . Chance is 50% ) , and thus a good opportunity to see the effects of attention . We implement feature-based attention in this network by modulating the activity of units in each feature map according to how strongly the feature map prefers the attended object category ( see Materials and methods , 'Tuning values' and 'How attention is applied' ) . A schematic of this is shown in Figure 3A . The slope of the activation function of units in a given feature map is scaled according to the tuning value of that feature map for the attended category ( positive tuning values increase the slope while negative tuning values decrease it ) . Thus the impact of attention on activity is multiplicative and bi-directional . The effects of attention are measured when attention is applied in this way at each layer individually ( Figure 3B; solid lines ) or all layers simultaneously ( Figure 3—figure supplement 1A , red ) . For both image types ( merged and array ) , attention enhances performance and there is a clear increase in performance enhancement as attention is applied at later layers in the network ( numbering is as in Figure 1A ) . In particular , attention applied at the final convolutional layer performs best , leading to an 18 . 8% percentage point increase in binary classification on the merged images task and 22 . 8% increase on the array images task . Thus , FSGM-like effects can have large beneficial impacts on performance . Attention applied at all layers simultaneously does not lead to better performance than attention applied at any individual layer ( Figure 3—figure supplement 1A ) . We also performed a control experiment to ensure that nonspecific scaling of activity does not alone enhance performance ( Figure 3—figure supplement 1C ) . Some components of the FSGM are debated , for example whether attention impacts responses multiplicatively or additively ( Boynton , 2009; Baruni et al . , 2015; Luck et al . , 1997; McAdams and Maunsell , 1999 ) , and whether the activity of cells that do not prefer the attended stimulus is actually suppressed ( Bridwell and Srinivasan , 2012; Navalpakkam and Itti , 2007 ) . Comparisons of different variants of the FSGM can be seen in Figure 3—figure supplement 2 . In general , multiplicative and bidirectional effects work best . We also measure performance when attention is applied using gradient values rather than tuning values ( these gradient values are calculated to maximize performance on the binary classification task , rather than classify the image as a given category; therefore technically they differ from those shown in Figure 2 , however in practice they are strongly correlated . See Materials and methods , 'Object category gradient calculations' and 'Gradient values' for details ) . Attention applied using gradient values shows the same layer-wise trend as when using tuning values . It also reaches the same performance enhancement peak when attention is applied at the final layers . The major difference , however , comes when attention is applied at middle layers of the network . Here , attention applied according to gradient values outperforms that of tuning values . In the previous section , we examined the best possible effects of attention by choosing the strength for each layer and category that optimized performance . Here , we look at how performance changes as we vary the overall strength ( β ) of attention . In Figure 4A we break the binary classification performance into true and false positive rates . Here , each colored line indicates a different category and increasing dot size represents increasing strength of attention . Ideally , true positives would increase without an equivalent increase ( and possibly with a decrease ) in false positive rates . If they increase in tandem , attention does not have a net beneficial effect . Looking at the effects of applying attention at different layers , we can see that attention at lower layers is less effective at moving the performance in this space and that movement is in somewhat random directions , although there is an average increase in performance with moderate attentional strength . With attention applied at later layers , true positive rates are more likely to increase for moderate attentional strengths , while substantial false positive rate increases occur only with higher strengths . Thus , when attention is applied with modest strength at layer 13 , most categories see a substantial increase in true positives with only modest increases in false positives . As strength continues to increase however , false positives increase substantially and eventually lead to a net decrease in overall classifier performance ( representing as crossing the dotted line in Figure 4A ) . Applying attention according to negated tuning values leads to a decrease in true and false positive values with increasing attention strength , which decreases overall performance ( Figure 4—figure supplement 1A ) . This verifies that the effects of attention are not from non-specific changes in activity . Experimentally , when switching from no or neutral attention , neurons in MT showed an average increase in activity of 7% when attending their preferred motion direction ( and similar decrease when attending the non-preferred ) ( Martinez-Trujillo and Treue , 2004 ) . In our model , when β= . 75 ( roughly the value at which performance peaks at later layers; Figure 4—figure supplement 1B ) , given the magnitude of the tuning values ( average magnitude: . 38 ) , attention scales activity by an average of 28 . 5% . This value refers to how much activity is modulated in comparison to the β=0 condition , which is probably more comparable to passive or anesthetized viewing , as task engagement has been shown to scale neural responses generally ( Page and Duffy , 2008 ) . This complicates the relationship between modulation strength in our model and the values reported in the data . To allow for a more direct comparison , in Figure 4B , we collected the true and false positive rates obtained experimentally during different object detection tasks ( explained in Materials and methods , 'Experimental data' ) , and plotted them in comparison to the model results when attention is applied at layer 13 using tuning values ( pink line ) or gradient value ( brown line ) . Five experiments ( second through sixth studies ) are human studies . In all of these , uncued trials are those in which no information about the upcoming visual stimulus is given , and therefore attention strength is assumed to be low . In cued trials , the to-be-detected category is cued before the presentation of a challenging visual stimulus , allowing attention to be applied to that object or category . The majority of these experiments show a concurrent increase in both true and false positive rates as attention strength is increased . The rates in the uncued conditions ( smaller dots ) are generally higher than the rates produced by the β=0 condition in our model , consistent with neutrally cued conditions corresponding to β>0 . We find ( see Materials and methods , 'Experimental data' ) , that the average corresponding β value for the neutral conditions is . 37 and for the attended conditions . 51 . Because attention scales activity by 1+βfclk ( where fclk is the tuning value ) , these changes correspond to a ≈5% change in activity . The first dataset included in the plot ( Ori-Change; yellow line in Figure 4B ) comes from a macaque change detection study ( see Materials and methods , 'Experimental data' for details ) . Because the attention cue was only 80% valid , attention strength could be of three levels: low ( for the uncued stimuli on cued trials ) , medium ( for both stimuli on neutrally-cued trials ) , or high ( for the cued stimuli on cued trials ) . Like the other studies , this study shows a concurrent increase in both true positive ( correct change detection ) and false positive ( premature response ) rates with increasing attention strength . For the model to achieve the performance changes observed between low and medium attention a roughly 12% activity change is needed , but average V4 firing rates recorded during this task show an increase of only 3 . 6% . This discrepancy may suggest that changes in correlations ( Cohen and Maunsell , 2009 ) or firing rate changes in areas aside from V4 also make important contributions to observed performance changes . Thus , according to our model , the size of experimentally observed performance changes is broadly consistent with the size of experimentally observed neural changes . While other factors are likely also relevant for performance changes , this rough alignment between the magnitude of firing rate changes and magnitude of performance changes supports the idea that the former could be a major causal factor for the latter . In addition , the fact that the model can capture this relationship provides further support for its usefulness as a model of the biology . Finally , we show the change in true and false positive rates when the threshold of the final layer binary classifier is varied ( a ‘receiver operating characteristic’ analysis , Figure 4B , gray line; no attention was applied during this analysis ) . Comparing this to the pink line , it is clear that varying the strength of attention applied at the final convolutional layer has more favorable performance effects than altering the classifier threshold ( which corresponds to an additive effect of attention at the classifier layer ) . This points to the limitations that could come from attention targeting only downstream readout areas . Overall , the model roughly matches experiments in the amount of neural modulation needed to create the observed changes in true and false positive rates . However , it is clear that the details of the experimental setup are relevant , and changes aside from firing rate and/or outside the ventral stream also likely play a role ( Navalpakkam and Itti , 2007 ) . Some of the results presented above , particularly those related to the layer at which attention is applied , may be influenced by the fact that we are using an object categorization task . To see if results are comparable using the simpler stimuli frequently used in macaque studies , we created an orientation detection task ( Figure 5A ) . Here , binary classifiers trained on full-field oriented gratings are tested using images that contain two gratings of different orientation and color . The performance of these binary classifiers without attention is above chance ( distribution across orientations shown in inset of Figure 5A ) . The performance of the binary classifier associated with vertical orientation ( 0 degrees ) was abnormally high ( 92% correct without attention , other orientations average 60 . 25% . This likely reflects the over-representation of vertical lines in the training images ) and this orientation was excluded from further performance analysis . Attention is applied according to orientation tuning values of the feature maps ( tuning quality by layer is shown in Figure 5B ) and tested across layers . We find ( Figure 5D , solid line and Figure 3—figure supplement 1B , red ) that the trend in this task is similar to that of the object task: applying attention at later layers leads to larger performance increases ( 14 . 4% percentage point increase at layer 10 ) . This is despite the fact that orientation tuning quality peaks in the middle layers . We also calculate the gradient values for this orientation detection task . While overall the correlations between gradient values and tuning values are lower ( and even negative for early layers ) , the average correlation still increases with layer ( Figure 5C ) , as with the category detection task . Importantly , while this trend in correlation exists in both detection tasks tested here , it is not a universal feature of the network or an artifact of how these values are calculated . Indeed , an opposite pattern in the correlation between orientation tuning and gradient values is shown when using attention to orientation to classify the color of a stimulus with the attended orientation ( see 'Recordings show how feature similarity gain effects propagate' , and Materials and methods , 'Oriented grating attention tasks' and 'Gradient values' ) . The results of applying attention according to gradient values is shown in Figure 5D ( dashed line ) . Here again , using gradient value creates similar trends as using tuning values , with gradient values performing better in the middle layers . Signal detection theory is frequently used to characterize the effects of attention on performance ( Verghese , 2001 ) . Here , we use a joint feature-spatial attention task to explore effects of attention in the model . The task uses the same two-grating stimuli described above . The same binary orientation classifiers are used and the task of the model is to determine if a given orientation is present in a given quadrant of the image . Performance is then measured when attention is applied to an orientation , a quadrant , or both an orientation and a quadrant ( effects are combined additively , for more , see Materials and methods , 'How attention is applied' ) . Two key signal detection measurements are computed: criteria and sensitivity . Criteria is a measure of the threshold that’s used to mark an input as positive , with a higher criteria leading to fewer positives; sensitivity is a measure of the separation between the two populations ( positives and negatives ) , with higher sensitivity indicating a greater separation . Figure 5E shows that both spatial and feature-based attention influence sensitivity and criteria . However , feature-based attention decreases criteria more than spatial attention does . Intuitively , feature-based attention shifts the representations of all stimuli in the direction of the attended category , implicitly lowering the detection threshold . Starting from a high threshold , this can lead to the observed behavioural pattern wherein true positives increase before false positives do . Sensitivity increases more for spatial attention alone than for feature-based attention alone , indicating that spatial attention amplifies differences in the representation of whichever features are present . These general trends hold regardless of the layer at which attention is applied and whether feature-based attention is applied using tuning curves or gradients . Changes in true and false positive rates for this task can be seen explicitly in Figure 5—figure supplement 1 . In line with our results , spatial attention was found experimentally to increase sensitivity and ( less reliably ) decrease criteria ( Hawkins et al . , 1990; Downing , 1988 ) . Furthermore , feature-based attention is known to decrease criteria , with lesser effects on sensitivity ( Rahnev et al . , 2011; Bang and Rahnev , 2017; though see Stein and Peelen , 2015 ) . A study that looked explicitly at the different effects of spatial and category-based attention ( Stein and Peelen , 2017 ) found that spatial attention increases sensitivity more than category-based attention ( most visible in their Experiment 3c , which uses natural images ) , and the effects of the two are additive . Attention and priming are known to impact neural activity beyond pure sensory areas ( Krauzlis et al . , 2013; Crapse et al . , 2018 ) . This idea is borne out by a study that aimed to isolate the neural changes associated with sensitivity and criteria changes ( Luo and Maunsell , 2015 ) In this study , the authors designed behavioural tasks that encouraged changes in behavioural sensitivity or criteria exclusively: high sensitivity was encouraged by associating a given stimulus location with higher overall reward , while high criteria was encouraged by rewarding correct rejects more than hits ( and vice versa for low sensitivity/criteria ) . Differences in V4 neural activity were observed between trials using high versus low sensitivity stimuli . No differences were observed between trials using high versus low criteria stimuli . This indicates that areas outside of the ventral stream ( or at least outside V4 ) are capable of impacting criteria ( Sridharan et al . , 2017 ) . Importantly , it does not mean that changes in V4 don’t impact criteria , but merely that those changes can be countered by the impact of changes in other areas . Indeed , to create sessions wherein sensitivity was varied without any change in criteria , the authors had to increase the relative correct reject reward ( i . e . , increase the criteria ) at locations of high absolute reward , which may have been needed to counter a decrease in criteria induced by attention-related changes in V4 ( similarly , they had to decrease the correct reject reward at low reward locations ) . Our model demonstrates clearly how such effects from sensory areas alone can impact detection performance , which , in turn highlights the role downstream areas may play in determining the final behavioural outcome . To explore how attention applied at one location in the network impacts activity later on , we apply attention at various layers and 'record' activity at others ( Figure 6A , in response to full field oriented gratings ) . In particular , we record activity of feature maps at all layers while applying attention at layers 2 , 6 , 8 , 10 , or 12 individually . To understand the activity changes occurring at each layer , we use an analysis from ( Martinez-Trujillo and Treue , 2004 ) that was designed to test for FSGM-like effects and is explained in Figure 6B . Here , the activity of a feature map in response to a given orientation when attention is applied is divided by the activity in response to the same orientation without attention . These ratios are organized according to the feature map’s orientation preference ( most to least ) and a line is fit to them . According to the FSGM of attention , this ratio should be greater than one for more preferred orientations and less than one for less preferred , creating a line with an intercept greater than one and negative slope . In Figure 6C , we plot the median value of the slopes and intercepts across all feature maps at a layer , when attention is applied at different layers ( indicated by color ) . When attention is applied directly at a layer according to its tuning values ( left ) , FSGM effects are seen by default ( intercept values are plotted in terms of how they differ from one; comparable average values from ( Martinez-Trujillo and Treue , 2004 ) are intercept: . 06 and slope: 0 . 0166 , but note we are using β=0 for the no-attention condition in the model which , as mentioned earlier , is not necessarily the best analogue for no-attention conditions experimentally . Therefore we use these measures to show qualitative effects ) . As these activity changes propagate through the network , however , the FSGM effects wear off , suggesting that activating units tuned for a stimulus at one layer does not necessarily activate cells tuned for that stimulus at the next . This misalignment between tuning at one layer and the next explains why attention applied at all layers simultaneously isn’t more effective ( Figure 3—figure supplement 1 ) . In fact , applying attention to a category at one layer can actually have effects that counteract attention at a later layer ( see Figure 6—figure supplement 1 ) . In Figure 6C ( right ) , we show the same analysis , but while applying attention according to gradient values . The effects at the layer at which attention is applied do not look strongly like FSGM , however FSGM properties evolve as the activity changes propagate through the network , leading to clear FSGM-like effects at the final layer . Finding FSGM-like behaviour in neural data could thus be a result of FSGM effects at that area or non-FSGM effects at an earlier area ( here , attention applied according to gradients which , especially at earlier layers , are not aligned with tuning ) . An alternative model of the neural effects of attention—the feature matching ( FM ) model—suggests that the effect of attention is to amplify the activity of a neuron whenever the stimulus in its receptive field matches the attended stimulus . In Figure 6D , we calculate the fraction of feature maps at a given layer that show feature matching behaviour ( defined as having activity ratios greater than one when the stimulus orientation matches the attended orientation for both preferred and anti-preferred orientations ) . As early as one layer post-attention , some feature maps start showing feature matching behaviour . The fact that the attention literature contains conflicting findings regarding the feature similarity gain model versus the feature matching model ( Motter , 1994; Ruff and Born , 2015 ) may result from this finding that FSGM effects can turn into FM effects as they propagate through the network . In particular , this mechanism can explain the observations that feature matching behaviour is observed more in FEF than V4 ( Zhou and Desimone , 2011 ) and that match information is more easily read out from perirhinal cortex than IT ( Pagan et al . , 2013 ) . We also investigated the extent to which measures of attention’s neural effects correlate with changes in performance ( see Materials and methods , 'Correlating activity changes with performance' ) . For this we developed a new , experimentally-feasible way of calculating attention’s effects on neural activity that is inspired by the gradient-based approach to attention ( that is , it focuses on classification rather than tuning ) . We show ( Figure 6—figure supplement 2 ) that this new measure better correlates with performance changes than the FSGM measure of activity changes , particularly at earlier layers . There is a simple experiment that would distinguish whether factors beyond tuning , such as gradients , play a role in guiding attention . It requires using two tasks with very different objectives ( which should produce different gradients ) but with the same attentional cue . An example is described in Figure 7 . Here , the two tasks used would be an orientation-based color classification task ( two gratings each with their own color and orientation are simultaneously shown , and the task is to report the color of the grating with the attended orientation ) and an orientation detection task ( report if the attended orientation is present or absent in the image ) . In both cases , attention is cued according to orientation . But gradient-based attention will produce different neural modulations for the two tasks , while the FSGM predicts identical modulations ( Figure 7C ) . Thus , an experiment that recorded from the same neurons during both tasks could distinguish between tuning-based and gradient-based attention .
In this work , we utilized a deep convolutional neural network ( CNN ) as a model of the visual system to probe the relationship between modulation of neural activity , as in attention , and performance . Specifically , we formally define the feature similarity gain model ( FSGM ) of attention ( the basic tenets of which have been described in several experimental studies ) as a multiplicative modulation of neuronal activity proportional to the neuron’s mean-subtracted feature tuning . This formalization allows us to investigate the FSGM’s ability to enhance a CNN’s performance on challenging visual tasks . We found that , across a variety of tasks , neural activity changes matching the type and magnitude of those observed experimentally can indeed lead to performance changes of the kind and magnitude observed experimentally . We used the full observability of the model to investigate the relationship between tuning and function . We compared attention applied according to feature tuning ( the FSGM ) with attention designed to optimally modulate activity to improve performance ( as determined by the gradient of performance with respect to the neural activity ) . Attention applied according to tuning does not successfully propagate from lower or middle to higher layers; that is , enhancing the activity of neurons that most prefer a given category at lower layers need not selectively enhance the activity of neurons preferring that category at higher layers . As a result , attention applied according to the FSGM performs poorly when applied at early to middle layers , while attention applied according to gradients at these layers performs better . Attention is most effective applied at later layers ( e . g . , layers 9–13 ) , where tuning and gradient values are better correlated . According to ( Güçlü and van Gerven , 2015 ) , these layers correspond most to areas V4 and LO . Such areas are known and studied for reliably showing attentional effects , whereas earlier areas such as V1 are generally not ( Luck et al . , 1997; Abdelhack and Kamitani , 2018 ) . In a study involving detection of objects in natural scenes , the strength of category-specific preparatory activity in object selective cortex was correlated with performance , whereas such preparatory activity in V1 was anti-correlated with performance ( Peelen and Kastner , 2011 ) . This is in line with our finding that feature-based attention effects at earlier areas can counter the beneficial effects of that attention at later areas ( Figure 6—figure supplement 1 ) . Our work raises the question: is attention applied simply according to tuning or is it targeted to best optimize function on a given task ? We suggested a simple experiment ( Figure 7 ) that would reveal whether non-tuning factors , such as gradients , guide attentional modulation . In ( Chelazzi et al . , 1998 ) the correlation coefficient between an index of tuning and an index of attentional modulation was . 52 for a population of V4 neurons , suggesting factors other than selectivity influence attention . Furthermore , many attention studies , including that one , use only preferred and anti-preferred stimuli and therefore don’t include a thorough investigation of the relationship between tuning and attentional modulation . ( Martinez-Trujillo and Treue , 2004 ) uses multiple stimuli to provide support for the FSGM , however the interpretation is limited by the fact that they only report population averages . ( Ruff and Born , 2015 ) investigated the relationship between tuning strength and the strength of attentional modulation on a cell-by-cell basis . While they did find a correlation ( particularly for binocular disparity tuning ) , it was relatively weak , which leaves room for the possibility that tuning is not the primary factor that determines attentional modulation . Local connectivity is also likely to play a role , as a correlation between normalization and attentional modulation has been shown ( Ni et al . , 2012 ) . A major challenge for understanding the biological implementation of selective attention is determining how such a precise attentional signal is carried by feedback connections . We believe that it is plausible that the visual system can learn the connections needed to carry out gradient-based attention . For example , if a high-level neuron related to the classification of an image sends a feedback connection to lower areas , an anti-Hebbian post-pre spike timing-dependent learning rule would strengthen the connection from the high level neuron to the low level one , if the lower level one causes the firing of the higher . In this way , neurons in later areas can learn to target the cells in earlier areas that caused them to fire . In contrast , it is actually more difficult to imagine how higher areas could learn the connections needed to target neurons according to their tuning , as in the FSGM . The machine learning literature on attention and learning may inspire other useful hypotheses on underlying brain mechanisms ( Xu et al . , 2015; Lillicrap et al . , 2016 ) . The concept of attention has been introduced in these models previously in the machine learning literature ( Mnih et al . , 2014 ) . Generally , this kind of attention relates to what would be called overt spatial attention in the neuroscience literature . That is , the attention mechanism serially selects areas of the input image for further processing , rather than modulating the activity of neurons representing those areas ( as in our model of spatial attention ) . Other work has been done using attention to selectively process image features ( Stollenga et al . , 2014 ) and it would be interesting to compare the workings of that model to the feature-based attention used in our study . While CNNs have representations that are similar to the ventral stream , they lack many biological details including recurrent connections , dynamics , cell types , and noisy responses . Preliminary work has shown that these elements can be incorporated into a CNN structure , and attention can enhance performance in this more biologically-realistic architecture ( Lindsay et al . , 2017 ) . Furthermore , while the current work does not include neural noise independent of the stimulus , the fact that a given image is presented in many contexts ( different merged images or different array images ) can be thought of as a form of highly structured noise that does produce variable responses to the same image . Another biological detail that this model lacks is 'skip connections , ’ where one layer feeds into both the layer directly after it and deeper layers after that ( He et al . , 2016; Huang et al . , 2017 ) as in connections from V2 to V4 or V4 to parietal areas ( Ungerleider et al . , 2008 ) . Our results regarding propagation of changes through the network suggest that synaptic distance from the classifier is a relevant feature—one that is less straight forward to determine in a network with skip connections . Because experimenters can easily control the image , defining a cell’s function in terms of how it responds to stimuli makes practical sense . However , it may be that thinking about visual areas in terms of their synaptic distance from decision-making areas such as prefrontal cortex ( Heekeren et al . , 2004 ) can be more useful for the study of attention than thinking in terms of their distance from the retina . Thus far , coarse stimulation protocols have found a relationship between the tuning of neural populations and their impact on perception ( Moeller et al . , 2017; DeAngelis et al . , 1998; Salzman et al . , 1990 ) . However , studies of the relationship between tuning and choice probabilities suggest that a neuron’s preferred stimulus is not always an indication of its causal role in classification ( Zaidel et al . , 2017; Purushothaman and Bradley , 2005 ) , though see ( Katz et al . , 2016 ) . Targeted stimulation protocols and a more fine-grained ability to determine both upstream drivers of , and downstream responses driven by , stimulated neurons will be needed to better address these issues .
The weights for the model ( 'VGG-16' ) came from Frossard ( 2017 ) ( RRID SCR_016494 ) . This work uses a deep convolutional neural network ( CNN ) as a model of the ventral visual stream . Convolutional neural networks are feed forward artificial neural networks that consist of a few basic operations repeated in sequence , key among them being the convolution . The specific CNN architecture used in the study comes from Simonyan and Zisserman , 2014 ( VGG-16D ) and is shown in Figure 1A ( a previous variant of this work used a smaller network ( Lindsay , 2015 ) . For this study , all the layers of the CNN except the final classifier layer were pre-trained using back propagation on the ImageNet classification task , which involves doing 1000-way object categorization ( weights provided by Frossard , 2017 ) . The training of the top layer is described in subsequent sections . Here we describe the basic workings of the CNN model we use , with details available in Simonyan and Zisserman , 2014 . The activity values of the units in each convolutional layer are the result of applying a 2-D spatial convolution to the layer below , followed by positive rectification ( rectified linear ’ReLu’ nonlinearity ) : ( 1 ) xijlk=[ ( Wlk⋆Xl−1 ) ij]+where ⋆ indicates convolution , and [x]+=x if x>0 , 0 otherwise . Wlk is the kth convolutional filter at the lth layer . The application of each filter results in a 2-D feature map ( the number of filters used varies across layers and is given in parenthesis in Figure 1A ) . xijlk is the activity of the unit at the i , jth spatial location in the kth feature map at the lth layer . Xl−1 is thus the activity of all units at the layer below the lth layer . The input to the network is a 224 by 224 pixel RGB image , and thus the first convolution is applied to these pixel values . Convolutional filters are 3 × 3 . For the purposes of this study the convolutional layers are most relevant , and will be referred to according to their numbering in Figure 1A ( numbers in parentheses indicate number of feature maps per layer ) . Max pooling layers reduce the size of the feature maps by taking the maximum activity value of units in a given feature map in non-overlapping 2 × 2 windows . Through this , the size of the feature maps decreases after each max pooling ( layers 1 and 2: 224 × 224; 3 and 4: 112 × 112; 5 , 6 , and 7: 56 × 56 . 8 , 9 , and 10: 28 × 28; 11 , 12 , and 13: 14 × 14 ) . The final two layers before the classifier are each fully-connected to the layer below them , with the number of units per layer given in parenthesis in Figure 1A . Therefore , connections exist from all units from all feature maps in the last convolutional layer ( layer 13 ) to all 4096 units of the next layer , and so on . The top readout layer of the network in ( Simonyan and Zisserman , 2014 ) contained 1000 units upon which a softmax classifier was used to output a ranked list of category labels for a given image . Looking at the top-5 error rate ( wherein an image is correctly labelled if the true category appears in the top five categories given by the network ) , this network achieved 92 . 7% accuracy . With the exception of the gradient calculations described below , we did not use this 1000-way classifier , but rather replaced it with a series of binary classifiers . The tasks we use to probe the effects of feature-based attention in this network involve determining if a given object category is present in an image or not , similar to tasks used in ( Stein and Peelen , 2017; Peelen et al . , 2009; Koivisto and Kahila , 2017 ) . To have the network perform this specific task , we replaced the final layer in the network with a series of binary classifiers , one for each category tested ( Figure 1B ) . We tested a total of 20 categories: paintbrush , wall clock , seashore , paddlewheel , padlock , garden spider , long-horned beetle , cabbage butterfly , toaster , greenhouse , bakery , stone wall , artichoke , modem , football helmet , stage , mortar , consomme , dough , bathtub . Binary classifiers were trained using ImageNet images taken from the 2014 validation set ( and were therefore not used in the training of the original model ) . A total of 35 unique true positive images were used for training for each category , and each training batch was balanced with 35 true negative images taken from the remaining 19 categories . The results shown here come from using logistic regression as the binary classifier , though trends in performance are similar if support vector machines are used . Once these binary classifiers are trained , they are then used to classify more challenging test images . Experimental results suggest that classifiers trained on unattended and isolated object images are appropriate for reading out attended objects in cluttered images ( Zhang et al . , 2011 ) . These test images are composed of multiple individual images ( drawn from the 20 categories ) and are of two types: 'merged’ and 'array’ . Merged images are generated by transparently overlaying two images , each from a different category ( specifically , pixel values from each are divided by two and then summed ) . Array images are composed of four separate images ( all from different categories ) that are scaled down to 112 by 112 pixels and placed on a two by two grid . The images that comprise these test images also come from the 2014 validation set , but are separate from those used to train the binary classifiers . See examples of each in Figure 1C . Test image sets are balanced ( 50% do contain the given category and 50% do not , 150 total test images per category ) . Both true positive and true negative rates are recorded and overall performance is the average of these rates . When neural networks are trained via back propagation , gradients are calculated that indicate how a given weight in the network impacts the final classification . We use this same method to determine how a given unit’s activity impacts the final classification . Specifically , we input a 'merged’ image ( wherein one of the images belongs to the category of interest ) to the network . We then use gradient calculations to determine the changes in activity that would move the 1000-way classifier toward classifying that image as belonging to the category of interest ( i . e . rank that category highest ) . We average these activity changes over images and over all units in a feature map . This gives a single value per feature map: ( 2 ) gclk=−1Nc∑n=1Nc 1HW∑i=1 , j=iH , W ∂E ( n ) ∂xijlk ( n ) where H and W are the spatial dimensions of layer l and Nc is the total number of images from the category ( here NC=35 , and the merged images used were generated from the same images used to generate tuning curves , described below ) . E ( n ) is the error of the 1000-way classifier in response to image n , which is defined as the cross entropy between the activity vector of the final layer ( after the soft-max operation ) and a one-hot vector , wherein the correct label is the only non-zero entry . Because we are interested in activity changes that would decrease the error value , we negate this term . The gradient value we end up with thus indicates how the feature map’s activity would need to change to make the network more likely to classify an image as the desired category . Repeating this procedure for each category , we obtain a set of gradient values ( one for each category , akin to a tuning curve ) , for each feature map: glk . Note that , as these values result from applying the chain rule through layers of the network , they can be very small , especially for the earliest layers . For this study , the sign and relative magnitudes are of more interest than the absolute values . In addition to attending to object categories , we also test attention on simpler stimuli . In the orientation detection task , the network detects the presence of a given orientation in an image . Again , the final layer of the network is replaced by a series of binary classifiers , one for each of 9 orientations ( 0 , 20 , 40 , 60 , 80 , 100 , 120 , 140 , and 160 degrees . Gratings had a frequency of . 025 cycles/pixel ) . The training sets for each were balanced ( 50% had only the given orientation and 50% had one of 8 other orientations ) and composed of full field ( 224 by 224 pixel ) oriented gratings in red , blue , green , orange , or purple ( to increase the diversity of the training images , they were randomly degraded by setting blocks of pixels ranging uniformly from 0% to 70% of the image to 0 at random ) . Test images were each composed of two oriented gratings of different orientation and color ( same options as training images ) . Each of these gratings were of size 112 by 112 pixels and placed randomly in a quadrant while the remaining two quadrants were black ( Figure 5A ) . Again , the test sets were balanced and performance was measured as the average of the true positive and true negative rates ( 100 test images per orientation ) . These same test images were used for a task wherein the network had to classify the color of the grating that had the attended orientation ( cross-featural task paradigms like this are commonly used in attention studies , such as Sàenz et al . , 2003 ) . For this , the final layer of the network was replaced with a 5-way softmax color classifier . This color classifier was trained using the same full field oriented gratings used to train the binary classifiers ( therefore , the network saw each color at all orientation values ) . For another analysis , a joint feature and spatial attention task was used . This task is almost identical to the setup of the orientation detection task , except that the searched-for orientation would only appear in one of the four quadrants . Therefore , performance could be measured when applying feature-based attention to the searched-for orientation , spatial attention to the quadrant in which it could appear , or both . This study aims to test variations of the feature similarity gain model of attention , wherein neural activity is modulated by attention according to how much the neuron prefers the attended stimulus . To replicate this in our model , we therefore must first determine the extent to which units in the network prefer different stimuli ( 'tuning values' ) . When attention is applied to a given category , for example , units’ activities are modulated according to these values . For the joint spatial-feature attention task ( Figure 5 ) , we calculated criteria ( c , 'threshold' ) and sensitivity ( d′ ) using true ( TP ) and false ( FP ) positive rates as follows ( Luo and Maunsell , 2015 ) : ( 7 ) c=−0 . 5 ( Φ−1 ( TP ) +Φ−1 ( FP ) ) where Φ−1 is the inverse cumulative normal distribution function . c is a measure of the distance from a neutral threshold situated between the mean of the true negative and true positive distributions . Thus , a positive c indicates a stricter threshold ( fewer inputs classified as positive ) and a negative c indicates a more lenient threshold ( more inputs classified as positive ) . The sensitivity was calculated as: ( 8 ) d′=Φ−1 ( TP ) −Φ−1 ( FP ) This measures the distance between the means of the distributions for true negative and two positives . Thus , a larger d′ indicates better sensitivity . To prevent the individual terms in these expressions from going to ±∞ , false positive rates of < . 01 were set to . 01 and true positive rates of > . 99 were set to . 99 . In Figure 6 , we examined the effects that applying attention at certain layers in the network ( specifically 2 , 6 , 8 , 10 , and 12 ) has on activity of units at other layers . Attention was applied with β= . 5 . The recording setup is designed to mimic the analysis of ( Martinez-Trujillo and Treue , 2004 ) . Here , the images presented to the network are full-field oriented gratings of all orientation-color combinations . Feature map activity is measured as the spatially averaged activity of all units in a feature map in response to an image . Activity in response to a given orientation is further averaged over all colors . We calculate the ratio of activity when attention is applied to a given orientation ( and the orientation is present in the image ) over activity in response to the same image when no attention is applied . These ratios are then organized according to orientation preference: the most preferred is at location 0 , then the average of next two most preferred at location 1 , and so on with the average of the two least preferred orientations at location 4 ( the reason for averaging of pairs is to match Martinez-Trujillo and Treue , 2004 as closely as possible ) . Fitting a line to these points gives a slope and intercept for each feature map ( lines are fit using the least squares method ) . FSGM predicts a negative slope and an intercept greater than one . To test for signs of feature matching behaviour , each feature map’s preferred ( most positive tuning value ) and anti-preferred ( most negative tuning value ) orientations are determined . Activity is recorded when attention is applied to the preferred or anti-preferred orientation and activity ratios are calculated . According to the FSGM , activity when the preferred orientation is attended should be greater than when the anti-preferred is attended , regardless of whether the image is of the preferred or anti-preferred orientation . According to the feature matching ( FM ) model , however , activity when attending the presented orientation should be greater than activity when attending an absent orientation , regardless of whether the orientation is preferred or not . Therefore , we say that a feature map is displaying feature matching behaviour if ( 1 ) activity is greater when attending the preferred orientation when the preferred is present versus when the anti-preferred is present , and ( 2 ) activity is greater when attending the anti-preferred orientation when the anti-preferred is present versus when the preferred is present . The second criteria distinguishes feature matching behaviour from FSGM . In Figure 6—figure supplement 2 , we use two different measures of attention-induced activity changes in order to probe the relationship between activity and classification performance . In both cases , the network is performing the orientation detection task described in Figure 5A and performance is measured only in terms of true positive rates . Because we know attention to increase both true and false positive rates , we would expect a positive correlation between activity changes and true positive performance , but a negative correlation between activity changes and true negative rates . This predicts that activity changes will have a monotonic relationship with true positive performance , but an inverted U-shaped relationship with total performance . Since we are calculating correlation coefficients of activity with performance , which measure a linear relationship , we use the rate of true positives as our measure of performance . The first measure is meant to capture feature similarity gain model-like behaviour in a way similar to the metric described in Figure 6B . The main difference is that that measure is calculated over a population of images of different stimuli , whereas the variant introduced here can be calculated on an image-by-image basis . Images that contain a given orientation are shown to the network and the spatially-averaged activity of feature maps is recorded when attention is applied to that orientation and when it is not . The ratio of these activities is then plotted against each feature map’s tuning value for the orientation . According to the FSGM , this ratio should be above one for feature maps with positive tuning values and less than one for those with negative tuning values . Therefore , we use the slope of the line fitted to these ratios plotted as a function of tuning values as an indication of the extent to which activity is FSGM-like ( with positive slopes more FSGM-like ) . The median slope over a set of images of a given orientation is paired with the change in performance on those images with attention . This gives one pair for each combination of orientation , strength ( β= . 15 , . 30 , . 45 , . 60 , . 75 , . 90 ) , and layer at which attention was applied ( activity changes are only recorded if attention was applied at or before the recorded layer ) . The correlation coefficient between these value pairs is plotted as the orange line in Figure 6—figure supplement 2C . The second measure aims to characterize activity in terms of its downstream effects , rather than the contents of the input ( 'Vector Angle' measure , see Figure 6—figure supplement 2A for a visualization ) . It is therefore more aligned with the gradient-based approach to attention rather than tuning , and is thus related to 'choice probability' measures ( Zaidel et al . , 2017; Purushothaman and Bradley , 2005 ) . First , for a particular orientation , images that both do and do not contain that orientation are shown to the network . Activity ( spatially-averaged over each feature map ) in response to images classified as containing the orientation ( i . e . , both true and false positives ) is averaged in order to construct a vector in activity space that represents positive classification for a given layer . To reduce complications of working with vectors in high dimensions , principal components are found that capture at least 90% of the variance of the activity in response to all images , and all computations are done in this lower dimensional space . The next step is to determine if attention moves activity in a given layer closer to this direction of positive classification . For this , only images that contain the given orientation are used . For each image , the cosine of the angle between the positive-classification vector and the activity in response to the image is calculated . The median of these angles over a set of images is calculated separately for when attention is applied and when it is not . The difference between these medians ( with-attention minus without-attention ) is paired with the change in performance that comes with attention on those images . Then the same correlation calculation is done with these pairs as described above . The outcome of these analyses is a correlation coefficient between the measure of activity changes and performance changes . This gives two values per layer: one for the FSGM-like measure and one for the vector angle measure . To determine if these two values are significantly different , we performed a bootstrap analysis . For this , correlation coefficients were recalculated using simulated data made by sampling with replacement from the true data . We do this 100 times and perform a two-sided t-test to test for differences between the two measures . Model results were compared to previously published data coming from several studies . In Lupyan and Ward , 2013 , a category detection task was performed using stereogram stimuli ( on object present trials , the object image was presented to one eye and a noise mask to another ) . The presentation of the visual stimuli was preceded by a verbal cue that indicated the object category that would later be queried ( cued trials ) or by meaningless noise ( uncued trials ) . After visual stimulus presentation , subjects were asked if an object was present and , if so , if the object was from the cued category ( categories were randomized for uncued trials ) . In Experiment 1 ( 'Cat-Drawings' in Figure 4B ) , the object images were line drawings ( one per category ) and the stimuli were presented for 1 . 5 s . In Experiment 2 ( 'Cat-Images' ) , the object images were grayscale photographs ( multiple per category ) and presented for 6 s ( of note: this presumably allows for several rounds of feedback processing , in contrast to our purelfeed forwardrd model ) . True positives were counted as trials wherein a given object category was present and the subject correctly indicated its presence when queried . False positives were trials wherein no category was present and subjects indicated that the queried category was present . In Lupyan and Spivey ( 2010 ) , a similar detection task was used . Here , subjects detected the presence of an uppercase letter that ( on target present trials ) was presented rapidly and followed by a mask . Prior to the visual stimulus , a visual ( 'Letter-Vis' ) or audio ( 'Letter-Aud' ) cue indicated a target letter . After the visual stimulus , the subjects were required to indicate whether any letter was present . True positives were trials in which a letter was present and the subject indicated it ( only uncued trials or validly cued trials—where the cued letter was the letter shown—were considered here ) . False positives were trials where no letter was present and the subject indicated that one was . The task in Koivisto and Kahila ( 2017 ) was also an object category detection task ( 'Objects' ) . Here , an array of several images was flashed on the screen with one image marked as the target . All images were color photographs of objects in natural scenes . In certain blocks , the subjects knew in advance which category they would later be queried about ( cued trials ) . On other trials , the queried category was only revealed after the visual stimulus ( uncued ) . True positives were trials in which the subject indicated the presence of the queried category when it did exist in the target image . False positives were trials in which the subject indicated the presence of the cued category when it was not in the target image . Data from trials using basic category levels with masks were used for this study . Finally , we include one study using macaques ( 'Ori-Change' ) wherein both neural and performance changes were measured ( Mayo and Maunsell , 2016 ) . In this task , subjects had to report a change in orientation that could occur in one of two stimuli . On cued trials , the change occurred in the cued stimulus in 80% of trials and the uncued stimulus in 20% of trials . On neutrally-cued trials , subjects were not given prior information about where the change was likely to occur ( 50% at each stimulus ) . Therefore performance could be compared under conditions of low ( uncued stimuli ) , medium ( neutrally cued stimuli ) , and high ( cued stimuli ) attention strength . Correct detection of an orientation change in a given stimulus ( indicated by a saccade ) is considered a true positive and a saccade to the stimulus prior to any orientation change is considered a false positive . True negatives are defined as correct detection of a change in the uncued stimulus ( as this means the subject correctly did not perceive a change in the stimulus under consideration ) and false negatives correspond to a lack of response to an orientation change . While this task includes a spatial attention component , it is still useful as a test of feature-based attention effects . Previous work has demonstrated that , during a change detection task , feature-based attention is deployed to the pre-change features of a stimulus ( Cohen and Maunsell , 2011; Mayo et al . , 2015 ) . Therefore , because the pre-change stimuli are of differing orientations , the cueing paradigm used here controls the strength of attention to orientation as well . In cases where the true and false positive rates were not published , they were obtained via personal communications with the authors . Not all changes in performance were statistically significant , but we plot them to show general trends . We calculate the activity changes required in the model to achieve the behavioural changes observed experimentally by using the data plotted in Figure 4B . We determine the average β value for the neutral and cued conditions by finding the β value of the point on the model line nearest to the given data point . Specifically , we average the β values found for the four datasets whose experiments are most similar to our merged image task ( Cat-Drawings , Cat-Images , Letter-Aud , and Letter-Vis ) . | Imagine you have lost your cell phone . Your eyes scan the cluttered table in front of you , searching for its familiar blue case . But what is happening within the visual areas of your brain while you search ? One possibility is that neurons that represent relevant features such as 'blue' and 'rectangular' increase their activity . This might help you spot your phone among all the other objects on the table . Paying attention to specific features improves our performance on visual tasks that require detecting those features . The 'feature similarity gain model' proposes that this is because attention increases the activity of neurons sensitive to specific target features , such as ‘blue’ in the example above . But is this how the brain solves such challenges in practice ? Previous studies examining this issue have relied on correlations . They have shown that increases in neural activity correlate with improved performance on visual tasks . But correlation does not imply causation . Lindsay and Miller have now used a computer model of the brain’s visual pathway to examine whether changes in neural activity cause improved performance . The model was trained to use feature similarity gain to detect an object within a set of photographs . As predicted , changes in activity like those that occur in the brain did indeed improve the model’s performance . Moreover , activity changes at later stages of the model's processing pathway produced bigger improvements than activity changes earlier in the pathway . This may explain why attention affects neural activity more at later stages in the visual pathway . But feature similarity gain is not the only possible explanation for the results . Lindsay and Miller show that another pattern of activity change also enhanced the model’s performance , and propose an experiment to distinguish between the two possibilities . Overall , these findings increase our understanding of how the brain processes sensory information . Work is ongoing to teach computers to process images as efficiently as the human visual system . The computer model used in this study is similar to those used in state-of-the-art computer vision . These findings could thus help advance artificial sensory processing too . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | How biological attention mechanisms improve task performance in a large-scale visual system model |
Calcium/calmodulin-dependent protein kinase II ( CAMK2 ) plays fundamental roles in synaptic plasticity that underlies learning and memory . Here , we describe a new recessive neurodevelopmental syndrome with global developmental delay , seizures and intellectual disability . Using linkage analysis and exome sequencing , we found that this disease maps to chromosome 5q31 . 1-q34 and is caused by a biallelic germline mutation in CAMK2A . The missense mutation , p . His477Tyr is located in the CAMK2A association domain that is critical for its function and localization . Biochemically , the p . His477Tyr mutant is defective in self-oligomerization and unable to assemble into the multimeric holoenzyme . In vivo , CAMK2AH477Y failed to rescue neuronal defects in C . elegans lacking unc-43 , the ortholog of human CAMK2A . In vitro , neurons derived from patient iPSCs displayed profound synaptic defects . Together , our data demonstrate that a recessive germline mutation in CAMK2A leads to neurodevelopmental defects in humans and suggest that dysfunctional CAMK2 paralogs may contribute to other neurological disorders .
Calcium/calmodulin-dependent protein kinase II ( CAMK2 ) is a calcium-activated serine/threonine kinase that is extremely abundant in the brain , comprising as much as 0 . 3% of the total brain protein content ( Bennett et al . , 1983 ) . CAMK2 is highly enriched at the synapses and is necessary for the process of long-term potentiation ( LTP ) , the activity-dependent strengthening and modulation of synaptic activity that is thought to be the molecular basis of some forms of learning and memory ( Kandel et al . , 2014; Lisman et al . , 2002 ) . In humans , there are four genes encoding distinct CAMK2 iso-enzymes . CAMK2A and CAMK2B are the predominant isoforms in the nervous system , with CAMK2A being expressed 3–4 times higher than CAMK2B ( Hanson and Schulman , 1992 ) . Each enzyme comprises a kinase domain , a regulatory domain and an association domain . Structurally CAMK2 holoenzymes are homo- or hetero-oligomers , consisting of 12 or 14 CAMK2 subunits ( Hudmon and Schulman , 2002b ) . The holoenzyme assembly requires the carboxy-terminal association domain , which forms stacked pairs of hexameric or heptameric rings with the regulatory and kinase domain projecting radially to interact with other essential proteins for CAMK2 function and localization ( Bhattacharyya et al . , 2016 ) . In the absence of calcium signalling , CAMK2 is inactive , as the regulatory domain inhibits kinase function ( Yang and Schulman , 1999 ) . This auto-inhibition is relieved when calcium-loaded calmodulin binds to the regulatory domain , thereby exposing the kinase domain and allowing it to phosphorylate target substrates ( Meador et al . , 1993 ) . Ca2+-Calmodulin binding also exposes Thr286 on the regulatory domain of CAMK2A , which becomes phosphorylated in trans by adjacent subunits . Once this residue is phosphorylated , the enzyme is persistently active , even in the absence of continual Ca2+-Calmodulin signaling ( Rich and Schulman , 1998; Stratton et al . , 2013 ) . This switch from auto-inhibition to autonomous , persistent activity is thought to constitute a biochemical form of memory , which marks the neuron for having experienced a previous calcium influx ( Bhattacharyya et al . , 2016; Stratton et al . , 2014; Stratton et al . , 2013 ) . Mice that are homozygous null for CAMK2A are viable and display impaired spatial memory and reduced LTP in the hippocampus ( Silva et al . , 1992a , 1992b ) . The heterozygous mutant ( Camk2a+/- ) show a significant deficit in spatial working memory and contextual fear memory ( Frankland et al . , 2001; Matsuo et al . , 2009 ) . More recently , it was demonstrated that mice with neuron-specific conditional knock-out of CAMK2A similarly displayed learning deficits and defects in LTP that were comparable to the complete knockout mice ( Achterberg et al . , 2014 ) . These findings suggest that neuron-intrinsic CAMK2A function is indispensable during the period of learning for memory formation . CAMK2 is conserved in invertebrates , such as D . melanogaster and C . elegans , where the kinase also plays critical roles in behavioral and cognitive traits ( Cho et al . , 1991; Hudmon and Schulman , 2002a; Reiner et al . , 1999; Rongo and Kaplan , 1999 ) . In C . elegans , the only CAMK2 is encoded by the unc-43 gene , which is essential for synaptic function ( Rongo and Kaplan , 1999 ) . Loss of unc-43 causes worms to have flaccid muscle tone , locomotion defects and spontaneous body contractions that resemble seizures ( Reiner et al . , 1999 ) . In pediatric care , global developmental delay in infants is defined as a significant functional delay in two or more developmental domains including gross and fine motor function , speech and language , cognition , social development and personal skills ( Quality Standards Subcommittee of the American Academy of Neurology et al . , 2003 ) . These defects are detected at an early age in children age five years or under , and can persist throughout life ( Shevell , 2008 ) . About 25–50% of identified case are caused by germline genetic changes , including chromosomal abnormalities , copy number variants and monogenic mutations ( Srour and Shevell , 2014; van Bokhoven , 2011 ) . For many patients with global neuro-developmental delay , the genetic etiology remains unknown .
Here , we report the identification of a consanguineous family from Jordan with two affected children manifesting global neuro-developmental delay with frequent seizures and convulsions . The two affected siblings had no dysmorphic features but failed to develop the ability to walk or speak ( Figure 1A , B , Figure 1—figure supplement 1A ) . They displayed progressive psychomotor retardation with hypotonic muscles ( Supplemental Material , Videos 1 and 2 ) . Electroencephalogram ( EEG ) analysis revealed abnormal epileptiform transients ( Figure 1C , Figure 1—figure supplement 1B ) , consistent with frequent myoclonic seizures . Magnetic resonance imaging ( MRI ) scan showed no major structural defects in the brain of proband II:4 ( Figure 1—figure supplement 1C ) . Serum metabolite levels were normal , ruling out potential lysosomal storage disorders . Assuming autosomal recessive inheritance , we performed identity-by-descent ( IBD ) homozygosity mapping using genomic DNA from both parents , two affected probands and three healthy siblings . Only one candidate locus greater than 5 cM on chromosome 5 spanning 28 Mb was delineated ( Figure 1D , Figure 1—figure supplement 1D ) . Whole-exome sequencing was subsequently performed on index case II:1 . After filtering for variants with low quality and low sequencing coverage , 72 homozygous variants were identified , out of which four lie within the Chr . 5 IBD region ( Table 1 ) . Three of the homozygous variants had been previously annotated as known polymorphisms with minor allele frequencies > 0 . 0001 . In addition , healthy individuals who are homozygous for the minor alleles had been identified in genomic sequencing databases such as dbSNP and gnomAD ( Figure 1—figure supplement 1E ) . We therefore filtered out these variants as non-pathogenic . The fourth homozygous variant within the IBD region mapped to the CAMK2A gene ( MIM: 114078 ) , resulting in a missense mutation p . His477Tyr that has never been observed in previous large-scale sequencing databases and our in-house ethnically-matched cohort ( Figure 1E , Figure 1—figure supplement 1E ) . Using Sanger sequencing , this private mutation was confirmed to segregate with the disease in all seven family members ( Figure 1A , D ) . CAMK2A is a neuron-specific , highly abundant serine/threonine kinase that plays critical roles in synaptic plasticity . To understand how neuronal function is affected due to the mutation in CAMK2A , we reprogrammed primary dermal fibroblasts from patient II . 4 into iPSCs , differentiated them into neurons and measured population-level neuronal activity using a multi-electrode array ( MEA ) system ( Figure 2A ) . Compared to H1 embryonic stem cell derived neurons and an unrelated CAMK2A wild-type iPSC control , the patient’s iPSCs were equally efficient in differentiating into mature neurons expressing neuronal markers TUJ1 and MAP2 after 21 days of in vitro differentiation ( Figure 2B ) . When these differentiated neurons were plated on MEA plates to measure spontaneous action potentials , we observed a significant reduction in both the total number of spontaneous spikes ( Figure 2C , left ) and the mean firing rate ( Figure 2C , right ) in the patient-derived neurons harboring the p . H477Y mutation compared to the wild-type controls , suggesting that CAMK2AH477Y causes a profound functional defect in cultured neurons . The CAMK2A enzyme consists of an N-terminal catalytic kinase domain , a Ca2+-calmodulin-binding regulatory domain and a C-terminal association domain ( AD ) that is necessary for the assembly of the 12- or 14-subunit holoenzyme . The identified mutation , p . H477Y is located in the association domain ( Figure 1E ) and affects a histidine residue that is invariant across all vertebrate and invertebrate CAMK2A homologues . It is also conserved in other human CAMK2 paralogs with a similar association domain such as CAMK2B , CAMK2D and CAMK2G ( Figure 3A , Figure 3—figure supplement 1A ) . In addition , this mutation is predicted to be deleterious by both SIFT , PolyPhen and MCAP algorithms . Structurally , His477 is located at the dimeric interface that forms part of the extensive interaction surface at the ‘equatorial’ plane of the CAMK2A holoenzyme ( Figure 3B ) . Based on prior findings that CAMK2A oligomerization through its association domain is indispensable for substrate phosphorylation and synaptic localization ( Bhattacharyya et al . , 2016 ) , we hypothesize that the p . H477Y allele is hypomorphic and that biallelic loss-of-function in CAMK2A is the direct cause for the neurodevelopmental phenotypes in the two probands . To measure the oligomerization potential of the CAMK2AH477Y mutant in cells , we transiently expressed FLAG-tagged wild-type CAMK2A and CAMK2AH477Y mutant in 293 T cells , which do not express endogenous CAMK2A . A third mutant CAMK2AH477X , which lacks amino acids 477–489 and thus encodes a truncated association domain , was used as an additional control . Using native lysis conditions that preserve non-covalent macromolecular interactions , we found that wild-type CAMK2A forms a prominent complex with an apparent molecular weight ~1 MDa ( Figure 3C , Native-PAGE , lane 2 ) , which is consistent with the 12- or 14- subunit CAMK2A holoenzyme ( Bhattacharyya et al . , 2016 ) . As compared to wild-type CAMK2A , the ~1 MDa , putatively oligomeric species was drastically reduced for the p . H477Y mutant and was undetectable for the p . H477X mutant ( Figure 3C , Native-PAGE , lane 3 and 4 vs . lane 2 ) . Next , we examined the ability of in vitro translated CAMK2A to self-oligomerize in a cell-free system . In contrast to the negative control protein GFP , wild-type FLAG-CAMK2A efficiently co-immunoprecipitated with HA-CAMK2A ( Figure 3D , Lane 10 vs 11 ) . This self-association was preserved between CAMK2AWT and CAMK2AH477Y ( Figure 3D , lanes 12 and 15 ) , but was completely lost between wild-type CAMK2A and the p . H477X mutant ( Figure 3D , lane 13 ) . In contrast , we could not detect any self-association between FLAG- CAMK2AH477Y and HA- CAMK2AH477Y ( Figure 3D , Lane 16 ) . Taken together , these results suggest that the missense p . H477Y partially disrupts the self-association between identical CAMK2A molecules , which had been shown to be required for holoenzyme assembly . The partial loss of function of the p . H477Y mutant , as compared to a more severe mutation p . H477X , is consistent with the observed autosomal recessive inheritance of the disease in the family , where the heterozygous carriers do not display apparent neuro-developmental symptoms . During the course of this analysis , we noticed that both p . H477Y and p . H477X mutants had reduced protein abundance . This effect on the p . H477Y mutant was , however , subtle and could not readily explain the difference in the CAMK2A oligomer observed in the native gel ( Figure 3C , SDS-PAGE ) . As it is known that in general fully assembled oligomeric complexes have enhanced stability in vivo compared to partially assembled complexes with disrupted folding like CAMK2AH477Y ( Lord , 1996; Oromendia et al . , 2012 ) , we hypothesized that the p . H477Y mutant may exhibit reduced stability and undergo proteasomal degradation . To test this , 293T cells were transfected with reporter constructs that encode GFP tagged wild-type and mutant CAMK2A followed by a self-cleaving peptide T2A and mCherry ( Figure 3E ) . The intensity of GFP fluorescence was used as a direct quantitative measure of CAMK2A stability with the mCherry as an internal control for transfection and translational efficiency . We observed a significant reduction in GFP intensity in cells expressing CAMK2AH477Y and CAMK2AH477X as compared to wild-type CAMK2A or GFP alone ( Figure 3E ) despite comparable mCherry fluorescence levels . This reduction was rescued when we treated the cells with MG132 , which blocked proteasomal degradation . MG132 treamentled to enhanced accumulation of the p . H477Y and p . H477X mutant , with a greater effect on p . H477X ( Figure 3F , lane 3 , 4 vs lane 7 , 8 , Figure 3—figure supplement 1B ) . By contrast , the level of the wild-type protein was reduced , likely due to the toxic effects of MG132 ( Figure 3F , lane 2 vs . lane 6 ) . These results suggest that in addition to causing reduced holoenzyme assembly , the p . H477Y mutation might also directly or indirectly reduce the overall CAMK2A levels by compromising its half-life . To unequivocally demonstrate the pathogenicity of the p . H477Y allele in vivo , we performed rescue experiments using C . elegans . A single ortholog of CAMK2 , encoded by unc-43 , is present in the worm genome and its functions in the nervous system are welldocumented ( Reiner et al . , 1999 ) . To study the neuronal defects caused by loss of unc-43 , we focused on motor neuron , DA9 , which has its cell body located in the pre-anal ganglion with a dendrite that extends anteriorly and a posteriorly oriented axon extending via a commissure into the dorsal nerve cord , where it proceeds anteriorly to form approximately 25 en passant synapses onto body wall muscles and reciprocal inhibitory neurons ( Figure 4A ) ( Klassen and Shen , 2007; Maeder et al . , 2014 ) . Using mCherry-tagged UNC-43 , we observed that UNC-43 could be found along the entire DA9 neuron but concentrated at synaptic boutons ( Figure 4B ) . The homologous patient-specific mutation p . H466Y in UNC-43 ( homologous to p . H477Y in human CAMK2A ) significantly disrupted the synaptic localization of UNC-43 and caused the protein to be dispersed throughout the entire axon ( Figure 4B ) . To test the consequence of UNC-43 mutation on DA9 synaptic function , we used the itr-1 pB promoter to express the fluorescently conjugated synaptic vesicle marker RAB-3 ( RAB3::GFP ) within DA9 in both the wild-type and unc-43 mutant background . In wild-type animals , RAB-3::GFP accumulated in discrete puncta along the axon of DA9 at stereotyped synaptic locations . In the canonical unc-43 ( e408 ) loss-of-function mutant , we observed a reduction in individual pre-synaptic puncta fluorescence intensity as compared to wild-type animals . This phenotype could be rescued by cell-autonomous expression of wild-type unc-43 in DA9 . However , expression of unc-43 harboring the homologous patient mutation UNC-43H466Y failed to rescue this defect ( Figure 4C , D ) . In addition , transgenic expression of human wild-type CAMK2A fully rescued this defect , confirming the high degree of functional conservation between CAMK2A homologs , while the patient-derived human CAMK2AH477Y failed to do so ( Figure 4C , bottom panels ) . These results suggest that the p . H477Y mutation is defective in supporting pre-synaptic function in C . elegans . Immediately posterior to the stretch of presynaptic puncta is an asynaptic domain within the DA9 axon that is devoid of any RAB-3::GFP fluorescence in wild-type animals ( Figure 4E , top panel ) . Loss of unc-43 results in the mislocalization of the synaptic marker RAB3::GFP into this asynaptic region ( Figure 4E , bottom panel ) . This defect could also be rescued by cell autonomous expression of either unc-43 or human CAMK2A . Patient derived mutation CAMK2AH477Y or the worm homologous mutation UNC-43H466Y both failed to rescue this phenotype ( Figure 4E ) . In addition , expression of UNC-43H466Y in wild-type animals did not cause any synaptic defect or mislocalizaton of RAB3::GFP , suggesting that the mutation does not have dominant negative effects ( Figure 4F ) . We further tested if the patient derived mutation in CAMK2A impacted worm locomotor behavior . Null mutants for unc-43 are flaccid in posture and move with a flattened uncoordinated waveform . The animals are variably convulsive , often spontaneously contracting and relaxing their body-wall muscles in brief repeating bursts that resemble seizures ( Reiner et al . , 1999 ) . We expressed either wild-type UNC-43 or UNC-43H466Y in the muscles and neurons of unc-43 ( e408 ) mutant worms and scored the behavior of young adults in a double-blind experiment . Only the wildtype UNC-43 was able to rescue the movement defects in unc-43 ( e408 ) animals , but not UNC-43H466Y , suggesting that UNC-43H466Y is not functional ( Figure 4G ) . Together , the data show that CAMK2AH477Y is a loss-of-function mutation which fails to support synaptic function in vivo . In summary , we have identified an autosomal recessive neurodevelopmental syndrome characterized by growth delay , frequent seizures and severe intellectual disability that is caused by a biallelic germline loss-of-function mutation in CAMK2A . Mechanistically this mutation disrupts CAMK2A self-oligomerization and holoenzyme assembly via its association domain . Our functional results are consistent with the high degree of evolutionary conservation of the affected residue H477 in CAMK2A orthologs , as well as previous structural data demonstrating that His477 is located in the interface between two stacked CAMK2A subunits , which together form the basic repeat unit of the ring-shaped holoenzyme ( Bhattacharyya et al . , 2016; Stratton et al . , 2014 ) . The pathogenicity of the biallelic p . H477Y mutation is furthered highlighted by rescue experiments in C . elegans , where in contrast to wild-type human CAMK2A the p . H477Y mutant failed to rescue the neuronal and behavioral defect in unc-43 ( CAMK2 ortholog ) null worms . Together , these data demonstrate that the loss of function of CAMK2A is the most plausible genetic cause for the neurodevelopmental defects observed in the two affected siblings .
CAMK2 plays important and evolutionary conserved roles in synaptic plasticity , neuronal transmission and cognition in near all model organisms examined , and several groups have shown that somatic mutations in human CAMK2 isoforms may contribute to neurological disorders ( Ghosh and Giese , 2015; Robison , 2014; Takemoto-Kimura et al . , 2017 ) . Notably , a de novo p . E183V mutation in the CAMK2A catalytic domain was shown to cause autism spectrum disorder ( Stephenson et al . , 2017 ) . This mutation was shown to act in a dominant-negative manner to reduce wild-type CAMK2A auto-phosphorylation and localization to dendritic spines . While this manuscript was under revision , S . Küry et al . reported multiple families with intellectual disability caused by de novo , heterozygous mutations in both CAMK2A and CAMK2B kinase and auto-regulatory domains , which disrupted CAMK2 phosphorylation and caused neuronal migratory defects in murine models ( Küry et al . , 2017 ) . Our discovery of a novel neuro-developmental syndrome caused by biallelic CAMK2A mutations further broadens the spectrum of human neurological disorders caused by the CAMK2 family of kinases . To the best of our knowledge , this represents the first Mendelian human disease caused by biallelic CAMK2A mutations . Our functional characterization of the novel mutation p . H477Y in vitro and in vivo also reveal novel insights on how the CAMK2A holoenzyme regulates neuronal function . In contrast to all previously reported mutations in CAMK2A in intellectual disability syndromes , the p . H477Y is located within the C-terminal association domain and results in a partial but significant disruption of self-oligomerization , suggesting that the assembly of CAMK2A oligomers , in addition to its kinase function is required for neuronal function . Interestingly , the CAMK2AH477Y mutant retains the ability to interact with wild-type CAMK2A but not with itself ( Figure 3D ) . CAMK2A displays very specific cellular and subcellular expression patterns that is important for regulating substrate phosphorylation in cells ( Liu and Murray , 2012; Tsui et al . , 2005 ) . The p . H477Y missense affects the subcellular localization in neurons and this may affect its ability to function efficiently . These biochemical results provide a mechanistic basis for the autosomal recessive nature of the disease in our family: the p . H477Y allele is hypomorphic and becomes pathogenic when recessively inherited in the homozygous state . We speculate that heterozygous carriers retain sufficient CAMK2A activity for proper neuronal function . As compared to the milder disease phenotypes reported by Kury et . al . , the symptoms afflicting the p . H477Y patients likely represent the most severe manifestation of the CAMK2A dysfunction in humans . Due to the high degree of conservation of CAMK2A across evolution , we employed the established C . elegans unc-43 mutant to prove the pathogenicity of the p . H477Y mutation . UNC-43 is the worm homologue of vertebrate CAMK2 . We demonstrated wild-type human CAMK2A could rescue the locomotive defects of the unc-43 mutants , while the p . H477Y mutant failed to do so , likely due to its inability to localize into neuronal synapses ( Figure 4 ) . We anticipate this in vivo functional assay in C . elegans to be widely applicable to assess the pathogenicity of newly discovered CAMK2 alleles found in human diseases .
Patients were identified and diagnosed by clinical geneticists at King Hussein Medical Centre , Amman , Jordan . Informed consent was obtained from the families for genetic testing in accordance with approved institutional ethical guidelines ( Institute of Medical Biology , Singapore , A*STAR , Singapore , NUS-IRB reference code 10–051 ) . For patients in Figure 1 , informed consent to publish photographs and videos was obtained from parents . Genomic DNA samples were isolated from saliva using the Oragene DNA collection kit ( OG-500 , DNAGenotek ) and a punch skin biopsy was taken from patient II . 4 . Whole exome sequencing of proband II-1 was performed using the Illumina TruSeq Exome Enrichment Kit for exome capture using 1 ug of genomic DNA . Illumina HiSeq2000 High-output mode was used for sequencing as 100 bp paired-end runs at the UCLA Clinical Genomics Centre and at the UCLA Broad Stem Cell Research Centre as previously described ( Hu et al . , 2014 ) . An average coverage of 34X was achieved across the exome with 87% of these bases covered at ≥10X . After filtering , a total of 73 homozygous , 125 compound heterozygous and 493 heterozygous variants were protein-changing variants with population minor allele frequencies < 1% . Both parents and their five children were genotyped using Illumina Humancore-12v1 BeadChips following manufacturer’s instructions . Call rates were above 99% . Gender and relationships were verified using Illumina BeadStudio . Mapping was performed by searching for shared regions that are homozygous and identical-by-descent ( IBD ) in the two affected children using custom programs written in the Mathematica ( Wofram Research , Inc . ) data analysis software ( Source code file 1 - IBD linkage program ) . Candidate regions were further refined by exclusion of common homozygous segments with any unaffected family members . The confidence criteria to identify IBD blocks were a minimum of 5 cM . We identified one shared candidate loci on chromosome 5 . HEK 293 T cells ( ATCC Cat# CRL-3216 , RRID:CVCL_0063 , from Lab Stock ) were cultured in DMEM media ( Gibco ) supplemented with 10% FBS . Cell line identity was authenticated by commercial human STR profiling with ATCC ( ATCC , #135-XV ) . All cell lines used in this paper tested negative for mycoplasma using Lonza MycoAlert ( Lonza LT07 ) . For transient transfection , 6 × 105 cells per well were seeded in 6 well plates 24 hr before being transfected with Lipofectamine 2000 ( ThermoFisher ) -complexed plasmids in OPTIMEM ( ThermoFisher ) . To construct the CAMK2A expression plasmids , human CAMK2A cDNA was PCR-amplified from ImageClone with AscI and PacI restriction sites and cloned into pCDNA3 . 1 with an N-terminal 3xFLAG or 3xHA tag . All CAMK2A mutants were generated using QuikChange XL ( Agilent ) . Cells were treated with MG132 ( Sigma , M8699 ) at 5 µM to block proteasome degradation . iPSCs-derived NPCs were differentiated into neurons for 21 days using a previously published protocol ( Xu et al . , 2017a ) . Briefly , NPCs were plated at a density of 50 , 000 cells/cm2 in a poly-L-ornithine and laminin-coated plates , cultured in N2B27 medium supplemented BDNF ( 20 ng/ml ) , GDNF ( 20 ng/ml ) , cAMP ( N6 , 2’-O-dibutyryladenosine 3’ , 5’-cyclic monophosphate; Sigma; 0 . 3 mM ) and ascorbic acid ( 0 . 2 mM ) . H1 embryonic stem cells ( Gift from Lawrence W . Stanton , WiCell , RRID:CVCL_C813 ) were also differentiated into neurons to use as controls . H1 embryonic stem cells were verified by karyotyping ( Cytogenetics Laboratory , KK Women's and Children's Hospital , Singapore ) and checked for pluripotency by differentiation into the three germ layers marked by Nestin ( ectoderm ) , AFP ( endoderm ) and ASM-1 ( mesoderm ) . For the immunofluorescence staining , neurons were fixed for 15 min in ice cold 4% ( w/v ) paraformaldehyde . Permeabilization using 0 . 3% ( v/v ) Triton-X in 1X PBS was performed for 10 min then incubated with 1:1000 mouse anti-Tuj1 ( Covance Research Products Inc Cat# MMS-435P , RRID:AB_2313773 ) , 1:1500 guinea pig anti-MAP2 ( Synaptic Systems Cat# 188 004 , RRID:AB_2138181 ) overnight at 4°C in 5% ( w/v ) BSA diluted with 1X PBS . For visualization , 1:1000 secondary antibody conjugated to Alexa Fluor 594 ( Thermo Fisher Scientific Cat# A-11076 , RRID:AB_2534120 ) or Alexa Fluor 488 ( Thermo Fisher Scientific Cat# A-11001 , RRID:AB_2534069 ) was applied . Counter staining for nuclei were performed using Dapi . Images were captured using the FV3000 Olympus confocal . For the multi-electrode array ( MEA ) recordings , neurons on day 21 were dissociated and replated on 0 . 1 polyethylenimine ( Sigma ) -coated 48 well MEA plates ( Axion Biosystems ) in BrainPhys media supplemented with BDNF , GDNF , cAMP and ascorbic acid as previously described ( Xu et al . , 2017b ) . Spontaneous neuronal activity was observed and recorded at 37°C for 5 min every 2–3 days using the Maestro MEA System ( Axion Biosystem ) . Independent measurements were taken from seven wells for each condition ( technical replicates ) . CAMK2A proteins were synthesized in vitro using TNT T7 Quick Coupled Transcription/Translation ( Promega ) with 1 μg of plasmids in 20 μl reaction volumes for 90 mins at 30°C . For co-immunoprecipitation , the reactions were diluted 10x in TBS ( 100 mM Tris-HCl , pH = 8 , 150 mM NaCl ) with 1% Nonidet P40 ( NP40 ) and incubated with 10 μl anti-FLAG M2 agarose beads ( Sigma ) at 4°C overnight . Proteins were eluted with 1x Laemmli buffer after 3 washes in the 1xTBS with 1% NP40 . Total protein lysate was quantified using a standard Bradford assay and 10 μg of lysate was used for immunoblotting experiments . For Blue Native PAGE , cells were lysed in 1x Sample Preparation buffer ( ThermoFisher ) containing 1% digitonin . 1% SDS was supplemented for SDS-PAGE . All proteins were transferred to PVDF membranes using TurboBlot ( Bio-rad ) at 2 . 5 mA for seven mins . The primary antibodies used were anti-FLAG ( M2 , Sigma-Aldrich Cat# F3165 , RRID:AB_259529 ) , anti-GAPDH ( Santa Cruz Biotechnology Cat# sc-47724 , RRID:AB_627678 ) and anti-HA ( Y-11 , Santa Cruz Biotechnology Cat# sc-805 , RRID:AB_631618 ) . The secondary antibodies were anti-rabbit IgG-HRP ( Jackson ImmunoResearch Labs Cat# 111-035-003 , RRID:AB_2313567 ) , anti-mouse IgG-HRP ( light chain specific ) ( Jackson ImmunoResearch Labs Cat# 205-032-176 , RRID:AB_2339056 ) and anti-mouse IgG-HRP ( Jackson ImmunoResearch Labs Cat# 115-035-003 , RRID:AB_10015289 ) . All strains were maintained at 20°C on OP50 E . coli nematode growth medium plates as described ( Brenner , 1974 ) . N2 Bristol strain worms ( WB Cat# CB4852 , RRID:WB-STRAIN:CB4852 ) were used as the WT reference , and the unc-43 ( e408 ) mutant was used . To visualize synaptic vesicles in DA9 neuron , wyIs85 [Pitr-1::GPF::RAB-3] was used ( Klassen and Shen , 2007 ) . OTL70 wyIs85[Podr-1::dsred , Pitr-1::gfp::rab-3]; jpnEx40[Podr-1::gfp , Pmig-13::unc-43] OTL71 wyIs85; jpnEx41[Podr-1::gfp , Pmig-13::unc-43] OTL72 unc-43 ( e408 ) ; wyIs85; jpnEx44[Podr-1::gfp , Pmig-13::unc-43 ( H466Y ) ] OTL73 unc-43 ( e408 ) ; wyIs85; jpnEx45[Podr-1::gfp , Pmig-13::unc-43 ( H466Y ) ] OTL74 unc-43 ( e408 ) ; wyIs85; jpnEx42[Podr-1::gfp , Pmig-13::unc-43] OTL75 unc-43 ( e408 ) ; wyIs85; jpnEx43[Podr-1::gfp , Pmig-13::unc-43] OTL76 unc-43 ( e408 ) ; wyIs85; jpnEx47[Podr-1::gfp , Pmig-13::CAMK2A] OTL77 unc-43 ( e408 ) ; wyIs85; jpnEx48[Podr-1::gfp , Pmig-13::CAMK2A] OTL78 unc-43 ( e408 ) ; wyIs85; jpnEx49[Podr-1::gfp , Pmig-13::CAMK2AH477Y] OTL79 unc-43 ( e408 ) ; wyIs85; jpnEx50[Podr-1::gfp , Pmig-13::CAMK2AH477Y]] OTL82: jpnEx54[Podr-1::gfp , Pmig-13::mcheery::unc-43] OTL83: jpnEx55[Podr-1::gfp , Pmig-13::mcherry::unc-43] OTL84: jpnEx56[Podr-1::gfp , Pmig-13::mcherry::unc-43 ( H466Y ) ] OTL85: jpnEx57[Podr-1::gfp , Pmig-13::mcherry::unc-43 ( H466Y ) ] OTL86: unc-43 ( e408 ) ; jpnEx58[Punc-122::dsred , Punc-104::mcherry::unc-43 , Phlh-1::mcherry::unc-43] OTL87: unc-43 ( e408 ) ; jpnEx59[Punc-122::dsred , Punc-104::mcherry::unc-43 , Phlh-1::mcherry::unc-43] OTL88: unc-43 ( e408 ) ; jpnEx60[Punc-122::dsred , Punc-104::mcherry::unc-43 ( H466Y ) , Phlh-1::mcherry::unc-43 ( H466Y ) ] OTL89: unc-43 ( e408 ) ; jpnEx61[Punc-122::dsred , Punc-104::mcherry::unc-43 ( H466Y ) , Phlh-1::mcherry::unc-43 ( H466Y ) ] Expression plasmids for transgenic worm lines were made using the pSM vector ( C . Bargmann ) , a derivative of pPD49 . 26 ( A . Fire ) . The mig-13 promoter was cloned between SphI/AscI , and C . elegans unc-43 isoform d or human CAMK2α was cloned between NheI/KpnI or AscI/NheI , respectively . P . H466Y and p . H477Y mutations were introduced by PCR-based mutagenesis using KOD-plus- high fidelity DNA polymerase ( TOYOBO , Tokyo , Japan ) . Transgenic worms were generated as described ( Mello , 1995 ) . Plasmids were injected into animals at 10 ng/μl ( in the case of Pmig-13::unc-43 ) and 50 ng/μl ( in the case of Pmig-13::CAMK2α ) together with coinjection markers at 90 ng/μl . All fluorescence images of DA9 were taken in live worms immobilized with 5% agar pad , 10 μM levamisol ( Sigma ) and 0 . 1 mm polystylene beads ( Polysciences , Inc . , Warrington , PA , USA ) with a 63×/1 . 4 NA objective on a Zeiss Axioplan 2 Imaging System or a Plan-Apochromat 63×/1 . 3 objective on a Zeiss LSM710 confocal microscope using similar imaging parameters for the same marker across different genotypes . Fluorescence quantification was determined using Image J software ( ImageJ , RRID:SCR_003070 ) . mcherry::unc-43 or mcherry::unc-43 ( H466Y ) were expressed in unc-43 ( e403 ) mutant worms using both unc-104 promoter ( neuron-specific promoter ) and hlh-1 promoter ( muscle-specific promoter ) . Two independent lines were established and analyzed . The behavioral phenotype of the transgenic worms was scored in a double-blind manner using a stereo microscope ( SZX16 , Olumpus , Tokyo , Japan ) . From the movement behavior on OP50 feeder NGM plates , each worm was classified to behave like ‘wild-type’ or ‘unc-43’ . 100 worms were observed for each genotype on the same day from two independently derived transgenic lines . | Each year , some children are born with developmental disorders and intellectual disabilities . These conditions are often caused by mutations in specific genes . Sometimes both copies of a gene – one inherited from each parent – need to be mutated for the symptoms to develop . These mutations are known as recessive mutations . Here , Chia , Zhong , Niwa et al . diagnosed two siblings in their clinical care with a new form of neurological disease that affects the development of the brain and leads to frequent seizures . To test whether the young patients shared a genetic mutation that could explain their condition , the researchers analyzed the DNA of the children and compared the results with the DNA from their parents and healthy siblings . The results showed that the two children with the condition had inherited a recessive mutation in a gene called CAMK2A . The protein this gene encodes helps nerve cells to form connections and communicate with each other , and it has been shown to be essential for learning and memory . The CAMK2A enzyme is made up of several identical subunits that form a complex . Chia et al . discovered that the mutation prevented these subunits from joining together properly , resulting in a faulty protein . CAMK2A and other related proteins are crucial for the health of the brain in a wide range of animals . Indeed , experiments in Caenorhabditis elegans , a roundworm commonly used to study neurons , confirmed that the mutation inherited by the children indeed caused similar neurological defects in the worms . Taken together , these experiments suggest that the children’s condition is caused by the mutation in both copies of the CAMK2A gene . For patients born with inherited diseases , it is often difficult to pinpoint exactly which mutation is responsible for the specific disorder . These findings could therefore help pediatric geneticists recognize this newly defined syndrome and reach the correct diagnoses . These results could also be the starting point for studies that look into restoring the activity of the defective CAMK2A protein . More broadly , identifying genes that are critical for the healthy development of the brain could shed light on common neurological conditions , such as epilepsy and autism . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | A homozygous loss-of-function CAMK2A mutation causes growth delay, frequent seizures and severe intellectual disability |
It is widely argued that personalized instruction based on individual differences in learning styles or genetic predispositions could improve learning outcomes . However , this proposition has resisted clear demonstration in human studies , where it is difficult to control experience and quantify outcomes . Here , we take advantage of the tractable nature of vocal learning in songbirds ( Lonchura striata domestica ) to test the idea that matching instruction to individual genetic predispositions can enhance learning . We use both cross-fostering and computerized instruction with synthetic songs to demonstrate that matching the tutor song to individual predispositions can improve learning across genetic backgrounds . Moreover , we find that optimizing instruction in this fashion can equalize learning differences across individuals that might otherwise be construed as genetically determined . Our results demonstrate potent , synergistic interactions between experience and genetics in shaping song , and indicate the likely importance of such interactions for other complex learned behaviors .
Recent studies in human populations have demonstrated strong genetic influences on academic achievement ( Branigan et al . , 2013; Lee et al . , 2018; Okbay et al . , 2016; Rietveld et al . , 2013 ) . This raises the question of whether genes place immutable bounds on achievement or whether experiential factors could amplify or dampen the impact of genetic variation . A particularly intriguing possibility is that customizing instruction based on individual differences in learning styles or genetic predispositions could improve learning outcomes ( Asbury and Plomin , 2013; Moser and Zumbach , 2018; Pashler et al . , 2009; Plomin , 2014 ) . However , despite widespread interest in the potential value of ‘personalized’ instruction , it has been difficult to evaluate this idea in human populations , where it is challenging to control genetic variation , manipulate experience and quantify outcomes . Birdsong affords an attractive system for studying how tailoring instruction based on genetic differences influences learning outcomes . Song , like human speech , is a complex vocal behavior that is learned during early life from adult vocal models ( Doupe and Kuhl , 1999 ) . Young birds listen to a 'tutor song’ ( usually that of their father ) , and through practice develop vocalizations that closely match this target ( e . g . Figure 1A ) . Although much prior work has focused on how experience shapes song learning , genetic predispositions also contribute to learned song structure at both the species and individual levels ( Fehér et al . , 2009; Gardner et al . , 2005; Marler and Peters , 1977; Marler and Peters , 1982; Marler and Peters , 1988; Mets and Brainard , 2018a; Mundinger , 1995; Mundinger and Lahti , 2014; Podos et al . , 1999; Soha and Marler , 2000; Thorpe , 1954 ) . For example , while many species can learn aspects of heterospecific song , birds will preferentially acquire species typical song structure ( spectral content and ordering of syllables ) when tutored with a combination of heterospecifc and conspecific songs ( Marler and Peters , 1977; Marler and Peters , 1982; Marler and Peters , 1988; Mundinger , 1995; Mundinger and Lahti , 2014; Podos et al . , 1999; Soha and Marler , 2000; Thorpe , 1958 ) . Consistent with this , deviations between a tutor song and species typical song can lead to poor copying of the tutor song ( Lahti et al . , 2011; Marler and Peters , 1988; Podos , 1997; Podos et al . , 2004 ) . Moreover , we have previously found that even within a single-species colony of Bengalese finches ( Lonchura striata domestica ) , there is a strong heritable predisposition for individuals to produce songs at differing tempos ( Mets and Brainard , 2018a ) . The presence of such heritable biases for learned song structure , together with the ease of controlling instructive experience and quantifying learning outcomes , renders song learning in Bengalese finches particularly suitable for testing whether tailoring instruction in accordance with individual genetic predispositions can enhance learning .
Within our genetically heterogeneous Bengalese finch colony , we found that there was a broad range in the quality of song learning . Many juveniles that were reared conventionally in their home nests and tutored by their genetic fathers learned to copy tutor song flexibly and with high fidelity ( e . g . Figure 1Ai–ii , tutor vs . learned song ) . However , across individuals , learning could range from nearly perfect ( e . g . Figure 1B , tutor song vs . tutee song , top ) to extremely poor ( e . g . Figure 1B , tutor song vs . tutee song , bottom; compare with isolate song ) . We quantified song learning using the Song Divergence ( SD ) , a measure that estimates how much of the spectral content of syllables in the tutor song is absent from the learned song ( Mets and Brainard , 2018b ) ; hence , an SD of 0 indicates a song that perfectly matches the tutor song , while increasing values of SD ( quantified in bits ) indicate songs that are progressively worse copies of the tutor . The SD is computed in an automated fashion from a large set of syllables randomly selected from each bird’s song , and provides a measure that corresponds well with human assessments of song learning across a broad range of learning quality ( see Materials and methods ) ( Mets and Brainard , 2018b ) . Prior work indicates that variation in the quality of song learning ( Figure 1C ) could be influenced by both experiential and genetic factors ( Chen et al . , 2016; Doupe and Kuhl , 1999; Fehér et al . , 2009; Gardner et al . , 2005; Marler and Peters , 1988; Marler and Peters , 1982; Marler and Peters , 1977; Mets and Brainard , 2018a; Mundinger , 1995; Mundinger and Lahti , 2014; Podos et al . , 1999; Soha and Marler , 2000; Tchernichovski et al . , 1999; Thorpe , 1958 ) . Here , we were interested in the possibility that a component of this variation could be explained by an interaction between these factors - specifically , whether matching instructive experience to genetic predispositions of individuals could improve learning outcomes . To investigate whether alignment between tutoring experience and individual genetic predispositions influences learning outcomes , we compared how well a tutor’s song was learned by his own genetic offspring relative to how well it was learned by birds that were cross-fostered from other nests ( Figure 2A ) . We reasoned that many of the genetic factors that contribute to the structure of a father’s song would also be passed on to his offspring . Hence , we expected that a tutor’s song would be better matched to the genetic predispositions of his own offspring than to those of cross-fostered birds ( Plomin et al . , 1977 ) . We hypothesized that if the correspondence between tutor song instruction and individual genetic biases influences learning , then home-reared birds would learn better than cross-fostered birds . Eight breeding pairs served as parents and provided tutoring to both their own home-reared progeny and to cross-fostered birds ( see Materials and methods ) . To ensure that cross-fostered birds were not exposed to the songs of their genetic fathers , we transferred eggs to the nests of foster parents within 36 hr of laying , prior to the development of the peripheral auditory system ( Murray et al . , 2013; Yamasaki and Tonosaki , 1988 ) . For both cohorts , juveniles were reared with their tutors from hatching until adulthood ( ~120 days of age ) , at which point their songs were recorded for analysis . In order to assess any learning differences in the context of the same tutor song and parental environment , we conducted paired comparisons of the median quality of learning within each nest for home-reared versus fostered juveniles . Across nests , genetic offspring ( home-reared birds ) learned the tutor song significantly better than cross-fostered birds ( Figure 2B , C; p<0 . 005 , Wilcoxon signed-rank test ) . This suggests that better alignment between the acoustics of the tutors’ songs and the individual genetic predispositions of their offspring improved learning outcomes for home-reared birds . However , learning in these experiments also could have been facilitated for home-reared birds by genetic contributions to other aspects of instructive experience , such as the amount or quality of tutoring directed at offspring versus fostered birds . To eliminate the confound of potential variability in individual interactions with the tutor , we next used a computer tutoring paradigm ( Mets and Brainard , 2018a; Tchernichovski et al . , 1999 ) to hold both the acoustic structure of the tutor song and the number of song exposures constant across all individuals ( Figure 3A ) . We limited exposure to auditory stimuli other than the computer tutor song by transferring eggs within 36 hr of laying to nests where hatchlings were reared to independence by non-singing female foster parents . At ~45 days post-hatch , juveniles were transferred to sound chambers where they were computer-tutored with identical songs , yielding a cohort of 20 birds from 13 different breeding nests , all with the same tutor song instruction . Despite exposure to controlled computer tutoring experience , the distribution of learning outcomes was similar to that generated by live tutoring; some birds learned songs for which the spectral content of syllables closely resembled the tutor stimulus ( Figure 3B , top row of 'tutee songs' , and 3C ) while other birds learned songs that had little resemblance to the tutor stimulus ( Figure 3B , bottom row of 'tutee songs' , and 3C ) . Thus , differences in parental behavior and in individual experience with the live tutor could not account for observed variation in the quality of learning . In contrast , we found that a significant amount of variation in the quality of learning could be explained by how well the tutor stimulus matched individual genetic biases of juveniles . Previous work ( Mets and Brainard , 2018a ) demonstrated that juveniles learn songs with tempos that are strongly biased towards the tempos of their fathers’ songs , even when they have never heard their fathers sing . We therefore estimated the genetic bias for tempo of individuals from different nests as the tempo ( in syl/s ) measured for their fathers’ songs . Across the juvenile birds that were computer tutored , the individual biases for tempo ranged from 5 . 5 syl/s to 12 . 5 syl/s . Hence , the computer tutor stimulus , which was presented at 8 . 5 syl/s , was better matched to the genetic biases of some individuals than for others . We measured the quality of learning for computer tutored birds in three groups: birds that had biases for tempo that were slower than ( <7 . 5 syl/s ) , similar to ( 7 . 5–9 . 5 syl/s ) , or faster than ( >9 . 5 syl/s ) , the tempo of the tutor song ( 8 . 5 syl/s ) . Across these groups , the birds that had biases for tempo that were most similar to the tutor song learned best ( Figure 3D ) . Thus , as with live tutoring , the quality of learning for spectral content of a controlled , synthetic song could be explained in part by the alignment between instructive experience and individual genetic bias . To more explicitly test whether matching the tempo of the tutor song to the genetic bias of individual birds could enhance learning , we carried out an additional computer tutoring experiment in which we assessed how varying the tempo of the tutor song influenced the quality of learning for birds from three different genetic backgrounds that were biased to sing at differing song tempos - one near the lower decile of song tempos produced in our colony ( 7 . 18 syl/s father song tempo ) , one near the median ( 7 . 89 syl/s ) , and one near the upper decile ( 10 . 31 syl/s ) . We tutored groups of juveniles from each of these ‘slow’ , ‘medium’ and ‘fast’ genetic backgrounds with songs that had identical spectral content but that were presented at ‘slow’ , ‘medium’ and ‘fast’ tempos ( 6 . 5 syl/s , 8 . 5 syl/s , and 10 . 5 syl/s; see Materials and methods ) ( n = 28 animals; Schematized , Figure 4A ) . Synthetic tutor songs that differed in tempo were constructed by varying the gaps between syllables , without altering either the durations or spectral content of the syllables themselves ( Figure 4—figure supplement 1 ) . Hence , every bird was tutored with stimuli that had the same spectral content and number of syllables . Correspondingly , in order to compare learning across groups tutored with different tempo songs , we used the Song Divergence , which quantifies how well syllable spectral content is learned independently of song temporal structure . The differences in quality of learning that we observed therefore reflect influences of the rate at which syllables are presented on an orthogonal aspect of song structure - how well the spectral content of syllables is learned . Joint consideration of tutor song tempo and individual genetic bias revealed a strong interaction between the two; for each of the three genetic backgrounds , the best learning was achieved if tutor song tempo was ‘matched’ to the genetic bias ( Figure 4A , B; example spectrograms illustrated in Figure 4—figure supplement 1 and individual groups quantified in Figure 4—figure supplement 2A ) . Within the ‘unmatched’ groups , birds learned better if they were tutored with a song slower than their genetic bias ( Figure 4B , p<0 . 02 , two-tailed t-test , Holm-Bonferroni correction for multiple comparisons ) . These three categories ( tutor song faster than father’s song , tutor song matched to father’s song , and tutor song slower than father’s song ) accounted for 39 . 8% of the variation in song learning outcomes ( ANOVA , p<0 . 008 , Holm-Bonferroni correction for multiple comparisons ) . In contrast , genetic background considered alone , and tutor song tempo considered alone , explained only 4 . 1% and 0 . 28% , respectively , of the variation in learning ( Figure 4—figure supplement 2B , C ) . These findings demonstrate that much of the quality of song learning can be explained by the interaction between tutor song experience and genetic bias . We considered whether the interaction between tutor experience and genetic bias could reflect a trade-off between learning song spectral structure and optimizing tempo . In particular , for some species , it has been suggested that individuals strike a balance between producing syllables with broadband spectral content and producing those syllables at a faster tempo ( Lahti et al . , 2011; Podos , 1997; Podos et al . , 2004 ) . This observation raises the possibility that worse SD scores in our study reflect in part a potentially advantageous sacrifice in the quality of spectral copying in order to optimize song tempo . To examine this possibility , we first tested whether SD scores were worse for individuals that sang faster songs than for individuals that sang slower songs , as might be expected if there was a trade-off between the quality of spectral copying and maximizing song tempo - a song feature that has been identified in some studies as more attractive to females and therefore favored ( Ballentine , 2004; Nowicki and Searcy , 2005 ) . We did not find this trend in our data ( Figure 4—figure supplement 3A ) . We then tested whether SD scores were worse for individuals that sang songs more closely matched in tempo to the tutor song , as might be expected if there was a trade-off between the quality of song spectral copying and the ability to match tutor song tempo . In contrast to this possibility , we found that there was significantly better copying of syllable spectral content in birds that also more closely matched song tempo to the tutor song ( Figure 4—figure supplement 3B ) . This is consistent with the idea that the ability of an individual to match the tutor song tempo facilitates the learning of syllable spectral content , and argues against the possibility that birds with poor learning of spectral content are optimizing some other song feature .
Together , the cross-fostering and computer tutoring experiments show that tailoring an instructive stimulus to match the genetic bias of an individual can enhance learning . It is noteworthy that our estimates of individual bias derived exclusively from the tempo of the father’s song; hence , a more accurate estimate of bias , incorporating other genetic factors such as maternal contributions , would potentially enable even more effective tailoring of instruction . Our results additionally indicate that a failure to take into account how experience and genetics interact can lead to erroneous conclusions about the immutability of genetic constraints on individual differences in learning . For example , absent consideration of this interaction , birds that are genetically biased to sing slow songs appear to be inherently worse learners than birds that are biased to sing fast ( e . g . averaged across all stimuli , slow birds learn worse than fast birds; see Figure 4—figure supplement 2 ) . However , when instruction is individually tailored , it is apparent that ‘slow’ birds can learn as well as , or better than , ‘fast’ birds ( Figure 4A and Figure 4—figure supplement 2; slow birds tutored with slow songs learned better than fast birds tutored with fast songs ) . Hence , customization of instruction can not only improve learning outcomes , but in so doing may also attenuate differences across individuals that otherwise might have been construed as genetically determined . These results also inform an understanding of how genetic factors shape the cultural transmission of song within a population of birds . Previous work has indicated that songbirds are subject to genetic constraints at the species level that bias birds to learn better from songs that are more species-typical ( Gardner et al . , 2005; Lahti et al . , 2011; Marler and Peters , 1988; Podos , 1997; Podos , 1996; Podos et al . , 2004 ) , and it has been proposed that such constraints might contribute to the long-term stability of a given species’ song . According to this idea , learning drives variation in songs across individuals , but genetic factors ‘pull’ all birds back toward a single species-specific song ‘template’ , thus reducing drift in the population level song structure over generations ( Fehér et al . , 2009; Lachlan et al . , 2018; Lachlan and Slater , 1999; Lynch , 1996 ) . The observation that individual Bengalese finches are biased to learn better from some songs than from others is broadly consistent with the idea that innate factors might constrain ‘drift’ away from species-typical song models . However , our finding that within this genetically heterogeneous population , each bird or nest may have a different genetic bias , indicates that there is no single song model toward which all individuals are drawn ( indeed , the median song tempo in our colony was 8 . 5 syl/s , yet this ‘species typical’ song is a poor model for birds from faster and slower nests ) . Rather , our findings suggest an alternative possibility - that whatever genetic biases toward different song structure are present between nests or families could be preserved or amplified across generations , potentially contributing to gradual divergence of song structure within distinct subpopulations of birds . These results also demonstrate that learning of complex skills like song can be shaped by both direct and indirect influences of parental genes . For home-reared birds , genetic factors bias the acoustic structure of the father’s song , including tempo , but also contribute to a similar bias in his offspring ( Mets and Brainard , 2018a ) ; this establishes an alignment of experience and genetics that we have shown can enhance learning . These findings support the idea that , for human families , a similar synergistic interaction between the home environments that are shaped by parental behavior , and the heritable predispositions of their children , contributes to the observed clustering within families of achievement in verbal , quantitative , musical and athletic domains ( Meredith , 1973; Plomin et al . , 1977; Tan et al . , 2014; Vinkhuyzen et al . , 2009 ) . Such interactions may be especially potent for behaviors like speech and other motor skills , where ‘tutoring’ in the early home environment plays a critical role in shaping the perceptual and motor systems underlying performance ( Kuhl , 2010 ) . More broadly , our findings highlight the critical role that gene-experience interactions can play in determining complex phenotypes . When considered alone , neither genes nor experience had a clear impact on song learning , but their interaction explained nearly 40% percent of variation in learning outcomes . Detection of this strong interaction required detailed knowledge of individual genetic predispositions and experience , information that is often hard to obtain in the context of human studies ( Halldorsdottir and Binder , 2017 ) . Nevertheless , a better understanding of such interactions between genetics and experience will likely be required to fully elucidate the mechanisms driving individual-to-individual variation in complex learned behaviors .
Subjects were male Bengalese finches ( Lonchura striata domestica ) . 239 birds were reared and tutored by live birds and 47 additional birds were reared by foster females and tutored by computer . These animals were bred in our colony . Other than efforts to maintain some separation between lineages , mating pairs comprised randomly selected male and female birds . All protocols were reviewed and approved by the Institutional Animal Care and Use Committee at the University of California , San Francisco . Data on tempo from a subset of the cross-foster and computer tutored birds was presented in a previous study ( Mets and Brainard , 2018a ) . For audio recording , birds were single-housed in sound isolation chambers ( Acoustic Systems ) . Songs were digitally recorded at a sampling frequency of 32 kHz , and a bit depth of 16 . Recording microphones were placed in a fixed position at the top of the cage housing the bird . Prior to further analysis , all songs were high-pass filtered at ~500 Hz using a digitally implemented elliptical infinite impulse response filter with a passband edge frequency of 0 . 04 radians . All recordings used for analysis were acquired during early adulthood ( 90–120 days post hatch ) . To create populations of birds that were tutored by a live bird but never heard the song of their genetic fathers , eggs were taken from parents within 36 hr of laying and transferred to foster nests where hatchlings were reared to adulthood . For each individual , the specific foster nest was randomly selected from a set of 8 foster nests within our breeding colony . All birds that were studied in these experiments were offspring of breeding pairs from our large and genetically heterogeneous colony . The breeding pairs themselves were established from birds that were acquired over a multi-year period from outside vendors , or were bred in house ( see for example Figure 2—figure supplement 1 ) . Given the structure of our colony , the cross-fostered birds were less related to their tutors than were the home-reared birds , but in many cases , the cross-fostered birds were more or less distant ‘cousins’ of their tutors . Our expectation is that effects we reported for these experiments ( consistently better learning for genetic-offspring than for less-closely related cross-fostered birds ) would have been even more pronounced had the cross-fostered birds been drawn from entirely different colonies or other outside sources . The quality of song learning was measured using Song Kullback-Leibler Divergence or Song Divergence ( Mets and Brainard , 2018b ) . This is a largely automated measure that estimates the degree to which the spectral content of a tutor song is copied by a tutee , and has been shown to have a good correspondence with human assessments of song learning ( Mets and Brainard , 2018b ) . Song Divergence is computed by creating a statistical model that captures the song spectral content of both tutor and tutee and estimating the divergence between these two models . This estimate of divergence is directional; the Song Divergence calculated in the tutor-tutee direction indicates the amount of song spectral content ( in bits ) present in the tutor song that is not present in the tutee song ( "missing content" ) but does not capture spectral content from the tutee song that is not present in the tutor song ( "improvisation" ) . The model for each bird in our study was generated from a corpus of 60 song bouts as follows . Songs were segmented into discrete units of sound ( syllables ) separated by silence using an automatically determined amplitude threshold . For each syllable , the spectral content was extracted by calculating a power spectral density ( PSD ) . Here , we used a single PSD per syllable , removing information about syllable temporal structure . Compared to other semi-automated methods for evaluating song learning , this allows assessment of the copying of song spectral content independently of song temporal structure ( Burkett et al . , 2015; Mandelblat-Cerf and Fee , 2014; Mets and Brainard , 2018b; Tchernichovski et al . , 2000 ) . These PSDs were transformed into a syllable-syllable similarity space and a Gaussian mixture model ( GMM ) was then fit to the distributions of these syllables . The Kullback-Leibler Divergence between two GMMs corresponding to songs from a tutor and a tutee is the Song Divergence . In the case of a synthetic tutor stimulus where there is only a single variation of the song , Song Divergence was calculated by comparing the song corpus of a tutee to the song corpus of a bird that had learned the tutor song especially well . The song corpus of the same ‘tutor’ bird was used for all such calculations . Song tempo was quantified as the average number of syllables produced per second of song , a measure we have previously used to identify song tempo as heritable ( Mets and Brainard , 2018a ) . Discrete units of sound separated by silence ( syllables ) were identified based on amplitude . First , an ‘amplitude envelope’ was created by rectifying the song waveform and then smoothing the waveform through convolution with an 8 ms square wave . Threshold crossings of this amplitude trace were then used to identify periods of vocalization . Thresholds were set heuristically to result in segmentation that corresponded with syllable onsets and offsets apparent in human examination of spectrograms . Once the threshold was established , ‘objects’ were identified as uninterrupted regions longer than 10 ms over which the amplitude envelope exceeded threshold . Any objects separated by a gap of 5 ms or less were merged producing a final set of objects that were defined as syllables . A series of syllables that had no gaps larger than 250 ms was considered a song bout . For each bird , tempo was then quantified as the number of syllables present in a song bout divided by the duration of the song bout , averaged across at least 60 bouts of singing . To create populations of birds that had controlled tutoring experience , eggs were taken from parents within 36 hr of laying , prior to neural development ( Murray et al . , 2013; Yamasaki and Tonosaki , 1988 ) , and were then raised by pairs of non-singing foster mothers housed in sound isolation chambers . While the female foster mothers do not sing , they produce unlearned calls and influence other aspects of a hatchling’s experience , including feeding and social interactions , that could plausibly influence hatchling development and the quality of subsequent computer driven song learning . In order to ensure that any such influences of female fostering on song learning did not contribute systematically to our results , we used 14 different female foster nests ( each with two foster mothers ) and randomly assigned eggs from experimental nests to these foster nests . In order to prevent any ‘batch effects’ , only a single hatchling was raised at a time in any given foster nest . Foster mothers raised the juveniles until they were able to feed themselves . At independence ( usually 35–40 days post hatch ) birds were moved to an acoustic isolation chamber with an audio recording system and a computer tutoring apparatus , based on an approach that previously has been demonstrated to drive song learning ( Mets and Brainard , 2018a; Tchernichovski et al . , 1999 ) . At 45 days post-hatch , the tutoring apparatus was activated , allowing birds to access a tutor song . The apparatus consisted of a perch activated switch that caused playback of a tutor stimulus ( see below ) . Each perch hop elicited a single playback of the tutor stimulus . Birds were allowed to playback 10 songs , three times a day ( morning , noon , and evening ) . Playback of tutor song was limited to 30 songs per day based on previous work indicating that this was near an optimal value to maximize the quality of song learning in this paradigm ( Tchernichovski et al . , 1999 ) . It sometimes took birds a few days to begin consistently actuating song playbacks . Nevertheless , all birds elicited at least 90% of the available song playbacks during the tutoring period ( median = 96% ) , and there was no relationship between the number of playbacks that an individual heard and the quality of song learning . The computer tutoring apparatus was implemented with custom LabView software ( National Instruments , "EvTutor" ) that is provided in supporting materials for this paper . Birds remained in the tutoring apparatus until 120 days post-hatch . For experiments involving different tutor song tempos , the tempo for an individual was randomly selected from three possible tempos ( see below ) . The synthetic stimulus used during playback tutoring was the same as that used in our previous work ( Mets and Brainard , 2018a; Mets and Brainard , 2018b ) . To create a naturalistic but controlled learning stimulus , a synthetic song used for computer tutoring was derived from songs sampled from our Bengalese finch colony . The synthetic song was composed of 9 categorically distinct syllables that were chosen to reflect a range of different syllable types found in Bengalese finch song ( i . e . short ‘introductory’ syllables , noisy syllables , syllables with harmonic structure and constant or modulated frequency , etc . ) . The tutor stimulus consisted of a series of introductory syllables followed by three repetitions of a stereotyped sequence of syllables , or ‘motif’ ( shown in Figure 3B ) . The gaps between syllables were chosen to reflect naturalistic means and standard deviations based on the distribution of gap durations found in normal Bengalese finch song . Correspondingly , the tutor song stimulus had a relatively natural prosody compared to a stimulus in which the time between syllable onsets is fixed . The 8 . 5 syl/s tutor song stimulus arose naturally out of this process , as 8 . 5 syl/s is close to the median song tempo present in our colony . The 6 . 5 and 10 . 5 syl/s tutor stimuli were created by proportionally increasing or decreasing only the inter-syllable gap durations , resulting in songs with identical spectral content presented at different tempos . All statistical testing in this study was carried out in consultation with the biostatistics consultancy of the UCSF Clinical and Translational Science Institute . No birds were removed from the study . As no subjective measurements were made , no blinding was performed . When used , statistical tests were appropriate to the data presented and the data , to the limit of detection , were consistent with the assumptions of the tests . For all ANOVA analyses , estimates of variance explained are Omega squared . For each experimental group , p-values for statistical tests were corrected for multiple testing within that group using the Holm-Bonferroni procedure , an extension of the Bonferroni correction which retains the same family-wise error rate while reducing the false negative rate relative to traditional Bonferroni ( Holm , 1979 ) . Unless otherwise indicated , all tests were two-tailed . For all cases , where a one-tailed test was conducted because of a directional hypothesis , the threshold p-value for the test was considered to be significant at the 0 . 025 level to reduce the false positive rate to be equivalent to a two-tailed test . For all tests , the statistical significance was not impacted by the use of a one-tailed versus a two-tailed test . For the cross-fostering experiments presented in Figure 2 , the median within-nest SD scores for home-reared birds were compared to the median within nest SD scores for cross-fostered juveniles . Median values were used as a summary statistic for this comparison because the SD scores were approximately gamma distributed ( Figure 1C ) . The effect of home-rearing on song learning was tested using a Wilcoxon signed-rank test , a paired test which tests the sign of the change across experimental conditions and not the magnitude of the change . This enables a test for the impact of home-rearing relative to cross-fostering while controlling for nest-specific or tutor-song-specific variables . | Some people do better at school than others , and some of this difference comes down to genes . But do genes place fixed limits on an individual's academic potential ? Or is it possible to increase or decrease the impact of genes by changing how a person is taught ? One possibility is that individuals learn best in different ways , and that tailoring instruction to suit individual learning styles could improve learning outcomes . But despite widespread interest in this idea , testing it systematically has proven challenging . Mets and Brainard have therefore taken a new approach by testing the idea in a songbird called the Bengalese finch . Birdsong is a complex behavior learned in a similar way to human speech . Young birds listen to a tutor song – usually that of their father – and learn to mimic it through trial and error . But some songbirds learn better than others . By swapping eggs between nests , Mets and Brainard show that genetic offspring often learn the father’s song more accurately than birds fostered in from another nest . This might be because the father and offspring share genetic characteristics that contribute to the sound of the father’s song . Birds with the same genes will thus find it easier to learn the same song . Alternatively , it could be that father birds spend more time teaching their genetic offspring than young they have fostered . To control for this possibility , Mets and Brainard played computer-generated songs to juvenile birds from different nests that had all been raised by non-singing females . Some of the songs had a fast tempo , others were slow , and a third set were in between . The results showed that juveniles learned most successfully when the training song had a similar tempo to their father’s song . This was true even though none of the birds had ever heard their father sing . The findings thus suggest that tailoring instruction to suit an individual's natural learning tendencies – which depend on their genes – can enhance learning . Without knowing about this effect , it would be easy to assume that some of the songbirds in the current study were simply poorer learners than others . But in fact , optimizing instruction for each individual’s genetic background reduced the differences between individuals . If learning in humans is similar to vocal learning in birds , there could be broad implications for education . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience",
"genetics",
"and",
"genomics"
] | 2019 | Learning is enhanced by tailoring instruction to individual genetic differences |
The budding yeast genome contains regions where meiotic recombination initiates more frequently than in others . This pattern parallels enrichment for the meiotic chromosome axis proteins Hop1 and Red1 . These proteins are important for Spo11-catalyzed double strand break formation; their contribution to crossover recombination remains undefined . Using the sequence-specific VMA1-derived endonuclease ( VDE ) to initiate recombination in meiosis , we show that chromosome structure influences the choice of proteins that resolve recombination intermediates to form crossovers . At a Hop1-enriched locus , most VDE-initiated crossovers , like most Spo11-initiated crossovers , required the meiosis-specific MutLγ resolvase . In contrast , at a locus with lower Hop1 occupancy , most VDE-initiated crossovers were MutLγ-independent . In pch2 mutants , the two loci displayed similar Hop1 occupancy levels , and VDE-induced crossovers were similarly MutLγ-dependent . We suggest that meiotic and mitotic recombination pathways coexist within meiotic cells , and that features of meiotic chromosome structure determine whether one or the other predominates in different regions .
The transition from the mitotic cell cycle to meiosis involves substantial changes in mechanisms of DNA double strand break ( DSB ) repair by homologous recombination ( HR ) . Most mitotic HR repairs spontaneous lesions , and most repair products are non-crossovers ( NCOs ) that do not involve exchange of flanking parental sequences ( Kadyk and Hartwell , 1992; Ira et al . , 2003; Pâques et al . , 1998 ) . In contrast , meiotic recombination is initiated by programmed DSBs ( Cao et al . , 1990; Sun et al . , 1989 ) that often are repaired as crossovers ( COs ) between homologous chromosomes ( homologs ) , with exchange of flanking parental sequences . Inter-homolog COs , combined with sister chromatid cohesion , create physical linkages that ensure faithful homolog segregation during the first meiotic division , avoiding chromosome nondisjunction and consequent aneuploidy in gametes ( reviewed by Hunter , 2015 ) . The DSBs that initiate meiotic recombination are formed by Spo11 in complex with a number of accessory proteins , and will be referred to here as Spo11-DSBs ( reviewed by Lam and Keeney , 2015 ) . Spo11-DSBs and resulting recombination events are non-uniformly distributed in the genomes of organisms ranging from budding yeast to humans ( Baudat and Nicolas , 1997; Blitzblau et al . , 2007; Buhler et al . , 2007; Fowler et al . , 2013; Gerton et al . , 2000; Hellsten et al . , 2013; Pratto et al . , 2014; Singhal et al . , 2015; Smagulova et al . , 2011; Wijnker et al . , 2013 ) . In budding yeast , this non-uniform distribution of Spo11-DSBs is influenced by meiosis-specific proteins , Red1 and Hop1 , which are components of the meiotic chromosome axis . The meiotic chromosome axis coordinates sister chromatids and forms the axial element of the synaptonemal complex , which holds homologs in tight juxtaposition ( Hollingsworth et al . , 1990; Page and Hawley , 2004; Smith and Roeder , 1997 ) . Spo11-DSBs form frequently in large ( ca 50–200 kb ) 'hot' domains that are also enriched for Red1 and Hop1 , and these 'hot' domains are interspersed with similarly-sized 'cold' regions where Spo11-DSBs are infrequent and Red1/Hop1 occupancy levels are low ( Baudat and Nicolas , 1997; Blat et al . , 2002; Blitzblau et al . , 2007; Buhler et al . , 2007; Panizza et al . , 2011 ) . Normal Spo11-DSB formation requires recruitment of Spo11 and accessory proteins to the meiotic axis ( Panizza et al . , 2011; Prieler et al . , 2005 ) , and Red1/Hop1 are also central to mechanisms that direct Spo11-DSB repair towards use of the homolog as a recombination partner ( Carballo et al . , 2008; Niu et al . , 2005; Schwacha and Kleckner , 1997 ) . Other eukaryotes contain Hop1 analogs that share a domain , called the HORMA domain ( Rosenberg and Corbett , 2015 ) , and correlations between these meiotic axis proteins and DSB formation are observed in fission yeast , nematodes and in mammals ( Fowler et al . , 2013; Goodyer et al . , 2008; Wojtasz et al . , 2009 ) . Thus , most meiotic interhomolog recombination occurs in the context of a specialized chromosome structure and requires components of that structure . Meiotic recombination pathways diverge after DSB formation and homolog-directed strand invasion . In budding yeast , about half of meiotic events form NCOs via synthesis-dependent strand annealing , a mechanism that does not involve stable recombination intermediates ( Allers and Lichten , 2001a; McMahill et al . , 2007 ) and is suggested to be the predominant HR pathway in mitotic cells ( Bzymek et al . , 2010; McGill et al . , 1989 ) . Most of the remaining events are repaired by a meiosis-specific CO pathway , in which an ensemble of meiotic proteins , called the ZMM proteins , stabilize early recombination intermediates and promote their maturation into double Holliday junction joint molecules ( Allers and Lichten , 2001a; Börner et al . , 2004; Lynn et al . , 2007; Schwacha and Kleckner , 1994 ) . These ZMM-stabilized joint molecules ( JMs ) are subsequently resolved as COs ( Sourirajan and Lichten , 2008 ) through the action of the MutLγ complex , which contains the Mlh1 , Mlh3 , and Exo1 proteins ( Argueso et al . , 2004; Khazanehdari and Borts , 2000; Wang et al . , 1999; Zakharyevich et al . , 2010 , 2012 ) . MutLγ does not appear to make significant contributions to mitotic COs ( Ira et al . , 2003 ) . A minority of events form ZMM-independent JMs that are resolved as both COs and NCOs by the structure-selective nucleases ( SSNs ) Mus81-Mms4 , Yen1 , and Slx1-Slx4 , which are responsible for most JM resolution during mitosis ( Argueso et al . , 2004; Santos et al . , 2003; De Muyt et al . , 2012; Ho et al . , 2010; Muñoz-Galván et al . , 2012; Zakharyevich et al . , 2012; reviewed by Wyatt and West , 2014 ) . A similar picture , with MutLγ forming most meiotic COs and SSNs playing a minor role , is observed in several other eukaryotes ( Berchowitz et al . , 2007; Holloway et al . , 2008; Plug et al . , 1998 ) . To better understand the factors that promote the unique biochemistry of CO formation during meiosis , in particular MutLγ-dependent JM resolution , we considered two different hypotheses . In the first , expression of meiosis-specific proteins and the presence of high levels of Spo11-DSBs results in nucleus-wide changes in recombination biochemistry , shifting its balance towards MutLγ-dependent resolution of JMs , wherever they might occur . In the second , local features of meiotic chromosome structure , in particular enrichment for meiosis-specific chromosome axis proteins , provides an in cis structural environment that favors MutLγ-dependent JM resolution . However , because Spo11-DSBs form preferentially in Red1/Hop1-enriched regions , and because these proteins are required for efficient Spo11-DSB formation and interhomolog repair , it is difficult to distinguish these two models by examining Spo11-initiated recombination alone . To test these two hypotheses , we developed a system in which meiotic recombination is initiated by the sequence- and meiosis-specific VMA1 derived endonuclease , VDE ( Gimble and Thorner , 1992; Nagai et al . , 2003 ) . VDE initiates meiotic recombination at similar levels wherever its recognition sequence ( VRS ) is inserted ( Fukuda et al . , 2008; Neale et al . , 2002; Nogami et al . , 2002 ) . VDE- catalyzed DSBs ( hereafter called VDE-DSBs ) form independent of Spo11 and meiotic axis proteins . However , like Spo11-DSBs , VDE-DSBs form after pre-meiotic DNA replication and are repaired using end-processing and strand invasion activities that also repair Spo11-DSBs ( Fukuda et al . , 2003; Neale et al . , 2002 ) . We examined resolvase contributions to VDE-initiated CO formation , and obtained evidence that local enrichment for meiotic axis proteins promotes MutLγ-dependent CO formation; while recombination that occurs outside of this specialized environment forms COs by MutLγ-independent mechanisms . We also show that CO formation at a locus , and in particular MutLγ-dependent CO formation , requires Spo11-DSB formation elsewhere in the genome .
The recombination reporter used for this study contains a VDE recognition sequence ( VRS ) inserted into a copy of the ARG4 gene on one chromosome , and an uncleavable mutant recognition sequence ( VRS103 ) on the homolog ( Figure 1 ) . Restriction site polymorphisms at flanking HindIII sites , combined with the heterozygous VRS site , allow differentiation of parental and recombinant DNA molecules . This recombination reporter was inserted at two loci: HIS4 and URA3 , which are 'hot' and 'cold' , respectively , for Spo11-initiated recombination and Red1/Hop1 occupancy ( Borde et al . , 1999; Buhler et al . , 2007; Panizza et al . , 2011; Wu and Lichten , 1995; also see Figure 4A and Figure 4—figure supplement 1 , below ) . Consistent with previous reports , Spo11- DSBs and the resulting crossovers , are about five times more frequent in inserts at HIS4 than at URA3 ( Figure 1—figure supplement 1A ) . When VDE is expressed , ~90% of VRS sites at both loci were cleaved by 7 hr after initiation of sporulation ( Figure 2A ) , consistent with previous reports that VDE cuts very effectively ( Johnson et al . , 2007; Neale et al . , 2002; Terentyev et al . , 2010 ) . Thus , in most cells , both sister chromatids are cut by VDE ( Gimble and Thorner , 1992; Neale et al . , 2002 ) . In contrast , Spo11-DSBs infrequently occur at the same place on both sister chromatids ( Zhang et al . , 2011 ) . While the consequences of this difference remain to be determined , we note that inserts at both HIS4 and URA3 are cleaved by VDE with equal frequency ( Figure 2A ) . Thus , any effects due simultaneous sister chromatid-cutting should be equal at the two loci . 10 . 7554/eLife . 19669 . 003Figure 1 . Inserts used to monitor VDE-initiated meiotic recombination . The HIS4 and URA3 loci are denoted throughout this paper in red and blue , respectively , and are in Red1/Hop1 enriched and depleted regions , respectively ( see Figure 4A and Figure 4—figure supplement 1 , below ) . ( A ) Left—map of VDE-reporter inserts at HIS4 , showing digests used to detect recombination intermediates and products . One parent ( P1 ) contains ARG4 sequences with a VDE-recognition site ( arg4-VRS ) , flanked by an nourseothricin-resistance module [natMX , ( Goldstein and McCusker , 1999 ) ] and the Kluyveromyces lactis TRP1 gene [KlTRP1 , ( Stark and Milner , 1989 ) ]; the other parent ( P2 ) contains ARG4 sequences with a mutant , uncuttable VRS site [arg4-VRS103 , ( Nogami et al . , 2002 ) flanked by URA3 and pBR322 sequences . Digestion with HindIII ( H ) and VDE ( V ) allows detection of crossovers ( CO1 and CO2 ) and noncrossovers ( NCO ) ; digestion with HindIII alone allows detection of crossovers and DSBs . P2 , CO1 and CO2 fragments are drawn only once , as they are the same size in HindIII digests as in HindIII + VDE digests . Right—representative Southern blots . HindIII-alone digests are probed with a fragment ( probe 2 ) that hybridizes to the insert loci and to the native ARG4 locus on chromosome VIII; this latter signal serves as a loading control ( LC ) . Times after induction of meiosis that each sample was taken are indicated below each lane . ( B ) map of VDE-reporter inserts at URA3 and representative Southern blots; details as in ( A ) . Strain , insert and probe details are given in Materials and methods and Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 00310 . 7554/eLife . 19669 . 004Figure 1—figure supplement 1 . Spo11-initiated events at the two insert loci . ( A ) Spo11-catalyzed DSBs are more frequent at HIS4 that at URA3 . Left—Southern blots of EcoRI digests of DNA from vde∆ strains , probed with pBR322 sequences , showing Spo11-DSBs in the Parent 2 insert ( see Figure 1 ) in resection/repair-deficient sae2∆ mutant strains . Right—location of DSBs and probe and DSB frequencies ( average of 7 and 8 hr samples from a single experiment; error bars represent range ) . Spo11-DSBs in the Parent 1 inserts at HIS4 and URA3 were at different locations within the insert , but displayed similar ratios between the two loci ( data not shown ) . ( B ) Southern blots of HindIII digests of DNA from vde∆ strains , to detect total Spo11-initiated crossovers . ( C ) Southern blots of HindIII-VDE double digests of the same samples , to determine the background contribution of Spo11-initiated COs in subsequent experiments measuring VDE-initiated COs , which will be VDE-resistant due to conversion of the VRS site to VRS103 . Probes were as shown in Figure 1 . ( D ) Quantification of data in panels B ( total COs; filled circles ) and C ( VDE-resistant COs; open circles ) . Data are from a single experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 00410 . 7554/eLife . 19669 . 005Figure 2 . VDE-initiated recombination occurs at similar levels at the two insert loci . ( A ) Cumulative DSB levels are similar at the two insert loci . The fraction of uncut VRS-containing chromosomes ( Parent 1 ) was determined by subtracting the amount of the NCO band in HindIII + VDE digests from the amount of the Parent 1 + NCO band in HindIII digests . ( B ) Non-cumulative VDE-DSB frequencies , measured as fraction of total lane signal , excluding loading controls , in HindIII digests . ( C ) Crossover ( average of CO1 and CO2 ) and noncrossover frequencies , measured in HindIII-VDE digests . Solid lines—recombinants from cells expressing VDE; dashed lines—Spo11-initiated crossovers from vde- strains , measured in HindIII-VDE digests and thus corresponding to VDE-resistant products ( see also Figure 1—figure supplement 1C ) . Values are the average of two independent experiments; error bars represent range . Representative Southern blots are shown in Figure 1 and Figure 1—figure supplement 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 00510 . 7554/eLife . 19669 . 006Figure 2—figure supplement 1 . 70–80% of VDE-DSBs are repaired . ( A ) Fraction of inserts remaining , calculated using HindIII digests ( see Figure 1 ) . For the arg4-VRS103 insert , the ratio ( Parent 2 + CO2 ) / ( 0 . 5 x LC ) was calculated at 9 hr , and was then normalized to the 0 hr value . For the arg4-VRS insert , a similar calculation was made: ( Parent 1 + NCO + CO1 ) / ( 0 . 5 x LC ) ( B ) Relative recovery of interhomolog recombination products , calculated using HindIII-VDE double digests ( see Figure 1 ) . The sum of CO ( average of CO1 and CO2 ) and NCO frequencies was divided by the frequency of total DSBs , as calculated in Figure 2A . Data are the average of two independent experiments; error bars represent range . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 006 DSBs appeared and disappeared with similar timing at the two loci ( Figure 2B ) , with measures of insert recovery ( Figure 2—figure supplement 1A ) and levels of interhomolog recombinants relative to cumulative VDE-DSB levels ( Figure 2—figure supplement 1B ) indicating that ~70% of VDE DSBs are repaired by interhomolog recombination . The remaining VRS-containing inserts appear to be lost , consistent with high levels of VDE activity preventing recovery of inter-sister recombinants . Thus , the two VDE recombination reporter inserts undergo comparably high levels of meiotic recombination initiation , regardless of the local intrinsic level of Spo11-initiated recombination . When VDE-DSBs are repaired by interhomolog recombination , VRS sequences are converted to VRS103 , and become resistant to digestion by VDE . We therefore used HindIII/VDE double digest to score recombinants that are resistant to VDE cleavage ( Figure 1 ) . Comparing the levels of such recombinants in VDE-expressing and vde∆ strains indicates that Spo11-initiated events comprise only a few percent of the recombinants scored in VDE-expressing strains ( Figure 2C , Figure 1—figure supplement 1 , data not shown ) . VDE-initiated recombinants formed at high frequencies at both HIS4 and URA3 , and NCOs exceeded COs by approximately twofold at HIS4 and threefold at URA3 ( Figure 2C ) . These values are within the range observed in genetic studies of Spo11-induced gene conversion in budding yeast ( Fogel et al . , 1979 ) , but differ from the average of near-parity between NCOs and COs observed in molecular assays ( Lao et al . , 2013; Martini et al . , 2006 ) . This is consistent with earlier findings , that cutting both sister chromatids at a DSB site is associated with a reduced proportion of COs among repair products ( Malkova et al . , 2000 ) . While VDE-initiated recombination occurred at similar levels in inserts located at HIS4 and at URA3 , we observed a marked difference between the two loci , in terms of the resolvase-dependence of CO formation ( Figure 3 ) . At the HIS4 locus , COs were reduced in mlh3∆ mutants , which lack MutLγ , by ~60% relative to wild type . In mms4-md yen1∆ slx1∆ mutants , which lack the three structure selective nucleases active during both meiosis and the mitotic cell cycle ( SSNs , triple mutants hereafter called ssn mutants ) , COs were reduced by ~30% , and by ~75% in mlh3 ssn mutants . Thus , like Spo11-initiated COs , VDE-initiated COs in inserts at HIS4 are primarily MutLγ-dependent , and less dependent on SSNs . In contrast , COs in inserts located at URA3 were reduced by only ~ 10% in mlh3 , by ~40% in ssn mutants , and by ~60% in mlh3 ssn mutants , so that the final level of residual COs was the same as at HIS4 . Thus , SSNs make a substantially greater contribution to VDE-initiated CO formation at URA3 than does MutLγ , and MutLγ’s contribution becomes substantial only in the absence of SSNs . 10 . 7554/eLife . 19669 . 007Figure 3 . Different resolvase-dependence of crossover formation at the two insert loci . ( A ) Crossover frequencies ( average of CO1 and CO2 ) measured as in Figure 2C from HIS4 insert-containing mutants lacking MutLγ ( mlh3 ) , structure-selective nucleases ( mms4-md yen1 slx1 ) or both resolvase activities ( mlh3 mms4-md yen1 slx1 ) . ( B ) Crossover frequencies in URA3 insert-containing strains , measured as in panel A . Values are the average of two independent experiments; error bars represent range . ( C ) Final crossover levels ( average of 8 and 9 hr values for two independent experiments ) , expressed as percent of wild type . Note that , in mlh3 mutants , crossovers in HIS4 inserts are reduced by nearly 60% , while crossovers in URA3 inserts are reduced by less than 10% . ( D ) Final noncrossover levels , calculated as in C , expressed as percent of wild type . Representative Southern blots are in Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 00710 . 7554/eLife . 19669 . 008Figure 3—figure supplement 1 . VDE-DSB and NCO frequencies in resolvase mutants . ( A ) VDE-DSB frequencies ( top ) , measured as in Figure 2B , and NCO frequencies ( bottom ) , measured as in Figure 2C , from HIS4 insert-containing strains . ( B ) As panel A , with strains containing inserts at URA3 . Data are the average of two independent experiments; error bars represent range . Representative Southern blots are in Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 00810 . 7554/eLife . 19669 . 009Figure 3—figure supplement 2 . Southern blots of HindIII and HindIII-VDE digests of DNA from HIS4 insert-containing strains ( top ) and from URA3 insert-contaning strains ( bottom ) . Probes and gel labels are as in Figure 1; JM—joint molecule recombination intermediates . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 009 At both insert loci , ssn and mlh3 ssn mutants accumulated DNA species with reduced electrophoretic mobility ( Figure 3—figure supplement 2 ) . These slower-migrating species contain branched DNA molecules , as would be expected for unresolved joint molecules ( D . M . , unpublished observations ) . Steady state VDE-DSB and final NCO levels were similar in all strains ( Figure 3D , Figure 3—figure supplement 1 ) , indicating that resolvases do not act during the initial steps of DSB repair , and consistent with most meiotic NCOs forming by mechanisms that do not involve Holliday junction resolution ( Allers and Lichten , 2001a; De Muyt et al . , 2012; Sourirajan and Lichten , 2008; Zakharyevich et al . , 2012 ) . The marked MutLγ-dependence and –independence of VDE-initiated COs in inserts at HIS4 and at URA3 , respectively , are paralleled by the levels of occupancy of the meiotic axis proteins Hop1 and Red1 ( Panizza et al . , 2011; Figure 4A , Figure 4—figure supplement 1A ) . To test further the suggestion that differential Hop1 occupancy is correlated with differences in CO formation at these loci , we examined the resolvase-dependence of VDE-initiated COs in pch2∆ mutants . Pch2 is a conserved AAA+ ATPase that maintains the nonuniform pattern of Hop1 occupancy along meiotic chromosomes ( Börner et al . , 2008; Joshi et al . , 2009 ) . The different Hop1 occupancies seen in wild type were preserved early in meiosis in pch2∆ mutants ( Figure 4A , Figure 4—figure supplement 1A ) , consistent with previous findings that , in pch2 cells , Spo11-DSB patterns are not altered in most regions of the genome ( Vader et al . , 2011 ) . By contrast , at later times ( 4–5 hr after initiation of meiosis ) , pch2∆ mutants displayed reduced Hop1 occupancy at HIS4 , more closely approaching the lower occupancy levels seen throughout meiosis at URA3 ( Figure 4A; Figure 4—figure supplement 1A ) . 10 . 7554/eLife . 19669 . 010Figure 4 . pch2∆ mutants display altered Hop1 occupancy and crossover MutLγ-dependence . ( A ) Hop1 occupancy at insert loci , determined by chromatin immunoprecipitation and quantitative PCR . Top—cartoon of insert loci , showing the location of primer pairs used . Bottom—relative Hop1 occupancy , expressed as the average ratio of immunoprecipitate/input extract for both primer pairs ( see Materials and methods for details ) . Values are the average of two independent experiments; error bars represent range . ( B ) VDE-initiated CO frequencies measured as in Figure 2C at HIS4 ( top ) and URA3 ( bottom ) in pch2∆ ( solid diamonds ) , pch2∆ mlh3∆ ( open diamonds ) , and pch2∆ mms4-md yen1 slx1 ( half-filled diamonds ) mutants . Crossovers from wild type ( solid line ) , mlh3∆ ( dotted line ) and mms4-md yen1 slx1mutants ( dashed line ) from Figure 3 are shown for comparison . Values are from two independent experiments; error bars represent range . Representative Southern blots are in Figure 4—figure supplement 2 . ( C ) Extent of CO reduction in mlh3∆ mutants , relative to corresponding MLH3 strains . ( D ) Extent of CO reduction in mms4-md yen1 slx1 ( ssn ) mutants , relative to corresponding MMS4 YEN1 SLX1 strains . For both ( C ) and ( D ) , PCH2 genotype is indicated at the top; values are calculated as in Figure 3C . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 01010 . 7554/eLife . 19669 . 011Figure 4—figure supplement 1 . Hop1 occupancy at non-insert loci , DSBs and NCOs in pch2∆ mutants . ( A ) Hop1 occupancy at corresponding loci lacking inserts , determined as in Figure 4A . Occupancy at HIS4 is from strains with inserts at URA3 , and vice versa . ( B ) DSBs and NCOs in inserts at HIS4 , determined as in Figure 2B and C , respectively . Symbols are as in Figure 4B . ( C ) DSBs and NCOs in inserts at URA3 , details as in panel B . Values are from two independent experiments; error bars represent range . Representative Southern blots are in Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 01110 . 7554/eLife . 19669 . 012Figure 4—figure supplement 2 . Southern blots of HindIII and HindIII-VDE digests of DNA from HIS4 insert-containing strains ( top ) and from URA3 insert-contaning strains ( bottom ) . Gel labels are as in Figure 1; JM—joint molecule recombination intermediates . In the gel with HinDIII digests of samples from a pch2∆ mm4-mn yen1∆ slx1∆ strain with inserts at URA3 , the 9 hr sample was originally loaded between the 4 and 5 hr samples; this image was cut and spliced as indicated by vertical lines for presentation purposes . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 012 The altered Hop1 occupancy in pch2∆ was accompanied by altered resolvase contributions to VDE-initiated COs ( Figure 4B , C , D ) . MutLγ contributions decreased at HIS4 and increased at URA3 , and the majority of COs were MutLγ-independent at both insert loci . In contrast , SSN contributions increased slightly at HIS4 , and remained unchanged at URA3 . Thus , in pch2∆ mutants , the similarity of Hop1 occupancy at later times in meiosis is paralleled by a shift towards more similar contributions of MutLγ to VDE-initiated COs at HIS4 and URA3 . Finally , VDE-induced DSB dynamics and NCO levels were similar in PCH2 and pch2∆ strains , except that NCO levels at both loci were reduced in pch2∆ mms4-md yen1∆ slx1∆ , suggesting a greater role for SSNs in NCO formation in the absence of Pch2 ( Figure 4—figure supplement 1B , C ) . All of the experiments reported above used cells with wild-type levels of Spo11-DSBs . While VDE-DSBs form at similar levels and timing in SPO11 and spo11 mutant cells ( Johnson et al . , 2007; Neale et al . , 2002; Terentyev et al . , 2010 ) , features of VDE-DSB repair , including the extent of end resection , are strongly influenced by the presence or absence of Spo11-DSBs ( Neale et al . , 2002 ) . To determine if other aspects of VDE-initiated recombination are also affected , we examined VDE-initiated recombination in a catalysis-null spo11-Y135F mutant , hereafter called spo11 . In spo11 mutants , VDE-DSB dynamics and NCO formation were similar in inserts at HIS4 and URA3 , were comparable to those seen in wild type ( Figure 5—figure supplement 1 ) , and were independent of HJ resolvase activities ( Figure 5—figure supplement 1 ) . In contrast , the absence of Spo11-DSBs substantially reduced VDE-induced COs , resulting in virtually identical CO timing and levels at the two loci ( Figure 5A ) . Unlike the ~60% reduction in COs seen at HIS4 in SPO11 mlh3∆ ( Figure 3C ) , final CO levels were similar in spo11 mlh3Δ and spo11 MLH3 strains , at both HIS4 and URA3 , and similar CO reductions were observed at both loci in spo11 ssn mutants ( Figure 5B , C ) . Thus , processes that depend on Spo11-DSBs elsewhere in the genome are important to promote VDE-initiated COs , and appear to be essential for MutLγ-dependent CO formation . 10 . 7554/eLife . 19669 . 013Figure 5 . VDE-initiated COs are reduced and are MutLγ-independent in the absence of Spo11 activity . ( A ) VDE-initiated crossover frequencies , measured as in Figure 2C in spo11-Y135F strains ( dark solid lines ) in inserts at HIS4 ( red ) and at URA3 ( blue ) . Data from the corresponding SPO11 strains ( dotted lines , from Figure 2C ) are presented for comparison . ( B ) COs in HIS4 inserts in spo11 strains that are otherwise wild-type ( spo11 ) or lack either Mutlγ or structure-selective nucleases . ( C ) As in B , but with inserts at URA3 . Values are from two independent experiments; error bars represent range . Representative Southern blots are in Figure 5—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 01310 . 7554/eLife . 19669 . 014Figure 5—figure supplement 1 . DSBs and recombinant products in spo11 strains . ( A ) Cumulative DSB levels , expressed as loss of the VRS-containing insert , calculated as in Figure 2A . ( B ) Relative recovery of recombination products , calculated as in Figure 2—figure supplement 1B . ( C ) VDE-DSB frequencies , as in Figure 2B . ( D ) NCO frequencies , as in Figure 2C . In all four panels , solid lines denote values from spo11 strains; values from wild type ( dotted lines , from Figure 2 and Figure 2—figure supplement 1 ) are presented for comparison . ( E ) DSB ( top ) and NCO ( bottom ) frequencies in spo11-Y135F strains with inserts at HIS4 . ( F ) DSB ( top ) and NCO ( bottom ) levels in spo11-Y135F strains with inserts at URA3 . For all panels , values are from two independent experiments; error bars represent range . Representative Southern blots are in Figure 5—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 01410 . 7554/eLife . 19669 . 015Figure 5—figure supplement 2 . Southern blots of HindIII and HindIII-VDE digests of DNA from spo11 strains with inserts at HIS4 ( top ) and at URA3 ( bottom ) . Gel labels are as in Figure 1; JM—joint molecule recombination intermediates . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 015
We examined the contribution of different Holliday junction resolvases to VDE-initiated CO-formation in recombination reporter inserts at two loci , HIS4 and URA3 , which are 'hot' and 'cold' , respectively , for Spo11-inititiated recombination and for occupancy by the meiotic chromosome axis proteins , Hop1 and Red1 . VDE-initiated COs at HIS4 are similar to those initiated by Spo11 , in that most depend on MutLγ . In contrast , VDE-initiated COs at the 'cold' locus , URA3 , more closely resemble mitotic COs , which are independent of MutLγ , but are substantially dependent on SSNs ( Ho et al . , 2010; Ira et al . , 2003; Muñoz-Galván et al . , 2012 ) . Locus-dependent differences in MutLγ-dependence are reduced in pch2∆ mutants , as are differences in Hop1 occupancy at later times in meiosis I prophase . Based on these findings , we suggest that local chromosome context exerts an important influence on the biochemistry of CO formation during meiosis , and that factors responsible for creating DSB-hot and -cold domains also create corresponding domains where different DSB repair pathways are dominant . An attractive hypothesis ( Figure 6 ) is that regions enriched for meiosis-specific axial element proteins create a chromosomal environment that promotes meiotic DSB formation , limits inter-sister recombination , preferentially loads ZMM proteins ( Joshi et al . , 2009; Serrentino et al . , 2013 ) , and is required for recruitment of MutLγ . In such regions , where most Spo11-dependent events occur , recombination intermediates will have a greater likelihood of being captured by axis-associated ZMM proteins , and consequently being resolved as COs by MutLγ . Regions with lower axial element protein enrichment are less likely to recruit ZMM proteins and MutLγ; DSB repair and CO formation in these regions are more likely to involve non-meiotic mechanisms . In short , the meiotic genome can be thought of as containing two types of environment: meiotic axis protein-enriched regions , where 'meiotic' recombination pathways predominate; and meiotic axis protein-depleted regions , in which recombination events more closely resemble those seen in mitotic cells . 10 . 7554/eLife . 19669 . 016Figure 6 . Different resolvase functions in different genome domains . ( A ) Early crossover decision model for meiotic recombination ( Bishop and Zickler , 2004; Hollingsworth and Brill , 2004 ) illustrating early noncrossover formation , a major pathway where recombination intermediates form in the context of ZMM proteins and are resolved by MutLγ to form crossovers , and a minor pathway where ZMM-independent intermediates are resolved by SSNs as both crossovers and noncrossovers . ( B ) Division of the meiotic genome into meiotic axis-protein-enriched 'hot' domains ( red ) that are enriched for Red1 and Hop1 , and 'cold' domains where Red1 and Hop1 are depleted . VDE DSBs ( yellow stars ) can be directed to form efficiently in either domain , but only VDE DSBs that form in 'hot' domains can be recruited to the meiotic axis . ( C ) DSBs in 'hot' domains can form joint molecules ( red star ) in the context of ZMM proteins and the synaptonemal complex , and thus can be resolved by MutLγ-dependent activities . DSBs in 'cold' domains form joint molecules ( blue star ) outside of this structural context , and are resolved by MutLγ-independent activities . DOI: http://dx . doi . org/10 . 7554/eLife . 19669 . 016 The observation that some COs at HIS4 are SSN-dependent , even though most are MutLγ-dependent ( Figure 3 ) , indicates that this division is not absolute . In addition , it is important to keep in mind that ChIP-based values for meiotic axis protein-enrichment and molecular measures of CO resolvase-dependence are both population-based averages , and do not detect cell-to-cell heterogeneity . It is possible that meiotic axis protein enrichment at HIS4 varies across a population , and most SSN-dependent COs form in cells where HIS4 is not meiotic axis protein-enriched . Alternatively , it is possible that meiotic axis protein enrichment at HIS4 is uniform across a population , but that MutLγ is recruited to JMs with less than unit efficiency , and that when MutLγ is not recruited , SSNs resolve JMs . Finally , it is important to recognize that , while meiotic axis protein occupancy is an attractive candidate as a determinant of resolvase contributions to VDE-induced CO formation , other explanations are possible . It is possible that the associations seen at HIS4 and URA3 , rather than being directly causative , reflect another underlying aspect of meiotic chromosome structure or function , and that other differences between these two loci cause the observed differences in resolvase usage . While the current study is the first to directly query the effect of chromosome context on JM resolution , others have obtained results that are consistent with an effect of local chromosome context on meiotic DSB repair . Malkova and coworkers used the HO endonuclease to initiate recombination in meiotic cells at LEU2 , also a ‘hot’ locus ( Panizza et al . , 2011; Wu and Lichten , 1995 ) . The resulting COs were dependent on Msh4 , a ZMM protein , to the same degree as are Spo11-induced COs , suggesting that these nuclease-induced COs at the axis enriched LEU2 locus were the products of ZMM/MutLγ-dependent JM resolution ( Malkova et al . , 2000 ) . Serrentino et al . ( 2013 ) showed that enrichment for the budding yeast ZMM protein , Zip3 , at DSB sites is correlated with interhomolog CO levels . Specialized chromosome elements also impact meiotic recombination in budding yeast: COs are differentially reduced relative to NCOs near telomeres ( Chen et al . , 2008 ) ; and interhomolog recombination is inhibited near centromeres ( Chen et al . , 2008; Lambie and Roeder , 1988 , 1986; Vincenten et al . , 2015 ) . Locus-specific differences in CO/NCO ratios also have been observed in mouse meiosis ( de Boer et al . , 2015 ) , locus-specific differences in partner choice have been reported in S . pombe ( Hyppa and Smith , 2010 ) , and crossover suppression by centromeres is observed in many species ( Talbert and Henikoff , 2010 ) . Consistent with the suggestion that different meiotic recombination uses different mechanisms in different regions , the meiotic genome also appears to contain regions that differ in terms of the response to DNA damage . Treatment of meiotic yeast cells with phleomycin , a DSB-forming agent , triggers Rad53 phosphorylation , as it does in mitotic cells , while Spo11-DSBs do not ( Cartagena-Lirola et al . , 2008 ) . This suggests that Spo11-DSBs form in an environment that is refractory to Rad53 recruitment and modification , but that there also are environments where exogenously-induced damage can trigger the mitotic DNA damage response . In light of this suggestion , it is interesting that the meiotic defects of spo11 mutants in a variety of organisms are often only partially rescued by DSBs caused by exogenous agents ( Bowring et al . , 2006; Celerin et al . , 2000; Dernburg et al . , 1998; Loidl and Mochizuki , 2009; Pauklin et al . , 2009; Storlazzi et al . , 2003; Thorne and Byers , 1993 ) . While other factors may be responsible for the limited rescue observed , we suggest that it reflects the random location of exogenously-induced DSBs , with only a subset forming in regions where repair is likely to form interhomolog COs that promote proper homolog segregation . Although we observe marked differences in the contributions of different resolvases to VDE-induced CO formation at HIS4 and at URA3 , there is no absolute demarcation between MutLγ and SSN activities at the two loci . At HIS4 , where MutLγ predominates , ssn mutants still display a modest reduction in VDE-initiated COs when MutLγ is active , but an even greater relative reduction in the absence of MutLγ . These findings are consistent with previous studies suggesting that , in the absence of MutLγ , SSNs serve as a back-up that resolves JMs to produce both COs and NCOs ( Argueso et al . , 2004; De Muyt et al . , 2012; Zakharyevich et al . , 2012 ) . Our current data indicate that the converse may also be true , since at URA3 , MutLγ appears to make a greater contribution to CO formation in the absence of SSNs than in their presence . However , in our studies , JMs are more efficiently resolved in mlh3∆ mutants than in ssn mutants , which display persistent unresolved JMs . Therefore , if MutLγ acts as a back-up resolvase , it can do so in only a limited capacity , possibly reflecting a need for a specific chromosome structural context in which MutLγ can be efficiently loaded and activated . The absence of such a meiosis-specific chromosome context may explain why MutLγ does not appear to contribute to CO formation during the mitotic cell cycle ( Ira et al . , 2003 ) , although lower MLH3 expression in mitotic cells ( Primig et al . , 2000 ) may also reduce its contribution . Both VDE-induced and Spo11-induced COs form at significant frequencies in mlh3∆ ssn mutants , which lack all four of the HJ resolvase activities thought to function during meiosis ( Figure 3; Argueso et al . , 2004; Zakharyevich et al . , 2012 ) . These residual crossovers may reflect the activity of a yet-unidentified JM resolvase; they may also reflect the production of half-crossovers by break-induced replication ( Ho et al . , 2010; Kogoma , 1996; Llorente et al . , 2008 ) or by other mechanisms that do not involve dHJ-JM formation and resolution ( Ivanov and Haber , 1995; Mazón et al . , 2012; Muñoz-Galván et al . , 2012 ) . Alternatively , long-tract NCO gene conversion events that include flanking heterologous sequences might be responsible for the products , scored in our molecular assays as COs , that are independent of both MutLγ and SSNs . In catalysis-null spo11-Y135F mutants , most VDE-DSBs are repaired by interhomolog recombination ( Figure 5 , Figure 5—figure supplement 2 ) , indicating that a single DSB can efficiently search the meiotic nucleus for homology . However , VDE-promoted COs are substantially reduced in spo11 mutants ( Figure 5 ) , as has been observed with HO endonuclease-induced meiotic recombination ( Malkova et al . , 2000 ) . Moreover , in spo11 mutants , virtually all VDE-initiated COs are MutLγ-independent ( Figure 5 , Figure 5—figure supplement 2 ) , and thus more closely resemble COs that form in mitotic cells . Because patterns of Hop1 occupancy are not markedly altered in spo11 mutants ( Franz Klein , personal communication ) , these findings indicate that , in addition to the local effects of meiotic chromosome structure suggested above , meiotic CO formation is affected by processes that require Spo11-DSBs elsewhere in the genome . Meiotic DSB repair occurs concurrently with homolog pairing and synapsis ( Börner et al . , 2004; Padmore et al . , 1991 ) , and efficient homolog synapsis requires wild-type DSB levels ( Henderson and Keeney , 2004 ) , indicating that multiple interhomolog interactions along a chromosome are needed for stable homolog pairing . To account for the reduced levels and MutLγ-independence of VDE-initiated COs in spo11 mutants , we suggest that a single VDE-DSB is not sufficient to promote stable homolog pairing , and that additional DSBs along a chromosome are needed to promote stable homolog pairing , which in turn is needed to form ZMM protein-containing structures that stabilize JMs and recruit MutLγ . However , the 140–190 Spo11-DSBs that form in each meiotic cell ( Buhler et al . , 2007 ) are also expected to induce a nucleus-wide DNA damage-response , and to compete with other DSBs for repair activities whose availability is limited , and both have the potential to alter recombination biochemistry at VDE-DSBs ( Johnson et al . , 2007; Neale et al . , 2002 ) . Thus , while we believe it likely that defects in homolog pairing and synapsis are responsible for the observed impact of spo11 mutation on VDE-initiated CO formation , it remains possible that it is due to changes in DNA damage signaling , repair protein availability , or in other processes that are affected by global Spo11-DSB levels . We have provided evidence that structural features of the chromosome axis , in particular the enrichment for meiosis-specific axis proteins , create a local environment that directs recombination to 'meiotic' biochemical pathways . In the remainder of the genome , biochemical processes more typical of mitotic recombination function . In other words , the transition to meiosis from the mitotic cell cycle does not involve a global inhibition of 'mitotic' recombination pathways . These 'mitotic' mechanisms remain active in the meiotic nucleus , and can act both in recombination events that occur outside of the local 'meiotic' structural context , and in recombination in spo11 mutants . It is well established that local chromosome context influences the first step in meiotic recombination , Spo11-catalyzed DSB formation ( Panizza et al . , 2011; Prieler et al . , 2005 ) . Our work shows that it also influences the last , namely the resolution of recombination intermediates to form COs . It will be of considerable interest to determine if other critical steps in meiotic recombination , such as choice between sister chromatid and homolog as a DSB repair partner , or the choice between NCO and CO outcomes , are also influenced by local aspects of interstitial chromosome structure . In the current work , we focused on correlations between local enrichment for the meiosis-specific axis protein Hop1 and Holliday junction resolution activity during CO formation . Other HORMA domain proteins , including HIM-3 and HTP-1/2/3 in C . elegans , ASY3 in A . thaliana and HORMAD1/2 in M . musculus , also have been reported to regulate recombination and homolog pairing ( Ferdous et al . , 2012; Fukuda et al . , 2010; Kim et al . , 2014; Wojtasz et al . , 2009 ) , suggesting that HORMA domain proteins may provide a common basis for the chromosome context-dependent regulation of meiotic recombination pathways in eukaryotes .
All yeast strains are of SK1 background ( Kane and Roth , 1974 ) , and were constructed by standard genetic crosses or by direct transformation . Genotypes and allele details are given in Supplementary file 1 . Recombination reporter inserts with arg4-VRS103 contain a 73nt VRS103 oligonucleotide containing the mutant VDE recognition sequence from the VMA1-103 allele ( Fukuda et al . , 2007; Nogami et al . , 2002 ) inserted at the EcoRV site in ARG4 coding sequences within a pBR322-based plasmid with URA and ARG4 sequences , inserted at the URA3 and HIS4 loci , as described ( Wu and Lichten , 1995 ) . Recombination reporter inserts with the cleavable arg4-VRS ( Neale et al . , 2002 ) were derived from similar inserts but with flanking repeat sequences removed , to prevent repair by single strand annealing ( Pâques and Haber , 1999 ) . This was done by replacing sequences upstream and downstream of ARG4 with natMX ( Goldstein and McCusker , 1999 ) and K . lactis TRP1 sequences ( Stark and Milner , 1989 ) respectively ( see Supplementary file 1 legend for details ) . The resulting arg4-VRS and arg4-VRS103 inserts share 3 . 077 kb of homology . VDE normally exists as an intein in the constitutively-expressed VMA1 gene ( Gimble and Thorner , 1993 ) , resulting in low levels of DSB formation in presporulation cultures ( data not shown ) , probably due to small amounts VDE incidentally imported to the nucleus during mitotic growth ( Nagai et al . , 2003 ) . To further restrict VDE DSB formation , strains were constructed in which VDE expression was copper-inducible . These strains contain the VMA1-103 allele ( Nogami et al . , 2002 ) , which provides wild type VMA1 function , but lacks the VDE intein and is resistant to cleavage by VDE . To make strains in which VDE expression was copper-inducible , VDE coding sequences on an EcoRI fragment from pY2181 ( Nogami et al . , 2002 ) ; a generous gift from Dr . Satoru Nogami and Dr . Yoshikazu Ohya ) were inserted downstream of the CUP1 promoter in plasmid pHG40 , which contains the kanMX selectable marker and a ~1 kb CUP1 promoter fragment ( Jin et al . , 2009 ) , to make pMJ920 , which was then integrated at the CUP1 locus . Yeast strains were grown in buffered liquid presporulation medium and shifted to sporulation medium as described ( Goyon and Lichten , 1993 ) , except that sporulation medium contained 10 uM CuSO4 to induce VDE expression . All experiments were performed at 30°C . Genomic DNA was prepared as described ( Allers and Lichten , 2000 ) . Recombination products were detected on Southern blots containing genomic DNA digested with HindIII and VDE ( PI-SceI , New England Biolabs ) , using specific buffer for PI-SceI . Samples were heated to 65°C for 15 min to disrupt VDE-DNA complexes before loading; gels contained 0 . 5% agarose in 45 mM Tris Borate + 1 mM EDTA ( 1X TBE ) and were run at 2 V/cm for 24–25 hr . DSBs were similarly detected on Southern blots , but were digested with HindIII alone as previously described ( Goldfarb and Lichten , 2010 ) , and electrophoresis buffer was supplemented with 4 mM MgCl2 . Gels were transferred to membranes and hybridized with radioactive probe as described ( Allers and Lichten , 2001a , 2001b ) , and were imaged and quantified using a Fuji FLA-5100 phosphorimager and ImageGauge 4 . 22 software . HindIII-VDE gel blots were probed with ARG4 sequences from −430 to +63 nt relative to ARG4 coding sequences ( Probe 1 , Figure 1 ) . To correct for the low level of uncut VDE sites present in all VDE digests ( see Figure 1 ) , NCO frequencies measured from these digests were adjusted by subtracting the frequency of apparent NCOs in 0 hr samples . HindIII gel blots were probed with sequences from the DED81 gene ( +978 to +1650 nt relative to DED81 coding sequence ) , which is immediately upstream of ARG4 ( Probe 2 , Figure 1 ) . Digests of sae2∆ strains ( Figure 1—figure supplement 1 ) were probed with nt 3149–4351 of pBR322 . Cells were formaldehyde-fixed by adding 840 μl of a 36 . 5–38% formaldehyde solution ( Sigma ) to 30 ml of meiotic cultures , incubating for 15 min at room temperature , and quenched by the addition of glycine to 125 mM . Cells were harvested by centrifugation , resuspended in 500 μl lysis buffer ( Strahl-Bolsinger et al . , 1997 ) except with 1 mg/ml Bacitracin and complete protease inhibitor cocktail ( one tablet/10 ml , Roche 04693116001 ) as protease inhibitors , and cells were lysed at 4°C via 10 cycles of vortexing on a FastPrep24 ( MP Medical ) at 4 M/sec for 40 s , with 5 min pauses between runs . Lysates were then sonicated to yield an average DNA size of 300 bp and clarified by centrifugation at 21 , 130 RCF for 20 min . 1/50th of the sample ( 10 µl ) was removed as input , and 2 μl of anti-Hop1 ( a generous gift from Nancy Hollingsworth ) was added to the remainder ( ~490 μl ) and incubated with gentle agitation overnight at 4°C . Antibody complexes were purified by addition of 20 μl of 50% slurry of Gammabind G Sepharose beads ( GE Healthcare 17088501 ) , with further incubation for 3 hr at 4°C , followed by pelleting at 845 RCF for 30 s . Beads were then processed for DNA extraction ( Blitzblau et al . , 2012; Viji Subramanian and Andreas Hochwagen , personal communication ) . Beads were washed with 1 ml lysis Buffer and once each with 1 ml high salt lysis buffer ( same as lysis buffer except with 500 mM NaCl ) , 1 ml ChIP wash buffer ( 10 mM Tris , 0 . 25M LiCl , 0 . 5% NP-40 , 0 . 5% sodium deoxycholate , 1 mM EDTA ) and 1 mL 10 mM Tris , 1 mM EDTA; all washes were done for 5 min at room temperature . DNA was then eluted from beads by adding 100 ml 10 mM Tris , 1 mM EDTA , 1% SDS and incubating at 65°C for 15 min . Beads were then pelleted by a short spin at 16 , 363 RCF and the eluate transferred to a fresh tube . Beads were washed again in 150 ml 10 mM Tris , 1 mM EDTA , 0 . 67% SDS , mixed and pelleted again . Both eluates were pooled and crosslinks reversed for both immunoprecipitated ( IP ) and input samples by incubating overnight at 65°C . 250 ml 10 mM Tris 1 mM EDTA , 4 ml 5 mg/ml linear acrylamide ( 20 mg ) and 5 ml 20 mg/ml Proteinase K ( 100 mg ) was added , and samples were incubated at 37°C for 30 min for immunoprecipitates and 2 hr for input samples . 44 µl 5M LiCl was then added to immunoprecipitates , and DNA was precipitated by adding 1 ml ice cold ethanol , incubating at −20°C for 20 min , and centrifugation at 21 , 130 RCF for 20 min . For input samples , 44 ml 5M LiCl was added , followed by extraction with an equal volume of phenol:chloroform:isoamyl alcohol ( 25:24:1 ) and centrifugation at 16 , 363 RCF for 10 min . The aqueous layer was transferred to a fresh tube and DNA was precipitated from input samples as with immunoprecipitate samples . qPCR analysis of purified DNA from input and immunoprecipitated samples used primer pairs that amplify two regions: chromosome III coordinates 65350–65547 and 68072–68271 , Saccharomyces Genome Database , flanking the HIS4 gene , and chromosome V coordinates 115119–115317 and 117728–117922 , flanking the URA3 gene ( see Figure 1—figure supplement 1 ) . Chromosome coordinates are from the Saccharomyce cerevisiae reference genome ( Engel et al . , 2014 ) . Primers and genomic DNA from input and immunoprecipitated samples were mixed with iQ SYBR green supermix ( Biorad ) and analyzed using a Biorad iCycler . Numerical values underlying all graphs are contained in Supplementary file 2 . | Inside the cells of many species , double-stranded DNA is packaged together with specialized proteins to form structures called chromosomes . Breaks that span across both strands of the DNA can cause cell death because if the break is incorrectly repaired , a segment of the DNA may be lost . Cells use a process known as homologous recombination to repair such breaks correctly . This uses an undamaged DNA molecule as a template that can be copied to replace missing segments of the DNA sequence . During the repair of double-strand breaks , connections called crossovers may form . This results in the damaged and undamaged DNA molecules swapping a portion of their sequences . In meiosis , a type of cell division that produces sperm and eggs , cells deliberately break their chromosomes and then repair them using homologous recombination . The crossovers that form during this process are important for sharing chromosomes between the newly forming cells . It is crucial that the crossovers form at the right time and place along the chromosomes . Chromosomes have different structures depending on whether a cell is undergoing meiosis or normal ( mitotic ) cell division . This structure may influence how and where crossovers form . Enzymes called resolvases catalyze the reactions that occur during the last step in homologous recombination to generate crossovers . One particular resolvase acts only during meiosis , whereas others are active in both mitotic and meiotic cells . However , it is not known whether local features of the chromosome structure – such as the proteins packaged in the chromosome alongside the DNA – influence when and where meiotic crossover occurs . Medhi et al . have now studied how recombination occurs along different regions of the chromosomes in budding yeast cells , which undergo meiosis in a similar way to human cells . The results of the experiments reveal that the mechanism by which crossovers form depends on proteins called axis proteins , one type of which is specifically found in meiotic chromosomes . In regions that had high levels of meiotic axis proteins , crossovers mainly formed using the meiosis-specific resolvase enzyme . In regions that had low levels of meiotic axis proteins , crossovers formed using resolvases that are active in mitotic cells . Further experiments demonstrated that altering the levels of one of the meiotic axis proteins changed which resolvase was used . Overall , the results presented by Medhi et al . show that differences in chromosome structure , in particular the relative concentration of meiotic axis proteins , influence how crossovers form in yeast . Future studies will investigate whether this is observed in other organisms such as humans , and whether local chromosome structure influences other steps of homologous recombination in meiosis . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2016 | Local chromosome context is a major determinant of crossover pathway biochemistry during budding yeast meiosis |
The mouse olfactory sensory neuron ( OSN ) repertoire is composed of 10 million cells and each expresses one olfactory receptor ( OR ) gene from a pool of over 1000 . Thus , the nose is sub-stratified into more than a thousand OSN subtypes . Here , we employ and validate an RNA-sequencing-based method to quantify the abundance of all OSN subtypes in parallel , and investigate the genetic and environmental factors that contribute to neuronal diversity . We find that the OSN subtype distribution is stereotyped in genetically identical mice , but varies extensively between different strains . Further , we identify cis-acting genetic variation as the greatest component influencing OSN composition and demonstrate independence from OR function . However , we show that olfactory stimulation with particular odorants results in modulation of dozens of OSN subtypes in a subtle but reproducible , specific and time-dependent manner . Together , these mechanisms generate a highly individualized olfactory sensory system by promoting neuronal diversity .
Mapping the neuronal diversity within a brain remains a fundamental challenge of neuroscience . Quantifying variance in a population of neurons within and between individuals first requires precise discrimination of cellular subtypes , followed by an accurate method of counting them . While this has been achieved in a simple invertebrate model containing hundreds of neurons ( White et al . , 1986 ) , applying the same approach to mammalian brains that encompass many millions of neurons represents a significant challenge ( Wichterle et al . , 2013 ) . The main olfactory epithelium ( MOE ) is an essential component of the olfactory sensory system . It contains olfactory sensory neurons ( OSNs ) that express olfactory receptors ( ORs ) , the proteins that bind odorants ( Buck and Axel , 1991; Zhao et al . , 1998 ) . The mouse genome codes for over a thousand functional OR genes , but each mature OSN expresses only one abundantly , in a monoallelic fashion ( Hanchate et al . , 2015; Saraiva et al . , 2016; Tan et al . , 2015; Chess et al . , 1994 ) . This results in a highly heterogeneous repertoire of approximately 10 million OSNs ( Kawagishi et al . , 2014 ) within the nose of a mouse , stratified into more than a thousand functionally distinct subpopulations , each one characterized by the particular OR it expresses . The monogenic nature of OR expression serves as a molecular barcode for OSN subtype identity . Thus , the MOE offers a unique opportunity to generate a comprehensive neuronal map of a complex mammalian sensory organ , and investigate the mechanisms that influence its composition and maintenance . To date only a few studies have quantified the number of OSNs that express a given OR ( Bressel et al . , 2016; Fuss et al . , 2007; Khan et al . , 2011; Rodriguez-Gil et al . , 2010; Royal and Key , 1999; Young et al . , 2003 ) . For the scarce data available ( <10% of the full repertoire ) reproducible differences in abundance have been observed between OSNs expressing different ORs ( Bressel et al . , 2016; Fuss et al . , 2007; Khan et al . , 2011; Young et al . , 2003 ) . This suggests variance in the representation of OSN subtypes exists within an individual , but the extent of variation between individuals is unknown . Moreover , the mechanisms that dictate the abundance of OSN subtypes are poorly understood . Most promoters of OR genes contain binding sites for Olf1/Ebf1 ( O/E ) and homeodomain ( HD ) transcription factors ( Young et al . , 2011 ) , and these are involved in determining the probability with which the OR genes are chosen for expression ( Rothman et al . , 2005; Vassalli et al . , 2011 ) . Enhancer elements also regulate the gene choice frequencies of nearby , but not distally located , ORs ( Khan et al . , 2011 ) . To date , these studies have focused only on a handful of OSN subtypes . In addition to OR gene choice regulation exerted by genetic elements , it is conceivable that the olfactory system adapts to the environment . The MOE is continually replacing its OSN pool and the birth of every neuron presents an opportunity to shape the proportion of different subpopulations . It is also possible that relative OSN abundances could be altered by regulating the lifespan of each OSN subtype . Indeed activation extends a sensory neuron’s life-span ( Santoro and Dulac , 2012 ) , suggesting that persistent exposure to particular odorants may , over time , increase the relative proportions of the OSNs responsive to them . Some OSN subtypes do reportedly increase in number in response to odor activation , but others do not ( Cadiou et al . , 2014; Cavallin et al . , 2010; Watt et al . , 2004 ) . Whether this variation reflects differences in the biology of OSN subtypes or experimental procedures is unclear . Here , we fully map OSN diversity in the MOE and characterize the influence of both genetic and environmental factors on its regulation . We show that RNA sequencing ( RNAseq ) is an accurate proxy for measuring the number of OSNs expressing a particular OR type , and use this approach to quantify , in parallel , the composition of 1115 OSN subtypes in the MOE . We report that , while the repertoire of OSN subtypes is stable across individuals from the same strain , it reproducibly and extensively differs between genetically divergent strains of laboratory mice . We show that under controlled environmental conditions , these stereotypic differences in OSN abundance are directed by genetic variation within regulatory elements of OR genes that predominantly act in cis and are independent of the function of the OR protein . However , we find that persistent , but not continuous , exposure to specific odorants can also subtly alter abundance of the OSN subtypes that are responsive to such stimuli . Taken together , these results show that the OSN repertoire is shaped by both genetic and environmental influences to generate a unique nose for each individual .
Previously , we characterized the transcriptional profile of the whole olfactory mucosa ( WOM ) in adult C57BL/6J animals ( hereafter termed B6 ) to generate hundreds of new , extended OR gene annotations ( Ibarra-Soria et al . , 2014 ) . As each OR gene is expressed in only a small fraction of cells within WOM , differences in their abundance are difficult to distinguish from sampling bias . We hypothesized that mapping RNAseq data to significantly extended OR transcripts should increase detection sensitivity . With these models , OR gene mRNA level estimates in adult WOM increase , on average , 2 . 3-fold , but some increase almost 20-fold ( Figure 1—figure supplement 1A ) . Despite this improvement , most OR mRNAs still have relatively low-expression values ( Figure 1A ) . Nevertheless , they show a dynamic range of abundance levels ( Figure 1A , inset ) that are consistent between biological replicates , as indicated by their very high correlation values ( median rho = 0 . 89 , p<2 . 2 × 10−16 ) . 10 . 7554/eLife . 21476 . 003Figure 1 . RNAseq is highly sensitive for OR mRNA detection and provides a measurement of OSN diversity . ( A ) Barplot of the mean normalized expression of 1249 OR genes from six biological replicates , accounting for gene length . Genes are ordered by decreasing abundance . The horizontal line is the median expression ( 32 . 06 ) and all the genes below it are shown in the inset . ( B ) Mean normalized mRNA expression values for the OR genes in chromosome 9 of the Olfr7 cluster deletion mouse line ( green; n = 3 ) . The corresponding abundances in wild-type animals ( orange ) are shown as a mirror image ( n = 3 ) . The break on the x-axis separates the two OR clusters . The dotted box encloses the deleted ORs . ( C ) Unequal RNAseq expression levels for different OR genes can be explained by two scenarios: ( left ) an OR gene with high RNAseq levels is expressed by a larger number of OSNs than a gene with low RNAseq abundance; and/or ( right ) an OR with high RNAseq values is expressed in the same number of OSNs as one with low RNAseq values , but at higher levels per OSN . ( D ) Comparison of the number of OSNs that express nine OR genes assessed by in situ hybridization ( ISH; x-axis ) to the corresponding RNAseq values ( y-axis ) . Error bars are the standard error of the mean ( ISH n = 4 , RNAseq n = 6 ) . The line is the linear regression and the Spearman’s correlation coefficient ( rho ) indicates a very strong correlation . Representative ISH images of two OR genes ( in red ) are shown . ( E ) In single-cell RNAseq experiments , 63 OSNs were randomly collected from the MOE . The distribution of OR mRNA expression in WOM samples is plotted ( left ) , alongside the equivalent values for the ORs that were present in the picked single-OSNs ( right ) . There is a significant enrichment ( p<6 . 44 × 10−9 ) toward collecting OSNs that express OR genes with high RNAseq counts in WOM . ( F ) Comparison of the normalized expression value for the highest OR detected in each of the 63 single-OSNs ( y-axis ) to the corresponding mean value in WOM ( x-axis , n = 3 ) . The line is the linear regression and the Spearman’s correlation coefficient ( rho ) indicates there is no correlation . See also Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 00310 . 7554/eLife . 21476 . 004Figure 1—figure supplement 1 . RNAseq expression values are a proxy for OSN number . ( A ) Scatter plot for the OR mRNA expression levels using Ensembl gene models ( x-axis ) or the extended gene models from Ibarra-Soria et al . ( 2014 ) ( y-axis ) . The red line is the 1:1 diagonal . A substantial proportion of the OR repertoire’s mRNA expression levels increase when mapped to more complete gene models . ( B ) Comparison of the number of OSNs in the MOE that express 10 particular OR genes ( x-axis ) as assessed by Bressel et al . ( 2016 ) , to the corresponding values in the RNAseq data ( y-axis ) . The line is the linear regression and the Spearman’s correlation coefficient is indicated . The high correlation between these measurements indicate that the RNAseq expression estimates are a proxy for the number of OSNs that express each OR gene . ( C ) Comparison of the number of OSNs in the MOE that express a particular set of OR genes ( x-axis ) , as assessed by in situ hybridization by Fuss et al . ( 2007 ) , to the corresponding values in the RNAseq data ( y-axis ) . The line is the linear regression and the Spearman's correlation coefficient is indicated . ( D ) Comparison of the number of OSNs in the MOE that express a particular set of OR genes ( x-axis ) , as assessed by in situ hybridization by Khan et al . ( 2011 ) , to the corresponding values in the RNAseq data ( y-axis ) . The line is the linear regression and the Spearman’s correlation coefficient is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 004 To assess whether these low OR mRNA expression values are biologically meaningful or if they represent low-level technical artifacts of RNAseq analysis , we sequenced RNA from WOM of a mouse strain that has a targeted homozygous deletion of the Olfr7 OR gene cluster on chromosome 9 ( Xie et al . , 2000; Khan et al . , 2011 ) , and compared their gene expression profile to control mice . From the 94 OR genes of the cluster that have been deleted , 83 ( 88 . 3% ) have no counts in any of the three biological replicates . The 11 remaining genes have just one or two fragments mapped in only one of the replicates ( Figure 1B ) , resulting in normalized counts of less than 0 . 4 . In contrast , the control mice have from 14 . 2 to 498 . 1 normalized counts for the same genes . Together these experiments demonstrate that the use of extended gene models significantly increases the sensitivity to detect OR mRNA expression in WOM , and that the full dynamic range of abundances reflects true measures of OR gene expression . The wide range of stereotypic OR gene expression can be explained by two scenarios , acting alone or in combination ( Figure 1C ) : either ( 1 ) OR genes with high-expression values are monogenically expressed in more OSNs than those with low-expression values; and/or ( 2 ) OR genes are consistently expressed at different levels per OSN . To differentiate between these possibilities , we performed in situ hybridization ( ISH ) of probes specific to nine OR genes with expression values distributed across the dynamic range . We then counted the number of OSNs in which each OR is expressed ( Figure 1D ) . We find very strong correlation between OSN number and RNAseq expression value ( rho = 0 . 98 , p=5 × 10−5 ) . We additionally compared OR gene RNAseq expression levels with three independent measures of the number of OSNs expressing the same ORs ( Bressel et al . , 2016; Fuss et al . , 2007; Khan et al . , 2011 ) . In all three cases , we find high correlations ( Figure 1—figure supplement 1B–D ) . We next collected 63 single mature OSNs from WOM , and determined the OR gene most abundantly expressed in each using a single-cell RNAseq approach ( Saraiva et al . , 2016 , unpublished data ) . If OR expression levels in WOM reflect the proportion of OSNs that express each receptor ( Figure 1C ) , the probability of isolating each OSN type is not equal . Indeed , we find a strong selection bias towards OSNs that express OR genes with high RNAseq levels in WOM ( hypergeometric test , p=6 . 44 × 10−9; Figure 1E ) , suggesting those OSN types are more numerous in the olfactory epithelium . Thus , consistent with a recent analysis in zebra fish ( Saraiva et al . , 2015 ) , OR RNAseq values are an accurate measure of the number of each OSN subtype in the mouse WOM ( scenario 1 ) . But do consistent differences in OR mRNA levels per cell also contribute ( scenario 2 ) ? To test this , we quantified the mRNA levels of the most abundant OR gene in each of the 63 single , mature OSNs , normalized to three stably expressed OSN marker genes ( Khan et al . , 2013 ) . We find OR mRNA levels vary within the single cells , but this does not correlate with expression levels across the WOM ( rho = −0 . 04 , p=0 . 7518 ) ( Figure 1F ) . Analysis of ERCC spike-ins confirmed that the levels of OR mRNAs in single OSNs are reliable . Moreover , the single OSN transcript levels also positively correlate with transcript levels in pools of millions of OSNs ( Saraiva et al . , 2016 ) . Together , these data demonstrate that OR mRNA levels obtained by RNAseq are an accurate proxy for quantifying the diversity of OSN subtypes that express each receptor . The relative proportion of each OSN subtype is stable between genetically identical animals . We have previously reported the expression of OR genes in B6 male and female mice ( Ibarra-Soria et al . , 2014 ) . By applying full gene models to these data , here we confirm their OSN distribution profiles are remarkably similar ( Figure 2A ) ; only 1 . 2% of the OR gene repertoire is significantly differentially expressed ( Figure 2B ) . To investigate whether this OSN distribution is a stereotypic feature of the species , we next reconstructed the WOM transcriptome of a different laboratory strain , 129S5SvEv ( hereafter termed 129 ) . The 129 genome has 4 . 4 million single nucleotide polymorphisms ( SNPs ) and 0 . 81 million small indels compared to B6 ( Keane et al . , 2011 ) , of which we find 13 , 484 SNPs and 1 , 936 indels within our extended OR gene transcripts . As OR genes are particularly variable in coding sequence between strains of mice ( Logan , 2014 ) , mapping RNAseq data from other strains to a B6 reference genome results in biases in OR gene expression estimates ( Figure 2—figure supplement 1A ) . We therefore generated a pseudo-129 genome on which to map the RNAseq data , by editing the reference genome at all polymorphic sites . We confirmed that the RNAseq expression estimates correlate with the number of OSNs that express the corresponding receptor genes in 129 animals , as judged by in situ hybridization ( rho = 1 , p=5 . 5 × 10−6; Figure 2—figure supplement 1B ) . From the 1 , 249 OR genes , we find 462 are significantly differentially expressed ( DE ) compared to B6 ( false discovery rate ( FDR ) < 5% ) , representing 37% of the whole repertoire ( Figure 2C , D ) . 10 . 7554/eLife . 21476 . 005Figure 2 . OSN diversity varies between mouse strains . ( A ) Mirrored barplot of the mean normalized RNAseq expression values for the OR genes in male ( dark blue , top ) and female ( light blue , bottom ) B6 animals ( n = 3 ) . ( B ) Scatter plot for the same data , with the Spearman’s correlation ( rho ) indicating a strong correlation . The red line is the 1:1 diagonal . Significantly differentially expressed ( DE ) OR genes are represented in blue , non-DE genes are in black . ( C ) Same as in ( A ) but with the average B6 expression in blue ( both males and females , n = 6 ) compared to the corresponding 129 expression values in yellow ( n = 3 ) . ( D ) Corresponding scatter plot , with the significant DE genes in green . ( E ) Same as in ( A ) but comparing the B6 expression in blue ( n = 6 ) to the CAST abundances in red ( n = 3 ) . ( F ) Corresponding scatter plot , with DE genes in purple . ( G ) Venn diagram illustrating the intersection of DE OR genes between the pairwise comparisons of the three strains . ( H ) An example of an OR gene , Olfr6 , that is DE in all strain comparisons . ( I ) Representative in situ hybridizations ( ISH ) on coronal slices of B6 and 129 MOEs for two OR genes , Olfr31 and Olfr736 , that are DE between these strains . ( J ) The quantification of OSNs expressing each OR gene in B6 ( blue ) and 129 ( yellow ) are plotted alongside the corresponding RNAseq counts . The log2 fold-changes between the strains are indicated . ( K ) Fold-change between the strains obtained from ISH data ( x-axis ) or RNAseq counts ( y-axis ) for four DE OR genes expressed higher in B6 ( blue ) , four expressed higher in 129 ( yellow ) and one expressed at equivalent levels in both strains ( grey ) ; these include Olfr31 and Olfr736 . The line is the linear regression and the Spearman’s coefficient ( rho ) indicates a strong correlation between OSN and RNAseq counts . See also Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 00510 . 7554/eLife . 21476 . 006Figure 2—source data 1 . OR expression data in three strains of mice . Excel workbook containing the normalized expression data for all OR genes in B6 , 129 and CAST , along with the results of the differential expression analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 00610 . 7554/eLife . 21476 . 007Figure 2—source data 2 . Novel OR alleles in the CAST genome . Text file with the fasta sequences of the novel alleles identified in the CAST genome . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 00710 . 7554/eLife . 21476 . 008Figure 2—figure supplement 1 . Differences in genetic background result in great variance in OSN subtype diversity . ( A ) The difference in mean expression values for OR gene mRNAs in 129 animals , as obtained by mapping to a pseudo-129 genome versus mapping to the B6 reference ( n = 3 ) . The genes are ordered by their decreasing mean expression value after mapping to the pseudo-129 genome . ( B ) Comparison of the number of OSNs that express nine OR genes assessed by in situ hybridization ( ISH; x-axis ) to the corresponding RNAseq values ( y-axis ) in 129 animals . Error bars are the standard error of the mean ( ISH n = 4 , RNAseq n = 3 ) . The line is the linear regression and the Spearman’s correlation coefficient ( rho ) indicates a very strong correlation . ( C ) Same as ( A ) but for CAST RNAseq data mapped to a pseudo-CAST genome or the B6 reference ( n = 3 ) . ( D ) Mirrored barplot of the mean normalized expression values for the OR gene mRNAs in 129 ( yellow ) and CAST ( red ) animals . ( E ) Scatter plot of the same data as in ( C ) , with the Spearman’s correlation value ( rho ) indicated . The red line is the 1:1 diagonal . Significantly differentially expressed ( DE ) genes are presented in orange . ( F ) Scatter plot of the mean OR raw RNAseq counts for the RNAseq CAST data mapped to the pseudo-CAST genome ( x-axis ) or to the pseudo-CAST genome plus additional OR alleles identified as CNVs ( y-axis , n = 3 ) . The red line is the 1:1 diagonal . The counts of the new alleles are represented in blue . Only mRNA from 36 other OR genes change their abundance; all lose counts that now map to the additional alleles . ( G ) Scatter plot for the differential expression analysis of the OR repertoire in B6 versus CAST . The mean mRNA expression for each OR gene is plotted against its fold-change between the strains . Those significantly DE are red and the rest grey . The horizontal red line indicates equal expression in both strains . Highlighted in black are the mRNAs from OR genes that , after accounting for the new OR alleles , lose their DE status; and in blue are those that become DE . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 008 To determine whether greater genetic diversity influences the variance in OSN repertoire , we repeated this experiment using a wild-derived strain from the Mus musculus castaneus subspecies ( CAST/EiJ , henceforth CAST ) . This strain has more than 17 . 6 million SNPs and 2 . 7 million indels relative to B6 ( Keane et al . , 2011 ) ; of these , we counted that 45 , 688 SNPs and 6 , 303 indels are found within our extended OR transcripts . After mapping to a pseudo-CAST genome ( Figure 2—figure supplement 1C ) , 634 OR genes are significantly differentially expressed ( FDR < 5% ) compared to B6 , constituting 50 . 8% of the whole OR repertoire ( Figure 2E , F ) . The changes in expression for some OR genes are dramatic: 132 genes have differences of at least eight fold . Taking all pairwise comparisons into account ( including 129 vs CAST , Figure 2—figure supplement 1D , E ) , 821 OR genes ( 65 . 7% ) are DE between at least two strains . One hundred and thirty-six of these are DE in all three pairwise comparisons ( Figure 2G ) ; for example , there are consistently different numbers of Olfr6-expressing OSNs in each strain ( Figure 2H ) . To determine if the DE OR genes between strains reflect differences in the proportions of OSN subtypes , we performed ISH with probes specific to OR genes with significantly different expression values between B6 and 129 ( Figure 2I ) . We then counted the number of OSNs that express nine different OR mRNAs , in each strain , and compared this with their RNAseq expression values ( Figure 2J ) . We find a high correlation between the difference in OSN number and the difference in RNAseq expression values between B6 and 129 ( rho = 0 . 98 , p=5 × 10−5; Figure 2K ) , demonstrating our RNAseq-based approach accurately measures the difference in OSN repertoires between strains . OR gene clusters are enriched in copy number variants ( CNVs ) between individual human ( Nozawa et al . , 2007; Young et al . , 2008 ) and mouse strain genomes ( Graubert et al . , 2007 ) . Thus , it is possible that variance in OSN subtype representations are a consequence of different numbers of highly similar OR genes between strains . To assess this , we mined CAST genome sequence data ( Keane et al . , 2011 ) for heterozygous SNPs within annotated OR genes . We identified 51 ORs that contain 10 or more heterozygous SNPs , an indication of additional alleles . Using genome sequencing data from these genes , we identified 30 novel or misassembled OR genes . We remapped the CAST RNAseq data to a pseudo-CAST genome incorporating these new OR alleles and re-estimated the expression of the OR repertoire . The overall abundance profile remains unchanged except for 36 genes ( Figure 2—figure supplement 1F ) . To assess whether this accounts for the observed differential expression between strains , we compared these estimates to B6 . Only 12 of 634 OR genes lose their DE status , and 4 OR genes now become DE ( Figure 2—figure supplement 1G ) . Thus , while differences in OR gene copy number minimally contribute to the diversity in OSN repertoire between three strains of mice , other mechanisms are responsible for most of the variation . Genetically divergent mouse strains produce different chemical odortypes in their urine ( Kwak et al . , 2012; Yamaguchi et al . , 1981 ) and amniotic fluid ( Logan et al . , 2012 ) . Therefore , each strain of mouse , when housed in homogeneous groups , is exposed to a unique pre- and post-natal olfactory environment . As odor exposure alters the life-span of OSNs in an activity-dependent manner ( François et al . , 2013; Santoro and Dulac , 2012; Watt et al . , 2004 ) , genetic variation could regulate OSN population dynamics either directly or indirectly , via odortype . We , therefore , devised an experiment to test and differentiate the influence of the olfactory environment from the genetic background . We transferred four to eight-cell stage B6 and 129 zygotes to F1 mothers to ensure they experienced an identical in utero environment . At birth , B6 litters were cross-fostered to B6 mothers and 129 litters to 129 mothers . In addition , B6 litters received a single 129 pup , and 129 litters a single B6 pup . Therefore , each litter experienced a characteristic olfactory environment , but one animal ( the alien ) had a different genetic background from the others ( Figure 3A ) . At 10 weeks of age , we quantified the OSN repertoires of six alien animals and six cage-mates using RNAseq . We found that the OSN repertoires cluster in two groups , clearly defined by genetic background ( Figure 3B ) . The correlation coefficient for any two B6 samples was on average 0 . 97 , with no significant difference between the environments ( t-test , p=0 . 09 ) . In contrast , the correlation for any B6 with a 129 sample had a mean of 0 . 89 , which is significantly lower ( t-test , p=3 . 8 × 10−12 ) . Five hundred and seven OR genes , among 5475 other genes are DE between these mice when grouped by strain ( Figure 3—figure supplement 1A ) . In striking contrast , across the whole transcriptome we find only mRNA from two genes that show differences in expression according to odor environment , both of which encode ORs ( Figure 3C , Figure 3—figure supplement 1B ) . These data demonstrate that the diversity in OSN repertoire we observe between strains is almost entirely dictated by direct genetic effects . In a controlled environment , the influence of odortype on the development and maintenance of the MOE is minimal , perhaps restricted to only a few OSN subtypes . 10 . 7554/eLife . 21476 . 009Figure 3 . OSN diversity is determined by the genetic background and not by the olfactory environment . ( A ) Experimental strategy to differentiate genetic from environmental influences on OSN diversity . B6 ( blue ) and 129 ( yellow ) embryos , depicted as circles , were transferred into F1 recipient mothers ( grey ) . After birth , the litters were cross-fostered to B6 and 129 mothers , respectively . Each B6 litter received one 129 pup ( the alien ) and vice versa . After 10 weeks , the WOM was collected for RNAseq from three aliens from each strain , and one cage-mate each . ( B ) Heatmap of the expression of the OR genes ( columns ) in all 12 sequenced animals ( rows ) . Samples cluster by the genetic background of the animals . The strain and environment of each mouse is indicated through shading ( right ) . ( C ) Differential expression analyses revealed mRNA from only two genes , Olfr875 and Olfr491 , that are significantly altered based on the olfactory environment . Expression values are shown for each group . Blue and yellow boxes indicate B6 or 129 animals respectively , and the background indicates the olfactory environment . See also Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 00910 . 7554/eLife . 21476 . 010Figure 3—figure supplement 1 . Genetic but not environmental factors regulate OSN subtype abundance . MA plots of the transcriptome-wide differential expression analysis between aliens and cage-mates , testing for the effect of ( A ) a different genetic background and ( B ) a different olfactory environment . Significantly DE ORs are represented in blue; other DE genes are in red . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 010 The indifference of the OSN repertoire to the olfactory environment suggests its development and maintenance is not influenced by the specific activity of OR proteins or , by inference , their protein coding sequence . To further test this , we analyzed the OSN repertoire of newborn pups . We identify the presence of 1 , 198 ( 95 . 9% ) OSN subtypes across a dynamic range of abundance ( Figure 4A ) . The differential proportions of OSNs expressing particular OR genes are therefore already present during the development of the MOE , suggesting that it is not dependent on the activity of the OSNs nor on differences in OSN life-span . 10 . 7554/eLife . 21476 . 011Figure 4 . OSN diversity is independent of OR activity and is controlled in cis . ( A ) Mean normalized expression of the OR mRNA in the WOM of newborn B6 animals , arranged from most to least abundant ( n = 3 ) . ( B ) Mean normalized mRNA expression of 134 annotated OR pseudogenes in the B6 adult WOM ( n = 6 ) . ( C ) A genetically modified mouse line was produced that contains the coding sequence ( CDS ) of Olfr2 in place of Olfr1507 ( Olfr2 > Olfr1507 ) . The strategy combined the use of CRISPR-Cas9 technology to create double-strand breaks on either side of the Olfr1507 CDS , and a DNA vector containing the Olfr2 CDS along with ~1 kb homology arms for homologous recombination ( HR ) . ( D ) Mirrored barplot of the mean normalized mRNA expression values for the OR repertoire in B6 animals ( light blue , top; n = 4 ) and in Olfr2 > Olfr1507 homozygous ( hom ) mutants ( dark blue , bottom; n = 4 ) . Olfr2 becomes the most abundant OR and Olfr1507 is no longer expressed in the genetically modified line . ( E ) Scatter plot of the mean normalized counts ( x-axis ) of OR genes versus the log2 fold-change between Olfr2 > Olfr1507 homozygotes and WT controls ( y-axis , n = 4 ) . OR genes that are significantly differentially expressed are represented in blue . Olfr2 and Olfr1507 are strikingly different , whereas the rest of the repertoire is equivalent or very slightly altered . ( F ) Comparison of the fold-change of the CAST versus B6 OR expression ( y-axis ) to the fold-change between the CAST and B6 alleles in the F1 ( x-axis ) . The genes fall largely on the 1:1 diagonal ( red line ) indicating the mRNA expression pattern observed in the parents is preserved in the F1 and thus OR abundance is controlled in cis . The concordance correlation coefficient ( ccc ) is indicated , which quantifies the correlation between the two fold-change estimates while correction for agreement on the x=y line . ( G ) Examples of the normalized mRNA expression in the parental strains ( top ) of an OR gene that is more abundant in CAST ( Olfr1535 ) or in B6 ( Olfr598 ) . The corresponding mRNA abundance of each allele in the F1 ( bottom ) is preserved . The log2 fold-change is indicated for each comparison . Error bars are the standard error of the mean . See also Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 01110 . 7554/eLife . 21476 . 012Figure 4—source data 1 . OR expression data in the Olfr2 > Olfr1507 mouse . Excel workbook containing the normalized expression data for all OR genes in the Olfr2 > Olfr1507 mouse line and B6 controls , along with the results of the differential expression analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 01210 . 7554/eLife . 21476 . 013Figure 4—source data 2 . Number and coordinates of OE and HD binding sites in OR gene promoters . Excel workbook detailing the number of Olf1/Ebf1 ( OE ) and homeodomain ( HD ) transcription-factor-binding sites predicted by RegionMiner , for 1 kb upstream of the transcription start site ( TSS ) of OR genes ( as reported in Ibarra-Soria et al . , 2014 ) in all three strains . The genomic coordinates of each motif are also provided . Note that while the coordinates for the B6 data are from the reference mouse genome ( GRCm38 ) , those for 129 and CAST are based on the constructed pseudo-genomes and will not match the reference genome annotation . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 01310 . 7554/eLife . 21476 . 014Figure 4—figure supplement 1 . Differentially represented ORs have more variation . ( A ) The proportion of OR genes that have no SNPs ( left ) or at least one SNP ( right ) , between the B6 and 129 ( top ) or B6 and CAST ( bottom ) genomes . Data are further subdivided by whether mRNA from the OR is significantly differentially expressed ( DE; red ) or not ( grey ) between the corresponding strains . Across both strains and all sequence windows , a smaller proportion of invariant ORs are DE , while in ORs with sequence variation a larger proportion are DE . ( B ) Comparison of the fold-change of the CAST versus B6 OR expression ( y-axis ) to the fold-change between the CAST and B6 alleles in an in silico generated F1 ( x-axis ) . The genes fall largely on the 1:1 diagonal ( red line ) , but some deviation is observed . Such deviations can be attributed to technical noise . The concordance correlation coefficient ( ccc ) is 0 . 95 . ( C ) The deviation from the x=y line in ( B ) was subtracted from the fold-change estimates from the real F1 data ( Figure 4F ) and the resulting estimates are plotted ( x-axis ) against the parental fold-change ( y-axis ) . The ccc increases from 0 . 86 to 0 . 92 suggesting that a proportion of the deviation from equality observed in Figure 4F is due to technical noise . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 014 Next , we analyzed the expression of ORs that are pseudogenized and do not produce receptor proteins capable of odor-mediated activity , but can be co-expressed with functional ORs ( Serizawa et al . , 2003; Saraiva et al . , 2016 ) . These are represented in OSNs with a very similar distribution to functional OR genes ( Figure 4B ) . Moreover , we analyzed the OR genes that encode identical protein-coding sequences between different strains . 36 . 3% of the OSN subtypes that express identical ORs are differentially represented between 129 and B6 . 44 . 8% are differentially represented between CAST and B6 . Together , these results suggest that the proportion of each OSN subtype is not dependent on its endogenous OR receptor activity . To directly test whether the abundance of a particular OSN subtype is influenced by the identity of the receptor protein it expresses , we used CRISPR-Cas9 to replace only the coding sequence of Olfr1507 with that of Olfr2 ( referred to as Olfr2 > Olfr1507 ) , in a pure B6 genetic background ( Figure 4C ) . OSNs expressing Olfr1507 are the most common subtype in B6 , while Olfr2 expressing OSNs are ranked 334th by decreasing abundance . Homozygous Olfr2 > Olfr1507 animals have 47 fold more Olfr2-expressing OSNs compared to controls , and is the highest subtype in these animals ( Figure 4D ) . DE analysis of OR genes supports the striking reciprocal differences in Olfr1507 and Olfr2-expressing OSNs in the Olfr2 > Olfr1507 animals , but we also find 118 other OSN subtypes with significant , albeit comparatively subtle , differences ( over 90% have fold-changes < 2 ) ( Figure 4E ) . Taken together , these data indicate that the extensive variance in OSN subtype composition we observe in mice is determined by the wider genetic architecture of the animal , and is independent of the function of the OR protein each subtype expresses . To investigate how genetic background influences OSN subtype abundances , we mined 129 and CAST whole genome sequences ( Keane et al . , 2011 ) for SNPs and short indels in regulatory regions of OR genes . We find that differentially represented OSN subtypes express OR genes with significantly greater amounts of variation in their coding sequence , whole transcript and regions of 300 bp or 1 kb upstream of the transcription start site , for both the 129 and CAST genomes ( Mann-Whitney one tail , p<0 . 02 for 129 and p<0 . 0002 for CAST; Figure 4—figure supplement 1A ) . Further , we scanned OR gene promoters for O/E and HD binding sites . In the CAST genome , 58 and 310 putative OR promoters have gains or losses of O/E and HD-binding sites respectively , compared to the B6 genome . In contrast , only 12 and 46 OR promoters show differences in the number of O/E and HD-binding sites , respectively , when comparing the 129 and B6 genomes . We therefore hypothesized that OSN subtype repertoires are generated via sequence variance in OR gene promoter and/or local enhancer elements , which dictate the frequency of OR gene choice . For two OR gene clusters , it has been demonstrated that enhancer/promoter interactions act in cis and do not influence the expression of the homologous OR allele on the other chromosome ( Fuss et al . , 2007; Khan et al . , 2011; Nishizumi et al . , 2007 ) . However , recent chromosome conformation capture experiments revealed interchromosomal interactions between OR enhancer elements ( Markenscoff-Papadimitriou et al . , 2014 ) . Moreover , the differential representation of 118 other OSN subtypes in the Olfr2 > Olfr1507 line ( Figure 4E ) , 108 of which express ORs that are located on a different chromosome from Olfr1507 , is consistent with the possibility that genetic modification of one OR locus directly influences the probability of choice in other ORs , in trans . To determine whether the genetic elements that instruct the whole OSN repertoire are cis- or trans-acting , we carried out an analysis at the OR allele level in B6 × CAST F1 hybrids . Following the logic of ( Goncalves et al . , 2012 ) , if the genetic elements act in cis then we would expect the OSN subtypes that differ between B6 and CAST to be maintained between OSNs expressing the corresponding B6 and CAST alleles within an F1 hybrid . On the other hand , if the elements act in trans the number of OSNs that express the B6 derived allele in the F1 would not differ from those that express the CAST allele . Within F1 mice , 840 OSN subtypes ( 67 . 2% ) expressed OR mRNAs that could be distinguished at the allelic level . The ratios between B6 and CAST OSN subtype abundance ( F0 ) strongly correlate with the ratios between alleles in the F1 hybrids at approximately 1:1 ( concordance correlation coefficient ( ccc ) = 0 . 86; Figure 4F ) . In other words , taken across over 800 OSN subtypes , those expressing a B6 OR allele in F1 animals have the same repertoire as the B6 parent , while the subtypes expressing the CAST OR allele match the CAST parent ( Figure 4G ) . To better understand the significance of any deviations from 1:1 concordance in specific OSN subtypes , we corrected for technical noise associated with distinguishing allelic expression at low counts ( see Materials and methods; Figure 4—figure supplement 1B , C ) . The corrected data has a stronger correlation ( ccc = 0 . 92 ) and the best fit of this data matches 1:1 concordance ( Cb = 0 . 999 ) . We therefore conclude that , collectively , the genetic elements dictating the abundance of over 800 OSN subtypes act in cis . However , we cannot exclude the possibility that a small contribution from trans-acting elements may account for subtle deviations from unity for some OSN subtypes . Taken together , these data are consistent with a predominant model where genetic variation in local , non-coding regulatory elements determines the probability with which each OR gene is chosen early in OSN neurogenesis . Previous studies have shown that OSNs activated by their cognate ligands have increased life-span ( François et al . , 2013; Santoro and Dulac , 2012; Watt et al . , 2004 ) . With time , longer survival rates should translate into enrichment in the neuronal population , compared to those OSN types that are mostly inactive ( Santoro and Dulac , 2012 ) . However , we found no evidence of different strain- or sex-derived odors influencing the OSN repertoire ( Figures 2A and 3B ) . Because these odor exposures were temporally constant , we hypothesized that the absence of an observed environmental influence on OSN repertoire could be due to olfactory adaptation ( a reduction of specific olfactory sensitivity due to prolonged odor exposure , reviewed in [Zufall and Leinders-Zufall , 2000] ) . To test this , we exposed mice to a mix of four chemically distinct odorants ( acetophenone , eugenol , heptanal and ( R ) -carvone ) . The odorant mixture was added to the drinking water supplied to the animals to avoid adaptation , such that they could smell the odor mixture when they approached the bottle to drink ( Figure 5A ) . We collected the WOM from animals exposed to the odorants for 24 weeks from birth , along with water-exposed controls , and performed RNAseq . DE analysis reveals 36 OR genes with significantly different mRNA levels ( FDR < 5% ) , with similar numbers more or less abundant in the exposed animals ( Figure 5B , Figure 5—figure supplement 1 ) . We selected seven OR genes with the biggest fold-changes in mRNA level for which specific TaqMan qPCR probes were available , and validated their expression levels in a larger cohort . The results indicate that all the tested genes have mRNA levels that are statistically significantly different from controls ( t-test , FDR < 5% ) and the direction of the expression changes are concordant with the RNAseq data ( Figure 5C ) . 10 . 7554/eLife . 21476 . 015Figure 5 . Acute but not chronic exposure to odors alters OR mRNA abundance . ( A ) Four different odorants were mixed together and used to stimulate B6 animals . In an acute paradigm , the odor mix was added to the drinking water supplied to the animals and WOM was collected at different time-points . ( B ) WOM from animals exposed for 24 weeks and matched controls were sequenced ( n = 6 ) . The plot shows the normalized mean mRNA expression value ( x-axis ) for each OR gene compared to its fold-change in exposed versus control samples ( y-axis ) . Genes highlighted in red or blue have significantly up- or downregulated mRNAs , respectively . OR genes represented by an asterisk were selected for further validation . ( C ) qRT-PCR validation of the DE genes highlighted in ( B ) . The mean fold-change between exposed and control samples is plotted for animals exposed for differing periods of time ( x-axis ) . After 24 weeks of exposure , all the genes are significantly DE ( n = 8–13 ) . ( D ) Animals were acutely exposed to the odor mix for four weeks and then the stimulus was removed for 6 weeks . After the recovery period none of the OR mRNAs are significantly different from controls ( n = 8 ) . ( E ) A chronic exposure paradigm was tested by presenting the odor mix on a cotton ball , placed in the cages of the animals for 24 hr a day . The WOM was collected at different time-points . The genes previously shown to be DE were tested by qRT-PCR and none show consistent changes in mRNA levels across time ( n = 3–10 ) . T-test , FDR < 5%; * < 0 . 05 , ** < 0 . 01 , *** < 0 . 001 . Error bars are the standard error of the mean . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 01510 . 7554/eLife . 21476 . 016Figure 5—source data 1 . OR expression data in odor-exposed mice . Excel workbook containing the normalized expression data for all OR genes in control and mice exposed to the mixture of ( R ) -carvone , heptanal , acetophenone and eugenol; along with the results of the differential expression analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 01610 . 7554/eLife . 21476 . 017Figure 5—figure supplement 1 . Specific OSN subtypes change in abundance upon olfactory stimulation . Significantly DE OR genes ( FDR < 5% ) after 24 weeks exposure to a mixture of ( R ) -carvone , heptanal , acetophenone and eugenol . The boxplots represent expression values for six controls ( grey ) and six exposed ( blue ) animals . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 017 To characterize the temporal dynamics of these OR mRNAs , we tested their expression after different periods of exposure ( 1 , 4 and 10 weeks ) in independent samples . After 1 week of treatment , none showed significant differences from controls , which is expected since young pups do not drink from the odorized water bottle . After 4 weeks , three of the OR genes are DE from controls , and at 10 weeks , five out of the seven receptors are DE ( t-test , FDR < 5%; Figure 5C ) . To assess the plasticity of these changes , we stimulated a group of animals for four weeks , and then removed the odor stimuli for an additional 6 weeks . In these mice , none of the OR genes are DE from controls ( Figure 5D ) . Thus , the abundance of specific OR types in WOM is increasingly altered , over a period of weeks to months , upon frequent environmental exposure to defined olfactory cues . These differences are reversible and require persistent stimulation to be maintained . To investigate whether olfactory adaptation blocks this effect , we presented the same odor mixture on a cotton ball inside a tea strainer ( Figure 5E ) , such that the stimuli are present in a sustained manner . None of the same seven OR genes are DE after 24 weeks , nor are any consistently dysregulated during the course of the exposure experiment ( t-test , FDR < 5%; Figure 5E ) . Therefore , when odorants are present in the environment in a constant manner ( similar to those differentially produced by gender or strains of mice ) , the OR mRNA abundance levels most responsive to acute exposure remain unchanged . If temporal differences in OR mRNA abundance are a consequence of odorant-specific activity , exposure to different odorants should lead to the differential expression of discrete subsets of OR genes . To test this , we odorized the drinking water with ( R ) -carvone alone , heptanal alone , or with the combination of both ( Figure 6A ) . After 10 weeks of exposure , we tested the expression of the seven DE OR mRNAs that were responsive to the four odor mix ( acetophenone , eugenol , heptanal and ( R ) -carvone ) , by TaqMan qRT-PCR . None of the genes are significantly DE in the animals exposed to ( R ) -carvone alone . However , four of the seven OR genes have mRNA levels significantly different in the animals exposed to heptanal , or to the combination of both odorants ( t-test , FDR < 5%; Figure 6—figure supplement 1A ) . We next carried out a transcriptome-wide analysis by RNAseq , finding 43 OR genes significantly DE in at least one of the conditions ( FDR < 5% ) of which 32 ( 74 . 4% ) are upregulated in the odor-stimulated animals ( Figure 6B ) . Exposure to ( R ) -carvone or heptanal resulted in a change in mRNA expression of 15 and 20 OR genes , respectively . These sets of receptors are almost completely independent , with only one OR mRNA significantly upregulated in both groups ( Olfr538; Figure 6C–D ) . The animals that were exposed to both odorants simultaneously showed significant changes in mRNA levels for 24 OR genes , 15 of which are shared with the individually exposed groups ( hypergeometric test , p=1 . 87 × 10−19 ) . Almost 40% of the ORs that show significant changes when exposed to all four odorants ( Figure 5B ) are also significantly altered in one or more of the groups exposed to ( R ) -carvone , heptanal or their combination . Thus , together these data demonstrate that acute environmental exposure to the odorants alters the global expression of around 1 . 2–1 . 6% of OR genes in the WOM . These changes are odor-specific and reproducible in isolation and in increasingly complex mixtures . 10 . 7554/eLife . 21476 . 018Figure 6 . Odor-mediated changes in OR mRNA abundance are receptor specific . ( A ) B6 animals were acutely exposed for 10 weeks to ( R ) -carvone , heptanal or both . ( B ) The fold-change of exposed compared to control animals based on RNAseq data ( y-axis ) is plotted against the OR genes mean mRNA abundance ( x-axis ) , for each of the experimental groups ( n = 6 ) . Genes in red or blue have significantly up or downregulated mRNAs , respectively . ( C ) Venn diagram showing the intersections of the DE OR genes in each of the exposure groups . Only one OR mRNA changed in all groups; all the other are specifically altered upon exposure to ( R ) -carvone or heptanal . ( D ) Examples of an OR mRNA that changes in all groups ( Olfr538 ) , one that is specific to stimulation with ( R ) -carvone ( Olfr902 ) and one that responds only to heptanal ( Olfr1182 ) . ( E ) Dose-response curve for HEK293 cells expressing Olfr538 ( black ) and challenged with increasing concentrations of ( R ) -carvone . HEK293 cells expressing a RHO-tag only ( grey ) were challenged with the same concentrations of ( R ) -carvone as a control . ( F ) Dose-response curve for cells expressing Olfr524 ( black ) and challenged with heptanal , control cell responses are represented in grey . Error bars are the standard error of the mean . ( G ) Comparison of DE genes identified after 10 weeks of acute exposure to heptanal to those found via an in vivo deorphanization strategy . On the x-axis is the fold change of acutely exposed versus control animals with the corresponding p-value on the y-axis . The horizontal line represents the cutoff for significance . Each dot is an OR gene; those called significantly DE in both assays are shown in black , while those responding in only one experiment are in blue and purple . Half ( 11/20 ) of all the DE genes in the acute exposure experiment are identified in the deorphanization assay , suggesting that the changes are indeed mediated by OSN activation by heptanal . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 01810 . 7554/eLife . 21476 . 019Figure 6—source data 1 . OR expression data in odor-exposed mice . Excel workbook containing the normalized expression data for all OR genes in control and mice exposed to ( R ) -carvone alone , heptanal alone or the combination of both; along with the results of the differential expression analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 01910 . 7554/eLife . 21476 . 020Figure 6—figure supplement 1 . Olfactory-induced changes in OSN abundance are odor-specific . ( A ) qRT-PCR expression estimates for seven OR genes that were previously validated as DE in animals exposed to a mix of four odorants ( Figure 5C ) , here in mice exposed singly to ( R ) -carvone ( left ) , to heptanal ( center ) or to the combination of both ( right ) . The mean fold-change in expression between the exposed ( n = 6 ) and control mice ( n = 6 ) is plotted . The horizontal red line represents equal expression in both groups . None of the genes are significantly DE in animals exposed to ( R ) -carvone , but four are significantly DE when exposed to heptanal or the combination of both . T-test , FDR < 5%; * < 0 . 05 , ** < 0 . 01 , *** < 0 . 001 . Error bars are standard error of the mean . ( B ) Dose-response curve for HEK293 cells expressing Olfr347 ( black ) challenged with increasing concentrations of ( R ) -carvone . Control cells ( grey ) do not respond to the same increase in odorant concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 21476 . 020 To investigate whether DE OR genes are directly activated by the environmental odorants , we expressed a subset ( Olfr538 , Olfr902 , Olfr916 , Olfr1182 , Olfr347 and Olfr524 ) in a heterologous system ( Zhuang and Matsunami , 2008 ) and challenged them with increasing concentrations of ( R ) -carvone and heptanal . Half of the DE ORs we tested were responsive in vitro ( Figure 6E–F , Figure 6—figure supplement 1B ) : for example , Olfr538 displayed a dose-dependent response to ( R ) -carvone ( Figure 6E ) and Olfr524 was responsive to heptanal ( Figure 6F ) . Some odorants , including heptanal , are known to be decomposed by enzymes present in the nasal mucus ( Nagashima and Touhara , 2010 ) such that in vivo exposure to an odorant may result in stimulation of the OSNs with chemically distinct byproducts . We therefore employed a recently published deorphanization system to identify the ORs that respond to heptanal stimulation in vivo ( Jiang et al . , 2015 ) . This strategy exploits the phosphorylation of the S6 ribosomal subunit when an OSN is activated . Thus , by coupling pS6-immunoprecipitation ( ps6-IP ) and RNAseq , the OR mRNAs expressed in the activated OSNs can be identified . We exposed mice to two concentrations of heptanal for an hour , and sequenced the mRNAs from OSNs after pS6-IP . Twelve and 210 OR mRNAs were significantly enriched ( FDR < 5% ) upon exposure to 1% and 100% heptanal , respectively , compared to controls . Over half of the DE ORs after 10-week acute exposure to heptanal ( Figure 6B ) are also DE in the pS6+ cells ( 11 out of 20; Figure 6G ) , which is significantly more than expected by chance ( hypergeometric test , p=0 . 0001 ) . Thus , using both in vitro and in vivo methods , we conclude that long-term odor-mediated changes in OR gene expression occurs via direct activation of OSNs expressing those receptors .
The process of OR gene choice , stabilization and exclusion during OSN maturation is poorly understood . Occasionally , it is referred to as a random process ( McClintock , 2010; Rodriguez , 2013 ) , suggesting there is no pattern or predictability to the outcome . However , our data indicates that , at the OSN population level , the result of this process is deterministic . A particular genetic background in controlled environmental conditions reproducibly generates an OSN population with fixed , unequal proportions of the different OSN subtypes . Thus , the process that generates this profile is more accurately described as stochastic . Despite divergence in the profiles generated by different genomes , all show a similarly shaped distribution: a small proportion of OSN subtypes are present at high levels with a rapid decay in abundance thereafter . In fact , 3 . 6% or less of the OSN subtypes contribute to 25% of the overall neuronal content of the WOM . We find that unequal OSN distributions are already present at birth ( Figure 4A ) , suggesting the genetic influence is on the process of OR gene choice/stabilization rather than modulating neuronal survival . Proximity to the H element , a cluster-specific enhancer , increases the frequency in which an OR is represented within the OSN population ( Khan et al . , 2011 ) . Therefore , the most highly represented OSN types may express OR genes located close to other strong enhancers . However , we propose that genetic variation in enhancers is not sufficient to account for the full diversity of differences in OSN subtypes between strains , as different ORs located adjacent to one another within a cluster are frequently represented very differently . Recently , it has been proposed that higher levels of OR transcription per cell may result in more OSNs expressing that receptor due to increased success in a post-selection refinement process ( Abdus-Saboor et al . , 2016 ) . Our measurements of OR mRNA expression levels in 63 single OSNs ( Figure 1F ) do not support this hypothesis . Instead , our data are consistent with a model where non-uniform probabilities of OR choice are instructed by genetic variation in both specific OR promoters and enhancers . Supporting this model , we identified many putative promoters for differentially represented ORs where genetic variation has altered the number of Olf1/Ebf1 ( O/E ) and homeodomain ( HD ) transcription factors binding sites between the mouse strains , both sequences known to influence the probability of OR choice ( D'Hulst et al . , 2016; Vassalli et al . , 2011 ) . Moreover , through analysis of F1 hybrids , we confirmed the finding that the probability of choice of OR genes linked to the H element are regulated in cis ( Fuss et al . , 2007 ) , and extended this to over 800 additional OR genes distributed throughout the genome that show little to no evidence in support of trans-acting regulation . Instead , the haploid CAST- and B6-derived OR alleles within an F1 are each regulated almost identically as they are in a diploid state within their original genetic backgrounds ( Figure 4F , G ) . Our data are inconsistent with trans-interactions of multiple enhancers acting additively to regulate the probability of OR choice ( Markenscoff-Papadimitriou et al . , 2014 ) . These trans interactions may , however , stabilize or maintain OR singularity after choice has been instructed in cis , and thus could also be necessary for the stereotypic representation of OSNs in a fixed genetic background . Many existing studies into OR gene choice , especially those utilizing transgenic mouse lines , use animals with a mixed 129/B6 genetic background . The remarkable diversity in the OSN repertoire between these strains ( Figure 2 ) suggests caution should be exercised in their interpretation . Here , we created a mouse line that carries the coding sequence of Olfr2 in the locus of Olfr1507 , the most frequently selected OR gene , in a pure B6 genetic background . Olfr2-expressing OSNs , which rank 334th across the repertoire in the original B6 strain , are the most abundant OSN subtype in this modified line , demonstrating the critical importance of the genetic context in the regulation of the probability of OR gene choice . Curiously , we also observed that ~10% of other OSN types show comparatively subtle but reproducible differences in abundance . The mechanism underlying these differences is unclear . One plausible hypothesis is that the transposition of Olfr2 OSNs to a different olfactory zone alters the dynamics and spatial organization of the glomeruli in the olfactory bulb . This in turn may impact on the proliferation and survival of proximal OSN subtypes . Alternatively , stabilizing trans interactions could be affected by the OR gene swap , resulting in slightly altered repertoires for some OSN subtypes . In this case , the sequence of an OR receptor gene would influence the relative abundance of other OSN subtypes , but our Olfr2 > Olfr1507 swap experiment suggests the sequence of an OR receptor does not dictate the abundance of its own OSN subtype . The main olfactory epithelium regenerates throughout the life of an animal . It has been suggested that activity-mediated mechanisms may shape the olfactory system by increasing OSN survival ( François et al . , 2013; Santoro and Dulac , 2012; Watt et al . , 2004; Zhao and Reed , 2001 ) , although other studies have found that the number of specific OSN subtypes decrease or are unaffected by odor-exposure ( Cadiou et al . , 2014; Cavallin et al . , 2010 ) . Each of these studies focused on one or two OSN subtypes and the odor exposure procedures varied significantly in frequency , persistence and length . Here , we took a comprehensive approach , measuring the response of over 1 , 000 ORs to four odorants , after different types of exposure from 1 week to 6 months . We find that mice living in stable chemical environments maintain the olfactory transcriptomes of their genetically dictated OSN repertoire . However , when frequently recurring odor stimulation is introduced , the abundances of responsive ORs are modified . We propose that this difference is a result of olfactory adaptation in the presence of continuous stimulation . However , other factors may also contribute , for example the continuous exposure odors were likely detected orthonasally , while the intermittently exposure odors were retronasally detected . The lengthy timeframe for odor-mediated differences to emerge is consistent with modulation of OSN lifespan . It is mechanistically unlikely that odor-activity could influence the probability of OR gene choice , but it could promote OR expression stabilization or singularity . Compared to the dramatic influence of genetic variation on OSN repertoire , odor-evoked changes are subtle , typically less than a two fold change after 6 months of exposure . The limited effect magnitude and long-time scales precluded a more detailed analysis to confirm a correlative alteration in OSN number . Interestingly , we identified ORs that became more abundant after exposure to specific odorants and , within the same animals , others that became less abundant . Both types were marked by phosphorylation of the S6 ribosomal subunit , a feature of activated OSNs ( Jiang et al . , 2015 ) , indicating that the differential expression is mediated by OSN subtype-specific olfactory stimulation . This may explain why very different conclusions were drawn from previous exposure studies on a small number of ORs ( Cadiou et al . , 2014; Cavallin et al . , 2010; François et al . , 2013; Watt et al . , 2004 ) . Short-term odor exposures ( 30 min to 24 hr ) result in a temporary down-regulation of activated OR mRNA ( von der Weid et al . , 2015 ) , presumably as part of the olfactory adaptation process . It is possible that our analyses are capturing this dynamic short-term response in addition to changes in OSN numbers resulting from long-term exposures . We could not identify any phylogenetic or chromosomal predictor of the ORs that responded with contrasting directional effects , and at this time , the logic underpinning the difference in the direction of expression changes remains unexplained . Genetic variation has great impact on individual phenotypic traits . Humans differ in up to a third of their OR alleles by functional variation ( Mainland et al . , 2014 ) , which contributes to an individually unique sense of smell ( Secundo et al . , 2015 ) . Segregating OR alleles have been functionally linked to perceptual differences of their odor ligands , by altering intensity , valence or detection threshold ( Jaeger et al . , 2013; Keller et al . , 2007; Mainland et al . , 2014; McRae et al . , 2013; Menashe et al . , 2007 ) . However , in most cases , these OR coding genetic variants explain only a small proportion of the observed phenotypic variance ( reviewed in Logan [2014] ) , suggesting that other factors contribute to individual differences in perception . Recently , it has been demonstrated that increasing the number of a particular OSN subtype in a mouse nose increases olfactory sensitivity to its ligand ( D'Hulst et al . , 2016 ) . Therefore , the very different OSN repertoires present between strains of mice are likely to result in significant phenotypic variation in olfactory thresholds , and thus contribute to the individualization of olfaction . Although it remains to be determined whether human OSN repertoires are as variable as the mice reported here , an array-based study of OR expression in 26 humans found unequal expression of ORs within and between individual noses ( Verbeurgt et al . , 2014 ) . Moreover , a recent systematic survey of olfactory perception in humans found high levels of individual variability in reporting the intensity of some odors ( for example , benzenethiol and 3-pentanone ) but not others ( Keller and Vosshall , 2016 ) . Further , a non-coding variant within an OR cluster associated with insensitivity to 2-heptanone has been shown to be dominant to the sensitive allele ( McRae et al . , 2013 ) . As OR genes are regulated monoallelicaly , this implies that a 50% reduction in the sensitive OR allele dosage is , in some cases , sufficient to influence perception . On the other hand , because many odorants activate multiple OSN subtypes ( Malnic et al . , 1999 ) , a differential representation of one subtype may have a limited influence on the overall perception of its odor . Further investigation into the functional consequence of diverse OSN repertoires will be necessary to determine the full extent to which they individualize the sense of smell .
Animal experiments were carried out under the authority of a UK Home Office license ( 80/2472 ) , after review by the Wellcome Trust Sanger Institute Animal Welfare and Ethical Review Board . All mice were housed in single sex groups within individually ventilated cages , with access to food and water ad libitum . All WOM samples were obtained from a single animal , except the pup WOM samples , which were the pool of three or four individuals . Details of the strain , age and sex of each animal sequenced can be found in Supplementary file 1 . MOEs were dissected and immediately homogenized in lysis RLT buffer ( Qiagen , Germantown , Maryland ) . Total RNA was extracted using the RNeasy mini kit ( Qiagen ) with on-column DNAse digestion , following the manufacturer’s protocol . mRNA was prepared for sequencing using the TruSeq RNA sample preparation kit ( Illumina , San Diego , California ) . All RNA sequencing was paired-end and produced 100-nucleotide-long reads . Sequencing data were aligned with STAR 2 . 3 ( Dobin et al . , 2013 ) to the GRCm38 mouse reference genome ( B6 ) or to pseudo-genomes created for the different strains using Seqnature ( Munger et al . , 2014 ) to impute the high-quality variants reported by the Mouse Genomes Project , release v3 ( http://www . sanger . ac . uk/science/data/mouse-genomes-project ) . The parameters used for mapping were as follows: --outFilterMultimapNmax 1000 --outFilterMismatchNmax 4 --outFilterMatchNmin 100 --alignIntronMax 50000 --alignMatesGapMax 50500 --outSAMstrandField intronMotif --outFilterType BySJout . The annotation used was from the Ensembl mouse genome database version 72 ( http://jun2013 . archive . ensembl . org/info/data/ftp/index . html ) , modified to include all reconstructed gene models for OR genes as reported in ( Ibarra-Soria et al . , 2014 ) . The numbers of fragments uniquely aligned to each gene were obtained using the HTSeq 0 . 6 . 1 package ( RRID:SCR_005514 ) , with the script htseq-count , mode intersection-nonempty ( Anders et al . , 2015 ) . Raw counts were normalized to account for sequencing depth between samples , using the procedure implemented in the DESeq2 package ( Love et al . , 2014 ) . Data analysis , statistical testing and plotting were carried out in R ( http://www . R-project . org ) . To compare OR expression levels between datasets , normalization to account for the number of OSNs present in the WOM samples was carried out subsequent to depth normalization . For this , we used a method proposed by Khan et al . ( 2013 ) that uses marker genes known to be stably expressed in mature OSNs only . This allows estimating the proportion of WOM RNA contributed by the OSNs . Five different marker genes were considered: Omp , Adcy3 , Ano2 , Cnga2 and Gnal ( except for the single-cell data , where Adcy3 and Ano2 were excluded because they are not expressed in many cells ) . To normalize for OSN number , we first computed the correlation coefficient between the expression of each marker gene and the total counts in OR genes; those marker genes with strong correlation values were used for normalization . Then , we calculated the geometric mean of all marker genes for each sample . The average of all geometric means was obtained , and divided by each individual mean; this results in the generation of size factors . Finally , the OR normalized counts were multiplied by the corresponding size factor . Normalized OR expression estimates for the three strains , the Olfr2 > Olfr1507 and the odor-exposed animals are provided in Figure 2—source data 1 , Figure 4—source data 1 , Figure 5—source data 1 and Figure 6—source data 1 . Differential expression analysis was performed with DESeq2 1 . 8 . 1 ( Love et al . , 2014 ) with standard parameters . To test for differential expression ( DE ) on the OR repertoire the double normalized counts ( accounting for OSN number per sample ) were provided directly , and the normalizationFactors function was used with size factors of 1 to turn off further normalization . Genes were considered significantly DE if they had an adjusted p-value of 0 . 05 or less . For the cross-fostering dataset , a likelihood ratio test ( nbinomLRT function in DESeq2 ) was used to compare the full model genetics+environment+genetics:environment to a reduced one accounting only for the genetics . Detailed results are provided in Figure 2—source data 1 , Figure 4—source data 1 , Figure 5—source data 1 and Figure 6—source data 1 . Probes were designed against nine OR genes chosen among genes covering the receptor expression dynamic range and DE between 129 and B6 strains ( Olfr24 , Olfr31 , Olfr78 , Olfr124 , Olfr323 , Olfr374 , Olfr543 , Olfr736 and Olfr1512 ) . The following gene-specific oligonucleotides were used to amplify by PCR an amplicon of ~1000 bp from each transcript: Olfr24 TGGCTTACGACCGGTTTGTG ( for ) GAAATTAATACGACTCACTATAGGGTTTACACAGCCCAGGATCACAG ( rev ) Olfr31 TTGCTACCTGCTCGTCTCAC GAAATTAATACGACTCACTATAGGGCTAGCACTCGGGAGGTTGGAG Olfr78 GAGGAAGCTCACTTTTGGTTTGG GAAATTAATACGACTCACTATAGGGCAGCTTCAATGTCCTTGTCACAG Olfr124 GGTAATATCTCCATTATCCTAGTTTCCC GAAATTAATACGACTCACTATAGGGTTGACCCAAAACTCCTTTGTTAGTG Olfr323 TATCCAAGGTCACGGAGTTTCAG GAAATTAATACGACTCACTATAGGGGAGGGCACTTCCTTTCACTCTG Olfr374 TTGACCTCCTACACACGCATC GAAATTAATACGACTCACTATAGGGCCAAGACTGGACAAGATTTGGTG Olfr543 ATTCATACAGTGGTGGCCCAG GAAATTAATACGACTCACTATAGGGCTAAGAATTCAACAAGTCATAGCAGC Olfr736 GGCAATTGTGTATGCAGTGTACTG GAAATTAATACGACTCACTATAGGGCTGTGAAAAGTTCCCATGTACCTG Olfr1512 TACATCCTGACTCAGCTGGGGAACG GAAATTAATACGACTCACTATAGGGCACATAGTACACAGTAACAATAGTC The reverse primer in each case includes the T7 RNA polymerase promoter sequence , and both oligos were designed to amplify a fragment with <75% sequence similarity to other OR genes in the mouse genome . Amplicons were purified from the PCR reactions using the Wizard PCR cleanup kit ( Promega , Fitchburg , Wisconsin ) and used in riboprobe in vitro transcription with T7 RNA polymerase ( ThermoFisher Scientific , Waltham , Massachusetts ) and DIG-labeled UTP . WOM from 10-week-old 129 or B6 mice were collected by dissection , fixed for 48 hr in 4% paraformaldehyde in 1x PBS and demineralized for 10 days in 0 . 45M EDTA pH 8 . 0/1x PBS . Samples were then cryoprotected in 0 . 45M EDTA pH 8 . 0/1x PBS/20% sucrose for 24 hr , before embedding in OCT medium and sectioning on a Leica CM1850 cryostat to produce slides containing 16 μm coronal MOE sections . Slides were air-dried for 10 min , followed by fixation with 4% paraformaldehyde for 20 min , and treated with 0 . 1M HCl for 10 min . Tissue acetylation proceeded in 250 mL of 0 . 1M triethanolamine ( pH 8 . 0 ) with 1 mL of acetic anhydride for 10 min , with gentle stirring . Two washes in 1× PBS were performed between incubations . Riboprobe hybridization was done with DIG-labeled probes ( 1000 ng/mL ) at 60°C in hybridization buffer ( 50% formamide , 10% dextran sulfate , 600 mM NaCl , 200 μg/mL yeast tRNA , 0 . 25% SDS , 10 mM Tris-HCl pH 8 . 0 , 1× Denhardt’s solution , 1 mM EDTA pH 8 . 0 ) overnight . Post-hybridization washes included one wash in 2× SSC , one wash in 0 . 2× SSC and one wash in 0 . 1× SSC at 55°C , for 30 , 20 and 20 min , respectively . Tissue permeabilization was performed in 1× PBS , 0 . 1% Tween-20 for 10 min , followed by two washes in TN buffer ( 100 mM Tris-HCl pH 7 . 5 , 150 mM NaCl ) for 5 min at room temperature , followed by blocking in TNB buffer ( 100 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 0 . 05% blocking reagent [Perkin Elmer , Waltham , Massachusetts] ) . Slides were then incubated overnight at 4°C with sheep anti-DIG-AP ( Roche , Germany ) diluted in TNB ( 1:800 ) , washed in TNT ( 100 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 0 . 5% Tween 20 ) six times , 5 min each , transferred to alkaline phosphatase buffer ( 100 mM Tris-HCl pH 9 . 8 , 100 mM NaCl , 50 mM MgCl2 , 0 . 1% Tween 20 ) twice for 5 min each . Signal development was performed in the same buffer containing 5% poly-vinyl alcohol ( Mowiol MW 31 , 000 , Sigma-Aldrich , St . Louis , Missouri ) , 50 μg/mL BCIP and 100 μg/mL NBT , until the purple precipitate was clearly visible with minimum background . Due to the large size of each MOE section , we collected serially scanned images with the 'Scan Large Image' function on NIS Elements software ( 3 . 22 version , Nikon Instruments ) , on a motorized upright Nikon Eclipse 90i microscope equipped with a planar PlanFluor 10x/0 . 30 DIC L/N1 objective ( Nikon , Japan ) . Background correction was applied on individual images with the NIS elements software and stitched together using Image Composite Editor ( Microsoft , Redmond , Washington ) with no projection . Linear image adjustments were performed on Photoshop , using the 'Brightness and Contrast' and 'Levels' functions , to equalize the background tone across images . OR-expressing cells were counted by visual inspection . For each gene , three to four animals were analyzed; two to four sections were counted for each animal , and the cell counts were collected independently for each MOE side ( hemi-section ) . The mean number of OR-positive cells per section was calculated ( from the two hemi-section counts ) , followed by calculation of the mean number of OR-positive cells per animal ( from the two to four counted sections ) . Mean and s . e . m . descriptive statistics were then calculated across the three to four animals analyzed . Venn diagrams with areas proportional to the number of elements represented were created using the eulerAPE version 3 software ( Micallef and Rodgers , 2014 ) . We mined the Mouse Genomes Project data ( Keane et al . , 2011 ) , release v5 ( http://www . sanger . ac . uk/science/data/mouse-genomes-project; RRID:IMSR_JAX:000928 ) . Regions with high numbers of het SNPs indicate multiple alleles being mapped to a single locus in the reference genome . The sequences in the C57BL/6J ( RRID:IMSR_JAX:000664 ) genome ( GRCm38 ) for OR genes with het SNPs in CAST were used to construct a neighbor-joining phylogenetic tree using MEGA6 ( Tamura et al . , 2013 ) . From the tree we selected 33 clades that contained the OR genes with highest number of het SNPs . Then we extracted all of the CAST/EiJ whole-genome Illumina sequencing reads produced by the Mouse Genomes Project that were mapped to these loci ( http://www . sanger . ac . uk/science/data/mouse-genomes-project ) ( Keane et al . , 2011 ) , realigned these to the members of the respective clade , and then extracted the read pairs poorly aligned to known Olfr members ( >2% mismatch ) . We created a de novo assembly of these reads to produce a set of contigs with Geneious R7 ( Kearse et al . , 2012 ) . The contigs were scaffolded and gap-filled using the llumina reads . From the resulting scaffold sequences , we identified putative new alleles in CAST/EiJ ( Supplementary file 1 ) . The new allele's sequences are reported in Figure 2—source data 2 . To dissect the influence of the genetic background from the olfactory environment , C57BL/6N and 129S5 four- to eight-cell stage embryos were transferred into F1 ( C57BL/6J×CBA ) pseudo-pregnant females . One day after birth , the C57BL/6N and 129S5 litters were cross-fostered to C57BL/6N and 129S5 wild-type mothers , respectively . Then , a single pup from the other strain was transferred to the cross-fostered litter ( the alien ) . At weaning , animals from the same sex as the alien animal were kept , always in a 4:1 ratio between strains . If not enough animals of the correct sex were available in the litter , surplus animals from other litters were used . At 10 weeks of age , the WOM was collected form the alien and a randomly selected cage-mate , and RNA was extracted and sequenced as described . CRISPR-Cas9 technology was used to generate double strand breaks on either side of the Olfr1507 coding sequence and facilitate homologous recombination . Two guideRNAs -with sequences AAACTAGATACTTGGCTCATAGG and CATATTCTAGACATTGTCATAGG- were produced with the Ambion T7 MEGAshortscript kit and the Cas9 RNA with the Ambion mMessage mMachine T7 Ultra kit ( ThermoFisher Scientific ) as specified by the manufacturer’s protocols . All RNA were purified with Ambion MegaClear columns , eluting with pre-heated ( 95°C ) elution solution . The eluate was then precipitated with ammonium acetate , and resuspended in ultrapure water ( Sigma-Aldrich ) . For homologous recombination , we produced a DNA vector containing the coding sequence of Olfr2 and homology arms for the Olfr1507 locus of ~1 kb . This was cloned into a modified pUC19 backbone via Gibson Assembly . The sequence-verified plasmid was purified with the NucleoBond Xtra Midi Plus EF Kit ( ThermoFisher Scientific ) following the manufacturer’s protocol . The plasmid was digested to remove the backbone and gel-purified with the QIAquick Gel Extraction Kit ( Qiagen ) following the kit’s protocol . The DNA was precipitated with sodium acetate and resuspended in ultrapure water ( Sigma-Aldrich ) . Finally , the DNA was spin through an Ultrafree-MC centrifugal filter ( Merck , Germany ) . All components were microinjected into the cytoplasm of 112 C57BL/6N zygotes at the following concentrations: 25 ng/µl for each gRNA , 100 ng/µl of Cas9 RNA and 200 ng/µl of vector DNA . Thirty-eight pups were born and four were positive for the homologous recombination event . One of these was the correct substitution , while the others contained several copies of the DNA vector . To map the RNAseq data from the Olfr2 > Olfr1507 homozygous mice , we modified the reference B6 mouse genome ( GRCm38 ) to substitute the Olfr1507 CDS with that of Olfr2 . Additionally , the Olfr2 CDS in the endogenous locus was removed to avoid multimapping . All the counts from both the endogenous Olfr2 UTRs and the modified Olfr1507 locus were added together and reported as the Olfr2 counts; Olfr1507 was set to zero . The WT controls were mapped to the unmodified reference genome . Data processing and DE analysis were performed as previously described . We used the RegionMiner tool from the Genomatix software suite ( https://www . genomatix . de/solutions/genomatix-software-suite . html ) to identify overrepresented transcription factor binding sites ( TFBSs ) in the regions 1 kb upstream of the transcription start site of OR genes as annotated in ( Ibarra-Soria et al . , 2014 ) , for all B6 , 129 and CAST sequences . We extracted the match details for the matrix families NOLF and HOMF ( Matrix Family Library version 9 . 3 ) , which correspond to Olf1/Ebf1 and homeodomain TFs , respectively . Ad hoc perl scripts were used to parse out the core sequence coordinates of each motif match ( defined as the central 10 nucleotides of the reported match ) , and then to compare the results for each promoter across the strains . We identified those OR genes that had differing number of predicted sites . The number and coordinates of the predicted TFBS are provided in Figure 4—source data 2 . The RNAseq data from the WOM of B6 x CAST F1 hybrids were obtained from a pre-publication release by the Wellcome Trust Sanger Institute ( ERP004533 ) . Data was processed as described above . Total expression estimates were obtained by mapping the RNAseq data to the B6 or pseudo-CAST genomes , with standard parameters . The expression estimates obtained with each genome were very highly correlated ( rho = 0 . 99 , p<2 . 2 × 10−16 ) . Therefore , the data mapped to the B6 reference was used in downstream analyses . To obtain allele-specific expression estimates , the RNAseq data was mapped to both the B6 and the pseudo-CAST genomes , without mismatches . Therefore , reads that span SNPs could only map to the genome corresponding to the allele they come from . Subsequent analyses were performed on the OR repertoire only . All reads mapped across each SNP were retrieved with SAMtools ( RRID:SCR_002105 ) ( Li et al . , 2009 ) ; ad hoc perl scripts were used to exclude reads that splice across the SNP or that were not uniquely mapped . Finally , the number of different reads mapping across all SNPs of each gene was obtained . To normalize for depth of sequencing , the total expression raw data was combined with the estimates from the parental strains , and normalized all together . The OR data was then further normalized to account for the number of OSNs , as described above . The same size factors were used to normalize the expression estimates from SNP positions . To deconvolve the total expression into allele-specific expression , a ratio of the expression of each allele was obtained from the counts in SNP positions by dividing the counts in B6 over the total counts in B6 and CAST . Then , the total expression normalized counts were multiplied by the ratio to obtain the B6 expression , and to the inverse of the ratio for the CAST-specific expression . Only those genes with normalized counts in SNP positions above the lowest quartile were used ( 840 OR genes ) . To estimate the amount of technical noise inherent to allelic expression estimation in genes with low counts such as the ORs , we created an in silico F1 dataset . To do this , we used SAMtools ( Li et al . , 2009 ) to downsample the F0 B6 and CAST samples to 50% of the median depth of the F1 samples; we used the three male B6 samples only . We combined each B6 sample with a CAST sample to create an F1 where the contribution from each allele was 50% . We processed the resulting F1 data as described above . The ratio of the two alleles in the F1 data should match the corresponding ratios in the parental calculations , and we can attribute any deviations from unity to technical noise . Thus , we computed the deviation form the x = y line for each gene in the in silico F1 and used it to correct the fold-change estimate between the allelic expression values in the real F1 data . We used the concordance correlation coefficient as a measurement of the correlation of the fold-change estimates between the parental and allelic expression values ( Lin , 1989 ) . Unlike a regular correlation coefficient , it corrects for the agreement on the x = y line , where the data should fall if regulation occurs in cis . Furthermore , it provides the parameter Cb which quantifies the deviation of the best fit to the data from the x = y line; Cb = 1 means no deviation . For both the in silico F1 and the real F1 data , the Cb was 0 . 99 , indicating that the OR repertoire as a whole lies on the diagonal . B6 mice were exposed to a mix of heptanal , ( R ) -carvone , eugenol and acetophenone at 1 mM concentration each , diluted in mineral oil ( all odorants from Sigma-Aldrich except acetophenone , from Alfa Aesar ) . For the acute exposure experiments , the odor mix was added to the water bottles of the animals; mineral oil alone was used for controls . Water bottles were replaced twice a week with freshly prepared ones . The exposure started from at least embryonic day ( E ) 14 . 5 and the WOM was collected from age-matched exposed and control groups at different time-points after the start of the treatment . For the chronic exposure experiments , the odor mixture or mineral oil only , were applied to a cotton ball with a plastic pasteur pipette; these were put into metal tea strainers that were then introduced into the cage of the animals . The cotton ball was replaced fresh daily . The exposure started from birth and the WOM was collected from age-matched exposed and control groups at different time-points after the start of the treatment . For the follow-up experiments , animals were acutely exposed only to ( R ) -carvone , or heptanal , or to the combination of both . The final concentration of each odorant was 1 mM . The odorants were directly added to the water bottles , without dilution in mineral oil . Therefore , the controls were kept with pure water . The water bottles were changed twice a week . The exposure started from at least E16 . 5 and the WOM was collected at 10 weeks of age . For qRT-PCR experiments , RNA from WOM was extracted as previously described . 1 μg of RNA was reversed-transcribed into cDNA using the High-Capacity RNA-to-cDNA kit ( Applied Biosystems , Waltham , Massachusetts ) with the manufacturer’s protocol . Predesigned TaqMan gene expression assays were used on a 7900HT Fast Real-Time PCR System ( ThermoFisher Scientific ) following the manufacturer’s instructions . Mean cycle threshold ( Ct ) values were obtained from two technical replicates , each normalized to Actb using the ΔCt method . Relative quantity ( RQ ) values were calculated using the formula RQ = 2ΔCt . Differential expression between groups was tested in R , by a two-tailed t-test , with multiple-testing correction by the Benjamini and Hochberg ( FDR ) method . For OR response in vitro , a Dual-Glo Luciferase Assay System ( Promega ) was employed using the previously described method ( Zhuang and Matsunami , 2008 ) . Modified HEK293T cells , Hana3A cells , were obtained directly from the Matsunami Laboratory . Cell line identity and negative mycoplasma status was confirmed by PCR after the completion of all experiments . Cells were plated on 96-well PDL plates ( ThermoFisher Scientific ) for transfection with 5 ng/well of RTP1S-pCI ( Saito et al . , 2004; Zhuang and Matsunami , 2007 ) , 5 ng/well of pSV40-RL , 10 ng/well pCRE-luc , 2 . 5 ng/well of M3-R-pCI ( Li and Matsunami , 2011 ) , and 5 ng/well of plasmids encompassing the six olfactory receptors of interest . ( R ) -carvone and heptanal ( Sigma-Aldrich ) were diluted to a 1 mM solution in CD293 ( ThermoFisher Scientific ) from 1M stocks in DMSO . 24 hr following transfection , we applied 10-fold serial dilutions of each odorant from 1 mM to 1 nM in triplicate . Luminescence was measured after a 4-hr odor stimulation period using a Synergy two plate reader ( BioTek , Winooski , Vermont ) . Transfection efficiency was controlled for by normalizing all luminescence values by the Renilla luciferase activity . The data were fit to a sigmoidal curve and every OR-odorant pair was compared to a vector-only control using an extra sums-of-squares F test ( significantly different from empty vector if p<0 . 05 , the s . d . of the fitted log ( EC50 ) was less than one log unit , and the 95% confidence intervals of the top and bottom parameters did not overlap ) . Data were analyzed with GraphPad Prism 7 . 00 and R . Three- to four-week old C57BL/6 mice were placed individually into sealed containers ( volume ≈ 2 . 7L ) inside a fume hood and allowed to rest for 1 hr in an odorless environment . For odor stimulus , 10 µl odor solution or 10 µl distilled water ( control ) was applied to 1cm × 1cm filter paper held in a cassette . The cassette was placed into a new mouse container into which the mouse was also transferred , and the mouse was exposed to the odor solution or control for 1 hr . Experiments were performed in triplicates or quadruplicates , and within each replication the experimental and control mice were littermates of the same sex . Following odor stimulation , the mouse was sacrificed and the OE was dissected in 25 ml of dissection buffer ( 1 × HBSS ( with Ca2+ and Mg2+ ) , 2 . 5 mM HEPES ( pH 7 . 4 adjusted with KOH ) , 35 mM glucose , 100 µg/ml cycloheximide , 5 mM sodium fluoride , 1 mM sodium orthovanadate , 1 mM sodium pyrophosphate , 1 mM beta–glycerophosphate ) on ice . The dissected OE was transferred to 1 . 35 ml homogenization buffer ( 150 mM KCl , 5 mM MgCl2 , 10 mM HEPES ( pH 7 . 4 adjusted with KOH ) , 100 nM Calyculin A , 2 mM DTT , 100 U/ml RNasin ( Promega ) , 100 µg/ml cycloheximide , 5 mM sodium fluoride , 1 mM sodium orthovanadate , 1 mM sodium pyrophosphate , 1 mM beta–glycerophosphate , protease inhibitor ( Roche , one tablet/10 ml ) ) and homogenized three times at 250 rpm and nine times at 750 rpm . The homogenate was transferred to a 1 . 5 ml lobind tube ( Eppendorf , Germany ) , and centrifuged at 4600 rpm for 10 min at 4°C . The supernatant was then transferred to a new 1 . 5 ml lobind tube , to which 90 µl 10% NP–40 and 90 µl 300 mM DHPC ( Avanti Polar Lipids , Alabaster , Alabama ) was added . The mixture was centrifuged at 13 , 000 rpm for 10 min at 4°C . The supernatant was transferred to a new 1 . 5 ml lobind tube , and mixed with 20 µl pS6 antibody ( Cell Signaling Technology , Danvers , Massachusetts ) . Antibody binding was allowed by incubating the mixture for 1 . 5 hr at 4°C with rotation . During antibody binding , Protein A Dynabeads ( ThermoFisher Scientific , 100 µl/sample ) was washed three times with 900 µl beads wash buffer 1 ( 150mM KCl , 5 mM MgCl2 , 10 mM HEPES ( pH 7 . 4 adjusted with KOH ) , 0 . 05% BSA , 1% NP–40 ) . After antibody binding , the mixture was added to the washed beads and gently mixed , followed by incubation for 1 hr at 4°C with rotation . After incubation , the RNA-bound beads were washed 4 times with 700 µl beads wash buffer 2 ( RNase free water containing 350 mM KCl , 5 mM MgCl2 , 10 mM HEPES ( pH 7 . 4 adjusted with KOH ) , 1% NP–40 , 2 mM DTT , 100 U/ml recombinant RNasin ( Promega ) , 100 µg/ml cycloheximide , 5 mM sodium fluoride , 1 mM sodium orthovanadate , 1 mM sodium pyrophosphate , 1 mM beta–glycerophosphate ) . During the final wash , beads were placed onto the magnet and moved to room temperature . After removing supernatant , RNA was eluted by mixing the beads with 350 µl RLT ( Qiagen ) . The eluted RNA was purified using RNeasy Micro kit ( Qiagen ) . Chemicals were purchased from Sigma if not specified otherwise . 1 . 5 µl purified RNA was mixed with 5 µl reaction mix ( 1× PCR buffer ( Roche ) , 1 . 5 mM MgCl2 , 50 µM dNTPs , 2 ng/µl poly–T primer ( TATAGAATTCGCGGCCGCTCGCGATTTTTTTTTTTTTTTTTTTTTTTT ) , 0 . 04 U/µl RNase inhibitor ( Qiagen ) , 0 . 4 U/µl recombinant RNasin ( Promega ) ) . This mixture was heated at 65°C for 1 min and cooled to 4°C . 0 . 3 µl RT mix ( 170 U/µl Superscript II ( ThermoFisher Scientific ) , 0 . 4 U/µl RNase inhibitor ( Qiagen ) , 4 U/µl recombinant RNasin ( Promega ) , 3 µg/µl T4 gene 32 protein ( Roche ) ) was added to each tube and incubated at 37°C for 10 min then at 65°C for 10 min . 1 µl ExoI mix ( 2 U/µl ExoI ( NEB ) , 1× PCR buffer ( Roche ) , 1 . 5 mM MgCl2 ) was added to each tube and incubated at 37°C for 15 min then 80°C for 15 min . 5 µl TdT mix ( 1 . 25 U/µl TdT ( Roche ) , 0 . 1 U/µl RNase H ( ThermoFisher Scientific ) , 1× PCR buffer ( Roche ) , 3 mM dATP , 1 . 5 mM MgCl2 ) was added to each tube and incubated at 37°C for 20 min then 65°C for 10 min . 3 . 5 µl of the product was added to 27 . 5 µl PCR mix ( 1× LA Taq reaction buffer ( TaKaRa , Japan ) , 0 . 25 mM dNTPs , 20 ng/µl poly–T primer , 0 . 05 U/µl LA Taq ( TaKaRa ) ) and incubated at 95°C for 2 min , 37°C for 5 min , 72°C for 20 min , then 16 cycles of 95°C for 30 s , 67°C for 1 min , 72°C for 3 min with 6 s extension for each cycle , and then 72°C for 10 min . The PCR product was purified by gel purification , and 50 ng of the purified product was used for library preparation with Nextera DNA Sample Prep kits ( Illumina ) . Libraries were sequenced on a HiSeq 2000/2500 ( 12 libraries pooled per lane ) to produce 50 base pair single-end reads . The sequencing data have been deposited in the Gene Expression Omnibus database ( https://www . ncbi . nlm . nih . gov/geo/ ) under accession GSE87695 . Short reads were aligned to the mouse reference genome mm10 using Bowtie ( Langmead et al . , 2009 ) . The reads mapped to annotated genes were then counted using BEDTools ( Quinlan and Hall , 2010 ) ; the gene models for OR genes were replaced by those reported in Ibarra-Soria et al . ( 2014 ) . A rescuing scheme was used as implemented in Jiang et al . ( 2015 ) ( code available at https://github . com/Yue-Jiang/RNASeqQuant ( Jiang , 2017 ) with a copy archived at https://github . com/elifesciences-publications/RNASeqQuant ) . The read count tables were then analyzed using EdgeR ( Robinson et al . , 2010 ) to identify differentially expressed ORs . | Smells are simply chemicals in the air that are recognized by nerves in our nose . Each nerve has a receptor that can identify a limited number of chemicals , and the nerve then relays this information to the brain . Animals have hundreds to thousands of different types of these nerves meaning that they can detect a wide array of smells . Smell receptors are proteins , and the genes that encode these proteins can be very different in two unrelated people . This could partly explain , for example , why some people find certain odors intense and unpleasant while others do not . However , having different genes for smell receptors does not by itself completely explain why some people are more sensitive than others to particular smells . The amounts of each nerve type in the nose might also differ between people and have an effect , but to date it has not been possible to accurately count them all . Ibarra-Soria et al . have now devised a new method to essentially count the number of each nerve type in the noses of mice from different breeds . The method makes use of a technique called RNA-sequencing , which can reveal which genes are active at any one time , and thus show how many nerves are producing each type of smell receptor . Ibarra-Soria et al . learned that different breeds of mice had remarkably different compositions of nerves in their noses . Further analysis revealed that this was due to changes to the DNA code near to the genes that encode the smell receptor . Next , Ibarra-Soria et al . sought to find out how the amount of each nerve type is controlled by giving mice water with different smells for weeks and looking how this affected their noses . These experiments revealed that a small number of the nerve types became more or less common after exposure to a smell . The altered nerves were directly involved in recognizing the smells , proving that the very act of smelling can change the make-up of nerves in a mouse’s nose . These results confirm that the diversity in the nose of each individual is not only dictated by the types of receptors found in there , but also by the number of each nerve type . The next challenge is to understand better how these differences change the way people perceive smells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience",
"genetics",
"and",
"genomics"
] | 2017 | Variation in olfactory neuron repertoires is genetically controlled and environmentally modulated |
Receptor tyrosine kinases ( RTKs ) signal through shared intracellular pathways yet mediate distinct outcomes across many cell types . To investigate the mechanisms underlying RTK specificity in craniofacial development , we performed RNA-seq to delineate the transcriptional response to platelet-derived growth factor ( PDGF ) and fibroblast growth factor ( FGF ) signaling in mouse embryonic palatal mesenchyme cells . While the early gene expression profile induced by both growth factors is qualitatively similar , the late response is divergent . Comparing the effect of MEK ( Mitogen/Extracellular signal-regulated kinase ) and PI3K ( phosphoinositide-3-kinase ) inhibition , we find the FGF response is MEK dependent , while the PDGF response is PI3K dependent . Furthermore , FGF promotes proliferation but PDGF favors differentiation . Finally , we demonstrate overlapping domains of PDGF-PI3K signaling and osteoblast differentiation in the palate and increased osteogenesis in FGF mutants , indicating this differentiation circuit is conserved in vivo . Our results identify distinct responses to PDGF and FGF and provide insight into the mechanisms encoding RTK specificity .
Receptor tyrosine kinases ( RTKs ) signal through a shared set of intracellular pathways , including extracellular signal-related kinase ( ERK ) and phosphatidylinositol 3-kinase ( PI3K ) , yet the in vivo functions directed by different RTKs can be quite distinct , raising the question of how specific cellular responses are elicited ( Lemmon and Schlessinger , 2010 ) . Several models have been put forth to explain how RTKs encode specificity ( Hunter , 2000; Simon , 2000; Pawson , 2004; Volinsky and Kholodenko , 2013 ) . In concept , distinct responses may be encoded by modulation of individual pathways downstream of receptor activation , with each pathway regulating a specific outcome . Alternatively , many cellular responses may require integration from multiple input pathways , and signal specificity could arise from this unique combination of pathways . Analysis of mice harboring point mutations to disrupt binding of specific effector proteins to RTKs suggests both these models may apply in vivo; platelet-derived growth factor ( PDGF ) Receptor α ( Pdgfra ) mutants display effector-specific phenotypes in line with the former model ( Klinghoffer et al . , 2002 ) , but PDGF Receptor β ( Pdgfrb ) mediated outcomes require combined output across multiple pathways , consistent with the latter model ( Tallquist et al . , 2003 ) . In addition , quantitative differences in the duration and magnitude of signal induction provide an added layer of regulatory complexity ( Marshall , 1995 ) . At the transcriptional level , one outcome of RTK activation is the expression of immediate early genes ( IEGs ) ( Cochran et al . , 1983; Lau and Nathans , 1987 ) . Studies in cell culture have suggested IEGs constitute a generic readout of RTK activation with minimal specificity at the receptor or intracellular effector level ( Fambrough et al . , 1999 ) , but genetic experiments in mice indicate a degree of IEG specificity ( Schmahl et al . , 2007 ) . Therefore , a central goal remains to delineate RTK responsive transcriptional programs , identify the key signaling parameters encoding their regulation , and determine how these gene expression profiles dictate cellular decisions . Many components of RTK signaling play important roles in mammalian craniofacial development ( Bentires-Alj et al . , 2006; Newbern et al . , 2008; Fantauzzo and Soriano , 2015 ) . In particular , PDGF and FGF signaling are both essential for midface development . In mice , loss of Pdgfra ( Soriano , 1997 ) or its ligands Pdgfa and Pdgfc ( Ding et al . , 2004 ) results in facial clefting , and mice harboring a mutation abrogating PI3K binding to PDGFRα mirror these craniofacial phenotypes , implicating PI3K as the main effector of PDGFRα signaling ( Klinghoffer et al . , 2002 ) . In addition , both Pdgfra ( Wnt1-Cre; Pdgfrafl/fl ) ( Tallquist and Soriano , 2003 ) and FGF receptor 1 ( Wnt1-Cre; Fgfr1 ) ( Trokovic et al . , 2003; Wang et al . , 2013 ) neural crest conditional mutants exhibit cleft face , indicating both pathways are required for normal development of the neural crest derived facial skeleton . At the intracellular pathway level , previous work has implicated ERK as a key effector downstream of FGF signaling ( Lanner and Rossant , 2010 ) . Furthermore , mutations in both PDGF and FGF signaling have been linked to craniofacial syndromes in humans ( Choi et al . , 2009; Miraoui and Marie , 2010; Rattanasopha et al . , 2012 ) . Interestingly , chimeric receptor experiments in mice have shown that the intracellular domain of Fgfr1 cannot compensate for Pdgfra during development , suggesting these two receptors transmit biologically distinct signals in vivo ( Hamilton et al . , 2003 ) . The midface thus offers a unique opportunity to interrogate the mechanisms of signal specificity between these two RTKs in a developmentally relevant system . Given the requirement for PDGF and FGF signaling in the development of the neural crest derived midface , we sought to compare the gene expression programs regulated by these two RTKs . The architecture of the transcriptional response to RTK activation consists of three stereotypic waves: an IEG response involving core transcriptional regulators ( Fos , Jun , Egr ) , a delayed response playing a feedback role ( phosphatases , RNA-binding proteins ) , and a late sustained response determining cellular outcome ( Amit et al . , 2007; Avraham and Yarden , 2011 ) . However , the degree of conservation between genes regulated in each wave across different RTK families is unclear . Furthermore , although classic feedback regulators of mitogen-activated protein kinase ( MAPK ) pathways have been described , such as dual-specificity phosphatases ( DUSPs ) for ERK and c-Jun N-terminal kinase ( JNK ) ( Li et al . , 2007; Owens and Keyse , 2007 ) , the extent of effector-dependent transcription genome-wide is not well characterized . In the present work , we compare the transcriptional response to PDGF and FGF signaling in E13 . 5 mouse embryonic palatal mesenchyme ( MEPM ) cells . Although both PDGF and FGF are required in the neural crest for craniofacial development , we find distinct transcriptional programs , effector dependencies , and cellular outcomes in response to each RTK . While many genes in the early wave are shared across the two RTKs , FGF induces a quantitatively stronger response than PDGF . In addition , the feedback control provided by the delayed transcriptional wave displays distinct characteristics in response to PDGF and FGF . By exploring the effect of MEK/ERK and PI3K inhibition on these RTK-regulated gene expression profiles , we find PDGF-mediated transcription displays greater PI3K dependence , while FGF-mediated gene expression programs predominantly require ERK activity . This relationship is conserved at the level of cellular outcome , with FGF driving proliferation but PDGF promoting PI3K-dependent differentiation . Finally , we show overlapping domains of PI3K signaling , PDGF target gene expression , and skeletal differentiation during palatogenesis in vivo , a process perturbed in Fgfr1 conditional mutants . Taken together , our studies suggest unique roles for PDGF and FGF during development of the facial skeleton , and more broadly , demonstrate that distinct transcriptional responses to RTK signaling are encoded through qualitative and quantitative differences in intracellular pathway activation .
Since neural crest conditional loss of either Fgfr1 or Pdgfra leads to clefting , we chose to perform RNA-seq on E13 . 5 MEPMs treated with either PDGFA or FGF1 + heparin to identify the gene expression programs regulated by each signaling pathway ( Figure 1A ) . MEPMs express many essential markers of the palatal mesenchyme and have been previously used to study responses to many pathways ( Bush and Soriano , 2010; Iwata et al . , 2012; Fantauzzo and Soriano , 2014 ) , including PDGF and FGF ( Vasudevan and Soriano , 2014 ) . We performed RNA-seq at 1 and 4 hr following ligand treatment in order to characterize both the early and late responses to PDGF and FGF signaling ( Supplementary File 1 ) . In the samples submitted for sequencing , both PDGF and FGF induced a robust phospho-ERK ( pERK ) response at 15 min ( Figure 1—figure supplement 1A ) , and MEPMs generated from Pdgfra-GFP ( Hamilton et al . , 2003 ) and Fgfr1-CFP ( to be described elsewhere ) knockin reporter embryos display expression of each receptor at the protein level in all cells ( Figure 1—figure supplement 1B ) , further validating MEPMs as a suitable system to study RTK responses . 10 . 7554/eLife . 07186 . 003Figure 1 . FGF and PDGF stimulation result in distinct transcriptional responses . ( A ) Mouse embryonic palatal mesenchyme ( MEPM ) cells were dissected from E13 . 5 embryos , passaged twice , and then serum starved overnight prior to stimulation with either PDGFAA or FGF1 + heparin . ( B ) Expression of all genes with FPKM ( fragments per kilobase of exon per million reads mapped ) >1 at 1 hr ( 11 , 217 genes ) and ( B' ) 4 hr ( 11 , 266 genes ) . Genes colored blue are significantly increased with fibroblast growth factor ( FGF ) treatment and genes colored red are significantly increased with platelet-derived growth factor ( PDGF ) treatment . Values plotted on log2 scale . ( C ) Fold change ( FC ) comparison for all differentially expressed ( DE ) genes at ( C ) 1 hr or ( C' ) 4 hr ( compared to serum starved sample ) shows high correlation between the transcriptional response to each growth factor at 1 hr but low correlation at 4 hr . ( D ) Protein–protein interaction ( PPI ) network for all genes upregulated at 1 hr contains many classic immediate early genes ( such as AP-1 components , Myc , and Srf ) . Genes are colored based on their primary reported role in transcriptional regulation ( 22 , 23 ) , with orange squares representing activators and gray circles representing repressors . ( E ) FC ( compared to untreated sample ) for selected genes upregulated at 1 hr . Genes in bold are induced >1 . 5-fold in response to the indicated growth factor . Although both PDGF and FGF regulate many shared targets , the induction in response to FGF exhibits greater magnitude ( Fos , Fosb , Junb , Atf3 , Egr1 , Egr2 ) and longer duration ( Fos , Fosb , Jun , Junb , Egr1 , Egr2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07186 . 00310 . 7554/eLife . 07186 . 004Figure 1—figure supplement 1 . Characterization of E13 . 5 MEPMs and transcriptional response to PDGF and FGF signaling . ( A ) In the samples submitted for RNA-seq , both PDGF and FGF induce robust pERK responses at 15 minutes . ( B ) MEPMs derived from either Pdgfra-GFP or Fgfr1-CFP knockin reporter embryos express each receptor in all cells , confirming MEPMs are a homogenous cell population at the level of PDGF and FGF receptor expression . ( C ) Principal component analysis ( PCA ) reveals distinct responses to PDGF and FGF signaling , with the first principal component ( 51 . 29% of variance ) segregating conditions based on ligand treatment and the second principal component ( 12 . 61% of variance ) separating samples based on duration of stimulation . ( D ) Gene ontology analysis ( molecular function ) on genes induced at 1 hr reveals an overrepresentation of transcription factors and MAP kinase phosphatases ( DUSPs ) . ( E ) Validation of selected transcription factors by qPCR confirms greater magnitude and duration of induction by FGF compared to PDGF . DOI: http://dx . doi . org/10 . 7554/eLife . 07186 . 004 We first plotted the expression of all genes with FPKM ( fragments per kilobase of exon per million reads mapped ) values >1 at both 1 hr ( Figures 1B and 4 hr ( Figure 1B' ) ; although only a small number of genes are differentially regulated between the 1-hr PDGF and 1-hr FGF samples ( Cuffdiff q < 0 . 1 , Supplementary File 2; Trapnell et al . , 2010 ) , the difference in the response to these two growth factors is much greater at 4 hr . Consistent with this observation , visualization of all replicates by principal component analysis ( PCA ) ( Figure 1—figure supplement 1C ) revealed that the 1-hr PDGF and 1-hr FGF samples cluster together , but the 4-hr FGF replicates are distinctly separate from the 4-hr PDGF samples . Comparing the stimulated MEPMs to untreated cells , genes differentially regulated at 1 hr by either PDGF or FGF ( Supplementary File 2 ) show high correlation ( r2 = 0 . 8173 , Figure 1C ) , but by 4 hr , the two RTK signals are divergent ( r2 = 0 . 2881 , Figure 1C' ) . In addition , the genes regulated by PDGF at 1 hr ( n = 40 ) form a subset of those genes regulated by FGF at 1 hr ( n = 159 ) , further highlighting the similarity within the early response to both growth factors . Gene ontology analysis ( Huang et al . , 2009 ) of the genes induced at 1 hr revealed an enrichment of transcription factors and MAP kinase phosphatases downstream of both RTKs ( Figure 1—figure supplement 1D , p < 0 . 001 ) , similar to previous descriptions of the response to RTK activation ( Amit et al . , 2007; Avraham and Yarden , 2011 ) . To better visualize the organization of these targets , we constructed a protein–protein interaction ( PPI ) network from the genes regulated at 1 hr; in constructing this network , we only included direct interactions between proteins ( path length = 1 ) and excluded predicted interactions ( Berger et al . , 2007; Chen et al . , 2012 ) . The resulting network contains 25 upregulated genes ( out of 113 total induced genes ) , including many classic components of the IEG response , such as activator protein-1 ( AP-1 ) subunits ( Fos , Jun ) and their regulators ( Zfp36 , Atf3 ) as well as Myc , Egr1-4 , and Srf ( Figure 1D ) . Closer inspection revealed that many shared target genes within this network are regulated to both a stronger magnitude and longer duration following FGF treatment compared to PDGF treatment ( Figure 1E ) . Indeed , Fos , Fosb , Jun , and Junb all exhibit differences in signal magnitude and/or duration in response to FGF , as validated by qPCR ( Figure 1—figure supplement 1E ) . In sum , the two RTKs induce a similar gene expression profile at 1 hr ( Figure 1C ) , as evidenced by the high-correlation coefficient and overlap between genes differentially regulated by PDGF and FGF at 1 hr . However , FGF drives a quantitatively stronger early response with many transcription factors showing both a stronger magnitude and greater duration of induction in response to FGF compared to PDGF ( Figure 1E ) , which may explain in part the divergent gene expression profiles observed at 4 hr . Given the importance of MEK/ERK and PI3K/Akt signaling downstream of these RTKs during development ( Klinghoffer et al . , 2002; Corson et al . , 2003; Lanner and Rossant , 2010; Fantauzzo and Soriano , 2014 ) , we analyzed pERK and pAkt activation following FGF and PDGF stimulation in MEPMs ( Figure 2A ) . Consistent with the stronger response to FGF treatment in the gene expression data , the FGF-induced pERK response displays both a higher magnitude and longer duration of activation compared to the PDGF-induced pERK signal . In contrast , both FGF and PDGF induce similar patterns of pAkt activation , but the magnitude of the PDGF-pAkt induction is slightly greater . The FGF-pERK signal is apparent up to 6 hr following growth factor treatment ( Figure 2—figure supplement 1A ) , and increasing the dose of PDGFA ligand did not alter the kinetics of pERK activation ( Figure 2—figure supplement 1B ) . To better understand differences in the response to PDGF and FGF , we performed gene ontology analysis for the molecular function of genes that are differentially expressed ( DE ) between 4-hr PDGF treatment and 4-hr FGF treatment ( Figure 2B , p < 0 . 001 ) ( Huang et al . , 2009 ) . The top results for genes enriched following FGF treatment are sets associated with modulation of signaling , such as protein kinases and GTPase regulators , which may function as activators of Ras to promote MEK/ERK signaling . Transcriptional feedback regulation of RTK signaling is well established , particularly the role of DUSPs providing negative feedback for MAPK signaling ( Amit et al . , 2007; Li et al . , 2007; Owens and Keyse , 2007 ) . Indeed , many DUSPs ( MAPK phosphatases ) are induced in response to both PDGF and FGF treatment at 1 hr ( Figure 1—figure supplement 1D ) , but FGF alone induces the expression of kinases and GTPase regulators at 4 hr , suggesting a distinct role for the FGF response in regulating MEK/ERK activity . We , thus , performed Western blots in the presence of cycloheximide following both PDGF and FGF treatment to determine the effect of inhibiting protein synthesis ( and consequently , the delayed transcriptional response ) on ERK activation . Consistent with previous work exploring the role of DUSP-mediated negative feedback ( Amit et al . , 2007 ) , cycloheximide treatment increased the duration of the PDGF-pERK response ( Figure 2C ) . However , we observed the opposite effect of cycloheximide treatment on the FGF-pERK response ( Figure 2C' ) , suggesting a positive feedforward loop in which FGF induces the expression of kinases and GEFs to modulate the ERK response in addition to activating ERK directly . To further explore the architecture of the late FGF response , we constructed a PPI network from genes increased at 4-hr FGF stimulation compared to 4-hr PDGF treatment . The FGF network ( Figure 2—figure supplement 1C ) contains Prkca as a highly connected node , which is interesting given reported roles for protein kinase C ( PKC ) in facilitating a sustained pERK response ( Bhalla et al . , 2002; Santos et al . , 2007 ) as well as its importance downstream of FGF in skeletal development ( Miraoui and Marie , 2010 ) . We found that inhibition of PKC decreased both the initial pulse and sustained activation of the FGF-mediated pERK response in MEPMs ( Figure 2—figure supplement 1C' ) , consistent with its potential function as a hub within the FGF response network . In addition , many of the kinases and GEFs transcriptionally regulated by FGF have reported roles in positively modifying MEK/ERK signaling ( Figure 2—figure supplement 1D ) , which may explain in part the residual pERK response to FGF in the presence of PKC inhibition . Collectively , these data support a model in which the balance between positive and negative transcriptional loops is critical for determining patterns of pERK response to different RTKs ( Figure 2D ) . 10 . 7554/eLife . 07186 . 005Figure 2 . The delayed transcriptional response provides differential regulation of pERK duration in response to FGF and PDGF signaling . ( A ) Signaling time course shows a more robust phospho-ERK ( pERK ) response to FGF ( blue ) than PDGF ( red ) and a similar pAkt response to both growth factors . ( B ) Gene ontology analysis ( molecular function ) of genes DE between the 4-hr PDGF and 4-hr FGF conditions indicates enrichment for kinases and GTPase regulators in response to FGF signaling . ( C ) Cycloheximide treatment has opposite effects on pERK duration following ( C ) PDGF and ( C' ) FGF stimulation , indicating the delayed transcriptional response ( dependent on protein synthesis and thus inhibited by cycloheximide ) can provide both negative and positive signals to modulate pERK kinetics . ( D ) Model depicting loops that regulate the duration of the pERK wave in response to receptor tyrosine kinase ( RTK ) signaling includes both negative ( dual-specificity phosphatases [DUSPs] ) and positive ( kinases , GEFs ) components from the delayed transcriptional response . Western blot quantification plotted as mean ± SEM , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 07186 . 00510 . 7554/eLife . 07186 . 006Figure 2—figure supplement 1 . Signaling kinetics and organization of the late transcriptional response in response to PDGF and FGF treatment . ( A ) Long timecourse of PDGF and FGF treatment followed by Western blot for pERK and pAkt shows the FGF mediated pERK response persists up to 6 hours following growth factor stimulation . ( B ) The duration of the pERK response to PDGFA treatment does not change with increasing ligand dosage , indicating 30 ng/mL is a saturating dose for pERK duration . ( C ) PPI network constructed from genes differentially expressed at 4 hours following FGF treatment reveals Prkca as a highly connected node , suggesting Protein Kinase C as a regulator of the pERK driven FGF response . ( C' ) Inhibition of PKC abolishes both the initial pulse and sustained wave of pERK in response to FGF treatment . ( D ) Many candidate kinases and GEFs , such as Prkca ( Bhalla et al . , 2002 ) , Stk40 ( Li et al . , 2010 ) , Trib1 ( Kiss-Toth et al . , 2004 ) , Sgk1 ( Won et al . , 2009 ) , Vav3 ( Caloca et al . , 2003 ) , and Bcar3 ( Gotoh , et al . 2000 ) , have reported roles in positively modifying MEK/ERK signaling . PrkcaDOI: http://dx . doi . org/10 . 7554/eLife . 07186 . 006 The differences in signaling pathway activation following PDGF and FGF stimulation led us to consider how inhibition of these pathways affected the two RTK-mediated transcriptional programs . Thus , we analyzed the effector dependence of the transcriptional response by performing RNA-seq in cells stimulated with either PDGF or FGF in the presence of PD325901 ( MEK inhibitor ) or LY294002 ( PI3K inhibitor ) ( Supplementary File 1 ) . PCA on all thirteen sequenced conditions ( Figure 3—figure supplement 1A ) segregates the samples based on growth factor treatment along PC1 ( 44 . 57% of the variance ) and on inhibitor treatment along PC2 ( 17 . 72% of the variance ) . Similarly , the correlation matrix for all sampled replicates mimics the PCA , with the 4 hr FGF and 4 hr FGF + LY showing a gene expression profile distinct from all other samples ( Figure 3—figure supplement 1B ) . We next directly tested the effect of pathway inhibition on both the shared gene expression program induced by the two RTKs at 1 hr ( 113 genes total ) and the genes regulated between FGF and PDGF treatment at 4 hr . Globally , FGF target genes show greater MEK/ERK dependence than PI3K dependence , while PDGF responsive genes exhibit the opposite relationship ( Supplementary File 3 ) mirroring the reported signaling requirements for each RTK . This trend is apparent at 1 hr ( Figure 3A , A' ) and striking for genes DE at 4 hr , where 52% of FGF responsive genes are MEK/ERK dependent ( Figure 3B ) but only 22% were PI3K dependent ( Figure 3B' ) . This dependence is inverted for PDGF , where 9% are MEK/ERK dependent ( Figure 3C ) but 28% are PI3K dependent ( Figure 3C' ) . Strikingly , MEK inhibition increases the expression of PDGF targets ( Figure 3C ) but not FGF targets ( Figure 3B ) , while PI3K inhibition increases the expression of FGF responsive genes ( Figure 3A' , B' ) , suggesting the FGF-ERK and PDGF-PI3K relationships are important for both gene induction and repression . These intracellular pathway dependencies are independent of the magnitude of induction/repression , as varying the FC threshold did not affect the PDGF-PI3K or FGF-ERK relationships ( Figure 3—figure supplement 1C , D' ) . 10 . 7554/eLife . 07186 . 007Figure 3 . FGF and PDGF transcriptional responses exhibit differential usage of intracellular pathways . ( A ) Volcano plots visualizing the effect of ( A ) MEK ( Mitogen/Extracellular signal-regulated kinase ) inhibition and ( A' ) phosphatidylinositol 3-kinase ( PI3K ) inhibition on the shared set of 113 genes upregulated at 1 hr reveal that FGF ( blue points ) shows increased dependence on MEK/extracellular signal-related kinase ( ERK ) signaling , while PDGF ( red points ) utilizes PI3K to a greater degree . X-axis plotted as log2 ( [1 hr ligand + inhibitor]/[1 hr ligand] ) . ( B ) Genes with increased expression at 4-hr FGF treatment show higher dependence on ( B ) MEK/ERK activity compared to ( B' ) PI3K . ( C ) In contrast , genes with increased expression at 4-hr PDGF treatment show greater usage of ( C' ) PI3K compared to ( C ) MEK/ERK . X-axis plotted as log2 ( [4 hr ligand + inhibitor]/[4 hr ligand] ) . Data analyzed using two sample t-test , and genes at p < 0 . 1 are colored significant . Black points represent genes not significant at this threshold in all plots . ( D ) A core set of MEK/ERK ( 20 genes ) and PI3K ( 16 genes ) are dependent on these pathways downstream of both PDGF and FGF . ( D' ) A minority of genes can be activated through either MEK/ERK or PI3K signaling in response to PDGF or FGF , indicating a degree of plasticity in intracellular pathway usage . ( E ) Scatter plot comparing effect of MEK/ERK and PI3K inhibition on all 4 hr DE genes reflects FGF-ERK and PDGF-PI3K dependencies . Data plotted as log2 ( [4 hr ligand + inhibitor]/[4 hr ligand] ) and capped at ±5 for visualization . ( E' ) 52% of FGF target genes at 4 hr are repressed by MEK/ERK inhibition , while 28% of PDGF responsive genes are repressed by PI3K inhibition . Interestingly , 12% of genes are repressed by MEK/ERK inhibition and ‘superinduced’ by PI3K inhibition , indicating crosstalk between these pathways . Furthermore , 43% of PDGF responsive genes and 34% of FGF responsive genes are not significantly affected by either inhibitor , which suggests either combinatorial requirement of MEK and PI3K or alternate intracellular pathways drive expression of these genes . DOI: http://dx . doi . org/10 . 7554/eLife . 07186 . 00710 . 7554/eLife . 07186 . 008Figure 3—figure supplement 1 . RTK target genes show distinct patterns of effector dependence in a threshold-independent manner . ( A ) PCA of all 13 conditions highlights effects of ligand stimulation and inhibitor treatment , with the first principal component ( 44 . 57% of variance ) segregating conditions based on ligand treatment and the second principal component ( 17 . 72% of variance ) separating samples based on inhibitor treatment . ( B ) Correlation matrix comparing the effect of inhibitor treatment across all PDGF and FGF stimulated replicates indicates distinct effects for PD and LY treatment . ( C ) The FGF-ERK and PDGF-PI3K relationships occur independently of FC threshold selection , indicating there is not a specific effect for ERK or PI3K inhibition based on magnitude . The number of genes repressed in the FGF-PD condition is greater than the number of the genes in the FGF-LY condition for all thresholds , while PDGF shows greater PI3K dependence than FGF , both MEK/ERK and PI3K signaling are utilized by PDGF across all thresholds . A subset of genes show significant ‘superinduction’ in the presence of inhibitor treatment , particularly for PI3K inhibition downstream of FGF . Each point on the x-axis represents a different threshold , and the value on the y-axis indicates the number of genes with a log2 ratio greater than that threshold . ( D ) The FGF-ERK and PDGF-PI3K relationships at 4 hr are also conserved independently of FC threshold selection . For any given threshold , ( D ) PI3K inhibition represses more PDGF target genes and MEK/ERK inhibition induces more PDGF target genes . Similarly , ( D' ) MEK/ERK inhibition has a stronger repressive effect on FGF target genes than PI3K inhibition at all thresholds . DOI: http://dx . doi . org/10 . 7554/eLife . 07186 . 008 We then considered the degree of overlap between ERK and PI3K targets . At 1 hr , many shared target genes show similar ERK or PI3K dependence , possibly constituting a core pathway specific gene set ( Figure 3D ) . On the other hand , cross-comparison of PI3K- and ERK-dependent genes reveals some overlap ( Figure 3D' ) , underscoring a degree of plasticity in pathway usage downstream of different RTKs . When directly comparing the effect of MEK and PI3K inhibition at 4 hr ( Figure 3E ) , the crosstalk between FGF targets is particularly striking , as a large number of genes repressed by MEK inhibition were regulated ( both positively and negatively ) by PI3K inhibition . Furthermore , while 23% of PDGF targets are specifically PI3K dependent , only 5% of FGF targets are repressed by PI3K alone ( Figure 3E' ) . Instead , many FGF-PI3K targets are also repressed by MEK inhibition ( 77% ) , and conversely , a number of FGF-ERK targets are induced by PI3K inhibition ( 23% ) , indicating the effect of PI3K inhibition downstream of FGF involves crosstalk with ERK signaling . We were intrigued by the subset of genes exhibiting significantly increased expression ( ‘superinduction’ ) following inhibitor treatment at both 1 and 4 hr . Both RTKs show increased Jun expression in response to PD325901 , while FGF and LY294002 treatment upregulates many classic ERK targets such as Fos , Fosb , and Dusp6 ( Supplementary File 3 ) , suggesting this ‘superinduction’ may reflect compensatory activation of other intracellular pathways . Many such examples of crosstalk between ERK and PI3K have been documented ( Mendoza et al . , 2011 ) . Thus , we assayed the activation of these pathways following MEK inhibition with PD325901 and PI3K inhibition with LY294002 . We found striking induction of pJNK upon pERK inhibition by PD325901 downstream of both FGF ( Figure 4A ) and PDGF ( Figure 4B ) stimulation . Interestingly , a moderate change in pAkt induction following PDGF treatment and MEK inhibition seems apparent , although this induction was not observed at the dose used for the RNA-seq experiment ( 1 μM PD325901 ) . Next , we investigated the effect of PI3K inhibition with LY294002 . We found that FGF-mediated pERK induction increased with LY294002 treatment ( Figure 4A' ) , but PDGF treatment did not produce this increase in pERK signal ( Figure 4B' ) , consistent with changes observed at the level of target gene expression . At the inhibitor doses used in the RNA-seq experiment ( 1 μM PD325901 , 10 μM LY294002 ) , both PDGF and FGF significantly induce JNK activation in the presence of MEK inhibition , while only FGF activates ERK when PI3K is inhibited ( Figure 4C , C' ) . Finally , we performed qPCR for a subset of target genes to confirm their response to pathway inhibition . Fos , Fosb , and Junb exhibit ‘superinduction’ downstream of FGF specifically in response to PI3K inhibition , confirming these genes are indeed MEK/ERK dependent and providing further evidence that LY294002-mediated induction of pERK can drive transcriptional changes ( Figure 4D ) . Similarly , Jun is ‘superinduced’ in the presence of PD325901 downstream of both RTKs . There is a modest increase in Jun induction following FGF treatment with PI3K inhibition , likely reflecting the compensatory induction of pERK in this condition and consequent crosstalk between ERK and JNK signaling . In addition , Fos induction is increased with LY294002 treatment even in the absence of growth factor , which may reflect a degree of growth factor independent crosstalk between PI3K and ERK . However , this effect synergizes with ligand treatment , indicating this compensation across intracellular pathways is at least partially dependent on receptor activation . 10 . 7554/eLife . 07186 . 009Figure 4 . Inhibition of effector activation results in compensatory induction of alternate signaling pathways detectable in the transcriptional response . ( A ) PD325901 dose response Western blots reveal induction of pJNK as pERK is progressively inhibited downstream of FGF signaling . ( A' ) Similarly , LY294002 dose response Western blots reveal increased pERK signal as pAkt is inhibited . ( B ) Inhibitor dose response Western blots in response to PDGF signaling show activation of c-Jun N-terminal kinase ( JNK ) in response to MEK/ERK inhibition but ( B' ) no activation of ERK following PI3K inhibition . ( C ) Quantification of effector activation in response to FGF ( blue ) or PDGF ( red ) at the doses used in the RNA-seq experiment reflects ( C ) increased pJNK activation when MEK/ERK signaling is inhibited and ( C' ) increased pERK induction when PI3K activity is blocked . Data plotted as mean ± SEM , n = 3 and compared using two sample , unpaired t-test to baseline of 100% ( no change ) . *p < 0 . 05; **p < 0 . 001 . ( D ) Gene expression reflects the crosstalk observed at the signaling level , as verified by qPCR for selected target genes . Canonical ERK targets such as Fos , Fosb , and Junb are ‘superinduced’ upon LY treatment , while the JNK target Jun is ‘superinduced’ with PD treatment . Interestingly , a degree of ‘superinduction’ is observed in the presence of inhibitor prior to growth factor addition , which may reflect RTK-independent crosstalk between intracellular pathways . Data plotted as mean ± SEM , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 07186 . 009 We next investigated the functional consequence of these differential effector activation patterns and transcriptional programs in response to PDGF and FGF signaling . Gene ontology for biological processes ( Huang et al . , 2009 ) enriched in genes differentially regulated by PDGF and FGF at 4 hr revealed an FGF-mediated proliferation program and a PDGF-regulated differentiation circuit ( Figure 5A , p < 0 . 001 ) . Interestingly , we also observed enrichment for regulators of Wnt signaling in the 4-hr PDGF condition , consistent with reports of antagonism between Wnt and FGF as regulating the balance between differentiation and proliferation in skeletal development ( Mansukhani et al . , 2005 ) . We further identified a set of differentiation genes ( Id1 , Id2 , Id3 , Mef2c , Atoh8 ) in the PPI network constructed from genes regulated by PDGF treatment at 4 hr ( Figure 5—figure supplement 1A ) and confirmed these genes by qPCR ( Figure 5—figure supplement 1A' ) . In line with PDGF directing a skeletal differentiation program , the Id genes ( Maeda et al . , 2004; Kee and Bronner-Fraser , 2005 ) and Mef2c ( Verzi et al . , 2007 ) are required for craniofacial skeleton development in vivo . In addition , mouse genome informatics mammalian phenotype analysis to determine overrepresented mouse phenotypes ( Chen et al . , 2013 ) identifies abnormal craniofacial bone development as the most enriched phenotype in the 4-hr PDGF target genes ( Figure 5—figure supplement 1B , B' ) . Plotting the expression of these FGF-proliferation and PDGF-differentiation genes emphasizes the distinct responses of these gene sets to these two growth factors ( Figure 5B ) . Furthermore , many proliferation genes are reduced upon MEK inhibition and induced upon PI3K inhibition , while the opposite is apparent for differentiation genes , consistent with the FGF-ERK and PDGF-PI3K dependencies observed globally at the transcriptional level . 10 . 7554/eLife . 07186 . 010Figure 5 . Distinct cellular outcomes are specified in response to PDGF and FGF signaling . ( A ) Gene ontology analysis of genes DE between the 4-hr FGF and 4-hr PDGF conditions shows enrichment for regulators of cell proliferation in genes upregulated by FGF treatment . In contrast , genes implicated in skeletal differentiation are overrepresented in the genes more highly expressed following PDGF treatment ( B ) Genes associated with cell proliferation are strongly upregulated by FGF treatment , while genes associated with cell differentiation are increased following PDGF stimulation . In addition , MEK inhibition represses many proliferation genes but induces differentiation genes , while PI3K inhibition has the opposite effect . Genes ordered by decreasing ratio of FGF:PDGF expression . ( C ) FGF induces a significantly more robust proliferation response than PDGF in MEPMs , although PDGF does promote a modest response compared to starved cells ( 0 . 1% FBS ) . Quantification plotted as mean ± SEM , n = 3 . Two tailed , unpaired t-test: *p < 0 . 05; **p < 0 . 005 ( D ) PDGF , but not FGF , treatment promotes alkaline phosphatase ( AP ) ( osteoblast marker ) positive cells . Furthermore , PD treatment drives AP staining , while LY treatment abolishes AP staining independent of growth factor stimulation . AP staining performed 8 hr following ligand treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 07186 . 01010 . 7554/eLife . 07186 . 011Figure 5—figure supplement 1 . PDGF-mediated differentiation responses exhibit a preference for PI3K signaling , while FGF-mediated effects on proliferation show greater MEK/ERK dependence . ( A ) . The PPI network for the genes upregulated at 4 hr following PDGF treatment shows a concentration of genes involved in controlling cell differentiation ( Id1 , Id2 , Id3 , Atoh8 , Mef2c ) , suggesting a PDGF-mediated differentiation program in MEPMs . ( A' ) Validation of genes in the PDGF target PPI network by qPCR confirmed these genes are increased following PDGF stimulation compared to FGF treatment . ( B ) Genes with higher expression at 4 hr following PDGF treatment show enrichment for craniofacial and skeletal phenotypes in mice , consistent with the enrichment for differentiation genes in the 4-hr PDGF condition . The top 10 terms are shown , and selected genes with craniofacial phenotypes are listed . ( B' ) Heatmap displaying expression of genes associated with craniofacial phenotypes and exhibiting increased expression at 4 hr PDGF stimulation . Many of these genes show increased expression upon MEK inhibition and decreased expression upon PI3K inhibition , consistent with a PDGF-PI3K-differentiation circuit . ( C ) Crystal violet assay demonstrates that a full FGF response ( blue lines ) requires both MEK/ERK and PI3K activity , although the effect of MEK/ERK inhibition is significantly greater . In contrast , the PDGF-mediated effect ( red lines ) on cell viability requires both MEK/ERK and PI3K activity , suggesting these pathways are utilized somewhat equally downstream of PDGF signaling . Data plotted as mean ± SEM , n = 3 . ( D ) Cleaved Caspase-3 staining reveals that inhibition of FGF-mediated MEK/ERK signaling leads to significantly greater apoptosis , while both MEK and PI3K inhibition downstream of PDGF signaling leads to cell death . Data plotted as mean ± SEM , with cell percentages quantified across three fields of view for three biological replicates . CC3: Cleaved Caspase-3 . *p < 0 . 1; **p < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 07186 . 011 To directly assay cell proliferation , we performed BrdU labeling at 4 hr following either FGF or PDGF treatment in E13 . 5 MEPMs; while PDGF induces a modest response compared to serum-starved cells , we found a significantly greater proliferative response to FGF ( Figure 5C ) . Crystal violet staining for cell viability confirms a greater effect of FGF than PDGF as well as the importance of MEK/ERK activity for this response ( Figure 5—figure supplement 1C ) . Furthermore , the PDGF-dependent effect on cell viability is significantly greater at day 3 compared to the 0 . 1% fetal bovine serum ( FBS ) treated cells , underscoring the role of PDGF in cell survival/growth in the absence of other growth factors . We next measured apoptosis following PDGF and FGF treatment in the presence of both inhibitors ( Figure 5—figure supplement 1D ) . Consistent with the cell viability results and phospho-JNK induction patterns , MEK inhibition following FGF stimulation results in a greater increase in apoptosis than inhibition of FGF-mediated PI3K signaling; in contrast , PI3K inhibition has a greater effect than MEK inhibition downstream of PDGF stimulation , although inhibiting either pathway results in increased apoptosis . Finally , we tested cell differentiation by alkaline phosphatase ( AP ) staining , a marker of osteoblast differentiation ( Wu et al . , 2008 ) . PDGF-treated MEPMs display a robust AP response , while FGF-treated cells show a striking lack of AP positive cells ( Figure 5D ) . In addition , MEK/ERK inhibition increases AP staining , but PI3K inhibition represses osteoblast differentiation ( Figure 5D ) . These experiments suggest the following model: FGF drives cell proliferation and represses cell differentiation in a MEK/ERK-dependent manner , while PDGF facilitates cell differentiation , at the expense of reduced proliferation , in a PI3K-dependent manner . Finally , we sought to investigate the FGF-ERK-proliferation and PDGF-PI3K-differentiation axes in vivo . We first examined the expression pattern of Fgfr1 in relation to Dusp6 ( ERK-specific ) , Dusp1 ( JNK-specific ) , and Dusp10 ( JNK-specific ) in the E13 . 5 palate by in situ hybridization ( Figure 6—figure supplement 1A ) . Although Dusp1 and Dusp10 are primarily epithelial , Fgfr1 and Dusp6 are co-expressed in the anterior palatal mesenchyme , consistent with previous work implicating ERK in proliferation within this region ( Bush and Soriano , 2010 ) . Similarly , immunohistochemistry ( IHC ) revealed pERK is scattered in the anterior palatal mesenchyme , with some epithelial staining ( Figure 6—figure supplement 1B ) . However , cell proliferation along the anterior–posterior axis is relatively uniform in the E13 . 5 palate ( Bush and Jiang , 2012 ) , complicating assignment of the observed expression patterns to a spatially restricted proliferation program . We next explored the relationship between PI3K signaling and osteoblast differentiation . Whole mount IHC for pAkt at E13 . 5 demonstrates expression restricted to the developing upper lip and middle to posterior palate , overlapping with AP in these structures ( Figure 6A ) . We further assayed the pattern of Runx2 expression ( Figure 6B ) in comparison to Pdgfra , Id1 , and Id3 in the E13 . 5 palate ( Figure 6B' ) , finding shared domains of expression and exclusion in the palate . Taken together , this correlation between PI3K activity , the Id genes , and regions of osteoblast differentiation supports the existence of a spatially restricted PDGF-PI3K-differentiation axis . 10 . 7554/eLife . 07186 . 012Figure 6 . In vivo correlation and perturbation of the RTK-mediated differentiation program during mouse craniofacial development . ( A ) At E13 . 5 , pAkt and AP domains co-localize in the middle to posterior secondary palate ( red arrowhead ) as well as in the developing upper lip ( open arrowhead ) . ( B ) Domain of Runx2 ( osteoblast marker ) expression overlaps with ( B’ ) Pdgfra , Id1 , and Id3 expression in the middle to posterior palate ( red arrowhead ) , with expression excluded along the midline ( black arrowhead ) . ( C ) Frontal sections from neural crest conditional Fgfr1 mutants ( Wnt1-Cre; Fgfr1fl/fl ) exhibit increased AP staining in the developing midface at E14 . 5 , supporting an in vivo role for FGF-mediated repression of osteoblast differentiation ( red arrowhead ) ( n = 3 ) . ( D ) FGF and PDGF signaling use different signaling pathways to instruct divergent cellular outcomes . FGF drives cell proliferation and represses cell differentiation in an ERK-dependent manner , consistent with a greater percentage of the FGF target genes being MEK/ERK dependent ( 50% ) than PI3K dependent ( 20% ) . In contrast , PDGF promotes cell differentiation in a PI3K-dependent manner , and PDGF target genes show greater PI3K dependence ( 30% ) than MEK/ERK dependence ( 10% ) . Furthermore , inhibition of PI3K signaling leads to an FGF specific induction of pERK ( green ) and consequently increased transcription of ERK targets such as Fos , Fosb , and Junb . On the other hand , MEK/ERK inhibition leads to pJNK induction ( orange ) and transcription of Jun , indicating multiple crosstalk mechanisms across different intracellular pathways in response to RTK activation . PP: primary palate; PS: palatal shelf; T: tongue; UL: upper lip . DOI: http://dx . doi . org/10 . 7554/eLife . 07186 . 01210 . 7554/eLife . 07186 . 013Figure 6—figure supplement 1 . Patterns of gene expression and pERK activity in the E13 . 5 palate . ( A ) In situ hybridization for Fgfr1 , Dusp6 , Dusp1 , and Dusp10 in the E13 . 5 palate indicates no obvious spatial correlation between receptor expression and putative domains of ERK and JNK activity . Fgfr1 and Dusp6 show enrichment in the anterior palatal mesenchyme ( red arrowhead ) and epithelium ( black arrowhead ) , while Dusp1 and Dusp10 expression are primarily epithelial ( black arrowhead ) . ( B ) Immunohistochemistry for pERK reveals broad expression in both the E13 . 5 anterior palatal mesenchyme ( red arrowhead ) and epithelium ( black arrowhead ) . E: epithelium; M: mesenchyme; PS: palatal shelf; T: tongue . DOI: http://dx . doi . org/10 . 7554/eLife . 07186 . 013 In MEPMs , FGF drives proliferation and represses differentiation . One prediction from this observation is that Fgfr1 mutants would have decreased repression of this program , and consequently , increased differentiation in the midface . We , therefore , performed AP staining on neural crest conditional Fgfr1 mutants ( Wnt1-Cre; Fgfr1fl/fl ) at E14 . 5 to investigate defects in osteoblast differentiation . We observed an increased domain of AP in the maxillary region of Fgfr1 conditional mutants compared to heterozygous controls ( Figure 6C ) . Collectively , these results suggest the FGF repressed , PI3K-dependent differentiation program identified in MEPMs is conserved in vivo .
Our studies show that PDGF and FGF signaling in MEPMs regulate different gene expression programs and phenotypic outputs , with PDGF mainly promoting cell differentiation through PI3K and FGF favoring cell proliferation through ERK . Although the initial wave of gene expression shows high overlap , FGF elicits a quantitatively stronger response in terms of both signal magnitude and duration , which is reinforced by a positive signal from the delayed transcriptional wave in response to FGF but not PDGF . Furthermore , FGF-responsive genes are predominantly ERK dependent , while PDGF targets exhibit greater PI3K dependence genome-wide , relationships mimicked at the level of cellular outcome . Finally , we observed correlation between PDGF-PI3K signaling and osteoblast differentiation at E13 . 5 as well as increased AP staining in Fgfr1 mutants , indicating the differentiation circuit repressed by FGF signaling in MEPMs is functional in vivo . The architecture of the transcriptional response to RTK signaling has been well described as three stereotypic waves: an IEG response , a delayed wave providing feedback control , and a late wave determining cellular outcome ( Amit et al . , 2007; Avraham and Yarden , 2011 ) . In MEPMs , the magnitude and duration of the IEG wave is much stronger in response to FGF compared to PDGF , suggesting one level of specificity may be achieved through quantitative differences in IEG induction . Indeed , transient and sustained pERK induction , resulting in distinct magnitudes of Fos mRNA expression , drive binary responses in c-Fos abundance and activity at the protein level ( Murphy et al . , 2002; Nakakuki et al . , 2010 ) , delineating one mechanism through which quantitative differences lead to a thresholded ‘ON-OFF’ response in downstream transcriptional activity . Thus , in addition to the observed PDGF-PI3K and FGF-ERK relationships , distinct patterns of IEG transcription factor expression may also contribute to the divergent gene expression profiles at later time points . While inhibition of the delayed transcriptional response prolongs the PDGF-pERK wave , the opposite effect is observed on FGF-pERK induction . This result is consistent with the diversity of feedback and feedforward regulation on the RTK response ( Avraham and Yarden , 2011; Volinsky and Kholodenko , 2013 ) . In addition to transcriptional feedback , many other mechanisms , including ligand identity and receptor endocytosis ( Francavilla et al . , 2013 ) , contribute toward specifying the signaling response to RTK activation . The complexity of this regulation underscores the intricate balance between positive and negative control systems , and it will be critical to determine how these regulatory mechanisms interact to produce the desired developmental outcomes . Given the pleiotropic roles of RTKs , a central question involves how a common set of signal transduction modules specifies distinct outcomes . Our study supports a model in which differential intracellular pathway activation is responsible for the distinct transcriptional responses and cellular processes mediated by RTK signaling in MEPMs ( Figure 6D ) . Pathway activation downstream of a single RTK can be affected by many parameters , such as receptor expression level ( Traverse et al . , 1994; Tallquist et al . , 2003 ) , suggesting that even a single receptor can regulate multiple downstream outputs and transcriptional programs across different contexts . The role of quantitative differences in the pERK response leading to distinct cellular outcomes is well characterized in PC12 neurons ( Marshall , 1995 ) , and the transcriptional response to the proliferative program mediated by sustained pERK activation in 3T3 fibroblasts has been reported ( Yamamoto et al . , 2006 ) . Building on these studies , the MEPM RNA-seq data provide insight into the differences in gene expression following both transient and sustained pERK induction as well as qualitatively specific target genes downstream of MEK/ERK and PI3K signaling , with the added advantage of profiling these responses within a system requiring these pathways for normal development . It is important to note that mRNA levels alone ( as measured by RNA-seq ) are not the only level of specificity in the transcriptional response , for example , differential cofactor recruitment can also specify distinct gene expression programs in response to FGF and PDGF signaling ( Vasudevan and Soriano , 2014 ) . Further work is necessary to identify the precise mechanisms regulating the activity of individual RTK target genes . Although we focused on MEK/ERK and PI3K signaling based on their reported roles downstream of PDGF and FGF during development ( Klinghoffer et al . , 2002; Corson et al . , 2003 ) , many other intracellular pathways are activated by RTKs , such as JNK , p38 , Src , PLCγ , and PKC ( Lemmon and Schlessinger , 2010 ) . As previously reported in other contexts ( Bhalla et al . , 2002; Santos et al . , 2007 ) , we found the FGF-pERK induction was dependent on PKC activity , and we further identified crosstalk in the presence of pathway inhibition between ERK , PI3K , and JNK . In addition , a subset of genes ( 34% FGF , 43% PDGF ) are not affected by MEK/ERK or PI3K inhibition , suggesting they either lie downstream of other intracellular pathways or require inhibition of both ERK and PI3K . Thus , while our studies demonstrate PI3K is necessary for differentiation and ERK is necessary for proliferation in MEPMs , the high connectivity across intracellular pathways suggests no single pathway in isolation is sufficient to drive transcriptional responses and cellular outcomes in their entirety . Rather , integrated output from multiple effector pathways likely contributes to the ultimate cellular outcome , and instead of the existence of linear PDGF-PI3K-differentiation and FGF-ERK-proliferation axes , we favor a model in which multiple signaling events converge on PI3K to promote differentiation and ERK to drive proliferation in the midface . The finding that effector inhibition increased the induction of many RTK target genes was surprising given their presumed primary function as positive effectors of signaling . The ‘superinduction’ of these transcriptional responses was due at least in part to compensatory activation of other pathways: inhibition of pERK downstream of both RTKs resulted in increased pJNK induction while inhibition of PI3K downstream of FGF , but not PDGF , resulted in increased pERK activation . This observation has implications for both genetic studies in mice and any system in which effector inhibition is used as an experimental or therapeutic agent . First , allelic series experiments in which adaptor binding is abrogated in order to abolish specific effector cascades may result in induction of alternate signaling pathways , complicating assignment of developmental function to a single intracellular pathway . Indeed , mice harboring a PI3K binding site mutation at the PDGFRα locus show increased SHP2 binding and altered pERK activation ( Klinghoffer et al . , 2002 ) . Second , chemical inhibition of these effectors may prime activation of other pathways , facilitating alternate signaling mechanisms and cellular outcomes . Many examples of such crosstalk between ERK and PI3K ( Mendoza et al . , 2011; Sun and Bernards , 2014 ) as well as ERK and JNK ( Lopez-Bergami et al . , 2007 ) have been described downstream of RTK signaling in cancer . The MEPM RNA-seq data provide a transcriptional signature for this crosstalk , offering an additional readout to measure activation of these pathways . However , it is important to caution that these responses likely vary based on cellular context , and although the core target genes and effector dependencies may be conserved , extrapolating this framework to other systems requires careful validation . In comparing our gene expression studies to mouse craniofacial development in vivo , we found co-localization of pAkt and AP in the upper lip and middle to posterior palate at E13 . 5 as well as overlapping domains of Runx2 and Id gene expression , indicating the PDGF-PI3K axis identified in MEPMs is correlated with osteoblast differentiation in the midface . It is interesting to note that the distribution of Pdgfra mRNA expression compared to pAkt activity and AP staining is not strictly one-to-one; we speculate other factors such as ligand distribution , availability of intracellular signaling proteins , and input from other pathways may contribute toward the spatially restricted domains of osteoblast differentiation . We further observed elevated AP expression in Fgfr1 conditional mutants at E14 . 5 , consistent with FGF-mediated repression of osteoblast differentiation . Many functions have been ascribed to FGF signaling in skeletogenesis ( Ornitz and Marie , 2002 ) , and a combination of parameters is likely responsible for the multiplicity of observed roles , such as ligand identity ( Francavilla et al . , 2013 ) and cellular context . During palatogenesis , Fgfr1 mutants have been previously reported to exhibit proliferation defects and increased BMP ( Bone Morphogenetic Protein ) signaling ( Wang et al . , 2013 ) , supporting the notion that FGF drives proliferation and represses differentiation in this system . In addition , the facial clefting phenotypes associated with neural crest conditional loss of RAF , MEK , or ERK ( Newbern et al . , 2008 ) suggest that the FGF-ERK axis is functionally relevant in vivo . One important point merits further discussion: Are PI3K and ERK induction interpreted the same independently of the stimulus driving effector activity ? There is evidence to suggest this is indeed the case , as IGF ( Insulin-like growth factor 1 ) -mediated PI3K/Akt activation is a key regulator of osteoblast differentiation in mesenchymal stem cells ( Xian et al . , 2012 ) , consistent with the PI3K-mediated differentiation outcome in MEPMs . However , this interpretation is likely restricted in large part by cellular context , as MEK/ERK signaling functions as a positive regulator of differentiation in embryonic stem cells ( Ying et al . , 2008 ) , in contrast to the role of MEK/ERK signaling in MEPMs . In skeletal differentiation , our findings of a MEK/ERK-mediated proliferation program and PI3K-mediated differentiation program are in line with previous work analyzing the role of these effectors ( Mansukhani et al . , 2005; Raucci et al . , 2008; Miraoui and Marie , 2010 ) . In addition , a FGF-PKC-AP-1 signaling axis in calvarial osteoblasts has been reported ( Miraoui et al . , 2010 ) , consistent with the MEPM data linking FGF to a robust AP-1 response and FGF-mediated pERK activation to PKC signaling . Delineating the hierarchy between these signals and pathways and determining the exact combinations sufficient to specify particular outcomes will be key questions for future studies . The present work provides a roadmap of the gene expression profiles underlying these cellular behaviors and a transcriptomic view of how two different RTKs lead to distinct outcomes .
All animal experiments were approved by the Institutional Animal Care and Use Committee at the Icahn School of Medicine at Mount Sinai . Wild-type C57B/6 mice were used to generate E13 . 5 MEPMs for RNA-seq . Pdgfratm11 ( EGFP ) Sor ( Hamilton et al . , 2003 ) , referred to as Pdgfra-GFP in the text , were maintained on a C57BL/6 background , and FGFR1-CFP mice ( to be described elsewhere ) , FGFR1tm5 . 1Sor mice ( Hoch and Soriano , 2006 ) , referred to as Fgfr1fl/fl in the text , and Tg ( Wnt1-Cre ) 11Rth mice ( Danielian et al . , 1998 ) , referred to as Wnt1-Cre in the text , were all maintained on a 129S4 background . Primary MEPM cells were isolated from E13 . 5 secondary palatal shelves ( day of plug: E0 . 5 ) as previously described ( Fantauzzo and Soriano , 2014; Vasudevan and Soriano , 2014 ) . Cells were grown in Dulbecco's modified Eagle’s medium ( GIBCO; Invitrogen , Carlsbad , CA ) with 10% fetal bovine serum ( FBS; HyClone Laboratories , Logan , UT ) , 50 U/mL penicillin ( GIBCO ) , 50 μg/mL streptomycin ( GIBCO ) , and 2 mM L-glutamine ( GIBCO ) . Cells were split twice to passage 2 for all experiments included in this study , and serum starvation was conducted in same media as above but with 0 . 1% FBS instead of 10% FBS . BrdU assays ( Bush and Soriano , 2010; Vasudevan and Soriano , 2014 ) and AP staining ( Wu et al . , 2008 ) were performed as described previously . Passage 2 E13 . 5 MEPMs were serum starved overnight in 0 . 1% FBS and then treated with either 30 ng/mL PDGFAA ( R&D Systems , Minneapolis , MN ) or 50 ng/mL FGF1 ( Peprotech , Rocky Hill , NJ ) + 1 μg/mL heparin ( Sigma-Aldrich , St . Louis , MO ) for the desired duration . When pathway inhibitors were used , cells were pretreated with either 1 μM PD325901 or 10 μM LY294002 for 30 min prior to growth factor addition . MEPMs generated from independent litters were used for each set of replicates , and RNA was collected using RNeasy Mini Kit ( Qiagen Inc . , Valencia , CA ) according to manufacturer's protocol before submission to the Mount Sinai Genomics Core ( http://icahn . mssm . edu/departments-and-institutes/genomics/about/resources/genomics-core-facility ) , where RNA was poly-A selected , libraries generated , and sequenced on Illumina HiSeq 2000 . Between 25 and 40 million reads per sample were obtained and mapped to the mouse genome ( version mm10 ) using TopHat ( Kim et al . , 2013 ) . Genes were tested for differential expression by Cuffdiff and considered significant at q <0 . 1 ( Trapnell et al . , 2010 ) . Data for untreated , 1-hr PDGF , and 1-hr FGF treated MEPMs are publicly available at the Gene Expression Omnibus ( GEO ) , Accession GSE61755 ( 42 ) . Data for 4 hr PDGF , 4 hr FGF , 1 hr PDGF + PD325901 , 1 hr PDGF + LY294002 , 4 hr PDGF + PD325901 , 4 hr PDGF + LY294002 , 1 hr FGF + PD325901 , 1 hr FGF + LY294002 , 4 hr FGF + PD325901 , and 4 hr FGF + LY294002 samples are publicly available at the GEO , Accession GSE66484 . E13 . 5 MEPMs were serum starved overnight in 0 . 1% FBS and then treated with desired growth factors and/or inhibitors as for RNA-seq . Total RNA was collected with the RNeasy Mini Kit ( Qiagen Inc . ) . First-strand cDNA was then synthesized using a ratio of 2:1 random primers: Oligo ( dT ) with SuperScript II RT ( Invitrogen ) . qPCR was performed using a Bio-Rad iQ5 Multicolor Real-Time PCR Detection System and analyzed with iQ5 Optical System Software ( version 2 . 0; Bio-Rad , Hercules , CA ) . Reactions were performed with PerfeCTa SYBR Green FastMix for iQ ( Quanta Biosciences Inc . , Gaithersburg , MD ) using 10 μM primers ( Integrated DNA Technologies Inc . , Coralville , IA ) . A list of qPCR primers used is available in Supplementary File 5 . The following cycling protocol was used: step 1: 95°C for 3 min; step 2: 95°C for 10 s; step 3: 60° for 40 s; repeat to step 2 39× ( total of 40 cycles ) , and a melting curve analysis was performed . In addition , PCR products were run on a 1 . 0% agarose gel to ensure correct amplicon size . β2m was used as an endogenous control . MEPMs were serum starved overnight and treated as for RNA-seq and then washed 3× in ice-cold PBS ( Phosphate-buffered saline ) before being harvested in NP-40 lysis buffer ( 20 mM Tris–HCl pH 8 , 150 mM NaCl , 10%glycerol , 1% Nonidet P-40 , 2 mM EDTA ( Ethylenediaminetetraacetic acid ) , 1× complete Mini protease inhibitor cocktail [Roche Applied Science , Indianapolis , IN] , 1 mM PMSF ( Phenylmethanesulfonylfluoride ) , 10 mM NaF , 1 mM Na3VO4 , and 25 mM β-glycerophosphate ) . Total cell lysates were sonicated briefly and then collected by centrifugation . Lysates were then resuspended in Laemmli buffer containing 10% β-mercaptoethanol , heated at 95°C for 5 min , and separated by SDS-polyacrylamide gel electrophoresis . The following inhibitors were used: LY294002 ( Sigma-Aldrich ) , PD325901 ( Stemgent , Cambridge , MA ) , cycloheximide ( Fisher Scientific , Waltham , MA ) , and Bim I ( Santa Cruz Biotechnology , Dallas , TX ) . The following antibodies were used: anti-phospho MAPK p42/p44 ( 9201; Cell Signaling Technologies , Danvers , MA; 1:1000 ) , anti-pAkt ( 9271; Cell Signaling Technologies; 1:1000 ) , and anti-pJNK ( 4671; Cell Signaling Technologies; 1:1000 ) . The anti-β tubulin E7 antibody ( 1:1000 ) developed by M . Klymkowsky was 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 . Immunofluorescence was performed as described previously ( Vasudevan and Soriano , 2014 ) . Briefly , cells were fixed in 4% formaldehyde at room temperature , blocked in 10% donkey serum , stained with primary antibody , and detected with secondary antibody , all for 1 hr at room temperature . The following antibodies were used: Anti-Cleaved Caspase-3 ( CST9661; Cell Signaling Technology; 1:400 ) . The anti-BrdU ( G3G4 , 1:500 ) developed by S . Kaufman was obtained from the Developmental Studies Hybridoma Bank , created by the NICHD of the NIH and maintained at The University of Iowa , Department of Biology , Iowa City , IA . IHC was performed as described previously ( Fantauzzo and Soriano , 2014 ) . For whole mounts , embryos were dissected onto ice-cold PBS , fixed overnight in 4:1 methanol:DMSO , cleared in 4:1:1 methanol:DMSO:H2O2 , and stored in 100% methanol . For sections , embryos were fixed overnight in 4% paraformaldehyde ( PFA ) and embedded in paraffin . Staining was done with primary antibody overnight and a goat anti-rabbit IgG peroxidase-conjugated secondary antibody ( Jackson ImmunoResearch Laboratories Inc . , West Grove , PA ) . Detection was performed using the Vector Laboratories SK-4100 kit ( Vector Laboratories Inc . , Burlingame , CA ) . Embryos were dissected in ice-cold PBS and fixed overnight in 4% PFA and embedded in paraffin or optimal cutting temperature compound for sectioning . In situ hybridization was performed according to standard protocols ( He and Soriano , 2013; Vasudevan and Soriano , 2014 ) . A list of probe sequences is provided in Supplementary File 5 . Generation of heatmaps , PCA , and other analysis were performed through custom code in R . PCA was done using the ‘prcomp’ function on centered , median normalized data , and correlation matrix was constructed using the ‘cor’ function on log2 ( FPKM + 1 ) transformed data . Hierarchical clustering and heatmaps were generated using the ‘heatmap . 2’ and ‘hclust’ functions . Pearson's correlation was used in the correlation matrix , and Euclidean distance was used as a distance metric for hierarchical clustering . PPI networks were constructed using the Expression2Kinase software ( Chen et al . , 2012 ) based on an updated version of Genes2Networks ( Berger et al . , 2007 ) , only direct connections ( path length = 1 ) were considered , and all published PPI databases except predicted PPIs were included . Networks were visualized and formatted in yEd ( www . yworks . com ) . For Gene ontology analysis ( Huang et al . , 2009 ) , only those terms at a significance threshold of p <0 . 001 were included . Redundant GO terms comprising an identical or fully shared subset of genes were removed . A full list of GO results is provided in Supplementary File 4 . | Cells produce many different proteins that play a variety of important roles . For example , proteins called receptor tyrosine kinases can detect particular molecules and send signals to other parts of the cell to regulate the activity ( or “expression” ) of genes involved in cell division , movement , and other processes . Humans have 58 receptor tyrosine kinases , and defects in these proteins have been linked to diseases such as cancer and diabetes . However , many different receptors regulate the activities of shared sets of genes , so it is not clear how an individual receptor can specifically control the genes involved in a particular process . Two receptor tyrosine kinases called PDGFR and FGFR are crucial for the development of the face , palate , and head in humans and other animals . Vasudevan et al . used a technique called RNA-sequencing to find out which genes are regulated by these receptors in mouse palate cells . The experiments show that there is a common set of genes whose activities change quickly—within 1 hour—in response to the activation of either PDGFR or FGFR . However , several hours later , cells in which PDGFR is activated have different patterns of gene expression compared to those with active FGFR . Vasudevan et al . also found that FGFR promotes cell division , while PDGFR promotes the changing of palate cells into different types with more specialized roles . These different outcomes arise because PDGFR and FGFR use different signaling pathways that involve distinct proteins . For example , a protein called PI3K is critical for changes in gene expression in response to PDGFR but not FGFR . These results suggest that PGDRF and FGFR control different cellular processes in the palate by sending distinct signals into the cell . Understanding the receptor tyrosine kinases and the networks of genes they activate will help us to identify the signals that are important for other processes , such as the development of the face . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"computational",
"and",
"systems",
"biology"
] | 2015 | Receptor tyrosine kinases modulate distinct transcriptional programs by differential usage of intracellular pathways |
Men and women may use alcohol to regulate emotions differently , with corresponding differences in neural responses . We explored how the viewing of different types of emotionally salient stimuli impacted brain activity observed through functional magnetic resonance imaging ( fMRI ) from 42 long-term abstinent alcoholic ( 25 women ) and 46 nonalcoholic ( 24 women ) participants . Analyses revealed blunted brain responsivity in alcoholic compared to nonalcoholic groups , as well as gender differences in those activation patterns . Brain activation in alcoholic men ( ALCM ) was significantly lower than in nonalcoholic men ( NCM ) in regions including rostral middle and superior frontal cortex , precentral gyrus , and inferior parietal cortex , whereas activation was higher in alcoholic women ( ALCW ) than in nonalcoholic women ( NCW ) in superior frontal and supramarginal cortical regions . The reduced brain reactivity of ALCM , and increases for ALCW , highlighted divergent brain regions and gender effects , suggesting possible differences in the underlying basis for development of alcohol use disorders .
Impaired affect regulation is a primary motive for the use of drugs , including alcohol ( Prescott et al . , 2004; Vaughan et al . , 2012 ) . Affective processing deficits have been linked to misinterpretation of environmental cues , irregularity in mood , and increased alcohol consumption and may be an underlying factor leading to the development and maintenance of alcohol use disorders ( AUD ) ( Gilman and Hommer , 2008; Thorberg et al . , 2009 ) . However , problem drinkers are a heterogeneous population . While alcohol and other GABAergic agents such as benzodiazepines typically are considered to be depressants because of their ability to decrease anxiety , tension , and inhibition , they also can function as a stimulant , generating feelings of euphoria and well-being ( Gilman et al . , 2008; Mukherjee et al . , 2008 ) . These effects can be experienced both by men and by women , but the appeal of alcohol for each gender subgroup of problem drinkers may be driven for contrasting reasons ( Buchmann et al . , 2010 ) . For example , on average , women might drink to decrease negative affect , and men might drink to enhance favorable emotional states ( Buchmann et al . , 2010; Buckner et al . , 2006; Crutzen et al . , 2013; Oscar-Berman et al . , 2014; Ruiz and Oscar-Berman , 2015 ) . Research unrelated to AUD has indicated that men and women process emotions differently ( Mareckova et al . , 2016; Proverbio et al . , 2009 ) , and there are differences between men and women in personality disorders and social impairments ( Becker et al . , 2017; Nixon et al . , 2014; Oscar-Berman et al . , 2009; Oscar-Berman et al . , 2014; Ruiz and Oscar-Berman , 2013 ) . Women also have been found to display different psychophysiological reactions to emotional stimuli ( Sawyer et al . , 2015 ) and to be more emotionally expressive than men ( Kring and Gordon , 1998 ) . Conversely , men on average have an increased tendency to repress emotional responses ( Birditt and Fingerman , 2003 ) . Additionally , alcoholic women ( ALCW ) are two to three times more likely to be diagnosed with anxiety and affective disorders than alcoholic men ( ALCM ) , while ALCM are twice as likely as ALCW to have antisocial personality disorders ( Merikangas et al . , 1996; Oscar-Berman et al . , 2009 ) . The presence of gender-specific deficits in emotional regulation may provide insight into what differentially motivates men and women to abuse alcohol ( Erol and Karpyak , 2015; Mosher Ruiz et al . , 2017; Regier et al . , 1990; Ruiz and Oscar-Berman , 2015; Valmas et al . , 2014 ) . Emotional processing is associated with activity within well-characterized network-based brain circuitries including prefrontal cortex , insula , cingulate cortex , and medial temporal lobe structures including the amygdala ( Davidson et al . , 1999; Proverbio et al . , 2009 ) . In functional magnetic resonance imaging ( fMRI ) studies measuring AUD-related abnormal brain responses during emotional processing ( Beck et al . , 2009; Chanraud-Guillermo et al . , 2009; Gilman and Hommer , 2008; Heinz et al . , 2007 ) , abstinent ALC individuals showed reduced fMRI activation in the amygdala , hippocampus , anterior cingulate , and medial frontal regions in response to viewing stimuli with a negative affective valence , compared to nonalcoholic control ( NC ) participants ( Marinkovic et al . , 2009; Padula et al . , 2015; Salloum et al . , 2007 ) ; in response to viewing stimuli with a positive affective valence , the ALC individuals showed an increase in activation in the anterior cingulate cortex , prefrontal cortex , ventral striatum , and thalamus ( Heinz et al . , 2007 ) . However , little is known about gender-specific persistent influences of alcoholism-related brain activation in response to affective materials , because little research has compared abstinent alcoholic men ( ALCM ) and women ( ALCW ) compared to nonalcoholic men ( NCM ) and women ( NCW ) ( e . g . , Salloum et al . , 2007 ) . Therefore , using fMRI in conjunction with measures of affective judgments , an important aim of the present exploratory study was to address the need for more research in this domain by examining gender differences in the processing of high-arousal emotional stimuli on brain and behavioral responses in ALC men and women compared to NC men and women . The study was designed within a conceptual model of emotional processing adapted from Halgren and Marinković ( 1995 ) . According to this model , when an emotionally salient stimulus is perceived , Emotional Event Integration and Evaluation takes place , and a response occurs in widespread and focal dynamic corticolimbic neural networks ( Figure 1—figure supplement 1 ) . These circuitries embody different functional systems that amalgamate cognitive with feeling aspects of emotions: ( 1 ) Attention and orientation to a salient stimulus occurs in insular , anterior cingulate , prefrontal , and posterior parietal cortices . ( 2 ) Emotional event appraisal , integration , and evaluation ( as influenced by the ongoing emotional context and the perceiver’s personality ) , takes place in posterior cingulate , orbital and medial prefrontal cortex , and other neocortical sites ( e . g . , fusiform gyrus and superior temporal sulcus ) , and limbic structures ( e . g . , hippocampus and amygdala ) . ( 3 ) Volition and decisions , which determine response choice , are generated in cingulate , precentral , premotor , and supplementary cortices . Using the above model as a guide , we analyzed brain activation and behavioral responses within a psychological task structure aimed to assess the subjective appraisal of valence of specific emotional categories . We chose to do this in order to disentangle how brain activity during the process of evaluation and interpretation of emotional content distinguished ALC from NC groups . To engage the Emotional Integration and Evaluation System , we asked participants to view complex , emotionally meaningful pictures ( aversive , erotic , gruesome , happy – and neutral for comparison ) , and to rate how the pictures made them feel ( good , bad , or neutral ) . We chose stimuli representing the contrasting valences , because findings from previous research indicated that one or more of those emotional categories were sensitive to deficits in emotional processing by abstinent ALC groups compared to NC groups ( Heinz et al . , 2007; Marinkovic et al . , 2009; Padula et al . , 2015; Salloum et al . , 2007 ) . It should be noted that the behavioral task is not , explicitly , either an emotion judgment task nor an emotion regulation measure . Instead , we expected the data to reflect neural responsivity as indirect measures of emotion processing and/or emotion regulation to a variety of emotionally salient stimuli . We were especially interested in responses to happy and aversive stimuli , because ( a ) they have been shown to be sensitive to gender effects in brain activation levels in abstinent ALC participants who viewed faces with different emotional expressions ( Padula et al . , 2015 ) , and ( b ) abnormalities in the evaluation of aversive stimuli ( which are associated with negative feelings such as fear , pain , and stress ) , play a crucial role in the transition to AUD or alcohol relapse ( Maleki and Oscar-Berman , 2019; Witkiewitz et al . , 2015 ) . Whereas brain activation alterations in emotional processes have been studied in relation to AUD ( Beck et al . , 2009; Chanraud-Guillermo et al . , 2009; Gilman and Hommer , 2008; Heinz et al . , 2007 ) , gender differences have not been explored in depth , and there is a need for more research in this domain ( Nixon et al . , 2014; Ruiz and Oscar-Berman , 2013 ) . Therefore , in accordance with the primary aim of the present exploratory study , we sought to determine how gender differences are manifested in the brain networks outlined by our conceptual model ( Event Integration and Evaluation ) . We hypothesized that AUD-related abnormalities in emotional evaluation ( i . e . , ratings and reaction times ) would differ by gender , and these processes would be reflected by gender differences in brain activity during emotional evaluation . Overall , we expected that the same brain regions as in the well-characterized system involved in emotion processing , as described above , would be involved in emotional processes; however , they would not be impacted in the same way for men and women . We hypothesized that ALCM would show dampened corticolimbic activation to stimuli from most of the emotional valence categories , thereby reflecting muted affect . For women , we postulated that the pattern of abnormalities associated with AUD would differ from that of men , by showing increased activation to emotional stimuli , indicative of hyper-sensitivity to affective input .
Demographics , alcoholism indices , neuropsychological and clinical assessment scores of the 88 participants are presented in Figure 2 ( and Appendix 1—tables 1 and 2 ) . Although the Hamilton Rating Scale for Depression ( HRSD; Hamilton , 1960 ) scores for the ALC men and women were higher than for the NC men and women ( p<0 . 01 ) , both groups’ scores were very low ( mean 3 . 6 vs . 1 . 1 ) : HRSD scores of 8 , 16 , and 25 or above indicate mild , moderate , or severe depression , respectively ( Zimmerman et al . , 2013 ) . The average number of daily drinks ( DD ) was significantly higher in ALCM compared to ALCW ( p<0 . 05 ) . The alcoholic participants were abstinent for extended lengths , on average for seven years . The ALCW and NCW had higher delayed memory scores than the ALCM and NCM ( Wechsler Memory Scale Delayed ( General ) Memory Index , p<0 . 01 ) . Of the 88 participants included in fMRI analyses , 12 were excluded from the analysis of behavioral ratings because of technical problems or incomplete data , leaving 76 subjects for the final analyses ( 21 ALCW , 15 ALCM , 21 NCW , 19 NCM ) . Overall , participants’ percentage ratings of good , bad , and neutral were generally consistent among ALC and NC men and women ( Figure 3 ) for the various conditions ( aversive , erotic , gruesome , happy , neutral ) . That is , the participants rated erotic pictures as mostly neutral and good; gruesome pictures as almost entirely bad; aversive pictures as bad , with a few neutral; happy pictures as mostly good , with some neutral; and neutral pictures as mostly neutral , with some good ( altogether representing a significant condition x rating interaction , Appendix 1—table 3 ) . While all groups ( ALC and NC men and women ) had a similar pattern , a significant group x condition x rating interaction revealed that the ALC group rated erotic pictures as good less frequently than the NC group . The gender x condition x rating interaction revealed that more men than women rated erotic pictures as good . As with the percentage ratings , evaluation times also were generally comparable for the four groups ( Figure 3—figure supplement 1 and Appendix 1—table 4 ) . There were significant interaction effects of condition x rating , rating x gender , and main effects of condition and rating ( p<0 . 001 for all ) . The evaluation time for gruesome and aversive stimuli were approximately 0 . 5 s longer than other conditions . The evaluation time for bad ratings were similarly shorter for gruesome and aversive stimuli . Women took approximately 0 . 25 s longer ( 14% ) to evaluate the good ratings than men , while the evaluation times for neutral and bad ratings were similar for men and women . Percentage ratings were significantly predicted by the interaction of Profile of Mood States ( POMS ) Depression x group x rating , but no post-hoc comparisons were significant after Bonferroni correction . For evaluation times , the following interactions were significant: VIQ x group x gender , and POMS Depression x group x condition x rating , but post-hoc comparisons were not significant for VIQ . The only significant post-hoc group comparison indicated that for the NC group , POMS Depression scores were positively related to evaluation times for neutral ratings in the happy condition ( 95% confidence interval: [62 , 157] ) , whereas they were not for the ALC group ( 95% confidence interval: [−19 , 40] ) . In other words , the NC participants with higher Depression scores were slower in rating happy stimuli as being neutral . For the ‘caudal middle frontal cluster 1’ and ‘superior frontal cluster’ obtained through analysis of the aversive contrast , percentage ratings were significantly predicted by the interaction of group x gender x rating x contrast effect size . However , post-hoc comparisons of the slopes of contrast effect size for each rating did not identify significant differences between the subgroups . That is , while we identified a different pattern in the relationships of percentage ratings to brain activity among the four subgroups , it was not clear how these relationships differed between the ALCW vs . NCW , and ALCM vs . NCM . The brain activity observed during the neutral condition was subtracted from aversive , erotic , happy , and gruesome conditions , yielding four main comparisons from the study . Overall , the ALC group exhibited lower brain activation values than the NC group for all four contrasts , but significant interactions of group x gender indicated striking differences in these abnormalities . That is , the general observation of lower activation values was evident for ALCM , while ALCW exhibited a different pattern; the values for each emotion vs . neutral contrast were shifted higher for ALCW . Table 1 identifies regions with significant group x gender interactions for each of the four contrasts . Because the pattern of these significant group x gender interactions was similar for all contrasts , we have chosen to exemplify the two most salient contrasts: erotic vs . neutral ( Figure 4 ) and aversive vs . neutral ( Figure 5 ) . A summary figure ( Figure 6 ) shows the group x gender interactions for all four contrasts . The contrast of erotic vs . neutral ( i . e . , erotic minus neutral ) is presented in Figure 4 , which shows that brain activity was greater in most subcortical brain regions for erotic than for neutral images ( for ALCW , ALCM , NCW , and NCM ) . The group x gender interaction revealed a significant cluster that encompassed limbic brain regions including the amygdala , thalamus , hippocampus , and parahippocampal cortex , as well as much of the cerebellum . The erotic and neutral pictures elicited less activation difference for ALCM than for NCM; this alcoholism-related abnormality was not observed for women . A complex pattern of gender-related alcoholism abnormalities in brain activity was revealed by the contrast of aversive vs . neutral conditions for several significant clusters ( Figure 5 ) . For some regions of the brain , activity was higher for aversive than neutral stimuli ( ‘aversive-responding’ regions ) , while for other regions of the brain , activity was higher for neutral than aversive ( ‘neutral-responding’ regions ) . The ALCM - NCM comparison resulted in negative values for both aversive-responding and neutral-responding regions , reflecting the following two situations: For aversive-responding regions , the aversive and neutral stimuli had less activation difference for the ALCM than for the NCM , while for neutral-responding regions , the aversive and neutral stimuli were more similar for NCM than for the ALCM . In four significant clusters , these negative values obtained from ALCM were significantly more negative than those obtained from ALCW . As shown in Figure 5 , three of the clusters were in left prefrontal cortex and one was in the inferior parietal gyrus; similar differences were found for the right hemisphere ( Table 1 ) . Interestingly , as can be seen in Table 1 , there also was a significant main effect in two adjoining medial prefrontal regions ( medial orbitofrontal and rostral anterior cingulate cortices ) , wherein alcoholics exhibited higher contrast than controls , and this was more evident in the men than in the women ( Figure 5—figure supplement 1 and Figure 5—figure supplement 3 ) . For men , this group difference was in the opposite direction as observed for the regions with significant group x gender interactions . In summary , we observed a similar pattern of significant group x gender results ( Figure 6 ) for each of the four contrasts ( aversive , erotic , gruesome , and happy — compared to neutral ) : ALCM demonstrated less activation for emotional stimuli compared with neutral images , whereas ALCW did not show these decreases and in some contrasts , demonstrated activation increases . For comparison with the observations revealed by the erotic contrast shown in Figure 4 ( which highlights the amygdala ) and the aversive contrast shown in Figure 5 ( cortical surface ) , Figure 6 shows all four of the contrasts , including gruesome and happy . For ALCW compared to NCW , significantly more positive brain activation contrasts were seen in superior frontal and supramarginal cortical regions . In ALCM as compared to NCM , the contrasts revealed more negative values across widespread areas throughout the brain , including the inferior parietal gyrus , anterior cingulate gyrus , and postcentral gyrus ( Table 1 and Figure 6 ) . Specifically , significant group x gender interactions were observed in the frontal ( superior frontal , rostral and caudal middle frontal ) , parietal ( inferior and superior parietal gyri , and precuneus ) , and occipital ( pericalcarine and cuneus ) lobes , as well as the caudal anterior cingulate , parahippocampal gyrus , and cerebellum . Happy and aversive contrasts were especially evident throughout widespread regions; the erotic contrast revealed a significant interaction for limbic structures and cerebellum; and the gruesome contrast revealed an interaction for the superior temporal sulcus .
Research on the relationship between AUD and emotional dysfunction has shown impairments in self-regulation of emotions , as well as deficits in the perception , identification , evaluation , and understanding of emotions of self and others . However , because little is known about the brain responses to emotional stimuli in ALCW as compared to ALCM , the present study combined fMRI neuroimaging with a sophisticated experimental design and advanced data analysis methods , to explore the relationship between gender and alcoholism in functional activation of brain regions as participants processed emotional stimuli of varying valences ( International Affective Picture System ) . As indicated in Table 1 , with the exception of two ventromedial prefrontal regions , our results showed consistently blunted brain activation responses to emotional stimuli vs . neutral stimuli in the ALC group compared to the NC group for men; this general pattern was not observed for women . Further , a significant interaction between gender and alcoholism indicated that the affective pictures elicited lower activation contrasts in ALCM relative to NCM , abnormalities that were significantly lower and more pervasive than those observed between ALCW and NCW . That is , by comparison , ALCW showed more positive activation contrasts than found for NCW , in regions including the superior frontal and supramarginal cortex . In the ALCM , the significant differences appeared in areas throughout the brain , including the inferior parietal gyrus , anterior cingulate gyrus , and postcentral gyrus . Table 1 ( and Appendix 1—tables 5 , 6 and 7 ) and Figures 4 , 5 and 6 show the extent and spread of the differences among ALCM , NCM , ALCW , and NCW . Emotional processing involves engaging multiple brain regions ( Davidson et al . , 1999 ) . In vivo neuroimaging studies as well as post-mortem pathological studies have shown that cortical loss in the frontal lobes is the most common damage observed both in association with AUD ( Oscar-Berman and Marinkovic , 2003 ) and in individuals having emotional disorders unrelated to AUD ( Bechara et al . , 2000; Young et al . , 2010 ) . Padula et al . ( 2015 ) used fMRI to compare gender effects in affective processing by abstinent alcohol dependent and healthy nonalcoholic individuals . Their stimuli were pictures of individual faces that displayed positive ( happy ) and negative ( sad , fearful ) emotional expressions . Similar to our approach , they examined contrasts in activation provoked by the emotion stimuli vs . the neutral stimuli . Of note , our present results are congruent with those reported by Padula et al . ( 2015 ) , who found significant group x gender interactions in frontal brain activation levels to positive and to negative emotional stimuli . Despite differences in experimental methods , results of both studies are consistent with the notion of gender-specific and alcoholism-related effects in affective processing , with an emphasis on frontal brain involvement . In our exploratory study , the frontal brain regions showing significant interactions between alcoholism and gender were precentral cortex , rostral and caudal middle frontal cortex , superior frontal cortex , and the caudal anterior cingulate cortex , for both happy and aversive stimuli . Previous fMRI studies have suggested that rostral middle frontal cortex may be involved in the implicit or uninstructed generation and perpetuation of emotional states ( Waugh et al . , 2010; Waugh et al . , 2014 ) . Moreover , in two studies ( Aldhafeeri et al . , 2012; Hägele et al . , 2016 ) , the investigators were consistent in their reports of significant increases in prefrontal and amygdala activation levels in response to pleasant and aversive IAPS pictures , respectively ( compared to neutral pictures ) . Given that in our study , ALCM showed lower activation compared to NCM in frontal , parietal , and temporal regions in response to most of the categories of emotional stimuli , our findings might reflect deficits in ALCM in maintaining positive and negative emotions . By comparison , our ALCW showed higher activation than NCW in superior frontal cortex in response to happy stimuli , and higher activation in the supramarginal gyrus to aversive stimuli , suggesting possible compensation for deficiency in maintaining positive and negative emotions . One of the other frontal brain regions that showed a significant gender x alcoholism interaction was the caudal anterior cingulate cortex , a region thought to be involved in appraisal and expression of negative emotion ( Etkin et al . , 2011 ) . However , for the regions anterior to the caudal anterior cingulate , we found a different pattern of group differences . The ALCM group had greater contrast values than the NCM group in the subcallosal regions of medial orbitofrontal cortex and rostral anterior cingulate cortex . The difference in the activation of these regions in ALCM was in the opposite direction to that observed for other regions , where group x gender interactions had been evident . As suggested by our conceptual model of emotional evaluation and integration ( Figure 1—figure supplement 1 ) , these frontal regions are involved in attending to and integrating cognitive and affective responses to external events ( Bush et al . , 2000; Margulies et al . , 2007; Oscar-Berman and Marinković , 2007; Riedel et al . , 2018 ) . Therefore , the increased responsivity in the ALCM group might indicate compensatory involvement in evaluating the emotional pictures ( Oscar-Berman and Marinković , 2007 ) . Additionally , significant interactions between gender and alcoholism were observed in cortical regions involved mainly in visual processing , including the cuneus and precalcarine regions , in response to happy stimuli ( Figure 6 ) . These significant interactions reflect higher contrast values for affective pictures compared to neutral pictures , more so in NCM than ALCM , whereas the effect was reduced for the two groups of women . In NC participants , we confirmed the greater activation in visual cortex while viewing emotional vs . neutral pictures that has been reported in prior studies , with some suggesting stronger responses by men to pleasant pictures and stronger responses by women to unpleasant pictures ( Sabatinelli et al . , 2004 ) . Inferior parietal cortex was another region that showed a significant interaction between gender and alcoholism , driven mainly by the blunted activation in the ALCM compared to the NCM men . Inferior parietal gyrus is involved in the perception of emotions in facial stimuli ( Sarkheil et al . , 2013 ) . Except for neutral pictures , most of the other pictures had a human face in them , and therefore , the interaction and lower activation in ALCM may represent an impairment in processing emotional facial expressions . In fact , previous research has shown that long-term abstinent ALCM showed less activation in temporal limbic areas , when viewing positive or negative emotional faces compared to controls ( Marinkovic et al . , 2009 ) . There also were significant interactions between gender and alcoholism in limbic and subcortical structures: In ALCM , brain activity for erotic and neutral pictures were relatively similar , leading to decreased differential activation , while NCM had stronger activity for erotic than neutral pictures , for parahippocampal cortex , hippocampus , amygdala , other limbic structures , and the cerebellum . This alcoholism-related abnormality was not observed for women: The ALCW had a slightly larger ( although not significant ) positive contrast between erotic and neutral pictures compared to NCW . The results of this exploratory study are to be considered in the context of several limitations . First , our results are based upon cross-sectional data , and as such , it is impossible to determine if chronic alcohol usage caused , or resulted from , the observed dysregulated emotional reactivity , or perhaps a combination of both . Further , these deficits could reflect differences in brain structure that influenced the emotional activity we observed . In that regard , our alcoholic participants were abstinent for extended lengths , on average for seven years , a variable that speaks to the persistent nature of emotion processing deficits in AUD populations . While it remains unclear whether these deficits predate or result from heavy drinking , or whether emotion processing deficits recover over the course of abstinence , a study of accuracy of decoding emotional facial expressions by short- and long-term abstinent alcoholic men and women ( Kornreich et al . , 2001 ) indicated that deficits in decoding accuracy for anger and disgust , and to a lesser degree sadness , continued with long-term abstinence . Nonetheless , the topic of persistence vs . recovery remains a promising direction for future studies . Second , we had limited information about the potentially confounding variable of smoking status , and therefore , it was not included in the analyses . Smoking abstinence has been associated with increased emotional reactivity in response to unpleasant stimuli ( Versace et al . , 2012 ) and interactions with alcoholism ( Durazzo et al . , 2013; Luhar et al . , 2013 ) , and therefore , may have influenced the results of the present study . Third , while there were peak regions of activation differences , these were observed against a background of broad regions identified that were different between each of the emotional conditions and the neutral condition , and the significant group x gender interactions reflected these broad differences in brain activity . We chose not to artificially suppress the display of these widespread effects in our figures by restricting the thresholds . Fourth , the erotic stimuli shown were identical for all participants in order to maintain a consistent experimental paradigm , while at the same time maximizing arousal . To do this , we selected erotic imagery based upon findings from studies measuring arousal levels to erotic stimuli in men and women ( Bradley et al . , 2001; Israel and Strassberg , 2007 ) . In those studies , men’s behavioral and electrophysiological responses to erotic photographs of women were , on average , much stronger than to erotic photographs of men , whereas responses by women to erotic imagery were similar for photographs of men and women . Therefore , of the 48 erotic pictures presented to the participants in our study , 23 were photographs of women , and 25 were photographs of men and women together . However , participants’ sexual orientation was not assessed , and tailoring the photographs to each individual participant might be more effective . Additionally , previous research ( Glöckner-Rist et al . , 2013 ) has suggested that direct measures of drinking motives might be helpful in interpreting our findings of gender differences in AUD . In the present study , we did not collect data to assess those variables . However , in a separate sample of abstinent alcoholic men and women with comparable drinking histories and demographic characteristics ( Mosher Ruiz et al . , 2017 ) , we did assess drinking motives , with Cooper’s DMQ-R scales ( Cooper , 1994 ) . Although Cooper’s scale is limited in scope , we found that the ALC group scored higher than the NC group on all of the drinking-motives scales , but the interactions between alcoholism , motives for drinking , and gender were not significant . Finally , as described in the Methods , the p-value thresholds used in this study in conjunction with the multiple-comparison cluster correction procedures employed have been shown to have higher false-positive rates than those specified ( Eklund et al . , 2016 ) . This lenient threshold is appropriate in the context of an exploratory study , both because it minimizes the chance of false negatives ( Type two error ) , and also because it allows for the size of the gender effects to be highlighted . However , we additionally conducted analyses with a cluster-forming p-value threshold of ( p<0 . 001 ) , which is commonly used for stronger control of false positives ( Type one error ) . The results of those analyses are shown in Appendix 1—table 8 and Figure 5—figure supplements 5 and 6 . Two clusters were identified consistent with the group x gender interaction effects highlighted in this exploratory study: left and right lateral frontal clusters for the contrast of aversive vs . neutral . Despite the above considerations , the findings from the present exploratory study highlight the need for continued research on the overlap between gender differences in processing of emotional stimuli and the development or maintenance of pathological alcohol consumption . While blunted emotional reactivity had been observed previously in alcoholics , earlier studies had focused either exclusively on men or had collapsed data across genders ( Gilman et al . , 2010; Marinkovic et al . , 2009; Salloum et al . , 2007 ) . Therefore , the present study provides additional insights into emotional processing in alcoholism by examining the influence of gender on brain activation . In our previous studies ( Rivas-Grajales et al . , 2018; Sawyer et al . , 2018; Sawyer et al . , 2017; Sawyer et al . , 2016; Seitz et al . , 2017 ) , we had reported gender differences in morphometry of cerebral and cerebellar subregions , and white matter integrity , in association with alcoholism history in men and women . In the current study , we reported functional abnormalities in cortical , subcortical , and cerebellar regions involved in emotional processing that were different in alcoholic men and women . Significant interactions between alcoholism and gender in several cortical regions in response to emotional stimuli were observed for the aversive and happy stimuli , as well as large differences between ALCM and NCM . Areas within the frontal lobes were among the brain regions evidencing the most profound alcoholism-related gender differences . The brain activity contrasts related to affective vs . neutral stimuli were dampened in ALCM in the current study , similarly to prior research showing that ALCM had blunted limbic activation to emotionally expressive faces ( Marinkovic et al . , 2009 ) . Women are traditionally believed to be more emotionally reactive than men ( Merikangas et al . , 1996 ) , and in the current study , whereas ALCM showed predominately decreased fMRI emotional responsivity , ALCW had similar or greater brain activity in response to emotional stimuli than NCW , leading to significant group x gender interaction effects . Future prospective research is advised in order to examine gender differences in emotional reactivity and subsequent drinking behavior , to determine the contributions of gender differences that precede AUD , as compared to gender differences that develop as a result of chronic alcoholism .
Prior to conducting the experiment , we computed estimates of sample size based upon Cohen’s d , which suggested approximately 20 participants per group were required to detect a medium to large effect size ( Cohen , 1988 ) , a number confirmed by fMRI-specific research ( Thirion et al . , 2007 ) . A total of 88 participants ( 25 ALCW , 17 ALCM , 24 NCW , and 22 NCM ) were included in the analyses . The characteristics of the participants , including alcoholism indices and neuropsychological test scores are presented in Figure 2 ( and Appendix 1—tables 1 and 2 ) of the Results section; data and code are available from Dryad ( https://doi . org/10 . 5061/dryad . 5fn0224 ) and GitLab ( https://gitlab . com/kslays/sawyer-iaps; copy archived at https://github . com/elifesciences-publications/sawyer-iaps ) . All participants were right-handed English speakers recruited from the Boston , MA ( USA ) area through flyers placed in facilities and in public places ( e . g . , churches , stores ) , and advertisements placed with local newspapers and websites . Selection procedures included an initial structured telephone interview to determine age , level of education , health history , and history of alcohol and drug use . Specifically , we investigated the stable and persistent sequelae of AUD that are independent of current drinking or withdrawal , by recruiting long-term abstinent participants with a history of heavy drinking . Eligible individuals were invited to the laboratory for further screening and evaluations ranging between five to eight hours over the course of one to three days . Prior to screening , written informed consent was obtained; the protocols and consent forms were approved by the Institutional Review Boards of the participating institutions: Boston University School of Medicine ( #H24686 ) , VA Boston Healthcare System ( #1017 and #1018 ) , and Massachusetts General Hospital ( #2000P001891 ) . Participants were reimbursed $15 per hour for assessments , $25 per hour for scans , and $5 for travel expenses . Participants underwent medical history interview and vision testing , plus a series of questionnaires ( e . g . , handedness , alcohol and drug use , HRSD ) to ensure they met inclusion criteria . Participants were given the computerized Diagnostic Interview Schedule ( Robins et al . , 2000 ) , which provides lifetime psychiatric diagnoses according to criteria established by the American Psychiatric Association . Participants were excluded from further participation if any source ( e . g . , hospital records , referrals , or personal interviews ) indicated that they had one of the following: Corrected visual acuity worse than 20/50 in both eyes; Korsakoff’s syndrome; cirrhosis , major head injury with loss of consciousness greater than 15 min unrelated to AUD; stroke; epilepsy or seizures unrelated to AUD; schizophrenia; HRSD score over 15; electroconvulsive therapy; history of illicit drug use more than once per week within the past five years ( except for one ALCW who had used marijuana more frequently but not during the six months preceding testing , and one ALCW who had used marijuana once per week for four years , ceasing four years before testing ) ; lifetime history of illicit drug use more than once per week for over 10 years or three times per week for over five years . Participants received a structured interview regarding their drinking patterns , including length of abstinence and duration of heavy drinking , that is more than 21 drinks per week ( one drink: 355 ml beer , 148 ml wine , or 44 ml hard liquor ) . For each participant , we calculated a Quantity Frequency Index ( Cahalan et al . , 1969 ) , which factors the amount , type , and frequency of alcohol usage ( ounces of ethanol per day , roughly corresponding to number of drinks per day ) over the last six months ( for the NC group ) , or over the six months preceding cessation of drinking ( for the ALC group ) . The ALC participants met criteria for alcohol abuse or dependence , and had over 21 drinks per week for at least five years in their lifetime; all had abstained from alcohol for at least 21 days . Importantly , to ensure stability in the sequelae of AUD , we investigated long-term abstinent participants with a history of heavy drinking and whose participation was independent of current drinking or withdrawal . None of the NC participants drank heavily ( 21 or more per week ) , except for one man who drank while serving in the army decades before the scan , but did not meet the criteria for alcohol dependence; social drinking patterns of the NC participants are reported in Figure 2 and Appendix 1—table 1 . We examined the group x gender interaction within a regression model for the demographics , alcoholism indices , neuropsychological and clinical assessment scores . We also conducted Welch’s t-tests to examine gender differences for each measure for the ALC and NC groups separately , and group differences for the men and women separately . Imaging data were acquired using a 3T Siemens ( Erlangen , Germany ) Trio Tim magnetic resonance scanner . Following automated shimming and scout image acquisition , two eight-minute 3D T1-weighted MP-RAGE sequences were obtained: TR = 2530 msec , TE = 3 . 45 msec , flip angle = 70 , FOV = 256 mm , 128 sagittal slices with in-plane resolution 1 × 1 mm , slice thickness = 1 . 33 mm . These two structural volumes were used for functional slice prescription , spatial normalization , and cortical surface reconstruction . Due to time constraints , only one MP-RAGE sequence was obtained for 23 subjects ( 11 NCM , 8 ALCM , 2 NCW , 2 ALCW ) . Functional whole-brain blood oxygen level-dependent ( BOLD ) images were obtained with a gradient echo T2*-weighted sequence: TR = 2 s , TE = 30 msec , flip angle = 900 , FOV = 200 mm , slice thickness = 3 . 0 mm , spacing = 1 . 0 mm , 32 interleaved axial-oblique slices aligned to the anterior-commissure/posterior-commissure line ( voxel size: 3 . 1 × 3 . 1 × 4 . 0 mm ) . The scans covered the entire cerebrum and the superior portion of the cerebellum . Participants were presented with blocks of pictures chosen to evoke emotional responses ( Figure 1 ) . The picture stimuli were from the International Affective Picture System ( Lang et al . , 1988 ) . Participants completed five runs ( except one NCW who completed only four runs ) , each including five conditions: aversive , erotic , gruesome , happy , and neutral pictures . As depicted in Figure 1 , each run contained three 24 s blocks of fixation plus eight 24 s blocks that each consisted of six pictures of one of the emotional conditions ( e . g . , happy pictures ) , for a total of 11 blocks per run . The five runs included a total of 40 blocks of emotional pictures with eight blocks for each of the five emotional picture conditions . Stimuli were presented only once , totaling 48 pictures per 264 s run ( 240 pictures in 22 min in total across the five runs ) . Within stimulus blocks , the six pictures were each serially presented against a black background for 3 s , followed by 1 s of fixation ( +++ ) . Participants were instructed to answer the question: ‘How does the picture make you feel ? ’ Following each image within a block , participants indicated feeling good , bad , or neutral , by using their index fingers to press buttons on a box . The left index finger indicated good , the right index finger indicated bad , and both center buttons indicated neutral; the left and right were counterbalanced across participants . Block order was counterbalanced across runs , and run order was counterbalanced across participants . The task was presented with the Presentation software package ( Neurobehavioral Systems , Albany , CA , USA ) . Behavioral response data were analyzed using R software mixed models ( Bates et al . , 2015; R Development Core Team , 2017 ) , with one model specified for reaction times , and one model specified for the percentage of pictures endorsed for each rating ( good , bad , neutral ) . For both reaction times and percentage models , independent intercepts were modeled for each participant , and full-factorial ANOVAs were calculated for the four factors of rating ( good , bad , neutral ) , condition ( aversive , erotic , gruesome , happy , neutral ) , group ( ALC , NC ) , and gender ( men , women ) . Full-factorial mixed models were employed to examine the relationships of percentage ratings and evaluation times to selected neuropsychological measures ( Wechsler Verbal and Performance IQ scores , and the Delayed Memory Index ) , affective measures ( the POMS Depression scale , and the Multiple Affect Adjective Check List [MAACL] Anxiety and Sensation Seeking scales ) , and brain activity ( i . e . , contrast effect size ) within the clusters identified to have significant group x gender interactions for aversive vs . neutral and erotic vs . neutral contrasts ( the two most salient contrasts ) . Separate mixed models were used for each measure ( three neuropsychological measures , three affect measures , and five clusters , for percentage rating and evaluation times , resulting in a total of 22 models ) . Outliers ( outside three standard deviations from the mean ) were removed prior to analyses; this resulted in the exclusion of 1 ALCW and 1 ALCM for POMS Depression , and 2 ALCW and 1 NCW for MAACL Anxiety . Models were examined for significant ( p<0 . 05 ) interactions of the measures with group or gender , and followed by planned comparisons: ALC vs . NC for group interactions , and subgroup differences ( ALCW vs . NCW , ALCM vs . NCM ) for group x gender interactions . Post-hoc comparisons examined the slope of each measure with percentage ratings or evaluation times , and Bonferroni correction was applied for the number of contrasts examined within the model . The imaging data were analyzed using FreeSurfer and FS-FAST v6 . 0 ( http://surfer . nmr . mgh . harvard . edu ) analysis packages ( Dale et al . , 1999; Fischl et al . , 1999a ) . Individual cortical surfaces were reconstructed using automatic gray and white matter segmentation , tessellation , and inflation . Images were registered with a canonical brain surface ( fsaverage ) based on sulcal and gyral patterns ( Fischl et al . , 1999b ) , and registered with a canonical brain volume ( MNI305 ) using a 12 degrees of freedom nonlinear transform . Gray and white matter surface accuracy was individually examined using automatically-generated quality control figures ( https://github . com/poldracklab/niworkflows ) , and no errors were detected for any of the subjects included in the analyses that would be likely to influence the outcomes of this project ( Waters et al . , 2018 ) . The fMRI data were corrected for motion and slice-time acquisition using FS-FAST preprocessing . Normalized motion and signal intensity spikes were obtained from the nipype rapidart algorithm ( https://www . nitrc . org/projects/rapidart/ , https://doi . org/10 . 5281/zenodo . 596855 ) , and blocks with motion over 1 . 5 mm , or signal intensity shifts over 3 . 0 standard deviations , were removed via a paradigm file covariate for each run . Subjects were removed from the study if this process excluded all but two or fewer blocks of any condition , a requirement that resulted in the exclusion of two additional NCW . Next , the FS-FAST process split the analysis into three spaces ( left and right surfaces , and subcortical volume ) , then data from each subject was spatially normalized ( co-registered with ) the fsaverage and MNI305 spaces , respectively; all subsequent analyses were performed in these three group spaces . Spatial smoothing was performed with a 5 mm full width at half maximum Gaussian kernel in 3D for the volume and in 2D for the surfaces . Condition-specific effects were estimated by fitting the amplitudes of boxcar functions convolved with the FSL canonical hemodynamic response function to the BOLD signal across all runs . Statistical maps were constructed from each contrast of stimulus conditions for each subject ( first level analyses ) . Four contrasts were examined: aversive vs . neutral , happy vs . neutral , erotic vs . neutral , and gruesome vs . neutral . These first-level analyses were concatenated , and second-level ( group level or between-subjects ) analyses were performed using random-effects models to account for inter-subject variance ( Friston et al . , 1999 ) , with weighted least squares effects incorporated from the variability measures from the first-level contrasts . We examined the overall main effect of group ( ALC vs . NC ) , the interaction of group x gender , and the effects of group for men and women separately , for each of the four contrasts ( each emotion condition vs . neutral condition ) . Cluster-level corrections for multiple comparisons were applied to cortical surface statistical contrast maps ( Hagler et al . , 2006 ) using 10 , 000 precomputed Z Monte Carlo simulations and applied to subcortical volumetric statistical contrast maps using gaussian random fields with a cluster forming threshold of p<0 . 05 and a cluster-wise threshold of p<0 . 05 ( further corrected to p<0 . 017 for the analysis of three spaces: left cortex , right cortex , and subcortical ) . While these procedures have been shown to have a false positive ( Type one error ) level higher than the one specified ( Eklund et al . , 2016 ) , the present exploratory study was designed to reveal the sizes of the effects , and balance minimizing the chance of a false negative ( Type two error ) with the goal of highlighting the broad regions where further investigation of gender differences may be warranted . Therefore , the p-value threshold was set to a value sufficiently liberal to achieve this goal . For comparisons with research using stricter p-values , we additionally conducted the same analyses using a cluster-forming threshold of p<0 . 001 , the results of which are discussed in the Limitations . Cortical surface cluster regions were identified by the location of each cluster’s peak vertex on the cortical surface ( Desikan et al . , 2006 ) , and subcortical cluster regions were identified by the MNI coordinates of each cluster’s peak voxel ( Fischl et al . , 2002 ) . | More than 100 million people worldwide are thought to have alcohol use disorder , also known as AUD , alcohol dependence or alcoholism . People who struggle to regulate their emotions tend to consume more alcohol than others . This suggests that impaired emotion processing may increase the risk of developing the disorder . Most studies of emotion processing in people with alcohol use disorder do not distinguish between men and women . But evidence suggests that men and women process emotions in different ways . Sawyer et al . set out to explore the possible relationships between emotion processing , gender and alcoholism . Four groups of volunteers took part in the study: abstinent men and women with the disorder , and control groups of men and women without a history of alcoholism . Each group contained between 15 and 21 participants . The two abstinent alcoholic groups had not consumed alcohol for at least 21 days . The average length of abstinence was 7 years . The volunteers viewed a mixture of emotionally charged and neutral images while lying inside a brain scanner . The emotionally charged images were of happy , erotic , gruesome or aversive scenes . Sawyer et al . measured the difference in brain responses to the emotionally charged images versus the neutral ones , and compared this measure across the four groups of participants . Abstinent alcoholic men showed muted brain responses to the emotionally charged images compared to their female counterparts . This effect was seen in brain regions involved in memory , emotion processing and social processing . The same pattern occurred for all four types of emotionally charged image . Abstinent alcoholic men also showed smaller brain responses to the emotionally charged images than non-alcoholic control men . By contrast , abstinent alcoholic women showed larger brain responses to the emotionally charged images than non-alcoholic control women . This suggests that abstinent alcoholic men and women differ in the way they process emotions . Future studies should investigate whether these differences emerge over the course of abstinence . They should also examine whether these differences might contribute to , or result from , differences in alcohol use disorder between men and women . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2019 | Alcoholism gender differences in brain responsivity to emotional stimuli |
The extracellular matrix ( ECM ) is a major component of tumors and a significant contributor to cancer progression . In this study , we use proteomics to investigate the ECM of human mammary carcinoma xenografts and show that primary tumors of differing metastatic potential differ in ECM composition . Both tumor cells and stromal cells contribute to the tumor matrix and tumors of differing metastatic ability differ in both tumor- and stroma-derived ECM components . We define ECM signatures of poorly and highly metastatic mammary carcinomas and these signatures reveal up-regulation of signaling pathways including TGFβ and VEGF . We further demonstrate that several proteins characteristic of highly metastatic tumors ( LTBP3 , SNED1 , EGLN1 , and S100A2 ) play causal roles in metastasis , albeit at different steps . Finally we show that high expression of LTBP3 and SNED1 correlates with poor outcome for ER−/PR−breast cancer patients . This study thus identifies novel biomarkers that may serve as prognostic and diagnostic tools .
With an estimated number of new cases in 2008 of 1 . 3 million worldwide ( http://globocan . iarc . fr ) , breast cancer is the second most frequent cancer worldwide and claimed the lives of over 39 , 000 patients in the United States in 2012 alone . Most cancer deaths ( ∼90% ) are due to the metastatic colonization of distant organs by the tumor cells . The development of novel diagnostic approaches , including novel screening and imaging methods to improve detection of smaller primary tumors and distant metastases , and the identification of novel prognostic markers to permit better classification of breast cancers according to their metastatic potential is imperative as is the development of novel therapeutic strategies . For the last decade , there has been increasing interest in the tumor microenvironment , which encompasses all the components ( cellular and acellular ) of a tumor in addition to the tumor cells themselves ( Joyce and Pollard , 2009; Egeblad et al . , 2010 ) . Indeed , in order to progress , a tumor needs to be surrounded by a permissive environment and , to metastasize , tumor cells need to find or create a favorable niche ( seed and soil theory ) ( Paget , 1889; Ribatti et al . , 2006; Erler and Weaver , 2009; Comen , 2012 ) . To create such a niche , tumor cells either directly alter the microenvironment , or instruct local or recruited stromal cells to do so . In the context of breast cancer , recent studies have shown that the cross-talk and interaction of the tumor cells with their surrounding microenvironment is necessary for tumor progression ( Tlsty and Coussens , 2006; Polyak et al . , 2009; Dvorak et al . , 2011; Boudreau et al . , 2012; Conklin and Keely , 2012; Fordyce et al . , 2012 ) . In addition , the nature of the tumor microenvironment or gene expression profile of the stromal cells has been used to define human breast cancer types ( Bergamaschi et al . , 2008; Finak et al . , 2008; Conklin et al . , 2011 ) . Recent studies have also demonstrated that treatment outcome depends on the tumor microenvironment ( vascularization , oxygenation , recruited normal cells , etc; Chauhan et al . , 2011; Jacobetz et al . , 2013 ) . The extracellular matrix ( ECM ) is the complex scaffold of proteins that provides the architectural support for cell and tissue organization ( Hynes and Naba , 2012 ) . Cells are in turn able to adhere to the extracellular matrix via different types of receptors including the integrins ( Hynes , 2002; Geiger and Yamada , 2011 ) . In addition to providing biophysical cues , the ECM provides biochemical signals that are major regulators of cell proliferation , survival , migration , and invasion . ( Hynes , 2009 ) . Pathologists have used excessive ECM deposition ( desmoplasia ) as a marker of tumors with poor prognosis long before the complexity of the ECM was even deciphered ( Anastassiades and Pryce , 1974 ) . Despite its great importance in physiological ( development , aging ) and pathological processes such as cancer , the extracellular matrix remains underexplored ( Wilson , 2010 ) . To define the composition of extracellular matrices of tissues and tumors , we have developed a proteomics-based method to enrich and identify ECM proteins and coupled it with a bioinformatic annotation of the ‘matrisome’ defined as the ensemble of ECM and ECM-associated proteins ( Naba et al . , 2012 ) . Using this approach , we characterized the extracellular matrices of normal murine tissues ( e . g . , lung and colon ) and demonstrated that each of these comprises over 100 proteins . In this study , we apply this proteomics approach to study the composition of the ECMs of poorly and highly metastatic human mammary carcinoma xenografts , and show that both the tumor cells and the stromal cells contribute in characteristic ways to the production of the tumor ECM . Moreover , we show that both tumor- and stroma-derived proteins differ between tumors of different metastatic potential . Importantly , we demonstrate functional roles for several specific tumor-cell-derived ECM proteins in promoting metastasis . Altogether , our results demonstrate that the proteomic analysis of the tumor ECM can be used to identify proteins playing causal roles in tumor progression that could be further developed as prognostic or diagnostic markers and potentially as therapeutic targets .
To identify ECM proteins important for breast cancer progression and metastasis formation , we used a xenograft model where human breast cancer cells were orthotopically injected into the mouse mammary fat pad . We used cell lines of differing metastatic potential . The poorly metastatic MDA-MB-231 cell line was established from cells isolated from a pleural effusion sample from a triple-negative breast cancer patient ( Cailleau et al . , 1978 ) . The highly metastatic MDA-MB-231-LM2 line ( denoted LM2 ) , was previously selected and characterized for increased metastatic potential to the lungs ( Minn et al . , 2005 ) . 6 . 5 weeks post-injection , the primary tumors were harvested , ECM proteins were enriched from tumors using the subcellular fractionation protocol described previously ( Naba et al . 2012 ) , and the composition of the ECM-enriched fractions obtained was characterized by mass spectrometry ( Figure 1A ) . 10 . 7554/eLife . 01308 . 003Figure 1 . Enrichment of extracellular matrix proteins from human mammary tumor xenografts . ( A ) The sequential extraction of intracellular components was monitored by immunoblotting for GAPDH ( cytosol ) , the transferrin receptor ( plasma membrane ) , and actin ( cytoskeleton ) . The remaining insoluble fraction was highly enriched for ECM proteins ( collagen I panel ) and largely depleted for intracellular components . The ECM-enriched fraction obtained is subsequently submitted to multidimensional proteomic analysis and the matrisome ( ECM composition ) of each tumor type is defined as the ensemble of proteins present in two replicate samples and with at least two peptides in one of the two replicates . ( B ) Venn diagram represents the comparison of the matrisomes of MDA-MB-231 and LM2 tumors . In addition to 118 ECM proteins detected in both tumor types , we identified 26 proteins specific to poorly metastatic ( MDA-MB-231 ) tumors and 43 proteins characteristic of highly metastatic tumors ( LM2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 003 We define the matrisome of a tumor as the ensemble of proteins detected in two independent biological replicates and by at least two peptides in one of the two replicates . According to this definition , the matrisome of MDA-MB-231 tumors is composed of 144 proteins and the matrisome of LM2 tumors is composed of 161 proteins ( Figure 1B , Figure 2 , Figure 2—source data 1 ) . Comparison of the matrisomes of MDA-MB-231 and LM2 tumors identified 118 proteins expressed by both tumor types . In addition , we detected 26 proteins specific to the poorly metastatic tumors and 43 proteins specific to highly metastatic tumors ( Figure 1B , Figure 2 , Figure 2—source data 1 ) . According to our previous bioinformatic definition of the matrisome ( Naba et al . , 2012 ) , we further categorized these ECM proteins into two categories: core matrisome proteins and matrisome-associated proteins . The core matrisome comprises ECM glycoproteins , collagens , and proteoglycans . Matrisome-associated proteins include ECM-affiliated proteins , ECM regulators ( ECM remodeling enzymes and their regulators ) , and ECM-associated secreted factors ( Figure 2 , Figure 2—source data 1 ) . One can note that the most abundant proteins for each of these categories were detected in both tumor types . Examples of these proteins are fibronectin ( FN1 ) , fibrinogen , laminins , HSPG2 or perlecan , collagens I , III , IV , V and VI , and transglutaminase 2 ( TGM2 ) . In fact , we have previously reported the expression of these proteins in several normal tissues ( Naba et al . , 2012 ) , which suggests that these proteins are widely expressed . Most of the matrisome-associated components were detected at lower abundance , reflecting their presence at lower molar ratios than structural core ECM proteins ( Figure 2 , Figure 2—source data 1 ) . Therefore , the analysis identified ECM signatures of poorly ( 26 proteins ) or highly ( 43 proteins ) metastatic tumors ( Figures 1B and 2 ) and the proteins that differ between the two tumor types belong mostly to the ECM glycoproteins and the ECM regulators categories . Among the 43 proteins characteristic of highly metastatic tumors ( Figure 3D ) , we identified several proteins that have previously been reported to play a role in breast cancer metastasis in diverse cell and tumor models . Examples of these are angiopoietin-like 4 ( ANGPTL4 , Padua et al . , 2008 ) , cathepsin B ( CTSB , Sevenich et al . , 2010 , 2011; Vasiljeva et al . , 2006 ) , insulin-like growth factor-binding protein 4 ( IGFBP4 , Ryan et al . , 2009 ) and lysyl-oxidase-like 2 ( LOXL2 , Barry-Hamilton et al . , 2010 ) . Several other proteins in our list have been shown to play roles in metastasis of other tumor types ( ‘Discussion’ ) . 10 . 7554/eLife . 01308 . 004Figure 2 . Comparison of the matrisomes of MDA-MB-231 tumors and LM2 tumors identifies ECM proteins characteristic of poorly and highly metastatic tumors . Color code represents the number of unique peptides for each protein from poorly metastatic ( MDA-MB-231 ) or highly metastatic ( LM2 ) human mammary tumors . Values used to generate the figure were extracted from Figure 2—source data 1 , columns P and AA ( number of peptides ) . Grayed cells indicate that no peptides were detected in either of the two replicate samples . A dash ( – ) indicates that the protein was detected in only one of the two replicate samples of a given tumor type or with only one peptide in both replicate samples . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 00410 . 7554/eLife . 01308 . 005Figure 2—source data 1 . Complete MS data set of ECM-enriched fraction from MDA-MB-231 and LM2 human mammary tumor xenografts . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 00510 . 7554/eLife . 01308 . 006Figure 2—figure supplement 1 . Validation of the differential expression of proteins identified by proteomics . ( A ) Immunoblots were performed on ECM-enriched fractions prepared from MDA-MB-231 or LM2 tumors to confirm the differential expression of some of the proteins identified by proteomics . ( B ) Immunohistochemistry was performed on MDA-MB-231 or LM2 tumor sections to evaluate the expression and localization of LTBP3 . LTBP3 is detected in LM2 but not MDA-MB-231 tumors . Note the extracellular distribution of LTBP3 . ( C ) RT-qPCR was used to monitor mRNA expression levels in LM2 tumors as compared to MDA-MB-231 tumors and was conducted in triplicate in two independent tumors per tumor type . Bar charts represent normalized expression of the genes in LM2 tumors ( black ) as compared to MDA-MB-231 tumors ( gray ) . Actin expression was used to normalize RT-qPCR data . Data are presented as means ±SEM . ( D ) RNA was extracted from normal murine mammary gland ( light grey ) or from autochtonous murine mammary tumors ( MMTV-PyMT transgenic mice: adenoma , dark grey or adenocarcinoma , black ) . RT-qPCR was conducted in triplicate . Bar charts represent normalized expression of the genes . Actin expression was used to normalize RT-qPCR data . Data are presented as means ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 00610 . 7554/eLife . 01308 . 007Figure 3 . The tumor extracellular matrix is secreted by both tumor cells and stromal cells and differs with the tumor’s metastatic potential . ( A ) Proteins expressed by both tumor types and by the same compartment in the two tumor types . ( B ) Proteins expressed by both tumor types but by different compartments . ( C ) Proteins secreted by MDA-MB-231 tumors and not by LM2 tumors . ( D ) Proteins secreted by LM2 tumors and not by MDA-MB-231 tumors . The number of peptides detected for each protein is indicated . For the proteins secreted by both the tumor cells and the stromal cells , the number of peptides listed corresponds to the number of human ( tumor-derived ) - or murine ( stroma-derived ) -specific peptides . For the proteins secreted by only one compartment , the number of peptides includes both species-specific and indistinguishable peptides . Proteins are sorted by tumor type and by their origins: tumor ( red ) , stroma ( blue ) , or both ( yellow: similar abundance of the human and mouse proteins , orange: human form is at least five times more abundant than the mouse form , green: the mouse form is at least five times more abundant ) . To determine the relative contributions of the tumor and stromal cells to the secretion of ECM proteins , human-to-mouse peptide abundance ratios were calculated using the values indicated in column P and AB for the MDA-MB-231 and LM2 tumors respectively ( Figure 3—source data 1 ) . Proteins for which different isoforms have been detected are indicated with an asterisk ( * ) and , for simplicity , isoforms are combined here , their UniProt accession numbers can be found in Figure 3—source data 1 , column AP . In a few instances , the origin of the protein could not be determined due to the lack of species-specific peptides; these proteins are indicated with a question mark ( ? ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 00710 . 7554/eLife . 01308 . 008Figure 3—source data 1 . Complete MS data set of ECM-enriched fraction from MDA-MB-231 and LM2 human mammary tumor xenografts taking into account the origin of matrisome proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 00810 . 7554/eLife . 01308 . 009Figure 3—source data 2 . Detailed list of all of the confidently identified peptide spectrum matches ( PSMs ) from the LC-MS/MS runs of the 11 fractions resulting from off-gel electrophoresis of each of the two MDA-MB-231 tumor replicates and each of the two LM2 tumor replicates . The four tables containing all of the confidently identified peptide spectrum matches ( PSMs ) from the LC-MS/MS runs of the 11 fractions resulting from off-gel electrophoresis of each of the four tumor samples can be downloaded from MassIVE ( http://massive . ucsd . edu ) using the identifier: MSV000078535 . The file should be accessible at ftp://MSV000078535:a@massive . ucsd . edu/results . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 009 A strength of human/mouse xenograft model systems coupled to mass spectrometry is that they allow the distinction of human ( tumor-derived ) protein sequences from their murine ( stroma-derived ) counterparts ( Naba et al . , 2012 ) . Therefore we could determine the relative contributions of tumor and stromal cells to the production of the tumor ECM . We found that the tumor extracellular matrix is secreted by both tumor cells and stromal cells; some proteins are exclusively secreted by the tumor cells or the stromal cells; and some proteins are secreted by both compartments: either in equal proportions or more abundantly by the tumor cells or by the stromal cells ( Figure 3 , Figure 3—source data 1 ) . We chose a conservative significance threshold of fivefold change in protein abundance , estimated as the summed precursor-ion intensity of the species-specific peptides . Because the tumor–stroma distinction relies on measurement of similar peptides of differing sequence ( with accompanying mass differences ) , it is impractical to use more accurate mass spectrometric relative quantitation approaches that incorporate isotopic labels . The majority of proteins ( 82 ) found in the matrisomes of poorly and highly metastatic tumors are secreted by the same compartment in both tumor types ( Figure 3A ) . As expected , ECM and ECM-associated proteins involved in hemostasis and found in the circulation are murine ( fibrinogen chains: α , β , γ , plasminogen , thrombin [Factor 2] , Factor XIII transglutaminase , vitronectin , etc ) . Structural collagens ( collagens I and V ) are also mostly secreted by the stroma . Components of the basement membranes were found to be expressed by the tumor cells only ( laminin chains α3 , β3 , γ2 ) , by both compartments ( laminin chains α5 , γ1 , COL6A3 , HSPG2 ) or by the stroma ( Nidogens 1 and 2 ) . In addition , 36 proteins were detected in both tumor types but their origin differed as a function of the tumor’s metastatic potential ( Figure 3B ) . In poorly metastatic tumors , fibronectin ( FN1 ) and periostin ( POSTN ) are secreted by the tumor cells , whereas in highly metastatic tumors , they are secreted by both the tumor cells and the stromal cells . Conversely , several basement membrane components ( laminin chains β1 and β2 , and the collagen chains 4A1 , 4A5 6A1 , 6A2 and 18A1 ) are secreted in poorly metastatic tumors mostly by the stroma , whereas in highly metastatic tumors , they are secreted by both the tumor cells and the stromal cells . As expected , many proteins specific to either poorly or highly metastatic tumors were secreted exclusively by the tumor cells . However , interestingly , we also observed that the stromal compartment also contributed to the differences detected ( Figure 3C , D , Figure 3—source data 1 ) . This indicates that tumor cells of differing metastatic potential not only synthesize distinct subsets of ECM proteins but also influence which ECM proteins are produced by the stroma . We sought to determine whether the expression of the ECM proteins characteristic of highly metastatic mammary tumors was controlled by common upstream regulators and whether , by probing the tumor ECM , we could identify signaling pathways contributing to tumor progression . To do so , we used Ingenuity Pathway Analysis ( Figure 4—figure supplement 1 ) . We found that 17 of the 43 ( nearly 40% ) ECM proteins found to be up-regulated in LM2 tumors are downstream of the TGFβ signaling pathway ( Figure 4 , left panel ) . The majority of these are tumor-derived , although several host proteins ( C1qa , C1qc and Vwf ) are also identified as being downstream of TGFβ . Interestingly , the LM2 cell line and its metastatic potential have previously been shown to be dependent on TGFβ ( Minn et al . , 2007 ) . 10 . 7554/eLife . 01308 . 010Figure 4 . The TGFβ and the HIF1α/VEGF pathways are up-regulated in highly metastatic mammary tumors . The 43 ECM and ECM-associated proteins ( tumor- and/or stroma-derived ) unique to highly metastatic mammary tumors were uploaded to the Ingenuity Pathway Analysis software and queried for common upstream regulators . The analysis revealed an enrichment of TGFβ and HIF1α/VEGF targets within our ECM signature of highly metastatic mammary tumors . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 01010 . 7554/eLife . 01308 . 011Figure 4—figure supplement 1 . Ingenuity Pathway Analysis . Identification of regulators ( cytokines , growth factors , G-protein-coupled receptors , transcription regulators , or transmembrane receptors ) controlling the expression of subsets of the 43 ECM genes characteristic of highly metastatic tumors was searched . Regulators upstream of at least three ECM genes are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 011 One of the other signaling pathways significantly represented within our data set is the HIF1α/VEGF pathway that influences the levels of 10 of the 43 ECM proteins identified in our study ( Figure 4 , right panel ) . We also noted that there is a large overlap between the targets of TGFβ and HIF1α/VEGF as the genes identified as being downstream of the HIF1α/VEGF pathway are all also downstream of the TGFβ pathway . Interestingly , a recent publication by Curran and Keely highlights the convergence of TGFβ and HIF1α signaling pathways as a key contributor to the cross-talk between breast tumor cells and stromal cells ( Curran and Keely , 2013 ) . Altogether , this analysis indicates that by interrogating the tumor extracellular matrix , one can identify signaling pathways up-regulated within the tumor cells or stromal cells and these represent potential targets for therapeutic intervention . As noted above , several of the proteins up-regulated in LM2 tumors have previously been implicated in metastasis . We next wanted to evaluate whether other proteins found to be differentially expressed between poorly and highly metastatic tumors might play any causal , functional roles in tumor progression . Accordingly , we selected a set of proteins ( LTBP3—Latent TGFβ Binding Protein 3; SNED1—Sushi , Nidogen and EGF-like Domains 1; EGLN1—Egl Nine homolog 1 and S100A2 ) based on the following criteria: they were ( 1 ) detected only in highly metastatic tumors; ( 2 ) secreted exclusively by the tumor cells; ( 3 ) in relatively high abundance ( Figures 2 and 3D , Figure 2—source data 1 , Figure 3—source data 1 ) ; ( 4 ) not previously investigated for any roles in metastasis . We focused on ECM proteins produced by the tumor cells as their expression can easily be manipulated in vitro . We also investigated the role of CYR61 , a tumor-derived ECM protein detected in both MDA-MB-231 and LM2 tumors but with a much greater abundance in LM2 tumors ( Figures 2 and 3 , Figure 2—source data 1 , Figure 3—source data 1 ) . For proteins against which antibodies were available ( LTBP3 , EGLN1 , and CYR61 ) we confirmed their differential expression by immunoblot and/or immunofluorescence analyses ( Figure 2—figure supplement 1A , B ) . We also performed RT-qPCR analyses for several other proteins in our lists despite the fact that it is well known that mRNA levels and steady-state protein levels do not correlate at all well ( Gygi et al . , 1999 ) . It can be seen that the mRNA changes for several of the proteins of interest ( CYR61 , EGLN1 , LTBP3 ) do not show mRNA changes corresponding with the marked differences in protein levels detected , although some do ( S100A2 , SNED1 ) ( Figure 2—figure supplement 1C ) . In addition , we show that Ltbp3 and Sned1 mRNAs are up-regulated during malignant progression in the autochthonous MMTV-PyMT murine mammary tumor model ( Figure 2—figure supplement 1D ) . To address the roles of these proteins in tumor progression , we knocked down each of the five proteins selected for analysis in LM2 cells using short hairpins . As control , we used an shRNA targeting the firefly luciferase ( sh-Cont . ) . For each gene of interest , two distinct hairpins giving efficient knockdown were selected ( Figure 5—figure supplement 1 ) . Control or knocked-down LM2 cells were orthotopically injected into the mouse mammary fat pad . At sacrifice ( 7 +/−0 . 5 weeks post-injection ) , the masses of the primary tumors were measured . We observed that none of the genes , when knocked down , significantly affected primary tumor growth ( Figure 5A ) . Consistently , we also observed no differences in tumor cell proliferation or apoptosis between control and knockdown tumors ( Figure 5—figure supplement 3 ) . Importantly , the primary tumors ( and the metastases deriving from them ) were still ZsGreen-positive , indicating that the shRNA construct was still strongly expressed ( Figure 5C ) . We also confirmed that the knockdowns persisted in vivo ( Figure 5—figure supplement 2 ) and that there was no compensatory up-regulation of these genes by the stroma ( data not shown ) . These data indicate that none of these genes is required for primary tumor growth . 10 . 7554/eLife . 01308 . 012Figure 5 . Tumor-cell-derived ECM proteins influence the metastatic dissemination of tumor cells to distant organs . Mice were injected orthotopically with control or knockdown LM2 cells . Tumors were allowed to grow for 7 ± 0 . 5 weeks . Number of mice per condition is indicated in Figure 5B . ( A ) ECM protein knockdown does not inhibit primary tumor growth . At sacrifice , control and knockdown primary tumors were weighed . Bar chart represents the mass of knockdown tumors as a percentage of that of control tumors ± SEM . Student’s t test was performed and none of the genes affected significantly and consistently primary tumor growth . ( B ) LM2 control tumors metastasize to the lungs , liver and spleen . The number of mice that presented with visible metastases in the indicated organs is indicated . ( C ) Representative pictures of whole left pulmonary lobe from LM2 ( control or knockdown ) -tumor-bearing mice with ZsGreen-positive metastatic foci . ( D ) Numbers of ZsGreen-positive metastatic foci in the left pulmonary lobe were counted . Data are presented as percentage of control ± SEM ( Student’s t test , *p<0 . 05 , and **p<0 . 01 ) . Number of animals per group is indicated in Figure 5B . ( E ) Lung sections were stained with a human-specific anti-vimentin antibody to detect human tumor cells in the murine lung . ( F ) Alu PCR was performed on genomic DNA extracted from the lungs of control or knockdown tumor-bearing mice . Data are presented as normalized Alu signal as compared to murine actin signal and as percentage of control ± SEM ( Student’s t test , *p<0 . 05 , **p<0 . 01 , ns: not significant , and nd: not determined ) . Number of animals per group is indicated in Figure 5B . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 01210 . 7554/eLife . 01308 . 013Figure 5—figure supplement 1 . Establishment of stable LM2 knockdown cells . RT-qPCR was used to monitor knockdown efficiency in cells in culture and was conducted for each cell line in triplicate on three independent sets of RNA collected at three consecutive cell passages . Bar charts represent normalized expression of the genes in knockdown cell lines relative to the control cell line ( sh-Cont . ) . Actin expression was used to normalize RT-qPCR data . Data are presented as means ± SEM . The two hairpins giving the most efficient knockdown are shown . When antibodies were available , knockdown efficiency was also monitored by immunoblotting ( see CYR61 and EGLN1 panels ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 01310 . 7554/eLife . 01308 . 014Figure 5—figure supplement 2 . Persistence of gene expression knockdown in tumors . RT-qPCR was used to monitor knockdown persistence in tumors and was conducted on RNA extracted from individual control or knockdown tumors and in triplicate for each sample . Bar chart shows normalized expression of the genes in knockdown tumors relative to control tumors . Actin expression was used to normalize RT-qPCR data . Data are presented as means ± SEM . The number of mice per group ranged from 3 to 13 and is indicated in Figure 5B . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 01410 . 7554/eLife . 01308 . 015Figure 5—figure supplement 3 . Tumor-cell-derived ECM proteins do not influence proliferation or apoptosis of primary mammary tumors . Primary tumor sections were stained with the anti-Ki67 antibody ( A ) to evaluate cell proliferation or anti-cleaved caspase 3 ( B ) to evaluate apoptosis . Knockdown of tumor cell-derived ECM proteins ( CYR61 , LTBP3 , SNED1 , EGLN1 , or S100A2 ) did not affect tumor cell proliferation ( Ki67 staining , A ) or apoptosis ( cleaved caspase 3 , B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 015 LM2 cells have been selected for their high metastatic potential and their tropism for the lungs ( Minn et al . , 2005 ) . In our system , LM2 cells are injected in the absence of exogenous ECM support ( Matrigel ) in immunodeficient mice ( NOD/SCID/IL2Rγ-null ) . We observed that the parental LM2 cells and the LM2 cells expressing the control hairpin , in addition to metastasizing to the lung , also could colonize the liver and the spleen from the primary site ( Figure 5B ) . We thus wanted to evaluate the capacity of the knockdown tumors to metastasize from the primary site ( mammary gland ) to distant organs . Whereas knocking down CYR61 , a protein that we detected by proteomics in both poorly and highly metastatic tumors , did not affect metastasis formation , knocking down any of the four genes characteristic of highly metastatic ( LM2 ) tumors , and not detected in the poorly metastatic ( MDA-MB-231 ) tumors , led to a significant decrease of the metastatic burden as compared with that of mice injected with control LM2 cells ( Figure 5B ) . None of the SNED1 or EGLN1 knockdown tumors disseminated to organs other than the lungs and the number of mice presenting visible lung metastases was dramatically decreased ( Figure 5B ) . LTBP3 and S100A2 knockdown also led to a very significant inhibition of metastatic dissemination , although we detected occasional visible metastases in the liver or spleen of mice injected with LTBP3 or S100A2 knockdown cells ( Figure 5B ) . Because of the pulmonary tropism of LM2 cells , we further analyzed the lungs of control or knockdown tumor-bearing mice . The analysis of the lungs of control or knockdown tumor-bearing mice for the presence or absence of ZsGreen-positive metastatic foci confirmed the significant decrease in the number of metastases formed by knockdown tumors ( Figure 5B–D ) . As the tumor cells are human , we monitored the presence of human cells within murine lungs . We used a human-specific anti-vimentin antibody and Alu PCR to detect human genomic DNA and both methods confirmed the significantly decreased metastatic burden in knockdown-tumor-bearing mice as compared to control mice ( Figure 5E , F ) . These results demonstrate that these four ECM proteins play significant functional roles in the metastatic dissemination of these mammary tumor cells . Metastatic dissemination is a multistep process that requires tumor cells to proliferate , invade the surrounding host tissue , intravasate , survive in the circulation , extravasate and eventually seed and proliferate in distant organs . To pinpoint which steps of the metastatic cascade are influenced by each metastasis-associated ECM protein , we first compared the vascularization , invasiveness , and degree of fibrosis in control and knockdown primary tumors . Knock-down of any of the five ECM genes selected did not affect significantly primary tumor vascularization ( Figure 6—figure supplement 1 ) . Control tumors markedly invaded the skin ( Figure 6 ) and the surrounding muscles ( not shown ) , and were significantly fibrotic , as indicated by staining tumor sections with Masson’s Trichrome ( Figure 6 ) . Cells in which CYR61 , EGLN1 , or S100A2 were knocked down formed highly invasive ( white arrow ) and desmoplastic tumors resembling the control tumors , suggesting that these proteins did not influence tumor invasion or fibrosis ( Figure 6 ) . Interestingly , LTBP3 and SNED1 knockdown tumors were characterized by deposition of collagen at the periphery of the tumor forming a thick capsule . Consistently , the LTBP3 and SNED1 tumors failed to invade the surrounding tissues , suggesting that LTBP3 and SNED1 are promoters of tumor invasiveness . 10 . 7554/eLife . 01308 . 016Figure 6 . Tumor-cell-derived ECM proteins influence the invasiveness of primary mammary tumors . Primary tumor sections were stained with Masson’s trichrome ( blue: collagen fibers , red: cells ) to evaluate fibrosis ( indicated by vertical double-dashed line ) and encapsulation ( bracket ) or invasiveness ( white arrowhead ) into the skin of the primary tumors . Note that tumors in which LTBP3 or SNED1 are knocked down are less invasive and more encapsulated than in the control tumors . CYR61 , EGLN1 , or S100A2 knockdown did not affect tumors’ invasiveness . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 01610 . 7554/eLife . 01308 . 017Figure 6—figure supplement 1 . Tumor-cell-derived ECM proteins do not influence vascularization of primary mammary tumors . Primary tumor sections were stained with an anti-CD31 ( endothelial cell marker ) to visualize tumor vasculature . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 017 To test whether each metastasis-associated ECM protein influenced later steps of the metastatic cascade ( survival in the circulation , extravasation and seeding , survival and proliferation in the lung ) , we injected the tumor cells directly into the circulation via the lateral tail vein . When injected into the blood circulation , control LM2 cells efficiently seeded and colonized the lungs and formed large metastases detectable by fluorescence ( Figure 7A , upper panel ) , by staining lung sections with hematoxylin and eosin ( Figure 7A , middle panel ) or with an anti-human vimentin antibody ( Figure 7A , lower panel ) . Knockdown of EGLN1 or S100A2 strongly inhibited the formation of pulmonary metastases ( Figure 7A–7C ) . In contrast , knockdown of LTBP3 or SNED1 in LM2 cells did not affect lung colonization ( Figure 7A , B ) , consistent with our finding that these proteins influence tumor cell invasion at the primary site ( Figure 6 ) . Although LTBP3 knockdown gave rise to the same number of metastases , the metastases were significantly smaller ( based on the size of the ZsGreen-positive foci ) and consistently , the Alu PCR signal was significantly lower than in the control lungs ( Figure 7C ) . Together , our data suggest that , although all these four proteins are required to increase metastasis formation , they do so by influencing different steps of the metastatic cascade . LTBP3 and SNED1 likely act early in the metastatic cascade by promoting tumor invasiveness . In addition , LTBP3 , but not SNED1 , appears to affect the growth of lung metastases . Our data also suggest that EGLN1 and S100A2 are likely to act at later stages of the metastatic cascade ( extravasation , seeding , survival , and/or growth of the tumor cells in secondary sites ) because , although the primary tumors are very invasive , they failed to colonize distant organs . 10 . 7554/eLife . 01308 . 018Figure 7 . Tail vein metastasis assay . Control or knockdown cells were injected via the tail vein and the formation of lung metastases was evaluated . ( A ) Upper panel: representative pictures of the whole left pulmonary lobe with ZsGreen-positive metastatic foci from mice injected with LM2 ( control or knockdown ) cells ( upper panel ) . Lung sections were stained with hematoxylin and eosin ( H&E , middle panel ) or with human-specific anti-vimentin antibody ( lower panel ) to visualize the metastatic foci . ( B ) Numbers of ZsGreen-positive metastatic foci in the left pulmonary lobe . Data are presented as percentage of control ± SEM ( Student’s t test , *p<0 . 05 , **p<0 . 01 , ns: not significant ) . Number of animals per group: sh-Cont . : 17 mice , sh-LTBP3: 10 mice , sh-SNED1: 10 mice , sh-EGLN1: 8 mice , sh-S100A2: 8 mice . ( C ) Alu PCR was performed to monitor the presence of human tumor cells in the murine lung . Data are presented as percentage of control ± SEM ( Student’s t test , *p<0 . 05 , **p<0 . 01 , ns: not significant ) . Number of animals per group: sh-Cont . : 17 mice , sh-LTBP3: 10 mice , sh-SNED1: 10 mice , sh-EGLN1: 8 mice , sh-S100A2: 8 mice . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 018 In addition to providing novel insights into the biology of breast cancer progression , we sought initial indications that any of the ECM proteins identified here by proteomics and shown to play causal roles in mammary carcinoma progression , might correlate with clinical outcome in breast cancer patients . Since equivalent proteomic data are not yet available for human patient material , we could only make comparisons with mRNA expression data , even though we know that correlation between protein and mRNA levels is known to be loose . We used Kaplan–Meier Plotter ( Györffy et al . , 2010 ) , an online-available meta-analysis tool for biomarker assessment . Using this tool , we tested whether the mRNA expression of any of the four genes we studied correlated with distant-metastasis-free survival for breast cancer patients . While expression levels for SNED1 and LTBP3 were not significantly correlated across all mammary cancer sub-types ( Figure 8 , right panels ) , they were significantly correlated ( p values of 0 . 0079 and 0 . 034 , respectively ) with poor prognosis for estrogen-receptor-negative and progesterone-receptor-negative ( ER−/PR− ) patients ( Figure 8 , left panels ) . In this context , it is significant that the MDA-MB-231 cells were isolated from the pleural effusion of a triple-negative breast cancer patient ( Cailleau et al . , 1978 ) . While we are not suggesting that the correlations between the expression of two ECM proteins and probability of metastasis in a limited subset of cancer patients is currently of predictive value , this analysis demonstrates that we can identify within our proteomics signatures , a subset of ECM proteins that are pro-metastatic in model systems and may have potential as prognostic biomarkers and are worthy of further investigation . 10 . 7554/eLife . 01308 . 019Figure 8 . Correlation between SNED1 and LTBP3 expression and distant-metastasis-free survival in breast cancer patients . We tested the predictive value of the four ECM genes studied using the online assessment tool Kaplan–Meier Plotter . Whereas the expression of none of the genes studied correlated with distant metastasis-free survival for the entire population of breast cancer patients , SNED1 and LTBP3 expression inversely correlated with the survival of estrogen-receptor-negative and progesterone-receptor-negative ( ER−/PR− ) breast cancer patients . DOI: http://dx . doi . org/10 . 7554/eLife . 01308 . 019
We demonstrate in this study that a proteomics-based discovery approach can define ECM signatures of tumors of differing metastatic potential . We show that the tumor ECM is derived from both the tumor and stromal cells and that tumors of differing metastatic potential differ in both the tumor- and stroma-derived ECM . Our study led to the identification of tumor-derived ECM proteins that play functional roles in tumor progression and metastatic dissemination . Indeed , a high proportion of the proteins we detected as up-regulated in highly metastatic tumors ( some reported previously , some new to this study ) contribute to metastatic dissemination . We further demonstrate that these proteins can affect different steps of the metastatic cascade: LTBP3 and SNED1 appear to promote early stages of invasion and dissemination , whereas EGLN1 and S100A2 act at later stages of the metastatic cascade . Finally , using gene expression data sets from breast cancer patients , we show that some of the proteins discovered by proteomics as differentially expressed are correlated with breast cancer patient outcomes . We studied in detail five ECM or ECM-associated proteins not previously clearly implicated in metastasis and demonstrate that four out of five of these proteins play roles at different steps of the metastatic cascade . CYR61 was detected in both poorly and highly metastatic tumors but with a greater abundance in highly metastatic tumors . Previous studies had reported the up-regulation of CYR61 in more aggressive tumor cell lines or lesions ( Tsai et al . , 2000 , 2002 ) . Our study demonstrates that although produced in greater abundance by highly metastatic tumors , CYR61 is not necessary to promote metastatic dissemination as knocking down its expression did not have any effect on the metastatic dissemination of LM2 cells . Consistent with the role of TGFβ in cancer progression ( Padua and Massagué , 2009; Massague , 2012 ) , we show that LTBP3 , a protein that regulates TGFβ secretion and bioavailability ( Chen et al . , 2002; Munger and Sheppard , 2011 ) promotes primary tumor invasion and metastasis . LTBP3 is also a protein that contributes to the formation of the fibrillar ECM network ( Ramirez and Rifkin , 2009; Todorovic and Rifkin , 2012 ) . Thus , whether LTBP3 contributes to tumor progression because of its signaling role or its architectural role remains to be elucidated . Of note , LTBP3 is a member of a family of proteins and the three other members of the family; LTBP1 , LTBP2 and LTBP4 were all detected in the matrisome of both poorly and highly metastatic tumors , LTBP1 and 4 being expressed by the tumor cells and not the stromal cells and LTBP2 being expressed by both the tumor cells and stromal cells ( Figures 2 and 3 , Figure 2—source data 1 , Figure 3—source data 1 ) . SNED1 ( Sushi , Nidogen , and EGF-like Domains 1 ) is a large ECM glycoprotein first identified as a potential component of the basement membrane ( Leimeister et al . , 2004 ) whose roles in normal physiology and pathology remain to be characterized . In this study , we report for the first time a functional role for SNED1 in tumor progression and metastasis and we now wish to understand the molecular mechanisms by which SNED1 contributes to tumor progression . Interestingly , SNED1 displays two potential integrin-binding sites , an RGD motif is found upstream of the amino-terminal NIDO domain and an LDV motif is upstream of the first EGF domain . EGLN1 ( also known as PHD2 , Prolyl Hydroxylase Domain-containing protein 2 ) is a prolyl-hydroxylase and , in normoxic conditions , catalyzes the hydroxylation of proline residues of the hypoxia-inducible factor 1 α ( HIF1α ) , rendering it susceptible to poly-ubiquitinylation and subsequent proteasomal degradation ( Berra et al . , 2003 ) . Klotsche-van Ameln et al . have reported that inhibiting EGLN1 had an anti-tumor effect by enhancing TGFβ anti-proliferative effect in a model of murine osteosarcoma ( Klotzsche-von Ameln et al . , 2011 ) . However , the extracellular role of this protein remains unknown . S100A2 is a calcium-binding protein that belongs to the S100 family of acute-phase response factors and , when overexpressed in non-small-cell lung carcinoma cells subsequently injected subcutaneously in mice , has been shown to promote metastasis to the lungs ( Bulk et al . , 2009 ) . The fact that we observed a decrease in the number of metastases formed by cells in which S100A2 was knockeddown , both when the cells were injected orthotopically and metastasized from the primary tumor and when the cells were injected in the circulation , suggests that S100A2 is important for the growth of tumor cells in distant organs , in particular the lungs . The proteomics-based discovery pipeline we developed identified , with high yield , proteins playing a causal role in tumor progression and metastatic dissemination . Indeed , in addition to the four proteins whose involvement in mammary carcinoma metastasis we report here , previous studies using various mammary carcinoma cell lines or tumor models had already implicated several other proteins in breast cancer progression ( ANGPTL4 , CTSB , IGFBP4 , LOXL2 ) or invasiveness of breast tumor cells ( ADAM9 , ADAM10 ) . In addition , analyses by Kaplan–Meier Plotter ( Györffy et al . , 2010 ) show that several proteins from our list of ECM proteins up-regulated in highly metastatic LM2 cells ( ADAM9 , LOXL2 , S100A2; data not shown ) show statistically significant correlations with poor prognosis for metastasis-free survival of breast cancer patients . Thus , in total , from the list of 43 ECM proteins that we find expressed only in the metastatic LM2 cells and not in MDA-MB-231 , 12 show suggestive evidence of functional involvement in metastasis of breast cancer . Also included are proteins previously implicated in tumor progression in other cancer types ( e . g . , CTGF , TIMP1 , S100A10 ) and proteins that we have identified in proteomics screens of other tumor models ( unpublished data ) . These concordances indicate that the proteins we identify here are relevant to breast cancer metastasis and their significance is likely not limited solely to the breast cancer lines we used in this study . To determine whether these proteins are regulated independently or might be coregulated , we conducted pathway analysis . Two main pathways , the TGFβ pathway and the HIF1α/VEGF pathway , appeared to be upstream of several of the proteins we identified . In addition to signal transduction pathways , microRNAs are another mechanism by which cells can rapidly control the expression or inhibition of sets of genes . Korpal et al . have identified in breast cancer patients the miR-200 family as being pro-metastatic via its inhibition of Sec23a , a protein involved in the secretion machinery ( Korpal et al . , 2011 ) . Interestingly , several of the proteins we found to be up-regulated in highly metastatic tumors ( AGRN , LTBP3 , EFEMP2 , IGFBP4 , SERPINE2 , SNED1 and TINAGL1 ) were shown in the Korpal study to be down-regulated in the conditioned medium of Sec23a knockdown cells and were thus postulated to be anti-metastatic . However , our study indicates that this set of proteins is pro-metastatic , which could point towards tumor-specific or context-dependent mechanisms of action for these proteins . Our study also revealed that not only the production of tumor-derived ECM proteins changes with a tumor’s metastatic potential but the stromal contribution to the tumor ECM also changes . This is consistent with the idea that the composition of the stroma ( possibly including the cell types present ) of a non-metastatic primary tumor differs from that of a highly metastatic primary tumor and that tumor cells of different metastatic potential may recruit different stromal cell types . In fact , we have observed that LM2 tumors are more vascularized than the MDA-MB-231 tumors ( data not shown ) . Our data also suggest that tumor cells of different metastatic potential can instruct local stromal cells to express different sets of ECM proteins . This interesting aspect of tumor–stroma interactions has not been extensively studied in vivo , although a recent paper notes that cancer-associated fibroblasts from different mammary tumor subtypes differ in their expression profiles ( Tchou et al . , 2012 ) . Finally , we believe that our approach offers the possibility of identifying proteins of clinical relevance . We showed that LTBP3 and SNED1 , identified by proteomics in our study as differentially expressed between poorly and highly metastatic mammary tumors , correlate at the transcript level with clinical outcome in a cohort of ER−/PR− breast cancer patient . We can envision three ways in which one could exploit the ECM to improve cancer diagnosis and prognosis and improve cancer patient care . As demonstrated for SNED1 and LTBP3 , other proteins that are part of the proteomic signatures of highly and poorly metastatic breast tumors described here , may prove to correlate with patient outcomes and may prove useful as novel prognostic and diagnostic biomarkers . Development of immuno- or mass spectrometry-based assays ( Gillette and Carr , 2013 ) for these proteins is therefore of interest since protein-level measurements may be of equivalent or greater predictive value than RNA expression analyses . This will require development of panels of specific antibodies against the ECM proteins of interest and their testing on larger numbers of well annotated patient samples to establish correlations with tumor stage and patient outcomes . ECM proteins are particularly favorable candidate biomarkers since they are abundant , are laid down in characteristic patterns and are readily accessible . ECM proteins can also serve as anchors to deliver selectively to tumors and/or metastases imaging probes to enhance the detection of metastases or anti-tumor agents to enhance therapy on the model of the work pioneered by Neri et al . ( Pasche and Neri , 2012 ) . Finally , another possible translational aspect is the identification of potential novel direct therapeutic targets . Whether by targeting the interactions between ECM proteins and their cellular receptors ( Goodman and Picard , 2012 ) or by disrupting the architecture of the ECM ( that acts as a barrier ) to allow more efficient drug delivery ( Yuan et al . , 2007; Whatcott et al . , 2011; Jacobetz et al . , 2013 ) , the proteins discovered in our differential proteomics-based approach have the potential to serve as novel anticancer therapeutic targets .
The human mammary carcinoma cells , MDA-MB-231 , and their highly metastatic variant MDA-MB-231_LM2 ( LM2 ) were a kind gift of Dr Joan Massague ( Memorial Sloan Kettering Cancer Center , New York , NY ) . The cells were grown in HyClone high-glucose Dulbecco’s modified Eagle’s medium ( Thermo Scientific , Pittsburgh , PA ) supplemented with 2 mM glutamine and 10% fetal bovine serum ( Invitrogen , Carlsbad , CA ) at 37°C in a 5% CO2 incubator . Tumor growth and spontaneous metastasis formation were assayed by transplanting tumor cells orthotopically into the mammary fat pad of 6 to 8-week-old female NOD/SCID/IL2Rγ-null mice ( Jackson Laboratory , Bar Harbor , ME ) . The mice were anesthetized by intraperitoneal ( IP ) injection of 125–250 mg/kg body weight of Avertin ( reconstituted in PBS ) , followed by IP injection of 100 μl of 12 μg/ml Buprenorphine for analgesia . A small incision was made in the right flank to expose the inguinal mammary gland , and 2 . 5 . 105 cells in suspension in 25 μl of Hank’s Balanced Salt Solution ( Invitrogen , Carlsbad , CA ) were injected . The mice received three additional IP injections of 100 μl of 12 μg/ml Buprenorphine at 12-hr intervals following the surgery . Surgery and mouse monitoring were performed according to the animal protocol approved by MIT’s Department of Comparative Medicine and Committee on Animal Care . Animals were sacrificed 7 ± 0 . 5 weeks post-injection and the tumors were dissected and weighed , flash frozen and kept at −80°C or fixed in 3 . 8% formaldehyde , imaged with a fluorescence microscope and subsequently embedded in paraffin and sectioned . In addition , secondary tumor sites ( lung , liver , spleen ) were collected and fixed in 3 . 8% formaldehyde for subsequent embedding in paraffin and sectioning . 5 . 104 cells in 100 μl of Hank’s Balanced Salt Solution were injected into the lateral tail vein of 6- to 8-week-old female NOD/SCID/IL2Rγ-null mice . The mice were sacrificed 4 weeks post-injection and lungs were inflated with 3 . 8% formaldehyde imaged with a fluorescence microscope and subsequently fixed overnight with 3 . 8% formaldehyde . Samples were kept in 70% ethanol prior to embedding and sectioning . ZsGreen-positive foci were counted in the left pulmonary lobe using the open-source software Cell Profiler 2 . 0 ( Lamprecht et al . , 2007; http://www . cellprofiler . org/ ) and counts were manually curated when needed . The Cell Profiler pipeline developed for this study is provided as a . cp file ( Supplementary file 1 ) . Student’s t test was performed to evaluate the statistical significance of the results . Sequential extractions of proteins from frozen tumor samples were performed using the CNMCS ( Cytosol/Nucleus/Membrane/Cytoskeleton ) compartmental protein extraction kit ( Cytomol , Union City , CA ) as previously described ( Naba et al . 2012 ) . This led to the extraction of intracellular soluble proteins and the enrichment of ECM proteins . Two independent biological replicates of each tumor type ( MDA-MB-231 or LM2 ) were analyzed . The ECM-enriched samples from each tumor were subsequently analyzed by mass spectrometry . 100–300 µg ECM-enriched pellets were solubilized and reduced in a solution of 8M urea , 100 mM ammonium bicarbonate pH 8 , 10 mM dithiothreitol , and incubated at 37°C for 30 min with continuous vortexing . After cooling to room temperature , cysteines were alkylated by adding iodoacetamide to 25 mM for 30 min . After diluting to 2M urea with 100 mM ammonium bicarbonate pH 8 . 0 , samples were deglycosylated with 1000–2000 units of PNGaseF ( New England BioLabs , Ipswich , MA ) and incubated at 37°C for 2 hr with continuous vortexing , followed by digestion with Lys-C ( Wako Chemicals USA , Inc . , Richmond , VA ) , at a ratio of 1:100 enzyme:substrate , with for 2 hr . Final digestion was done using trypsin ( Sequencing Grade , Promega , Madison , WI ) , at a ratio of 1:50 enzyme:substrate at 37°C overnight with continuous vortexing , followed by a second aliquot of trypsin , at a ratio of 1:100 enzyme:substrate , and an additional 2 hr of incubation . Solutions that began cloudy upon initial reconstitution were clear after overnight digestion . Samples were acidified and desalted using 10 mg HLB Oasis Cartridges ( Waters Corp . , Milford , MA ) eluted with 60% acetonitrile , 0 . 1% trifluoroacetic acid ( TFA ) , followed by concentration in a Speed-Vac . Approximately , 50 µg samples of peptide digest were fractionated using an Agilent 3100 OFFGEL Fractionator ( Agilent Technologies , Wilmington , DE ) and 13 cm Immobiline Drystrips pH 3–10 ( GE Healthcare BioSciences AB , Uppsala , Sweden , 17-6001-14 ) . Fractionation was performed according to the Agilent instruction manual . Briefly , peptides were diluted in IPG buffer , pH 3–10 ( GE Healthcare , 17-6000-87 ) , containing 5% glycerol . 150 µl of peptide solution were loaded into each of 12 wells and focused for 20 kV hr with a maximum current of 50 µA and power of 200 mW ( 24–36 hr ) . Focused solutions were pipetted out of each well and the wells were re-extracted with 30% acetonitrile/0 . 1% TFA . Fractions 9 and 10 were combined , yielding 11 total fractions for subsequent LC-MS/MS analysis . Fractions were acidified with TFA , cleaned-up using stage tips , that is , pipette tips packed with reversed-phase membrane disks ( Empore C-18 #2215 , 3M Corporation , St Paul , MN ) , eluted with 60% acetonitrile , 0 . 1% TFA , and then concentrated in a Speed-Vac ( Rappsilber et al . , 2003 ) . Tryptic digests were analyzed with an automated nano LC-MS/MS system , consisting of an Agilent 1100 nano-LC system ( Agilent Technologies , Wilmington , DE ) coupled to either an LTQ-Orbitrap or an LTQ Orbitrap XL Fourier transform mass spectrometer ( Thermo Fisher Scientific , San Jose , CA ) equipped with a nanoflow ionization source ( James A Hill Instrument Services , Arlington , MA ) . Peptides were eluted from a 10-cm column ( Picofrit 75 um ID , New Objectives ) packed in-house with ReproSil-Pur C18-AQ 3 μm reversed phase resin ( Dr Maisch , Ammerbuch Germany ) using either a 120 min gradient at a flow rate of 200 nl/min to yield ∼20 s peak widths . Solvent A was 0 . 1% formic acid and solvent B was 90% acetonitrile/0 . 1% formic acid . The elution portion of the LC gradient was 3–6% solvent B in 2 min , 6–31% B in 75 min , 31–60% B in 13 min , 60–90% B in 1 min , and held at 90% B for 5 min . Data-dependent LC-MS/MS spectra were acquired in ∼3 s cycles; each cycle was of the following form: one full Orbitrap MS scan at 60 , 000 resolution followed by 8 MS/MS scans in the ion trap on the most abundant precursor ions using an isolation width of 3 m/z . Dynamic exclusion was enabled with a mass width of ± 25 ppm , a repeat count of 1 and an exclusion duration of 45 s . Charge-state screening was enabled along with monoisotopic precursor selection and non-peptide monoisotopic recognition to prevent triggering of MS/MS on precursor ions with unassigned charge or a charge state of 1 . Normalized collision energy was set to 30 with an activation Q of 0 . 25 and activation time of 30 ms . All MS data was interpreted using the Spectrum Mill software package v4 . 1 beta ( Agilent Technologies , Santa Clara , CA ) . Similar MS/MS spectra acquired on the same precursor m/z within ± 60 s were merged , MS/MS spectra with precursor charge >4 and poor quality MS/MS spectra , which failed the quality filter by not having a sequence tag length >0 ( i . e . , minimum of two masses separated by the in-chain mass of an amino acid ) were excluded from searching . MS/MS spectra were searched against a UniProt database containing both human ( 78 , 369 entries ) and mouse ( 53 , 448 entries ) sequences . All sequences ( including isoforms and excluding fragments ) were downloaded from the UniProt website on 30 June 2010 . To each database a set of common laboratory contaminant proteins ( 73 entries ) was appended . Initial search parameters included: ESI linear ion-trap scoring parameters , trypsin enzyme specificity with a maximum of two missed cleavages , 35% minimum matched peak intensity , ± 20 ppm precursor mass tolerance , ± 0 . 7 Da product mass tolerance , and carbamidomethylation of cysteines and possible carbamylation of N-termini as fixed/mix modifications . Allowed variable modifications were oxidized methionine , deamidation of asparagine , pyro-glutamic acid modification at N-terminal glutamine , and hydroxylation of proline with a precursor MH + shift range of −18 to 97 Da . Hydroxyproline was only observed in the proteins known to have it ( collagen and proteins containing collagen domains; emilins , etc ) and only within the expected GXPG sequence motifs . Figure 3—source data 2 containing the detailed peptide spectral matches might have some examples not in the expected motif when there is either a proline near the motif for which the spectrum could have had insufficient fragmentation to confidently localize the mass change to a particular residue , or a nearby methionine in the peptide and the spectrum had insufficient fragmentation to localize the mass change to oxidized methionine or hydroxyproline . When the motif nX[ST] occurs in a peptide in Figure 3—source data 2 , this is likely to indicate a site where N-linked glycosylation was removed by the PNGaseF treatment of the sample . While a lowercase n indicates a gene-encoded asparagine residue detected in aspartic acid form , possible mechanisms of modification such as acid-catalyzed deamidation during sample processing vs enzymatic conversion during deglycosylation cannot be explicitly distinguished . Identities interpreted for individual spectra were automatically designated as confidently assigned using the Spectrum Mill autovalidation module to apply target-decoy-based , false-discovery rate ( FDR ) scoring threshold criteria via a two-step auto-threshold strategy at the spectral and protein levels . First , peptide mode was set to allow automatic variable range precursor mass filtering with score thresholds optimized to yield a spectral level FDR of 1 . 6% for each of the precursor charge states 2 , 3 , and 4 in each LC-MS/MS run . Second , protein mode was applied to further filter all the peptide-level validated spectra combined from the 22 LC-MS/MS runs from both replicates of an experiment using a minimum protein score of 20 and a maximum protein-level FDR of zero . Since the maximum peptide score is 25 , the protein-level step filters the results so that each identified protein is comprised of multiple peptides unless a single excellent scoring peptide was the sole match . The above criteria yielded false discovery rates of <1 . 0% for each sample at the peptide–spectrum match level and <1 . 2% at the distinct peptide level as estimated by target-decoy-based searches using reversed sequences . In calculating scores at the protein level and reporting the identified proteins , redundancy is addressed in the following manner: the protein score is the sum of the scores of distinct peptides . A distinct peptide is the single highest scoring instance of a peptide detected through an MS/MS spectrum . MS/MS spectra for a particular peptide may have been recorded multiple times , ( i . e . , as different precursor charge states , isolated from adjacent OGE fractions , modified by deamidation at Asn or oxidation of Met ) but are still counted as a single distinct peptide . When a peptide sequence >8 residues long is contained in multiple protein entries in the sequence database , the proteins are grouped together and the highest scoring one and its accession number are reported . In some cases , when the protein sequences are grouped in this manner there are distinct peptides which uniquely represent a lower scoring member of the group ( isoforms , family members , and different species i . e . , mouse vs human ) . Each of these instances spawns a subgroup and multiple subgroups are reported and counted towards the total number of proteins and in Figure 3—source data 1 , they are given related protein subgroup numbers ( see column AQ and AR I Figure 3—source data 1A , e . g . , Tenascin C ( TNC ) , the murine and human forms and are listed as subgroup members 25 . 1 and 25 . 2 respectively ) . Our in silico matrisome list was then used to categorize all of the identified protein subgroups as being ECM-derived or not . The reporting of the number of peptides contributing to each subgroup can be altered by enabling the subgroup-specific option in Spectrum Mill . This was done to report separately the species-specific peptides , the peptides common to both human and mouse , and the total of common and species-specific peptides . Relative abundances of proteins were determined using either the number of peptides or using the peptide abundance ( see legends for Figures 2 and 3 ) . When we sought to determine the relative abundance of human-derived and murine-derived proteins , we used extracted ion chromatograms ( XIC’s ) for each peptide precursor ion in the intervening high resolution FT-MS scans of the LC-MS/MS runs . An individual protein’s abundance was calculated as the sum of the ion current measured for all quantifiable peptide precursor ions with MS/MS spectra confidently assigned to that protein . Peptides were considered not quantifiable if they were shared across multiple subgroups of a protein or the precursor ions had a poorly defined isotope cluster ( i . e . , the subgroup-specific and exclude poor isotope quality precursor XIC’s filters in Spectrum Mill were enabled ) . Proteins were considered quantifiable if they were represented in two independent samples and represented by at least two distinct peptides in one of the two samples . The peak area for the XIC of each precursor ion subjected to MS/MS was calculated automatically by the Spectrum Mill software in the intervening high-resolution MS1 scans of the LC-MS/MS runs using narrow windows around each individual member of the isotope cluster . Peak widths in both the time and m/z domains were dynamically determined based on MS scan resolution , precursor charge and m/z , subject to quality metrics on the relative distribution of the peaks in the isotope cluster vs theoretical . Although the determined protein ratios are generally reliable to within a factor of twofold of the actual ratio , numerous experimental factors contribute to variability in the determined abundance for a protein . These factors may include incomplete digestion of the protein; widely varying response of individual peptides due to inherent variability in ionization efficiency and interference/suppression by other components eluting at the same time as the peptide of interest , differences in instrument sensitivity over the mass range analyzed , and inadequate sampling of the chromatographic peak between MS/MS scans . The original mass spectra may be downloaded from MassIVE ( http://massive . ucsd . edu ) using the identifier: MSV000078535 . The data should be accessible in the ‘Spectrum’ folder at ftp://MSV000078535:a@massive . ucsd . edu . 97-nucleotide miR30-based shRNAs targeting each gene of interest ( Supplementary file 2 ) were designed using the shRNA design software available through the Cold Spring Harbor Laboratory website ( http://katahdin . mssm . edu/siRNA/RNAi . cgi ? type=shRNA ) , and cloned into the MSCV-ZSG-2A-Puro-miR30 vector retroviral vector as described previously ( Stern et al . , 2008; Lamar et al . , 2012 ) . This vector expresses the miR30-based shRNA in the 3’UTR of a single transcript encoding the ZsGreen reporter gene and the puromycin-resistance gene . Retroviruses were packaged in 293FT cells and LM2 cells were transduced as previously described ( Stern et al . , 2008; Lamar et al . , 2012 ) . For qPCR , RNA was isolated from cell or tumor lysates using RNeasy kit ( Qiagen , Germantown , MD ) and cDNA was synthesized by reverse transcription using the First-Strand cDNA Synthesis Kit ( Promega , Madison , WI ) . qPCR reactions were performed using Bio-Rad SYBR Green Supermix ( Bio-Rad , Hercules , CA ) according to the manufacturer’s instructions . PCR conditions were 95°C for 10 min , followed by 40 cycles of 95°C for 20 s , 58°C for 30 s , and 72°C for 30 s . qPCR data analysis was performed using Bio-Rad CFX Manager Software . Human and murine PCR primers used are listed in Supplementary file 3 . Tumor samples were formaldehyde-fixed and paraffin-embedded . Sections were dewaxed and rehydrated following standard procedures . When required , heat-induced epitope retrieval was performed by incubating sections in 10 mM sodium citrate buffer ( pH6 . 0 ) heated at 95°C for 20 min . Sections were cooled down at room temperature and were then stained using appropriate Vectastain ABC kits ( Vector Laboratories , Burlingame , CA ) . Primary antibodies used were: mouse anti-human vimentin antibody ( Leica , Davie , FL ) , mouse anti-Ki67 antibody ( Vector Laboratories ) , rabbit anti-cleaved caspase 3 ( Cell Signaling , Danvers , MA ) , rabbit anti-CD31 ( PECAM ) antibody ( Abcam , Cambridge , MA ) , rabbit anti-LTBP3 ( Santa Cruz Biotechnology , Dallas , TX ) , Sections were subsequently counterstained with methyl green or hematoxylin . Hematoxylin and eosin and Masson’s trichrome stainings were performed following standard procedures . Tumor samples were lysed in Laemmli buffer , proteins were separated by SDS-PAGE and immunoblotting was performed using the following antibodies: rabbit anti-collagen I ( Millipore , Billerica , MA ) , mouse anti-transferrin receptor ( Invitrogen ) , mouse anti-GAPDH ( Millipore ) , rabbit anti-LTBP3 ( Santa Cruz Biotechnology ) , rabbit anti-EGLN1 ( Cell Signaling ) , the rabbit anti-actin antibody was generated in the laboratory . The rabbit anti-CYR61 antibody was a kind gift of Dr Lester Lau . DNA was extracted from formaldehyde-fixed and paraffin-embedded tissues using the QIAamp DNA FFPE Tissue Kit ( Qiagen ) . Quantitative PCR was conducted on 1 ng of genomic DNA . Primers used are: Alu Forward: 5′GTGAAACCCCGTCTCTACTAAAAATACAAA3′ , Alu Reverse: 5′GCGATCTCGGCTCACTGCAA3′ . qPCR reactions were performed using Bio-Rad SYBR Green Supermix ( Bio-Rad ) according to the manufacturer’s instructions . PCR conditions were 95°C for 10 min , followed by 40 cycles of 95°C for 20 s , 58°C for 30 s , and 72°C for 30 s qPCR data analysis was performed using Bio-Rad CFX Manager Software . Murine actin served as a reference to normalize qPCR data . Ingenuity Pathway Analysis ( http://www . ingenuity . com/products/ipa ) was used to identify common upstream regulators ( ‘Results’ ) . Kaplan–Meier Plotter ( http://www . kmplot . com/ ) , an online-available meta-analysis tool ( Györffy et al . , 2010 ) , was used to test possible correlations between expression of the genes encoding proteins identified by proteomics in our study ( ‘Results’ ) with distant metastasis-free survival . We interrogated the data from all patients or only the subset of estrogen-receptor-negative and progesterone-receptor-negative patients included on the multiple clinical data sets available . | Metastasis is the process whereby tumor cells spread within the body and is the cause of most deaths from cancer . This complex process involves several steps: first the cancer cells invade the tissues that surround the tumor; second , the cancer cells enter the blood stream and travel throughout the body; and third , the cancer cells seed the growth of new tumors in distant organs . Within tissues , the extracellular matrix forms a complex scaffold of proteins that surrounds cells , to support and organize them: it also provides signals that control how much cells can multiply , how likely cells are to stick together or migrate , and even a cell’s chances of survival . Pathologists have used an accumulation of extracellular matrix proteins in tumors as a sign that the outcome of the disease will likely be unfavorable for a patient , and that treatment will be challenging . However , we still do not have a clear picture of the composition of the tumor extracellular matrix and we do not know all the details of how it affects tumor growth and metastasis . Now , Naba et al . have explored these questions by injecting different types of human breast tumor cells into mice . Some of the cells were capable of spreading throughout the body and were said to have a high ‘metastatic potential’; others were less capable of spreading and were said to have a low metastatic potential . Naba et al . then analyzed the proteins that made up the extracellular matrix of the tumors that grew in the mice . Some proteins were found in both types of tumor; whereas some proteins were only found in the tumors with low metastatic potential and some were only found in the highly metastatic tumors . Naba et al . also demonstrated that both cancer cells and non-cancer cells—which are also found within the tumors—contributed to the production of the extracellular matrix in the tumor . Moreover , and somewhat surprisingly , the contributions from the non-cancer cells in the two types of tumors were also different . Computational analysis predicted that the production of several extracellular matrix proteins in the highly metastatic tumors was under the control of signaling pathways that are involved in cancer progression . Furthermore , Naba et al . also demonstrated that several of the extracellular matrix proteins specific to highly metastatic tumors were required for the cancer to spread . These proteins are involved in different stages of the metastatic process , and some of them are commonly over-produced in tumors from patients with some of the worst chances of recovery . If similar results are consistently observed in clinical samples from humans , the work of Naba et al . could help doctors to discriminate between tumors that will spread and those that will not , which should lead to improved patient care . The proteins and pathways associated with the highly metastatic tumors could be also investigated as potential drug targets . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"cancer",
"biology"
] | 2014 | Extracellular matrix signatures of human mammary carcinoma identify novel metastasis promoters |
A central feature of most stem cells is the ability to self-renew and undergo differentiation via asymmetric division . However , during asymmetric division the role of phosphatidylinositol ( PI ) lipids and their regulators is not well established . Here , we show that the sole type I PI transfer protein , Vibrator , controls asymmetric division of Drosophilaneural stem cells ( NSCs ) by physically anchoring myosin II regulatory light chain , Sqh , to the NSC cortex . Depletion of vib or disruption of its lipid binding and transfer activities disrupts NSC polarity . We propose that Vib stimulates PI4KIIIα to promote synthesis of a plasma membrane pool of phosphatidylinositol 4-phosphate [PI ( 4 ) P] that , in turn , binds and anchors myosin to the NSC cortex . Remarkably , Sqh also binds to PI ( 4 ) P in vitro and both Vib and Sqh mediate plasma membrane localization of PI ( 4 ) P in NSCs . Thus , reciprocal regulation between Myosin and PI ( 4 ) P likely governs asymmetric division of NSCs .
Understanding how neural stem cells divide asymmetrically is central for stem cell and cancer biology . Drosophila neural stem cells ( NSC ) , or neuroblasts , are an excellent model for understanding stem cell asymmetry and homeostasis . Each neuroblast divides asymmetrically to generate a self-renewing neuroblast and a differentiating daughter cell . It is the latter that ultimately produces neurons/glia ( Doe , 2008 ) . Disruption of asymmetric division in neuroblasts may result in the formation of ectopic neuroblasts , leading to brain tumor development ( Caussinus and Gonzalez , 2005; Lee et al . , 2006; Wang et al . , 2006; Wang et al . , 2007; Chabu and Doe , 2009; Wang et al . , 2009; Chang et al . , 2010; Wang et al . , 2011 ) . Asymmetric division of neuroblasts depends on the polarized distribution of proteins and their asymmetric segregation into different daughter cells . Apical proteins such as atypical protein kinase C ( aPKC ) , Par-6 ( Partitioning defective 6 ) , Par-3/Bazooka ( Baz ) , Inscuteable ( Insc ) and Partner of Inscuteable ( Pins ) control the localization of basal proteins , as well as the orientation of the mitotic spindle ( Knoblich , 2010; Chang et al . , 2012 ) . Basal cell fate determinants Numb , Prospero ( Pros ) , Brain tumor ( Brat ) , and their adaptor proteins Partner of Numb ( Pon ) and Miranda ( Mira ) are critical for neuronal differentiation upon asymmetric segregation into the differentiating daughter cell ( Gonzalez , 2007; Doe , 2008; Knoblich , 2010 ) . Asymmetric localization of Mira was thought to be achieved via a series of linear inhibitory regulations from aPKC to Mira via Lethal giant larvae ( Lgl ) and nonmuscle Myosin II ( Kalmes et al . , 1996; Barros et al . , 2003; Betschinger et al . , 2003 ) . aPKC was shown later to directly phosphorylate both Numb and Mira to polarize them in neuroblasts ( Smith et al . , 2007; Atwood and Prehoda , 2009 ) , while Lgl directly inhibits aPKC in neuroblasts , rather than displacing Mira on the cortex ( Atwood and Prehoda , 2009 ) . However , how myosin regulates asymmetric division of neuroblasts remains not well understood . Despite having identified phosphatidylinositol ( PI ) lipids as critical components of cellular membranes important for cell polarity of various cell types ( Krahn and Wodarz , 2012 ) , the role of PI lipids and their regulators are not well established in Drosophila neuroblasts . Phosphoinositides are phosphorylated derivatives of PI , and their synthesis is catalyzed by lipid kinases; specially , phosphoinositide 4-phosphate [PI ( 4 ) P] production is catalyzed by PI 4-OH kinases ( PI4K ) . Interestingly , these lipid kinases are inherently inefficient enzymes and production of biologically sufficient amounts of PI ( 4 ) P for efficient signaling requires the stimulation of these enzymes by phosphatidylinositol transfer proteins ( PITPs ) . PITPs leverage their lipid exchange activities to present PI as a superior substrate to PI4Ks ( Bankaitis et al . , 2010; Grabon et al . , 2015 ) . Classical PITPs bind to PI and phosphatidylcholine ( PC ) in a mutually exclusive manner , with PI being the preferred binding substrate ( Wirtz , 1991 ) . Classical PITPs are also divided into two different subtypes , type I and type II , both of which contain PITP domains with type II PITPs containing additional N-termini of large membrane-associated modules ( Nile et al . , 2010 ) . Loss of PITP function is associated with neurological disorders , including cerebellar ataxia ( Hsuan and Cockcroft , 2001; Cockcroft and Garner , 2011 ) . These deficits are consistent with the fact that PITPs are particularly highly expressed in the brain and cerebellum ( Utsunomiya et al . , 1997 ) . In mammals , type I PITPs consist of PITPα and PITPβ . Mouse hypomorphic mutants for PITPα/Vibrator exhibit a progressive , and ultimately fatal , whole-body tremor that reflects neurodegenerative disease , whereas mice with the PITPα structural gene deleted survive to term but die shortly after birth ( Hamilton et al . , 1997; Alb et al . , 2003 ) . Attempted rescue of PITPα-null mice with a PITPα mutant specifically defective in PI-binding exhibit phenotypes indistinguishable from those of the null mutant ( Alb et al . , 2007 ) . In Drosophila , a single type I PITP has been identified , named Vibrator/Giotto , which plays a prominent role in cytokinesis of spermatocytes and neuroblasts ( Gatt and Glover , 2006; Giansanti et al . , 2006 ) . However , the role of PITPs in neuroblast polarity is unknown . Herein , we show that the type I PITP Vibrator regulates asymmetric division of neuroblasts through anchoring nonmuscle myosin II regulatory light chain ( RLC ) , Sqh , to the cell cortex . We further demonstrate that depletion of PI4KIIIα ( a PI4 kinase that is essential for PI ( 4 ) P synthesis ) also leads to Sqh delocalization and asymmetric division defects . Thus , Vibrator and PI4KIIIα likely promote synthesis of a plasma membrane PI ( 4 ) P pool that , in turn , binds and anchors myosin to the cell cortex . Importantly , we show that Sqh is able to bind to phospholipid PI ( 4 ) P and facilitates its membrane localization in neuroblasts . Taken together , the data identify a reciprocal dependence between Myosin and PI ( 4 ) P localization in neuroblasts .
To identify novel regulators of neuroblast homeostasis , we performed a genetic screen on a collection of chromosome 3R mutants induced by ethyl methane sulfonate ( EMS ) mutagenesis ( CT Koe , F Yu and H Wang , unpublished ) . We identified two independent alleles of vibrator , which were designated as vibrator133 ( vib133 ) and vibrator1105 ( vib1105 ) . While the mutation ( s ) in vib1105 is/are yet to be identified , vib133 contains a mutation ( AG->TG ) at the splice acceptor site of its last intron ( Figure 1A ) , presumably causing a premature stop codon . Hemizygous vib133 or vib1105 over deficiency Df ( 3R ) BSC850 survived to 2nd instar larval stage , similar to hemizygous vibj7A3 , a reported protein null allele , over a deficiency Df ( 3R ) Dl-BX12 ( Gatt and Glover , 2006 ) . Both hemizygotes , vib133/Df ( 3R ) BSC850 and vib1105/Df ( 3R ) BSC850 , die earlier than hemizygous vibj5A6 , a P-element insertion allele , over Df ( 3R ) BSC850 , which survived to 3rd instar larval stage . Similar to the Vib protein levels in vibj7A3/Df ( 3R ) BSC850 ( 12% ) , Vib protein levels with the predicted size 32 KDa in the larval brains of hemizygous vib133/Df ( 3R ) BSC850 and vib1105/Df ( 3R ) BSC850 were both significantly reduced to 9% compared with control ( Figure 1—figure supplement 1A , n = 2 ) . The detection of trace amounts of Vib in vibj7A3/Df ( 3R ) BSC850 was likely due to the higher sensitivity of the new anti-Vib antibody we generated compared with previously reported antibody . Truncated Vib protein with predicted size of 26 KD was not detected in vib133/vibj5A6 protein extracts ( Figure 1—figure supplement 1A ) , suggesting that the truncated form might be unstable . Taken together , these results suggest that vib133 and vib1105 are most likely strong hypomorphic alleles . To analyze neuroblast homeostasis in these vib alleles , we generated MARCM clones of both type I and type II neuroblasts ( Lee et al . , 2000 ) . Type I wild-type control clones always contain one neuroblast that expresses pan-neural genes Deadpan ( Dpn ) and Asense ( Ase ) ( Figure 1B , n = 38 ) , while type II clones possesses one neuroblast that expresses Dpn but not Ase ( Figure 1C , n = 34 ) . By contrast , we observed that 56% ( n = 41 ) of type I vib133 clones and 61% ( n = 41 ) of type II vib133 clones showed ectopic neuroblasts ( Figure 1B , C ) . In addition , 28 . 7% ( n = 115 ) vib133 clones lost neuroblasts ( Figure 4—figure supplement 2A ) . Similarly , 28 . 3% ( n = 99 ) vib1105 clones were devoid of neuroblasts , with ectopic neuroblasts observed in 20% ( n = 35; data not shown ) of type I vib1105 clones and 50% ( n = 36 ) of type II vib1105 clones ( Figure 1B , C ) . Loss of neuroblasts upon Vib depletion is unlikely due to cell death , because active Caspase-3 is absent in both wild-type control and vib133 neuroblasts ( Figure 1—figure supplement 2A ) . Rather , it is likely partially due to premature differentiation of neuroblasts as none ( n = 15 ) of type I neuroblast control clones expressed nuclear Pros , a differentiation factor , while 66 . 7% ( n = 12 ) of type I vib133 neuroblast clones exhibited nuclear Pros ( Figure 1—figure supplement 2B ) . Furthermore , in vib133/j5A6 and vib1105/j5A6 mutants , 50% ( n = 158 ) and 56 . 8% ( n = 243 ) of interphase neuroblasts exhibited nuclear Pros expression , respectively ( Figure 1—figure supplement 2C; control , no nuclear Pros , n = 121 ) . Given that the nuclear Pros observed in vib- neuroblasts was very weak , we cannot exclude the possibility that other yet-to-be-identified factors are responsible for the premature differentiation of vib- neuroblasts . The average size of vib133 neuroblasts ( 7 . 76 μm , n = 27 ) is smaller than neuroblasts from control MARCM clones ( 10 . 79 μm , n = 37 ) ( Figure 1—figure supplement 2D ) . We measured the ratio of nucleolar to nuclear size in neuroblasts , which is an indicator of cellular growth . With nucleolus: nucleus ratio in wild-type control neuroblasts normalized as 1 ( n = 29 ) , this ration in vib133 neuroblasts was reduced slightly to 0 . 81 ( n = 28 ) ( Figure 1—figure supplement 2E ) . This result suggest that cell growth in vib- mutant neuroblasts is mildly reduced , likely due to the nuclear localization of Pros resulting in earlier termination of proliferation as well as premature differentiation . We also observed cytokinesis defects as previously reported in vib mutants ( Gatt and Glover , 2006; Giansanti et al . , 2006 ) ( Figure 4—figure supplement 1A and data not shown ) . Our evidence suggest that ectopic neuroblasts are unlikely caused by cytokinesis defects ( refer to page 13–14 ) . Quantification of neuroblast growth was performed on those clones without obvious cytokinesis block . We fused wild-type Vib with Venus at the C-terminus and generated transgenic flies expressing Vib-Venus . Ectopic neuroblasts in either mutant vib alleles were well rescued by expressing Vib-Venus . Ectopic neuroblasts were rescued in 91 . 3% of type I vib133 ( n = 23 ) , 100% of type II vib133 ( n = 14 ) , 100% of type I vib1105 ( n = 46 ) and 96% of type II vib1105 ( n = 25 ) neuroblast lineages ( Figure 1B and Figure 1—figure supplement 1B ) . Together , these results suggest that Vib regulates neuroblast homeostasis in both type I and type II neuroblast lineages . By performing pairwise sequence alignment ( EMBOSS Needle ) between Vib with rat PITPs , we observed 59 . 6% and 61 . 8% identity between Vib with rat PITPα and PITPβ , respectively . Given this sequence homology , we next explored if mammalian PITP homologs could replace Vib function in Drosophila neuroblasts . To this end , we generated transgenic flies expressing either rat PITPα or PITPβ , which were each fused with Venus at the C-terminus . Expression of PITPα completely rescued ectopic neuroblasts in both type I ( n = 43 ) and type II ( n = 26 ) MARCM clones of vib133 ( Figure 1—figure supplement 1C ) . Likewise , expression of PITPβ in vib133 fully rescued ectopic neuroblasts in both type I ( n = 100 ) and type II ( n = 39 ) MARCM clones ( Figure 1—figure supplement 1C ) . These results suggest that mammalian PITPs may play a conserved role in neuroblast homeostasis . Vib was previously reported to localize to both the cleavage furrow and spindle envelope in spermatocytes and only to the spindle envelope in neuroblasts ( Gatt and Glover , 2006; Giansanti et al . , 2006 ) . Since the anti-Vib antibody did not work consistently in neuroblasts , we analyzed Vib-Venus expression under the control of insc-Gal4 . Vib-Venus was strongly localized to the cell cortex and weakly to the spindle envelope throughout the cell division cycle ( Figure 1—figure supplement 1D ) . In telophase neuroblasts , Vib-Venus was cortically localized and slightly enriched at the cleavage furrow ( Figure 1—figure supplements 1D , 20% , n = 10 ) . To determine if Vib-Venus is enriched at the basal side , we measured the pixel intensity of GFP at both the apical and basal cortex of neuroblasts . At metaphase and telophase , neuroblasts showed an average VenusBasal: VenusApical ratio of 1 . 26 ( n = 32 ) and 1 . 39 ( n = 33 ) , respectively . In addition , the average ratio of Vib-Venus intensity at cleavage furrow to that at the apical cortex ( termed VenusCF: VenusApical ) of telophase neuroblasts was 1 . 33 ( n = 33 ) ( Figure 1—figure supplement 1F ) . The slight enrichment of Venus-Vib at the basal side was likely due to the attachment of multiple GMCs surrounding the neuroblasts , which is commonly seen for membrane proteins . Given that neuroblast homeostasis is disrupted in vib mutants , we assessed if Vib is required for asymmetric division of neuroblasts . In wild-type control , metaphase neuroblasts localize Par complex , Insc and the Pins-Gαi complex to the apical cortex , while displacing Mira-Brat-Pros and Numb-Pon complexes to the basal cortex ( Figure 2A–D , Figure 2—figure supplement 1A–C , and data not shown; 100% , n = 42 ) . By contrast , in 76% of both vib133/j5A6 and vib1105/j5A6 metaphase neuroblasts , aPKC was delocalized from the apical cortex to the cytoplasm , and exhibited weak or punctate crescent profiles , or was uniformly distributed on the cell cortex ( Figure 2A , C; n = 42 for both trans-heterozygotes ) . Consistent with the delocalization of aPKC , we also observed autophosphorylated aPKC ( p-aPKC ) dispersed from the apical cortex into the cytoplasm in vib133/j5A6 ( 57% , n = 21 ) and vib1105/j5A6 ( 53% , n = 32 ) metaphase neuroblasts ( Figure 2—figure supplement 1A , C ) . Since aPKC is important for neuroblast self-renewal ( Rolls et al . , 2003 ) , delocalization of aPKC into the cytoplasm is likely responsible for the loss-of-neuroblast phenotype observed in vib mutants . Similarly , 56% ( n = 39 ) and 57 . 4% ( n = 47 ) of metaphase neuroblasts with vib133/j5A6 and vib1105/j5A6 , respectively , showed delocalization of Par-6 into the cytoplasm , or exhibited as weak/punctate crescent ( Figure 2A , C ) . Membrane localization of apical protein Baz is mediated by direct interaction of its C-terminal region with two phosphatidylinositol phosphates ( PIPs ) , namely PI ( 3 , 4 , 5 ) P3 and PI ( 4 , 5 ) P2 ( Heo et al . , 2006; Krahn et al . , 2010 ) . Surprisingly , Baz localization remained unaffected in vib133/j5A6 ( Figure 2A , C , 97 . 5% , n = 68 ) and vib1105/j5A6 ( Figure 2A , C , 100% , n = 65 ) metaphase neuroblasts . This observation suggests that Vib controls localization of aPKC and Par-6 via a mechanism largely independent of Baz . An alternative possibility is that sufficient PI ( 4 , 5 ) P2 and PI ( 3 , 4 , 5 ) P3 remained in those vib mutant neuroblasts to support apical localization of Baz . Furthermore , Gαi was partially delocalized to the cytoplasm ( Figure 2—figure supplements 1A , C , 18 . 6% , n = 59 for vib133/j5A6 and 21 . 3% , n = 80 for vib1105/j5A6 ) , while localization of Insc ( n = 38 for vib133/j5A6 , n = 78 for vib1105/j5A6 ) , Pins ( n = 34 for vib133/j5A6 , n = 50 for vib1105/j5A6 ) and Cdc42 ( n = 24 , in vib133/j5A6 ) was largely unaffected ( data not shown ) . Taken together , the data demonstrate that Vib is particularly important for the localization of aPKC and Par-6 in neuroblasts . Next , we examined the localization of basal proteins in vib mutant neuroblasts . While Mira was asymmetrically localized in 100% of wild-type neuroblasts during metaphase ( Figure 2B , D , n = 60 ) , its localization was severely disrupted in metaphase neuroblasts of vib133/j5A6 and vib1105/j5A6 . It was either delocalized to the cytoplasm or mis-localized to the mitotic spindle ( Figure 2B , D , vib133/j5A6 , 85 . 6% , n = 118; vib1105/j5A6 , 90 . 4% , n = 42 ) . In wild-type neuroblasts , Pros and Brat , two cargo proteins of Mira , were basally localized ( Chang et al . , 2012 ) ( Figure 2B , D , Figure 2—figure supplements 1B , C , 100% , n = 50 ) . However , Pros ( Figure 2B , D , vib133/j5A6 , 47% , n = 36; vib1105/j5A6 , 49% , n = 47 ) and Brat ( Figure 2—figure supplement 1B , C , vib133/j5A6 , 42% , n = 102; and vib1105/j5A6 , 33 . 3%; n = 33 ) were delocalized to the cytoplasm or mitotic spindle in vib metaphase neuroblasts . In wild-type controls , cell fate determinant Numb is basally localized in metaphase neuroblasts ( Figure 2B , D , n = 53 ) . However , 65 . 5% ( n = 63 ) of vib133/j5A6 and 65% ( n = 98 ) of vib1105/j5A6 metaphase neuroblasts showed punctate crescent or cytoplasmic localization of Numb ( Figure 2B , D ) . Furthermore , 30% ( n = 10 ) and 22 . 5% ( n = 80 ) of metaphase neuroblasts showed cytoplasmic localization of PON in vib133/j5A6 and vib133/j5A6 , respectively ( Figure 2—figure supplement 1B , C ) . At telophase , mis-localized apical or basal proteins can be corrected and localize asymmetrically in a phenomenon named ‘telophase rescue’ ( Peng et al . , 2000 ) . While apical localization of aPKC was largely restored at telophase in vib neuroblasts ( vib133/j5A6 , 97 . 7% , n = 43; and vib1105/j5A6 , 100% , n = 23 ) , Mira ( 94%; n = 101 for vib133/j5A6 , 95 . 6%; n = 23 for vib1105/j5A6 ) , Pros ( Figure 4—figure supplements 2B , C , 54 . 8%; n = 31 for vib133/j5A6 , 71 . 4%; n = 21 for vib1105/j5A6 ) and Brat ( 42%; n = 19 for vib133/j5A6 ) remained on the mitotic spindle , in addition to their basal localization ( Figure 2E and data not shown ) . Consistently , 44% of metaphase neuroblasts of vibj5A6/Df ( 3R ) BSC850 hemizygotes showed cytoplasmic aPKC localization ( Figure 2—figure supplement 1D , n = 50 ) and 94% showed Mira delocalization ( Figure 2—figure supplement 1D , n = 50 ) . Mira remained mis-localized to the mitotic spindle in 85% of vibj5A6/Df ( 3R ) BSC850 telophase neuroblasts ( Figure 2—figure supplement 1D , n = 27 ) . The ratio of segregated apical/basal proteins determines whether cells adopt neuroblast or GMC fate following the neuroblast division ( Cabernard and Doe , 2009 ) . Given that Mira/Pros/Brat was still observed at the basal side of the cortex , it is unclear whether the formation of ectopic neuroblasts is partially contributed by the reduced amount of basal determinants eventually segregated into basal daughter cell due to their mislocalization on the spindle . Taken together , we conclude that Vib is required for localization of apical and basal protein and their faithful segregation in dividing neuroblasts . We ascertained whether spindle orientation was affected in mitotic neuroblasts of vib mutants . To this end , we measured the angle between apicobasal axis inferred by the Insc cortical crescent and mitotic spindle axis in metaphase neuroblasts . In wild-type control metaphase neuroblasts , the mitotic spindle is always parallel to the apicobasal polarity ( Figure 2F , n = 66 ) . By contrast , we observed that 44 . 2% of metaphase neuroblasts in vib133/j5A6 showed mis-orientation of the mitotic spindle axis with respect to neuroblast polarity ( Figure 2F , n = 86 ) . Likewise , 40 . 4% of metaphase neuroblasts in vib1105/j5A6 showed mitotic spindle mis-orientation ( Figure 2F , n = 47 ) . We observed orthogonal division in vib133/j5A6 metaphase neuroblasts ( Figure 2F; 2 . 3% , n = 86 ) , suggesting that spindle orientation defects likely contributed , at least partially , to the altered sibling cell fate observed in vib mutants . With the known function of Gαi in mitotic spindle orientation ( Schaefer et al . , 2001; Yu et al . , 2003 ) , partial delocalization of Gαi ( Figure 2—figure supplement 1A , C ) likely contributed to the spindle orientation phenotype in vib mutants . Centrosomal Mud localization was largely unaffected in vib133/j5A6 metaphase neuroblasts ( Figure 2—figure supplement 1E; 96 . 3% , n = 27; Control , n = 23 ) . Centrosomes in vib133/j5A6 neuroblasts were unaffected , as Cnn localization is normal in both control ( n = 24 ) and vib133/j5A6 ( n = 47 ) neuroblasts ( Figure 2—figure supplement 1F ) . Likewise , the spindle architecture is unaffected in vib133/j5A6 ( Figure 2—figure supplement 1G; n = 43; Control , n = 25 ) . They were able to assemble normal-looking mitotic spindle and astral microtubules that were labeled by α-tubulin . Taken together , Gαi delocalization , but not spindle architecture or centrosomal defects , most likely caused the spindle orientation defects observed in vib mutants . The non-muscle myosin II regulatory light chain protein Spaghetti-squash ( Sqh ) regulates both asymmetric localization of basal cell fate determinants in neuroblasts and cytokinesis of dividing cells ( Karess et al . , 1991; Barros et al . , 2003 ) . In particular , Mira was delocalized to the mitotic spindle in sqh- embryonic neuroblasts ( Barros et al . , 2003 ) . In the null allele , sqhAx3 , neuroblasts failed to divide in MARCM clones due to severe cell division defects , precluding us from analyzing their asymmetric division . To circumvent this problem , we knocked down sqh by RNAi under insc-Gal4 . Upon sqh RNAi knockdown Mira was delocalized to the cytoplasm in larval neuroblasts ( Figure 3—figure supplements 1A , 14 . 7% , n = 34 ) . We also observed aPKC delocalization ( Figure 3—figure supplements 1A , 16 . 2% , n = 37 ) in sqh RNAi neuroblasts . Our data confirmed the role of Sqh in regulating asymmetric division of larval brain neuroblasts . We next investigated whether the localization of Sqh is dependent on Vib function . To this end , we examined localization of GFP-tagged Sqh under its endogenous promotor ( Sqh::GFP ) in vib133/j5A6 and vib1105/j5A6 . In wild-type larval brains , Sqh::GFP was largely cytoplasmic at interphase with 38% cortical localization ( Figure 3—figure supplements 1B , 38% , n = 95 ) , while localizing uniformly to the cortex of neuroblasts at metaphase ( Figure 3A , B , 100% , n = 53 ) . As mitosis progressed , its distribution was restricted to the cleavage furrow at telophase ( Cabernard et al . , 2010 ) ( Figure 3A , B ) . However , in vib133/j5A6 , Sqh::GFP was observed in the cytoplasm in 92% of interphase neuroblasts ( Figure 3—figure supplement 1B , n = 123 ) and 42 . 6% of metaphase neuroblasts ( Figure 3A , B , n = 61 ) . In 29 . 4% of vib133/j5A6 telophase neuroblasts , Sqh::GFP failed to be accumulated at the cleavage furrow and became uniformly cortical and cytoplasm ( Figure 3A , B , n = 34 ) . Likewise , in vib1105/j5A6 , cortical localization of Sqh::GFP was diminished in 46 . 8% of metaphase neuroblasts ( Figure 3A , B , n = 79 ) , while 24 . 2% of telophase neuroblasts had increased cytoplasmic localization of Sqh::GFP ( Figure 3A , B , n = 33 ) . Consistently , Sqh localization examined by Sqh antibody revealed that its cortical localization at metaphase ( Figure 3—figure supplements 1C , 75% , n = 24 ) was disrupted in vib133/j5A6 neuroblasts . Myosin activity is regulated by phosphorylation of myosin regulatory light chain ( MRLC ) ( Ikebe and Hartshorne , 1985; Ikebe et al . , 1986; Ikebe et al . , 1988 ) . Drosophila MRLC Sqh has two phosphorylated forms , monophosphorylated at Ser21 ( Sqh1P ) or diphosphorylated at both Thr20 and Ser21 ( Sqh2P ) ( Zhang and Ward , 2011 ) . In imaginal discs , Sqh1P largely localizes to the adherens junction , while Sqh2P localizes to the apical domain ( Zhang and Ward , 2011 ) . To determine if the localization of active form of Sqh is affected in vib mutant , we examined Sqh1P in vib133/j5A6 , as the Sqh2P antibody is no longer available . During interphase , similar to Sqh::GFP , Sqh1P was largely cytoplasmic in both control and vib133/j5A6 neuroblasts ( Figure 3C , D ) . In wild-type mitotic neuroblasts , Sqh1P was seen at both the cell cortex ( 69% , n = 52 ) and as diffuse cytoplasmic staining at metaphase ( 31% , n = 52 ) , and enriched at the cleavage furrow at telophase ( Figure 3C , D , 84% , n = 25 ) . By contrast , Sqh1P was cytoplasmic in the vast majority of metaphase neuroblasts ( 92 . 5% , n = 40 ) and it failed to enrich at the cleavage furrow in telophase neuroblasts ( 62 . 5% , n = 24 ) of vib133/j5A6 ( Figure 3C , D ) . Taken together , these observations indicate that Vib is required for the localization of Sqh to the cell cortex in mitotic neuroblasts . In neuroblasts apical Rho kinase ( Rok ) phosphorylates and activates Sqh , ensuring cleavage furrow positioning ( Tsankova et al . , 2017 ) . To test whether Vib is necessary for the correct localization and activation of Rok , we first determined localization of Rok-GFP in vib mutant neuroblasts . In controls ( Ubi-Rok-GFP heterozygous ) , neuroblasts expressed cortical Rok-GFP at metaphase ( 93 . 5% , n = 46 ) and showed enriched Rok-GFP at the cleavage furrow in telophase ( 91 . 7% , n = 24 ) . Similarly , in vib133/j5A6 we observed cortical Rok-GFP localization in 91 . 6% of metaphase neuroblasts ( n = 24 ) and 85 . 7% ( n = 14 ) of telophase neuroblasts showed enriched Rok-GFP at the cleavage furrow ( Figure 3—figure supplement 1D ) . This result suggests that Vib is not essential for Rok localization in neuroblasts . Next , we explored whether increased Rok activity could rescue the asymmetric cell division defects observed in vib mutants . We overexpressed the catalytic domain of Rok ( Rok-CAT ) , a previously characterized transgene ( Winter et al . , 2001 ) that is known to increase phosphorylation of Sqh , and found that it was unable to recuse Mira delocalization in vib133/j5A6 ( Figure 3—figure supplement 1E; metaphase , 85 . 7% , n = 42; telophase , 83 . 3% , n = 12 ) , rather showing similar Mira delocalization to vib133/j5A6 trans-heterozygous mutant ( metaphase , 90 . 2% , n = 34; telophase , 84 . 6% , n = 26 ) . Expression of RokCAT in wild-type neuroblasts had no influence on Mira localization in neuroblasts ( Figure 3—figure supplement 1E; n = 27 for metaphase; n = 13 for telophase ) . These observations suggest that Vib regulates Sqh localization unlikely via localizing/activating Rok . Given that the signature biochemical activity of Vib is the heterotypic exchange of PI and PC , we investigated whether this lipid exchange activity is important for its role in regulating asymmetric division and homeostasis of neuroblasts . The lipid headgroup binding cavity of PITPs comprised of several conserved residues that are important for interaction with PI and PC ( Cockcroft and Garner , 2011 ) . Among these residues , mammalian Threonine 59 , the Thr63 equivalent in Drosophila Vib , has been extensively studied ( Cockcroft and Garner , 2011 ) . The Thr59 to Ala substitution of mammalian PITPα reduces PI binding specifically , whereas the Glu substitution specifically ablates the PI binding/transfer activity without affecting PC binding/transfer activities ( Alb et al . , 1995; Morgan et al . , 2004; Cockcroft and Garner , 2011 ) . We purified VibT63A ( Thr63 to Ala ) and VibT63E ( Thr63 to Glu ) and determined their PI and PC binding and transfer capacities in vitro . In vitro PI and PC binding and transfer activities were measured by monitoring the PyrPtdCho or [3 hr]-PtdCho and PyrPtdIns or [3 hr]-PtdIns transport from donor to acceptor liposomes as previously described ( Somerharju et al . , 1987; Bankaitis et al . , 1990; Schaaf et al . , 2008 ) . Similar to rat PITPα and PITPβ , wild-type Vib protein bound both PC and PI ( Figure 4A–B ) . In addition , Vib transferred both PI and PC robustly , albeit with slightly lower capacity than mouse PITPs and the major yeast PI/PC-transfer protein Sec14 ( Figure 4C , D ) . By contrast , mutation of Thr63 to Ala abolished both PI and PC binding and transfer capacity ( Figure 4A–D ) . While PI binding and transfer were abolished in VibT63E , the mutant protein retained reduced PC binding and transfer capability ( Figure 4A–D ) . Similar to the mutant mammalian PITPα ( Tilley et al . , 2004; Shadan et al . , 2008 ) , WF to AA mutation at position 202 and 203 of Vib , which are situated in the membrane binding region of Vib abolished both PI and PC binding as well as their transfer ( Figure 4A–D ) . Together , these data demonstrate that , similar to mammalian PITPα , Drosophila Vib processes PI and PC binding and transfer activity , with VibT63E including additional defects in PC binding and transfer . We next determined the ability of VibT63A and VibT63E to rescue asymmetric division defects observed in vib133/j5A6 . We generated transgenic flies expressing Venus tagged VibT63A ( VibT63A::Venus ) and Venus tagged VibT63E ( VibT63E::Venus ) in larval brains ( Figure 4—figure supplement 1A ) . Expression of VibT63A::Venus alone in larval brain had no influence on Mira and aPKC localization in mitotic neuroblasts ( Figure 4E , F , n = 44 and data not shown ) . Mira was still delocalized in 83 . 7% ( Figure 4E , F , n = 43 ) of metaphase neuroblasts and 86 . 2% of telophase neuroblasts ( Figure 4E , F , n = 29 ) with VibT63A::Venus expression in vib133/j5A6 transheterozygotes , indistinguishable from vib133/j5A6 control neuroblasts ( Figure 4E , F , 88% , n = 69 for metaphase; 85 . 7% , n = 42 for telophase ) . Furthermore , expression of VibT63A::Venus also failed to rescue ectopic neuroblasts in type I ( 47 . 6% , n = 21 ) and type II MARCM clones ( 57 . 7% , n = 26 ) of vib133 ( Figure 4—figure supplement 1B , C ) . Similarly , expression of VibT63E::Venus in larval brains of vib133/j5A6 still resulted in strong delocalization of Mira ( Figure 4E , F , 91% , n = 45 for metaphase; 86 . 3% , n = 38 for telophase ) and aPKC ( Figure 4—figure supplements 1D , 51% , n = 90 for metaphase; telophase rescue , n = 47 for telophase ) . Furthermore , in vib133/j5A6 transheterozygotes expressing VibT63E , 60 . 6% ( n = 33 ) of metaphase neuroblasts showed cytoplasmic Pros localization , similar to the delocalization of Pros in vib133/j5A6 metaphase neuroblasts ( Figure 4—figure supplement 2B , C; 53 . 3% , n = 45 ) . At telophase , 52 . 6% of vib133/j5A6 neuroblasts expressing VibT63E showed Pros delocalization at the mitotic spindle ( Figure 4—figure supplement 2B , C , n = 21 ) , undistinguishable from vib133/j5A6 telophase neuroblasts ( Figure 4—figure supplement 2B , C; 54 . 8% , n = 31 ) . We next tested whether lipid binding/transfer activity of Vib is critical for Sqh localization in neuroblasts . In vib133/j5A6 neuroblasts expressing VibT63A::Venus , 88 . 1% ( n = 59 ) of metaphase neuroblasts and 71 . 4% ( n = 21 ) of telophase neuroblasts showed delocalization of Sqh . Similarly , in vib133/j5A6 neuroblasts expressing VibT63E::Venus , 88 . 4% of metaphase neuroblasts ( n = 43 ) displayed cytoplasmic Sqh localization and 53 . 3% of telophase neuroblasts ( n = 15 ) failed to enrich Sqh at the cleavage furrow ( Figure 4—figure supplement 3A , B ) . These defects were similar to those observed in vib133/j5A6 ( Figure 4—figure supplement 3A , B ) . Likewise , Sqh1P delocalization observed in vib133/j5A6 neuroblasts failed to be rescued by overexpressing either VibT63A or VibT63E ( Figure 4—figure supplement 3C , D ) . At metaphase , 80% ( n = 65 ) of vib133/j5A6 neuroblasts expressing VibT63A::Venus were cytoplasmic and only 45% ( n = 20 ) of telophase neuroblasts were localized at cleavage furrow . Likewise , 85 . 7% ( n = 70 ) of vib133/j5A6 neuroblasts expressing VibT63E::Venus displayed cytoplasmic Sqh1P at metaphase and only 38 . 5% ( n = 26 ) of neuroblasts had Sqh1P localization at the cleavage furrow at telophase ( Figure 4—figure supplement 3C , D ) . Taken together , these observations indicate that lipid binding and transfer activity of Vib is critical for its role in regulating asymmetric division of neuroblasts . Vib has a previously known role in cytokinesis of spermatocytes and neuroblasts ( Gatt and Glover , 2006; Giansanti et al . , 2006 ) . In MARCM clones , 78 . 6% of vib133 neuroblasts displayed a cytokinesis defect ( Figure 4G , Figure 4—figure supplement 1B , n = 70 ) . Surprisingly , expression of VibT63E::Venus rescued the cytokinesis defect in 98 . 7% of vib133 clones ( n = 75 ) , while excess neuroblasts or depletion of neuroblasts persisted ( Figure 4G , Figure 4—figure supplement 1B , C , and Figure 4—figure supplement 2A ) . 24% of vib133 MARCM clones expressing VibT63E were depleted of neuroblasts ( Figure 4—figure supplement 2A , n = 99 ) . Similar to vib133/j5A6 neuroblasts ( Figure 4—figure supplements 2B , C , 41 . 6% , n = 327 ) , Pros was ectopically localized to the nucleus in interphase vib133/j5A6 neuroblasts expressing VibT63E ( Figure 4—figure supplements 2B , C , 41% , n = 315 ) , suggesting that nuclear Pros might be partly contributed to premature differentiation and subsequent loss of neuroblasts in these mutant brains . By contrast , expression of VibT63A::Venus , in which both PI and PC binding and transfer capacity were abolished , did not rescue the cytokinesis defects observed in vib133 neuroblasts ( 85% , n = 27; data not shown ) . Since VibT63E retains partial PC binding and transfer activity ( Figure 4A–D ) , overexpression of VibT63E shown in Figure 4—figure supplement 1A might restore sufficient PC binding and transfer activity to rescue cytokinesis defects in vib mutants . Thus , despite the pleotropic phenotypes observed in vib neuroblasts , asymmetric division defects are unlikely a consequence of cytokinesis failure as these two phenotypes could be uncoupled in the above experiments . Given that Sqh is targeted to the neuroblast cortex by Vib , a lipid binding and transfer protein , we examined whether Sqh was able to bind to phospholipids . To perform lipid-binding assays , Myc-tagged Sqh was expressed in S2 cells ( Figure 5A ) and protein extracts were incubated with lipid strips . While the control did not bind to phosphoinositides , Myc-Sqh was able to bind to PI ( 4 ) P , PI ( 4 , 5 ) P2 , PI ( 3 , 4 , 5 ) P3 and phosphatidylethanolamine ( Figure 5A ) . Therefore , we provide the initial evidence that Sqh binds to PIs in vitro . PITPs stimulate PI4K activity for PI ( 4 ) P production ( Cockcroft and Carvou , 2007 ) ( Figure 5—figure supplement 1A ) . In wild-type larval brain neuroblasts , PI ( 4 ) P was barely detected using an anti-PI ( 4 ) P antibody in wild-type larval brains ( Forrest et al . , 2013 ) , presumably due to its low abundance . In addition , PI ( 4 ) P marked by PH domain of FAPP ( RFP::PH-FAPP ) associates with Golgi apparatus in Drosophila tissues ( Polevoy et al . , 2009 ) . To probe the plasma membrane pool of PI ( 4 ) P more specifically , we took advantage of a PI ( 4 ) P-GFP reporter 2xOshPH::GFP [PI ( 4 ) P-GFP] , the GFP-tagged PH domain of yeast Osh2 , which is known to detect multiple pools of PI ( 4 ) P in yeast and mammalian fibroblasts ( Roy and Levine , 2004 ) . Strikingly , when it was expressed under insc-Gal4 driver in neuroblasts , the intense signal of PI ( 4 ) P reporter was uniformly cortical from interphase to anaphase ( Figure 5B and Figure 5—figure supplement 1B ) . Interestingly , PI ( 4 ) P appeared to be enriched at cleavage furrow in telophase neuroblasts ( Figure 5B–D , 73 . 1% , n = 78 ) . These observations suggest that PI ( 4 ) P is localized to the plasma membrane in neuroblasts . Next , we investigated whether the localization of PI ( 4 ) P in neuroblasts depended on Vib function . To this end , PI4P-GFP reporter was introduced into vib133/j5A6 trans-heterozygous larval brains . Strikingly , PI ( 4 ) P-GFP in 44% of interphase vib133/j5A6 neuroblasts localized to the cytoplasm and forming large aggregates ( Figure 5B , C , n = 157 ) . Cytoplasmic Mira aggregation was also observed in these neuroblasts ( Figure 5B , 57 . 3% , n = 129 ) . At metaphase , PI ( 4 ) P-GFP reporter in vib133/j5A6 was largely unaffected , likely due to its high abundance at this stage ( data not shown ) . However , the enrichment of PI ( 4 ) P at the cleavage furrow was lost in 51 . 9% of telophase neuroblasts ( Figure 5B , C , n = 54 ) . This observation is supported by the measurement of PI ( 4 ) P-GFP pixel intensity . Normalized against the intensity at the apical cortex , wild-type neuroblasts showed an average of 2 . 07-fold enrichment ( n = 34 ) of PI ( 4 ) P-GFP at the cleavage furrow , while in vib133/j5A6 telophase neuroblasts , enrichment of PI ( 4 ) P-GFP decreased to 1 . 44-fold ( n = 18 ) at cleavage furrow ( Figure 5D ) . There was also a decrease of basal enrichment of PI ( 4 ) P-GFP at the basal cortex of vib133/j5A6 neuroblasts , from 52 . 9% ( average of 1 . 56 fold , n = 34 ) in wild-type neuroblasts to 33 . 3% ( average of 1 . 26 fold , n = 18 ) in vib133/j5A6 neuroblasts ( Figure 5D ) . Since we observed that Vib is required for the localization of both Sqh and PI ( 4 ) P on the neuroblast cortex and that Sqh binds to PI ( 4 ) P in vitro , we examined the role of Sqh in regulating PI ( 4 ) P localization in neuroblasts . We analyzed the localization of PI ( 4 ) P-GFP in sqh1 hemizygotes , a hypomorphic allele of Sqh that survived to the third instar larval stage . Remarkably , 23 . 3% of interphase neuroblasts ( n = 313 ) and 36 . 8% of metaphase neuroblasts ( n = 19 ) displayed distinct PI ( 4 ) P-GFP aggregates in the cytoplasm ( Figure 5E ) . Given the known role of Sqh in regulating cytokinesis , we could not analyze the localization of PI ( 4 ) P-GFP in telophase sqh1 neuroblasts . Next , we knocked down sqh using insc-Gal4 and examined the localization of PI ( 4 ) P-GFP reporter in neuroblasts . While PI ( 4 ) P-GFP reporter was uniformly cortical in interphase control neuroblasts ( Figure 5F , n = 105 ) , sqh knockdown resulted in the formation of many distinct PI ( 4 ) P-GFP aggregates in interphase neuroblasts ( Figure 5F , 30% , n = 40 ) . We conclude that Sqh facilitates the membrane localization of PI ( 4 ) P in neuroblasts . Given that Vib is necessary for Sqh localization to the cell cortex , with both proteins mediating PI ( 4 ) P membrane localization in neuroblasts , we explored whether Vib physically interacts with Sqh . Due to Vib’s robust PI/PC transfer activity , it is likely that interaction of Vib with its interacting proteins will be transient . Therefore , we adopted bimolecular fluorescence complementation ( BiFC ) assay ( Gohl et al . , 2010 ) , a widely used method to probe protein-protein interactions that are transient or weak , due to the irreversibility of the BiFC complexes ( Shyu et al . , 2008 ) . We generated chimeric proteins , Vib-Myc-N termini of YFP ( Vib-Myc-NYFP ) , and C-termini of YFP ( CYFP ) -HA-Sqh constructs . Expression of these constructs with their respective negative controls , comprising of the matching half-YFP ( refer to the legend ) , in S2 cells by actin-Gal4 did not result in any fluorescence signals ( Figure 6—figure supplement 1 ) . Remarkably , we observed intense YFP signals upon co-transfection of both constructs in S2 cells with actin-Gal4 ( Figure 6—figure supplement 1 ) . On the contrary , we did not observe YFP signals in S2 cells that co-express either Vib-Myc-NYFP with an unrelated control WAVE-HA-CYFP ( Gohl et al . , 2010 ) or CYFP-HA-Sqh with WAVE-Myc-NYFP ( Gohl et al . , 2010 ) ( data not shown ) . As an additional control , Vib did not interact with Mira or Baz in BiFC assays ( data not shown ) . This observation suggests that Vib and Sqh were in sufficiently close proximity to each other to allow the two-halves of YFP to merge in trans and reconstitute a functional fluorescence protein . To validate this interaction , we generated two truncated Vib proteins , Vib N-terminus ( VibN ) and Vib C terminus ( VibC ) , and tested either truncated form of Vib abolished the interaction with Sqh in BiFC assays . Indeed , co-expression of either VibN-Myc-NYFP or VibC-Myc-NYFP with CYFP-HA-Sqh , driven by actin-Gal4 in S2 cells , did not result in any YFP fluorescence ( Figure 6—figure supplement 1 ) . As expected , neither of the negative control for these truncated proteins displayed YFP in the same BiFC assay ( Figure 6—figure supplement 1 ) . Next , to examine if Vib and Sqh interact in vivo , we generated transgenes expressing NYFP-Myc-Vib and CYFP-HA-Sqh . Co-expression of NYFP-Myc with CYFP-HA-Sqh ( metaphase , n = 44; telophase , n = 8 ) and NYFP-Myc-Vib with CYFP-HA ( metaphase , n = 32; telophase , n = 10 ) under the neuroblast driver , insc-Gal4 , did not result in YFP fluorescence ( Figure 6A ) . By contrast , co-expression of both NYFP-Myc-Vib and CYFP-HA-Sqh resulted in YFP fluorescence in neuroblasts ( Figure 6A ) . Despite NYFP-Myc-Vib being observed in the cytoplasm and CYFP-HA-Sqh at both cell cortex and spindle envelope in neuroblasts , YFP was found in the cytoplasm during interphase , then localized to the cortex as early as prophase and concentrated at the cleavage furrow at telophase ( Figure 6A; metaphase , n = 44; telophase , n = 15 ) . These data suggest that Vib interacts with Sqh at the cell cortex and cleavage furrow in neuroblasts . To further validate the physical association between Sqh and Vib , we performed a proximity ligation assay ( PLA ) , which allows detection of protein-protein interaction with high specificity and sensitivity in the form of fluorescence signal in situ ( Fredriksson et al . , 2002 ) . We co-stained S2 cells with anti-Flag and anti-Myc antibodies to visualize the co-expression of proteins tagged with Flag or Myc . Quantifications of PLA foci were carried out only in cells with co-expression of Flag and Myc-tagged proteins ( Figure 6C , D ) . In S2 cells expressing Flag and Myc controls , we did not observe any PLA fluorescence signal ( Figure 6C , D; n = 96 ) . In S2 cells that were co-transfected with Flag-Vib and Myc controls , 76 . 6% cells ( n = 188 ) had no PLA signal and 19 . 2% cells displayed weak PLA fluorescence signal ( ≤10 foci ) , with the remaining 3 . 7% and 0 . 5% cells displaying PLA foci between 11 and 30 and >30 , respectively . In control expressing Myc-Sqh with Flag , 86 . 6% cells ( Figure 6C , D , n = 149 ) had no PLA signal and 13 . 4% of cells with weak PLA signal ( ≤10 foci ) . By contrast , 82 . 9% ( Figure 6C , D , n = 216 ) of S2 cells expressing both Flag-Vib and Myc-Sqh showed PLA fluorescence: 15 . 3% of cells displayed strong PLA signal ( >30 foci ) , 24 . 5% of cells showed moderate PLA signal ( 11–30 foci ) and 43 . 1% of cells showed weak PLA signal ( <10 foci ) . On average , controls Flag-Vib with Myc and Myc-Sqh with Flag resulted in 1 . 7 and 0 . 3 PLA foci per cell , respectively . By contrast , co-expressing Flag-Vib and Myc-Sqh resulted in 15 . 2 PLA foci per cell . These data reinforce that Vib and Sqh physically interact . There are three PI4-kinases in Drosophila , namely PI4KIIIα , PI4KIIIβ and PI4KIIα . PI4KIIIα is required for actin organization and cell polarity during oogenesis ( Tan et al . , 2014 ) . PI4KIIIβ , also named Four wheel drive ( Fwd ) , regulates cytokinesis of spermatocytes by localizing PI ( 4 ) P and Rab11 to the plasma membrane ( Polevoy et al . , 2009 ) . PI4KIIα regulates membrane trafficking during the formation of secretory granules in the salivary gland ( Burgess et al . , 2012 ) . The role of these PI4Ks during neuroblast asymmetric division was previously unknown . Toward this end , we examined neuroblast homeostasis and asymmetric division in mutants or RNAi lines for these three PI4Ks . Loss of function of PI4KII did not perturb neuroblast homeostasis or cortical polarity of aPKC and Mira ( Figure 7—figure supplement 1A , B ) . Similarly , neuroblast homeostasis and cortical polarity were unaffected upon loss of PI4KIIIβ/Fwd ( Figure 7—figure supplement 1C and data not shown ) . PI4KIIIαC is an EMS-induced mutant that carries a nonsense mutation at Trp855 , leading to the generation of a truncated protein with its C-terminal half including the kinase domain deleted ( Yamamoto et al . , 2014 ) . We observed ectopic neuroblasts in PI4KIIIαC MARCM neuroblast clones , as evident by ectopic Dpn+Ase+ type I neuroblasts ( Figure 7A , 44 . 8% , n = 29 ) and Dpn+Ase- type II neuroblasts ( Figure 7A , 30% , n = 20 ) . Furthermore , 35 . 5% ( n = 76 ) of PI4KIIIαC MARCM clones were devoid of neuroblasts ( data not shown ) , suggesting that neuroblast homeostasis was disrupted . Neuroblast homeostasis defects of PI4KIIIαC clones were fully rescued by a PI4KIIIα genomic rescue construct ( Figure 7—figure supplement 1D , type I clones , n = 81; type II clones , n = 36 ) . We next investigated whether PI4KIIIα is required for asymmetric division of neuroblasts . Depletion of PI4KIIIα appears to result in a delay of mitotic entry , as very few mitotic neuroblasts were found in MARCM clones generated . Apical proteins aPKC ( 42% , n = 12 ) and Par-6 ( 50%; n = 20 ) delocalized to the cytoplasm ( Figure 7B , C ) . The delocalization of aPKC into cytoplasm presumably causes depletion of neuroblasts in PI4KIIIα mutants . Likewise , basal protein Mira either delocalized to the cytoplasm or mis-localized to the mitotic spindle in in PI4KIIIα mutants ( Figure 7B , C , 77% , n = 13 ) . Similarly , Numb crescent became punctate or delocalized in metaphase neuroblasts of PI4KIIIαC ( Figure 7B , C , 40%; n = 20 ) . The delocalization of basal proteins appears to cause ectopic neuroblasts observed in PI4KIIIα mutants . These observations suggest a role of PI4KIIIα is required for asymmetric division and homeostasis of neuroblasts . To determine whether Sqh was a common target of Vib and PI4KIIIαduring asymmetric division of neuroblasts , we sought to investigate whether Sqh localization is dependent on PI4KIIIα in neuroblasts . To this end , we examined the localization of Sqh::mCherry in PI4KIIIαC MARCM clones . Since Sqh::mCherry is mostly cytoplasmic during interphase , we focused our analysis on mitotic neuroblasts . In control MARCM clones , 46% of metaphase neuroblasts showed strong cortical localization of Sqh::mCherry , 23% was weakly cortical and 31% was cytoplasmic ( Figure 7D , n = 11 ) . However , a vast majority of metaphase neuroblasts from PI4KIIIαC MARCM clones displayed either cytoplasmic ( 36 . 4% ) or weakly cortical localization ( Figure 7D , 54 . 5% , n = 11 ) . This observation suggests that similar to Vib , PI4KIIIαis important for anchoring Sqh to the cell cortex in dividing neuroblasts . Consistently , Sqh1P cortical localization also depended on PI4KIIIαin neuroblasts ( Figure 7—figure supplement 1E ) . In the controls , while 12% were weakly cortical and the remaining 20% were cytoplasmic , 68% of metaphase neuroblasts showed cortical Sqh1P localization ( Figure 7—figure supplement 1E , n = 25 ) . By contrast , only 5 . 6% ( n = 18 ) of metaphase neuroblasts of PI4KIIIαC showed obvious cortical localization and the rest of them were either cytoplasmic ( 44 . 4% , n = 18 ) or weakly cortical ( Figure 7—figure supplements 1E , 50% , n = 18 ) . At telophase , Sqh1P localization at the cleavage furrow was only observed in 44 . 4% of PI4KIIIαC neuroblasts ( Figure 7—figure supplement 1E , n = 9 ) , while majority of control telophase neuroblasts showed Sqh1P localization at the cleavage furrow ( Figure 7—figure supplements 1E , 84 . 2% , n = 19 ) . These observations strongly suggest that PI4KIIIαplays an important role in localizing Sqh to the cell cortex in neuroblasts . Next , we assessed the effect of pharmacological inhibition of PI4KIIIα on asymmetric division of neuroblasts by phenylarsine oxide ( PAO ) , an inhibitor of PI4-kinases at a low concentration of 2 . 5 μM ( Bryant et al . , 2015 ) . In the mock treatment ( DMSO ) to wild-type larval brains , 82 . 4% of metaphase Sqh::mCherry were cortically localized and only 17 . 6% of neuroblasts showed cytoplasmic distribution ( Figure 7E , n = 51 ) . By contrast , with PAO treatment , Sqh::mCherry became cytoplasmic in 59% ( Figure 7E , n = 117 ) of metaphase neuroblasts . Likewise , compared with basal Mira crescent ( 100% , n = 48 ) in the mock treatment , 25% ( n = 64 ) of neuroblasts upon PAO treatment showed cytoplasmic Mira localization at metaphase ( Figure 7F ) , which was consistent with above observations in PI4KIIIαC neuroblasts . However , Rok-GFP localization was unaffected following PAO treatment . Similar to mock ( DMSO treatment ) , 92% ( n = 112 ) of metaphase neuroblasts with cortical Rok-GFP localization and 76 . 5% ( n = 17 ) of telophase with Rok-GFP enriched at the cleavage furrow were observed ( Figure 7—figure supplement 2 ) . This result suggests that Rok localization is likely independent of PI4KIIIα . Taken together , these observations reinforce the role of PI4KIIIα in Sqh localization to the cell cortex as well as asymmetric division .
Here , we show a new role for Drosophila PITP , Vibrator/Giotto in asymmetric cell division and homeostasis of neuroblasts by interacting and anchoring non-muscle myosin II light chain , Sqh . Importantly , lipid binding/transfer activities , particularly PI binding and transfer activities of Vib , are critical for asymmetric division . We also show that Sqh binds to phosphoinositides including PI ( 4 ) P and that a pool of PI ( 4 ) P is localized at the cell cortex of neuroblasts in a Vib and Sqh-dependent manner . Finally , a PI4-kinase PI4KIIIα , an essential kinase for PI ( 4 ) P synthesis , is required for Sqh localization and asymmetric division , similar to Vib . Given the lipid-binding and transfer function of Vib , it is conceivable that it anchors Sqh to the cell cortex prior to the asymmetric localization of Sqh and basal proteins in neuroblasts , rather than directly regulating asymmetric localization of Sqh or basal proteins . In addition , Vib’s localization in neuroblast cortex lacks apparent asymmetry , while it is required for the asymmetric localization of apical proteins aPKC and Par-6 , but not Baz . We speculate that Vib might also play a role in membrane localization of aPKC or Par-6 in neuroblasts . We also show that Vib likely interacts with Sqh to regulate asymmetric division through its stimulation of PI4KIIIα to produce a plasma membrane pool of PI ( 4 ) P that , in turn , specifies cortical targeting of Sqh in neuroblasts . This reciprocal regulation between Myosin and lipid transfer/synthesis may be important for asymmetric division of neuroblasts . The asymmetric division defects of vib-depleted neuroblasts were not a consequence of cytokinesis failure , because the VibT63E mutation , which rendered Vib deficient in PI binding/transfer with reduced PC binding/transfer , was able to rescue the defects of cytokinesis , but not neuroblast polarity , observed in vib-depleted neuroblasts . This finding suggests that PC binding/transfer function of Vib may be more important for cytokinesis , while PI binding/transfer may play a more prominent role in neuroblast polarity . Therefore , the pleotropic functions of Vib may be separable by cooperating with different downstream proteins during cytokinesis and asymmetric division of neuroblasts . Supporting this notion , depletion of PI4KIIIαresulted in loss of neuroblast polarity without blocking cytokinesis . Thus , it is conceivable that Vib and PI4KIIIα-dependent cortical localization of Sqh regulates asymmetric division , but not cytokinesis . PI ( 4 ) P is a functionally diverse inositol lipid in cells ( Gassama-Diagne and Payrastre , 2009; Hammond et al . , 2009 ) . Apart from being the major precursor of PI ( 4 , 5 ) P2 biosynthesis , PI ( 4 ) P also plays numerous crucial functions such as regulation of protein trafficking at the level of the Golgi complex , and sphingolipid and sterol biosynthetic trafficking ( Matsuoka et al . , 1998; Blumental-Perry et al . , 2006; Olkkonen and Li , 2013 ) . We show that PI ( 4 ) P can be detected on the cell cortex of neuroblasts with cleavage furrow enrichment at telophase . Moreover , we show that the distribution of PI ( 4 ) P on the plasma membrane is mediated by both Vib and Sqh in neuroblasts . Supporting this notion , Sqh binds to PI ( 4 ) P in vitro . Although the apical protein Baz , an apical protein , is known to be targeted to the plasma membrane via direct interactions with PI ( 3 , 4 , 5 ) P3 and PI ( 4 , 5 ) P2 ( Heo et al . , 2006; Krahn et al . , 2010 ) . Interestingly , the role of Vib in neuroblast asymmetric division that we report here is substantially independent of Baz function . Mammalian PI4-kinases are responsible for the generation of PI ( 4 ) P within cells and exhibit different subcellular distribution ( Clayton et al . , 2013 ) . While type II PI4-kinases and type III PI4KIIIβ localize to intracellular membranes including trans-Golgi network ( TGN ) , endoplasmic reticulum ( ER ) , endosomal and trafficking vesicles , PI4KIIIα localizes to the plasma membrane , early cis-Golgi compartments and nucleolus ( D'Angelo et al . , 2008 ) . The known distribution of various PI4 kinases and the fact that PI4KIIIα is the major kinase responsible for plasma membrane PI ( 4 ) P biosynthesis ( Roy and Levine , 2004; Balla et al . , 2005 ) well support the unique role of PI4KIIIα among three Drosophila PI4-kinases in neuroblast asymmetry . Drosophila PI4kIIIα is also required for plasma membrane integrity and oocyte chamber polarity potentially through regulating actin organization ( Tan et al . , 2014 ) . Interestingly , we found that actin localization labeled by Phalloidin was unaffected in both vib and PI4KIIIα mutant neuroblasts ( data not shown ) , suggesting that both Vib and PI4KIIIα likely regulate neuroblast polarity independent of actin organization . Mammalian PI4-Kinases are associated with cancer and neurological diseases including schizophrenia ( Furuta et al . , 2003; Jungerius et al . , 2008; Clayton et al . , 2013 ) . Thus , our findings on the role of Drosophila PI4KIIIα in neuroblast asymmetric division may shed light on the mechanisms underlying these diseases . Drosophila non-muscle myosin II RLC Sqh regulates asymmetric localization of basal cell fate determinants in neuroblasts ( Barros et al . , 2003 ) . Similarly , C . elegans non-muscle myosin II exhibits polarized distribution and mediates asymmetric localization of Par-3 to the anterior cortex during the asymmetric division of the one-cell stage embryo ( Severson and Bowerman , 2003; Ou et al . , 2010 ) . Here , we show that Vib is critical for anchoring Sqh to the neuroblast cortex . The asymmetric localization of Sqh in neuroblasts has been shown to be dependent on apical protein Pins , but not aPKC , Baz or canonical centralspindlin pathway including Pavarotti ( Pav ) ( Gatt and Glover , 2006 ) . The novel role of Vib in anchoring Sqh appears to be independent of Pins , since Vib is not important for Pins cortical polarity in neuroblasts . Together with our finding on the role of PI4KIIIα in Sqh localization , we demonstrate that the Vib and PI4KIIIα-dependent inositol lipid PI ( 4 ) P is important for Sqh cortical localization in neuroblasts . Myosin is known to associate with lipids indirectly through F-actin and the direct link between myosin and lipids was unclear . Our data on the association between Vib and Sqh provides a novel link between PITP-dependent PI ( 4 ) P pools and myosin localization in dividing cells . In addition , we show the binding of Sqh to phospholipids including PI ( 4 ) P , directly linking myosin to phospholipids . Very recently , mammalian non-muscle myosin II was shown to bind to liposomes containing multiple acidic PIPs through its RLC-binding site , but not to liposomes that contain only PC ( Liu et al . , 2016 ) . Our finding that Sqh mediates membrane localization of PI ( 4 ) P suggests that Sqh may reinforce the cortical localization of itself through an interaction with Vib and PI ( 4 ) P in neuroblasts . Whether this reciprocal regulation between Myosin and lipid synthesis is similarly employed in other cell types will be of great interest in future studies .
Fly stocks and genetic crosses were raised at 25°C unless otherwise stated . Fly stocks were kept in vials or bottles containing standard fly food ( 0 . 8% Drosophila agar , 5 . 8% Cornmeal , 5 . 1% Dextrose and 2 . 4% Brewer’s yeast ) . The following fly strains were used: vib133 , vib1105 , UAS-NYFP-Myc , UAS-CYFP-HA , UAS-NYFP-Myc-Vib , UAS-CYFP-HA-Sqh , UAS-Vib::Venus , UAS-VibT63A::Venus , UAS-VibT63E::Venus , UAS-PITPα::Venus and UAS-PITPβ::Venus ( this study ) , fwd3 , P{w+ , PI4KIIIα} FRT80B and PI4KIIΔ ( Julie , A . Brill ) , sqhAx3; p[w;sqh-GFP42] , UAS-Cdc42::GFP ( Akira Chiba ) , baz , FRT19A , w , Ubi-Rok-GFP ( I ) and w;; Ubi-Rok-GFP ( III ) ( Bellaiche , Y ) , UAS-WAVE-HA-CYFP and UAS-Abi-Myc-NYFP ( Bogdan , S ) . The following stocks were obtained from Bloomington Drosophila Stock Center: insc-Gal4 ( BDSC#8751 ) , elav-Gal4 ( BDSC#8765 ) , P{lacW}vibj5A6/TM6B , Tb ( BDSC#12144 ) , P{lacW}vibj7A3/TM3 , Sb ( BDSC#10308 ) , Vib deficiency w; Df ( 3R ) BSC850/TM6C , Sb , cu ( BDSC#27922 ) , sqhAX3 , FRT19A/FM7c ( BDSC#25712 ) , sqh RNAi ( BDSC#33892 ) , w; Sqh-mCherry ( BDSC#59024 ) , PI4KIIIαC , FRT19A/FM7c ( BDSC#57112 ) , w; UAS-2xOsh2PH-GFP ( BDSC#57353 ) , w; UAS-2xOsh2PH-GFP/CyO; Pri/TM6B , Tb ( BSDC#57352 ) . Crosses for RNAi knockdown and overexpression were incubated at 25°C for 24 hr before transferring to 29°C , where the larvae were aged for 72 hr . Wandering third instar larvae were dissected and processed for immunohistochemistry staining . Immunohistochemistry on third instar larval brains was performed as previously described ( Wang et al . , 2006 ) . The primary antibodies used were guinea pig anti-Dpn ( 1:1000 ) , rabbit anti-Ase ( 1:500 ) , rat anti-CD8 ( 1:200; Invitrogen ) , rabbit anti-aPKCζ C20 ( 1:100; Santa Cruz Biotechnology , Inc . ) , rabbit anti-PKC iota phospho-T555 + T563 ( 1:100; abcam ) , rabbit anti-Insc ( 1:1 , 000 ) , guinea pig anti-Gαi ( 1:200; F . Yu ) , rat anti-Brat ( 1:100; R . P . Wharton ) , mouse anti-α-tubulin ( 1:200; Sigma-Aldrich ) , mouse anti-Mira ( 1:40; F . Matsuzaki ) , guinea pig anti-Baz ( 1:500; A . Wodarz ) , guinea pig anti-Numb ( 1:100 ) , rabbit anti-PON ( 1:100; Y . N . Jan ) , rabbit anti-GFP ( 1:1 , 000; Molecular Probes ) , rabbit anti-RFP ( 1:100; Abcam ) , rabbit anti-Sqh ( 1:250; Adam C . Martin ) , guinea pig anti-Sqh1P ( 1:500 , Robert E . Ward IV ) , mouse anti-Myc ( 1:200 , Abcam ) and rabbit anti-HA ( 1:100 , Sigma-Aldrich ) . The secondary antibodies used were conjugated with Alexa Fluor 488 , 555 or 647 ( Jackson laboratory ) . To extract protein from larval brains , 50 wandering third instar larvae were dissected and brains were homogenized using RIPA buffer 50 mM Tris-HCl , pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 1% Triton X-100 , 0 . 5% sodium deoxycholate , and 0 . 1% SDS ) . Protein lysate were subjected to SDS-PAGE and western blotting . Antibodies used for western blotting were rabbit anti-Vib ( this study; 1:2 , 000 ) , mouse anti-α-tubulin ( 1:10 , 000; Sigma-Aldrich ) , mouse anti-Flag ( 1:2 , 000; Sigma-Aldrich ) , mouse anti-Myc ( 1:2 , 000; abcam ) , donkey α-mouse IgG HRP ( 1:5000 , Pierce , SA1-100 ) and donkey α-rabbit IgG HRP ( 1:5000 , Pierce , SA1-200 ) . MARCM clones were generated as previously described ( Lee and Luo , 1999 ) . Briefly , larvae were heat shocked at 37°C for 90 min at 24 hr ALH and at 10–16 hr after the first heat shock . Larvae were further aged for 3 days at 25°C , and larval brains were dissected and processed for immunohistochemistry . In our analysis on neuroblast homeostasis in clones , neuroblasts with obvious cytokinesis defects ( polyploidy cells ) were excluded . Neuroblasts with two ( Dpn+ ) nuclei in the same cytoplasm were also excluded in our analysis . We only counted ectopic neuroblasts that were clearly separated cells based on CD8-GFP , which marked the outline of cells . Mitotic spindle orientation was quantified for metaphase neuroblasts labeled with Insc , α-tubulin and DNA ( Topro3 ) . Apicobasal axis was inferred by a line that is perpendicular to the Insc crescent whereas the mitotic spindle represents the spindle axis . Spindle orientation which is denoted by the angle between the apicobasal and spindle axis , was measured and quantified . The pixel intensity of PI ( 4 ) P-GFP at the apical cortex , basal cortex and cleavage furrow of telophase neuroblasts were measured using ImageJ . Corresponding intensity was recorded and used to tabulate the ratio of PI ( 4 ) P-GFP intensity at cleavage furrow ( CF ) to PI ( 4 ) P-GFP intensity at apical cortex ( CF:Apical ) and the ratio of PI ( 4 ) P-GFP intensity at basal cortex to PI ( 4 ) P-GFP intensity at apical cortex ( Basal:Apical ) , respectively . Wandering third instar larvae were dissected and the harvested brains were incubated in supplemented Schneider’s medium ( supplemented with 10% FBS , 50 U/ml penicillin , 50 μg/ml Streptomycin , 0 . 02 mg/ml insulin , 20 mM glutamine and 0 . 04 mg/ml glutathione ) containing 2 . 5 μM PAO ( Sigma-Aldrich ) for 2 hr at RT . Mock treatment was done in supplemented Schneider medium added with DMSO . Larval brains were rinsed twice with PBS and processed for immunohistochemistry . For Ubi-Rok-GFP , larval brains were washed twice with PHEM buffer ( 60 mM PIPES , 25 mM HEPES , 10 mM EGTA and 4 mM MgSO4 ) , fixed with 4% formaldehyde in PHEM for 22 min and processed for immunohistochemistry . Expressed-sequence tags ( EST ) SD01527 ( Vib ) , LD14743 ( Sqh ) , and SD12145 ( PI4KIIIα ) were obtained from Drosophila Genomics Resource Center ( DGRC ) . Full length coding sequence of Vib and Sqh were amplified by PCR and cloned into pENTR/D-TOPO vector ( Invitrogen ) or pDONR221 vector ( Invitrogen ) using BP Clonase II ( Invitrogen ) according to the manufacturer’s protocol . Entry clone of Vib and Sqh were subsequently cloned into various destination vectors using LR clonase II ( Invitrogen ) according to manufacturer’s protocol . To remove the RfB cassette from the pUAS-NYFP-Myc-RfB cassette and pUAS-CYFP-HA-RfB cassette vector , a short oligo was first cloned into pDONR221 using BP clonase II and subsequently cloned into pUAS-NYFP-Myc and pUAS-CYFP-HA using LR clonase II . Primers used in this study are listed in Table 1 . The destination vectors used were pTW , pTWV , pTVW , pAFW and pAMW , which were obtained from DGRC . BiFC destination vectors , pUAST-NYFP-Myc , pUAST-CYFP-HA , pUAST-Myc-NYFP and pUAST-HA-CYFP were generous gifts from Bogdan , S . UAS-Vib , UAS-Vib::Venus , UAS-VibT63A::Venus , UAS-VibT63E::Venus , UAS-PITPα::Venus , UAS-PITPβ::Venus , UAS-NYFP-Myc , UAS-CYFP-HA , UAS-NYFP-Myc-Vib and UAS-CYFP-HA-Sqh transgenic flies were generated by P-element-mediated transformation ( BestGene Inc . ) . BDSC 8622 [yw; P{CaryP}attP2] was used as the injection stock for site-specific insertion of UAS-NYFP-Myc , UAS-CYFP-HA , UAS-NYFP-Myc-Vib and UAS-CYFP-HA-Sqh into chromosomal location 68A4 ( BestGene Inc . ) . Full length of Vib tagged with N-terminal His was expressed in bacterial cells , purified and used to immunize rabbits to raise polyclonal antibody , followed by affinity purification ( Genscript ) . Drosophila S2 cells ( CVCL_Z232 ) originally from William Chia’s laboratory ( with a non-authenticated identity but have been used in the laboratory for the past 10 years ) were cultured in Express Five serum-free medium ( Gibco ) supplemented with 2 mM Glutamine ( Thermo Fisher Scientific ) . S2 cell culture used in this study is free of contamination of Mycoplasma , due to the absence of small speckles of DAPI staining outside of the cell nucleus . For transient expression of proteins , S2 cells were transfected using Effectene Transfection Reagent ( QIAGEN ) according to manufacturer’s protocol . S2 cells were harvested 48 hr after transfection and were homogenized using lysis buffer ( 25 mM Tris pH8/27 . 5 mM NaCl/20 mM KCl/25 mM sucrose/10 mM EDTA/10 Mm EGTA/1 mM DTT/10% ( v/v ) glycerol/0 . 5% Nonidet P40 ) with Complete Proteases inhibitors ( Roche ) for 30 min at 4°C . Cell lysates were subjected to SDS-PAGE and western blotting . In vitro bimolecular fluorescence complementation assay ( Gohl et al . , 2010 ) was performed using S2 cells . 1 × 106 cells were seeded onto Poly-L-lysine coated coverslips ( Iwaki ) . S2 cells were transfected using Effectene Transfection Reagent ( QIAGEN ) with act-Gal4 and the BiFC constructs each at 0 . 2 μg per well , respectively . The BiFC constructs used were: pUAS-NYFP-Myc control and pUAS-CYFP-HA control , pUAS-NYFP-Myc-Vib , pUAS-Vib-Myc-NYFP , pUAS-VibN-Myc-NYFP , pUAS-VibC-Myc-NYFP , pUAS-CYFP-HA-Sqh , pUAS-NYFP-Myc and pUAS-CYFP-HA ( this study ) , pUAS-WAVE-HA-CYFP , pUAS-WAVE-Myc-NYFP , pUAS-Abi-HA-CYFP , pUAS-Abi-Myc-NYFP and the pUAST-BiFC vectors ( Bogdan , S . ) as control . 48 hr after transfection , growth medium was removed and S2 cells were rinsed with cold PBS before fixing with 4% EM grade Formaldehyde in PBS for 15 min . Fixed S2 cells were rinsed three times with PBS-T ( 1xPBS + 0 . 1% Triton-X100 ) and blocked with 5% BSA in PBS-T for 1 hr before primary antibody incubation at RT for 2 hr . S2 cells were then rinsed three times with PBS-T and were incubated with secondary antibodies in PBS-T for 1 hr at RT . Coverslips coated with immuno-stained S2 cells were mounted on to glass slides using vector shield ( Vector Laboratory ) for confocal microscopy . In vivo bimolecular fluorescence complementation was performed by expressing the BiFC vectors and constructs using insc-Gal4 . Crosses were set up and incubated at 25°C for 24 hr before transferring to 29°C . Larvae were aged for 72 hr and wandering third instar larvae were dissected and processed for immunohistochemistry staining . L-α-phosphatidylcholine , ( chicken ) egg PtdCho and L-α-phosphatidic acid ( chicken ) egg were purchased from Avanti Polar Lipids ( Alabaster , AL ) . 1-palmitoyl-2-decapyrenyl-sn-glycero-3-phospho-choline , PyrPtdCho and 1-palmitoyl-2-decapyrenyl-sn-glycero-3-phosphoinositol , PyrPtdIns , were generous gifts from Dr . Pentti Somerharju ( Helsinki University ) ; synthesis as described ( Gupta et al . , 1977; Somerharju and Wirtz , 1982 ) . 2 , 4 , 6-Trinitrophenylphosphatidyl-ethanolamine , TNP-PtdEtn was prepared from phosphor-ethanolamine as previously described ( Gordesky and Marinetti , 1973 ) and purified by silica gel column chromatography . The concentration of all phospholipid solutions was determined as previously described ( Rouser et al . , 1970 ) . PyrPtdCho and PyrPtdIns concentrations were determined spectroscopically using ε = 42 , 000 . The PyrPtdCho and PyrPtdIns binding and transfer measurements were essentially done as previously described ( Somerharju et al . , 1987 ) . For the binding measurements donor vesicles were made from eggPtdCho , PyrPtdCho/PyrPtdIns and TNP-PtdEtn . Solvent was evaporated and the lipid film was re-suspended in 10 µl of EtOH . The solution was then injected into 2 mL of low phosphate buffer ( 25 mM Na2HPO4 , 300 mM NaCl , pH 7 . 5 ) at 37°C and after 5–10 min incubation the solution was titrated with 0 . 1 nmol of indicated protein and fluorescence intensity was measured ( Horiba Ltd . Kyoto , Japan ) . For transfer measurements , the donor vesicles were injected into low phosphate buffer containing acceptor vesicles . The resulting lipid film was then hydrated and sonicated on ice for 10 min . After donor vesicle addition , the solution was incubated for 5–10 min at 37°C . After which the fluorescence intensity was recorded as a function of time . To initiate protein mediated lipid transfer total 9 µg of protein was injected to achieve binding site saturation . The end-point assays measuring transport of [3 hr]-PtdIns were performed as previously described ( Bankaitis et al . , 1990; Schaaf et al . , 2008 ) . The membrane lipid strips ( P-6002 , Echelon Biosciences ) were blocked in 3% fatty acid free BSA ( Sigma , A7030 ) in PBS-T [0 . 1% Tween 20 ( T ) ] at 4˚C overnight . S2 cells transfected with the indicated plasmids were lysed in lysis buffer ( 50 mM Tris-HCl , pH7 . 8 , 150 mM NaCl and 0 . 5% Triton X100 ) for 30 min at 4°C . The strips were incubated with S2 cell lysate in 3% fatty-acid-free BSA in PBS-0 . 05% T for 1 hr at RT . Strips were washed in PBS-0 . 1% T for 10 min thrice , incubated with mouse anti-Myc ( 1:2 , 000; abcam ) and subsequently proceeded similarly to western blot . The rational of PLA is as follow: Secondary antibodies conjugated with PLA PLUS or PLA MINUS probe bind to anti-Flag and anti-Myc antibodies , respectively . During ligation , connector oligos hybridize to PLA probes and T4 ligases catalyze to form a circularized template . DNA polymerase amplifies circularized template , which is bound by fluorescently labeled complementary oligos allowing the interaction to be observed as a PLA foci within cells ( Adopted from Duolink PLA , Merck ) . Proximity ligation assay was performed on S2 cells that were transfected with the indicated plasmids using Effectene Transfection Reagent ( QIAGEN ) . The plasmids used were: control Myc , control Flag , Flag-Vib and Myc-Sqh . S2 cells were washed thrice with cold PBS , fixed with 4% EM grade formaldehyde in PBS for 15 min and blocked in 5% BSA in PBS-T ( 0 . 1% Triton-X100 ) for 45min . Cells were then incubated with primary antibodies at RT for 2 hr before proceeding with Duolink PLA ( Sigma-Aldrich ) according to manufacturer’s protocol . After primary antibodies incubation , S2 cells were incubated with PLA probes at 37˚C for 1 hr . Cells were washed twice with Buffer A for 5 min each at RT , followed by ligation of probes at 37˚C for 30 min . Amplification was performed at 37˚C for 100 min , followed by two washes with Buffer B for 10 min each at RT . Cells were washed once with 0 . 01x Buffer B before incubating with primary antibodies diluted in 3% BSA in PBS for 2 hr at RT . Cells were washed twice with 0 . 1% PBS-T and incubated with secondary antibodies for 1 . 5 hr at RT before mounted with the provided in situ mounting media with DAPI ( Duolink , Sigma-Aldrich ) . | Stem cells are cells that can both make copies of themselves and make new cells of various types . They can either divide symmetrically to produce two identical new cells , or they can divide asymmetrically to produce two different cells . Asymmetric division happens because the two new cells contain different molecules . Stem cells drive asymmetric division by moving key molecules to one end of the cell before they divide . Asymmetric division is key to how neural stem cells produce new brain cells . Many studies have used the developing brain of the fruit fly Drosophila melanogaster to understand this process . Errors in asymmetric division can lead to too many stem cells or not enough brain cells . This can contribute to brain tumors and other neurological disorders . Fat molecules called phosphatidylinositol lipids are some of chemicals that cause asymmetry in neural stem cells . Yet , it is not clear how these lipid molecules affect cell behavior to turn stem cells into brain cells . The production of phosphatidylinositol lipids involves proteins called Vibrator and PI4KIIIα . Koe et al . examined the role of these two proteins in asymmetric cell division of neural stem cells in fruit flies . The results show that Vibrator activates PI4KIIIα , which leads to high levels of a phosphatidylinositol lipid called PI ( 4 ) P within the cell . These lipids act as an anchor for a group of proteins called myosin , part of the machinery that physically divides the cell . Hence , myosin and phosphatidylinositol lipids together control asymmetric division of neural stem cells . Further experiments used mouse proteins to compensate for defects in the equivalent fly proteins . The results suggest that the same mechanisms are likely to hold true in mammalian brains , although this still needs to be proven . Nevertheless , given that human equivalents of Vibrator and PI4KIIIα are associated with neurodegenerative disorders , schizophrenia or cancers , these new findings are likely to help scientists better to understand several human diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2018 | Vibrator and PI4KIIIα govern neuroblast polarity by anchoring non-muscle myosin II |
Insects have evolved diverse and remarkable strategies for navigating in various ecologies all over the world . Regardless of species , insects share the presence of a group of morphologically conserved neuropils known collectively as the central complex ( CX ) . The CX is a navigational center , involved in sensory integration and coordinated motor activity . Despite the fact that our understanding of navigational behavior comes predominantly from ants and bees , most of what we know about the underlying neural circuitry of such behavior comes from work in fruit flies . Here , we aim to close this gap , by providing the first comprehensive map of all major columnar neurons and their projection patterns in the CX of a bee . We find numerous components of the circuit that appear to be highly conserved between the fly and the bee , but also highlight several key differences which are likely to have important functional ramifications .
Honeybees and desert ants are iconic navigators ( Reviewed in Collett , 2019 ) . Honeybees can forage tens of kilometers away from their nest and can also communicate the distance and direction of foraging locations to fellow workers ( Frisch , 1967 ) . Desert ants forage hundreds of meters away from their nest in landscapes sometimes entirely bereft of visual landmarks ( Wehner , 2020 ) . Both insects preferentially make use of the most reliable visual cues , such as polarized skylight , but can also make use of an arsenal of backups to ensure a robust compass signal , including windflow ( Müller and Wehner , 2007 ) and magnetoreception ( Collett and Baron , 1994; Fleischmann et al . , 2020 ) . Although their navigational behavior is less well characterized than that of honeybees and desert ants , bumblebees are also impressive foragers , as well as generally charismatic insects that play vital ecological roles ( Goulson , 2003 ) . Bumblebees forage hundreds to thousands of meters away from their nest and have been shown to be capable of homing from novel locations at distances up to 9 . 8 km ( Goulson and Stout , 2001; Prŷs-Jones and Corbet , 2011 ) . Like honeybees , bumblebees rely on celestial cues such as polarized light for homing ( Wellington , 1974 ) and have been shown to compensate for wind drift ( Riley et al . , 1999 ) . Bumblebees also show a remarkable capacity to learn novel tasks ( Loukola et al . , 2017; Chittka , 2017 ) . While behaviorally , navigation and orientation are best understood in ants and bees , our understanding of how insect brains control navigation behavior has been largely obtained in a set of different species , most notably the desert locust Schistocerca gregaria , the fruit fly Drosophila melanogaster and more recently , Monarch butterflies ( Danaus plexippus ) and dung beetles ( Scarabaeus sp . ; Reviewed in Honkanen et al . , 2019; Turner-Evans and Jayaraman , 2016 ) . Yet none of these species show the sophisticated homing behavior of hymenopteran insects and the question remains how these tiny brains enable such flexible and impressive navigational skills . Work in other insects has revealed that one region of the insect brain lies at the heart of navigational control: the central complex ( CX; Honkanen et al . , 2019 ) . The CX is highly conserved across insects and plays important functional roles in multi-sensory integration , coordinated motor activity , and sleep . The CX receives external and internal sensory information that informs the insect about its visual environment , mechanosensory cues , and self-motion cues ( Seelig and Jayaraman , 2013; el Jundi et al . , 2014; Okubo et al . , 2020 ) . CX cells use these cues to generate sensory maps that encode the heading direction ( Heinze and Homberg , 2007; Seelig and Jayaraman , 2015; Varga and Ritzmann , 2016 ) and traveling direction ( Lu et al . , 2020; Lyu et al . , 2020; Hulse et al . , 2020 ) . The CX also likely compares current heading with a desired goal direction to generate appropriate motor commands ( Stone et al . , 2017; Rayshubskiy et al . , 2020; Hulse et al . , 2020 ) . A function of the CX in the context of motor control has indeed been firmly established ( Martin et al . , 2015; Pfeiffer and Homberg , 2014 ) . Across insects , the CX comprises four distinct neuropils ( five in the fruit fly Wolff and Rubin , 2018 ) : the protocerebral bridge ( PB ) , fan-shaped body ( FB; also called the upper division of the central body; CBU ) , the ellipsoid body ( EB; lower division of the central body; CBL ) , and the noduli ( NO ) . These neuropils are highly interconnected by the arborizations of three classes of neurons . First , columnar cells are central to most computations carried out by the CX . They link the four adjacent CX neuropils and provide the basis for intrinsic computations , but also generate output to other brain areas . Input to the CX moves predominantly via the second class , tangential neurons , which connect many different brain regions to distinct compartments of the CX . The third class comprise pontine cells , including hΔ cells and , in flies , vΔ cells ( Hulse et al . , 2020 ) . These are FB interneurons whose input and output processes are confined to the FB . For each class of cell , there exist numerous cell types , each of which is defined by their detailed morphology , polarity , and ultimately , by their connectivity to other neurons . One of the keys to understanding how this brain region is involved in so many different processes , ranging from motor control , sensory encoding , to sleep control , is the tight relation between structure and function , facilitated by the almost crystalline anatomical layout of this brain area . This link has inspired much research on the CX , which has culminated in the recent release of the first ever comprehensive map of all CX neurons and their connectivity , achieved for the fruit fly CX ( Scheffer et al . , 2020 ) . Even in this previously highly studied and well-understood model organism , connectomics analysis revealed the presence of many new cell types and produced novel insights into neural pathways and computational circuit motifs ( Hulse et al . , 2020 ) . Yet , the question arises how much of this Drosophila connectomics data is representative for other species and how much of it is specific for the fruit fly and its ecology . Neuroanatomical and physiological work outside of the fruit fly has clearly revealed differences in the anatomy and physiology in the CX between species ( Heinze and Homberg , 2008; Heinze et al . , 2013; Stone et al . , 2017; El Jundi et al . , 2018; Hensgen et al . , 2021; von Hadeln et al . , 2020 ) . However , the anatomical data from all species except the fruit fly is generally based on sparse labeling methods for identifying neurons , mostly intracellular dye injections and immunohistochemistry . While these methods yield beautiful morphological features of individual neurons , they are not comprehensive and often biased toward the largest neurons . Importantly , these studies will never be able to answer which neural elements are missing , no matter how many neurons are reported , and can neither provide definite data on cell quantities , projectivity motifs and connectivity . To address this issue , a counterpart of the Drosophila CX connectome will be needed . Obtaining similar data in a bee species will have the power to validate computational principles identified in the fly but also expand our understanding of how more complex behavioral abilities , that is , those not found in flies , are controlled by the CX . Here , we provide the first step in such an endeavour by reporting the first comprehensive map of neural projections in the bumblebee CX ( i . e . ‘projectome’; Kasthuri and Lichtman , 2007 ) . Using cellular resolution serial block-face electron microscopy ( SBEM ) , we traced the main neurites of all major CX columnar cells and FB interneurons , amounting to over 1300 neural skeletons . From these reconstructions , we established information about the numbers of cells , cell type identity and their projection patterns . Due to the tight structure function links in the CX , this detailed account of projection patterns yields a first approximation of the information flow principles in the bee CX . We additionally identified several cell types that may be unique to the bee , providing starting points for more detailed investigations . Most importantly , our findings highlight a conserved core circuitry of the CX and thus contribute to the notion that the CX contains an ancestral circuit , highly preserved across insects , but with additional layers or modifications adapted to species-specific ecologies and behaviors .
As in other insects , the bumblebee CX spans the mid-line of the protocerebrum in the central brain and is composed of four adjacent neuropils ( Figure 1a–c ) , when labeled with antibodies against the presynaptic marker synapsin ( Klagges et al . , 1996 ) . In order from posterior to anterior , these are the protocerebral bridge ( PB ) , the paired noduli ( NO ) , the fan-shaped body ( FB; alternatively named upper division of the central body ) , and the ellipsoid body ( EB; alternatively named lower division of the central body ) ( Figure 1c ) . The FB lays over the top of the EB and together they form the central body ( CB ) . As no internal structure was visible based on anti-synapsin labeling alone , we used labeling against neurotransmitters to highlight further sub-compartments . Immunohistochemical staining with antibodies raised against serotonin ( 5HT ) and tyrosine hydroxylase ( TH ) revealed horizontal layering in the CB ( Figure 1d–e’’ ) . In preparations labeled with anti-TH , at least three layers can be identified in the FB; one that contains very fine processes , another with blebbed fibers , and a third lacking any TH immunoreactivity ( Figure 1d ) . In samples stained against 5HT , a similar three-layer pattern emerged , containing zones with either fine branches , blebbed fibers , or no immunoreactivity at all ( Figure 1e ) . The EB contained at least two layers , a dorsal layer filled with fibers from dopaminergic ( TH ) neurons and a ventral layer innervated by serotonergic processes . To see how the observed layering in the CB corresponded across preparations and to electron microscopy data , whole brains stained with either TH or 5HT were imaged , manually segmented , and registered to the surface mesh of the 126 nm SBEM data set , which was used as reference volume ( Figure 1d’ , e’; see Materials and methods ) . The resulting registrations revealed at least four distinct layers in the FB , two of which were TH and 5HT immunoreactive ( layers 1 and 3a; Figure 1d” , e” ) , one which was only occupied by TH fibers ( layer 3b; Figure 1d” ) , and a fourth that appeared to only contain 5HT-positive processes ( layer 2; Figure 1e’’ ) . Interestingly , the PB was devoid of both TH- and 5HT-positive processes and synapsin immunolabeling did not reveal any glomeruli-like structure as is the case in the fly ( Wolff et al . , 2015 ) . Only two compartments could be discerned in the NO , one small and clearly demarked in synapsin stained tissue termed here NOs , and the other large , containing a coarse mesh of beaded , serotonergic processes ( NOm; see Results further on regarding the NO ) . Based on projection patterns of individual columnar neurons , the CX was further divided into vertical columns ( discussed further in the next section ) . However , no indication of columns was evident based on immunohistochemical labeling . All columnar cells originated from cell bodies situated dorsally of the CX . Their fibers extended ventrally to the PB where they branched to make contacts with their first synaptic partners . Following the PB , columnar cells fasciculated into prominent bundles . Eight in total , these bundles spanned the width of the CX and were bilaterally symmetric about the mid-line with four bundles in each hemisphere . The resulting formation was that of a chiasm , where neurons from a bundle in one hemisphere projected in the direction of the contralateral hemisphere ( Figure 2b ) . From lateral to medial , these bundles are generally referred to as the W , X , Y , and Z bundles ( Williams , 1975 ) and can be associated with either the right or left hemisphere with the addition of L or R abbreviation ( i . e . RW or LZ; Figure 2b , h ) . In locusts and flies these bundles correspond to the developmental origin of a neuron from one of four neuroblasts on either side of the midline ( Boyan and Reichert , 2011; Boyan et al . , 2017 ) , and therefore they served as our prime reference when describing columnar arborization patterns . Within the PB , each of these bundles generally gave rise to two neighboring columnar arborization domains . An exception to this rule was the layout of the innermost bundle ( Z-bundle ) , in which traced neurons innervated three adjacent arborization domains in each hemisphere ( Figure 2b ) . Therefore , our data demonstrated that the bumblebee PB contains 18 columns ( nine in each hemisphere ) , which are connected to nine corresponding columns in the CB ( Figure 2c , d ) . Columns in the PB were numbered L1/R1 to L9/R9 starting at the midline , while columns of the CB were numbered C1-C9 from right to left ( Figure 2d ) . Upon reaching the CB , columnar cells defasciculated from their bundle and branched into discrete zones , usually , but not always , within a well-defined column . Finally , except for one cell type discussed below , columnar cells projected further , linking the PB and CB with the NO or another tertiary neuropil outside of the CX . We next determined the innervated columns for all neurons of a given bundle , i . e . all neurons in the identity-giving W , X , Y , and Z bundles . All columnar cells followed one of four projection patterns ( Figure 2e , f ) . Columnar cell types that followed the ‘default’ projection pattern had no offset in the CB relative to the bundle they originate from . That is , neurons of the lateralmost PB-column ( L9/R9 ) connected to the lateralmost CB column ( CB1/CB9 ) and neurons from each of the following columns innervated a CB column shifted medially by one column . Thus , all neurons of one PB hemisphere projected to corresponding regions covering the entire width of the CB . Interestingly , neurons of the medial-most columns of the PB ( L1/R1 ) are swapped between the hemispheres ( Figure 2c ) , creating an interlocked layout of the neuronal systems that belong to either hemisphere . Three alternative projection patterns were observed that could be derived from the described default pattern by shifting either the column of origin in the PB or by shifting the target projection in the CB ( Figure 2e , f ) . Cells which are ‘CB-shifted’ were shifted by one CB column in the direction of the contralateral hemisphere , whereas cells that are ‘PB-shifted’ were shifted by one PB column toward the ipsilateral hemisphere . Finally , cells labeled here as ‘both-shifted’ were shifted by one PB column in the direction of the contralateral hemisphere and one CB column in the ipsilateral hemisphere ( Figure 2f ) . In these alternative projection patterns , neurons did not exist in all PB columns . Interestingly , while most neurons followed the default pattern , all cell types that projected to the NO were characterized by the CB-shifted pattern , while neurons that are likely to be main CX output neurons Stone et al . , 2017; Rayshubskiy et al . , 2020; Hulse et al . , 2020 followed a pattern with shifted PB innervation . We identified columnar cell types based on their projection fields within CX neuropils , their heterolateral projection patterns in the CB ( i . e . the column they project to relative to the bundle they originate from ) , their neurite paths , their population size ( number of cells per PB column ) , and by the tertiary neuropil they innervate ( Figure 3 ) . Overall , we found eleven columnar cell types , two that innervated the EB ( Figure 3a ) and nine which innervated the FB ( Figure 3b ) . For a list of corresponding neuron names historically used in other insects , see Table 1 . For additional information on nomenclature and the neuron naming scheme , see Methods . Quantities of each cell type per PB column can be found in Figure 3—figure supplement 1 . Columnar cell types projecting to the EB included EPG/PEG cells and PEN cells ( Figure 3a ) . EPG/PEG cells arborized in the PB , the EB , and the gall , while PENs sent projections through the PB , the EB , and a small compartment in the NO ( NOs; Figure 6a–c ) . All FB columnar cells contained processes that arborized in both the PB and the FB . Several FB cell types sent projections toward the lateral complex ( LX ) . These included PFL1 , 3 , PFL2 , PFLx , and PFx2 cells ( Figure 3b ) . PFx1 cells sent their fibers away from the CX toward regions in the inferior protocerebrum ( INP ) and lateral accessory lobes ( LAL ) , while PFx2 had descending fibers toward the LX , and PFx3 and PFx4 neurons contained fibers that extended anteriorly toward regions in the superior medial protocerebrum ( SMP ) and INP . Lastly , one cell type , PF1 , contained arborizations in only the PB and FB . The morphology , quantity and projections of these cells are discussed further in the sections that follow . In flying insects , the paired NO are two spherical units situated ventrally in the CX , posterior to the CB ( Figure 1c ) . In bees and flies , the NO receive mixed arborizations from PFN ( Figure 4a–e ) and PEN cells ( Figure 4d–e ) , which propagate signals to and from the CB and PB columns in the contralateral hemisphere , as well as input from large tangential cells ( Figure 4f; Stone et al . , 2017; Hensgen et al . , 2021; von Hadeln et al . , 2020; Hulse et al . , 2020 ) . To reveal the detailed cellular architecture of the bumblebee NO , we traced all input neurons as well as all columnar neurons that supply the NO in our high-resolution data ( 24 nm resolution ) . Based on these analyses , the projections from tangential neurons delineated three domains in the bumblebee NO: the small unit ( NOs ) , main unit ( NOm ) , and the cap ( NOc; Figure 4f–g ) . These regions were supplied by distinct types of tangential cells . Two small types ( LNO4 and LNO5 ) innervated specifically the NOs , while two larger neurons densely innervated the entire NOm with numerous fibers ( LNO1 and LNO2; Figure 4f ) . A fifth large neuron , entering the NO together with the LNO4/5 cells , innervated a subcompartment of the NOm overlapping with the projection fields of LNO1/2 cells ( LNO3 ) . Finally , a set of 15–20 smaller input cells projected to the NOc region . These cells did not have neurites projecting from the direction of the ipsilateral lateral accessory lobes ( LAL ) , but emerged from a fiber bundle that vertically passed the EB and originated in the contralateral anterior protocerebrum ( Figure 4d ) . The same bundle contained numerous tangential cells of the FB , suggesting that the tangential neurons supplying the NOc region also belong to this group of neurons . This proposed morphology was confirmed by an intracellular dye fill from Megalopta ( Figure 4—figure supplement 1d ) and we named these neurons FB-NOc cells . With respect to columnar cells our reconstructions revealed 12–20 PFN neurons with large to medium fiber diameters in the NOm , while PEN cells were restricted to the NOs ( Figure 4d–e , g ) . Additional cells with smaller fiber diameters originated alongside the larger PFN neurons , and mostly projected exclusively to the NOc region . Those cells which projected to the NOc were termed PFNc cells and their collective fiber bundles were adjacent to the main PFN cells along the entire neurite path toward the PB ( Figure 4e ) . They thus most likely constitute the only identifiable subtype of PFN cells . Given their projection fields , these cells overlapped only with the FB-NOc cells but not with LNO neurons ( Figure 4e–f; Figure 4—figure supplement 1a–b ) . Compared to other columnar cells of the CX , PFN cells are by far the most numerous cell type . In our low resolution data , we identified 202 PFN cells total , while in the 100 nm data we identified a minimum of 160 PFN neurons in a single hemisphere , with many still left to be traced . Finally , with the 24 nm data set , and including the PFNc neurons , we were able to identify a total of 427 PFN neurons in one hemisphere ( Figure 3—figure supplement 1 ) , with a maximum number of 100 PFN cells projecting from a single column in the PB/CB ( Figure 4—figure supplement 1d; R4 ) . Based on these numbers , we estimate the total numbers of PFN cells to be at least 854 for the entire CX , double the number found in Drosophila ( Hulse et al . , 2020 ) . Interestingly , approximately half of the found PFN cells belong to the PFNc subtype ( 201 in the analyzed hemisphere ) . At least two thirds of PFN neurons had very small fiber diameters , making them impossible to trace using the 126 nm resolution data set . Additionally , as branches within the FB could only be traced over short distances and no layering was present in the NO , we were unable to subtype this large group of neurons further , leaving PFNc and PFN ( main ) as the only obvious subtypes . However , clear and consistent differences in the diameter of the main PFN neurites in each columnar bundle demonstrated that the population of PFN cells is not homogeneous and many subtypes likely exist . With regard to their columnar projection pattern , PFN cells are laterally shifted by one column in the FB and are entirely absent from the innermost PB columns ( Figure 4c ) . In this regard , they are identical to the PEN cells that connect the NOs to the EB ( see next section; Figure 6d ) . In the bumblebee EB we found two principle types of columnar cells , EPG/PEG cells ( Figure 5 ) and PEN cells ( Figure 6a–c ) . While in other species , the former can be further classified into two subtypes based on their opposite polarity , this distinction could not be made in our data , owing to the limited resolution of the data set . EPG/PEG cells as well as PEN cells both carry information between the PB and EB . While PEN neurons additionally arborized in the NOs , where their fibers overlapped with tangential LNO4/LNO5 cells ( Figure 6c ) , EPG/PEG cells extended an axon into a small region positioned laterally and anteriorly to the CX , close to the border of the protocerebrum and the antennal lobe ( Figure 5a–a’ ) . Based on the morphology of EPG/PEG neurons in other species , we concluded that this target region corresponds to the gall , despite the lack of obvious neuropil boundaries in synapsin labeled preparations . In the fruit fly , PEN and EPG/PEG cells ( hereon referred to as EPGs for simplicity ) form the core of the heading direction circuit ( Turner-Evans et al . , 2020 ) . Akin to a biological compass , EPGs track the flies rotational movement as a single bump of activity in the EB and two bumps of activity in the PB , one in each hemisphere ( Seelig and Jayaraman , 2015 ) . Tangential EB neurons feed predominantly visual information from the anterior visual pathway to EPGs in the EB , tethering the bump to features of the environment ( Fisher et al . , 2019; Kim et al . , 2019 ) . At the same time , rotational self motion cues are sent to the PEN cells , predominantly via the NO and probably also directly via the PB ( Turner-Evans et al . , 2017; Green et al . , 2017 ) . While activity in the EB input cells directly generates an activity bump in the EPG network that is based on allothetic information , idiothetic information laterally shifts the bump position via the PEN activity . Importantly , this circuit relies on an anatomical offset between the EB projections of PEN and EPG cells: EPGs follow a default projection pattern , whereas PENs are shifted by one column in the EB . Therefore , EPGs projecting to any given column in the EB will be flanked on either side by PEN fibers , connecting to neighboring EPG cells . This offset has the effect whereby PEN activation results in shifting the activity bump either to the left or to the right , thereby translating clockwise or counterclockwise body rotations into counterclockwise or clockwise movements of the neural activity in the EPG population ( Turner-Evans et al . , 2017; Green et al . , 2017 ) . An identical offset in projections between EPG/PEG and PEN cells was also present in the bumblebee CX: EPG/PEG cells followed the default projection pattern ( Figure 5e ) , whereas PEN cells were shifted by one column in the CB ( Figure 6d–f; Figure 6—figure supplement 1 ) . Moreover , in the bee the total number of EPG/PEG and PEN cells per CX column closely approximated their homologues in the fly . In total , 69 EPG/PEG cells spanned the width of the bee CX with four neurons present in each PB column except for the innermost columns , which only contained three neurons each ( Figure 5b–d; Figure 3—figure supplement 1 ) . In the 126 nm data set , we identified 34 PEN cells at two per PB column , with the exception of the outermost columns that contained three PEN cells and the innermost columns that were entirely devoid of PEN neurons ( Figure 6a–b , d; Figure 3—figure supplement 1 ) . Interestingly however , we found 20 PEN cells projecting from a single hemisphere in the 24 nm data set , which suggest a total of 40 PEN cells are present in the entire bee CX , nearly identical to numbers to those found in Drosophila ( Hulse et al . , 2020 ) . In the fly , 74 EPG/PEGs and 42 PENs span the width of the CX with a similar , but slightly different distribution of these cells per PB column ( See Hulse et al . , 2020 ) . Notably , we found one set of EPG/PEG cells with arborizations in the innermost PB columns of the contralateral hemisphere ( ‘EPG_L1’ and ‘EPG_R1’ in Figure 5c ) . In contrast to the rest of the EPG/PEGs , these cells contained branching fibers in the outermost EB columns of the ipsilateral hemisphere relative to their arbor in the PB ( Figure 6—figure supplement 1 ) . These cells may provide a pathway for an activity bump to ‘jump’ from one lateral end to the other in the bumblebee EB , which compared to Drosophila is structurally an open loop ( see Discussion ) . The FB is the largest and arguably the most functionally mysterious of the CX neuropils . In our data set , the majority of FB columnar cell types were found to connect the PB with the FB and onward to tertiary regions . They possessed arborizations in the FB where they commingle among themselves , FB tangential neurons , and interneurons that were confined solely within the FB , the hΔ cells , or pontine cells ( Figure 11; e . g . Hanesch et al . , 1989 ) . As with the EB , we focused our reconstructions on the columnar cells with large neurite diameters . We revealed at least nine types of FB columnar cells in the bumblebee CX ( Figure 3b ) . One group of cells , the already mentioned PFN cells , innervated the posterior portion of the FB and represented the most numerous columnar neurons of the FB . Even considering only the largest PFN cells obtained in the low resolution data ( 202 individuals ) their number exceeded the total count of all remaining columnar FB cells combined ( 163 neurons ) . This ratio is likely even more biased toward PFN cells when considering that the 202 PFN cells likely comprise only one fourth of the total number of these cells ( as identified in our high-resolution data set ) . As the extent of the fine processes within the FB was not traceable , the exact layer these cells project to could not be identified . Although fiber diameters of different sizes present in each columnar bundle of PFN cells demonstrates the existence of subtypes , these could not be reliably established . The limits imposed by the unresolved neuronal terminals apply to all neuron types of the FB and were amplified by the fact that neural processes outside the imaged tissue block were truncated . Nevertheless , all identified cell types were clearly distinct from each other based on their columnar projection patterns ( Figure 2e–f ) , the fiber trajectories within the FB ( Figure 2b’ ) , the exit point they left the FB , and the location of the main branches . Two types of neurons had main neurites located near the dorsal edge of the bundle connecting the PB with the FB and possessed characteristically large fiber diameters when leaving the PB ( Figure 7; Figure 8 ) . This main neurite entered the FB dorsally giving rise to numerous thin branches . While passing through the FB in a ventral direction this neurite dramatically , but consistently , thinned before turning toward the contralateral lateral complex ( Figure 7f; Figure 8e ) . Due to this thinning , many cells were lost at that point . All these cells followed a projection pattern that exhibited shifted innervations of the PB . Based on these features as well as a single dye filled example from the halictid sweat bee Megalopta genalis these cells were identified as PFL neurons , the main output cells of the CX . More detailed comparisons of the projection patterns with data from other insects ( Heinze et al . , 2013; Hensgen et al . , 2021; Hulse et al . , 2020 ) allowed defining them as PFL1/3 as well as PFL2 neurons . While the latter has a bifurcating neurite in all other species , the bifurcation was not identified in our data , very likely because of the extremely thin neurite diameter at the point where the bifurcation should be located . PFL2 neurons were unique in other aspects as well . Although not shown in the idealized schematics ( Figure 8d ) , these cells did not typically stick within well defined columns in the PB . For instance , PFL2 fibers that arborized in the innermost PB columns showed considerable overlap , and in some cases PFL2 fibers were found covering a span of around three columns in the PB , relative to the projection domains of EPG/PEG cells ( see for example Figure 14—figure supplement 1 ) . Two more groups of columnar FB cells were identified , all of which followed the default projection pattern , identical to that shown by EPG/PEG cells of the EB . The first set of cells was a second type of columnar cell intrinsic to the CX , as no fibers were found that exited the CX after branching in the FB ( Figure 9a–d ) . These cells were termed PF1 cells and existed in two copies per PB column . Whereas fibers within the PB were not resolvable in each individual , in some columns these fibers were sufficiently clear to suggest their existence across all individuals of this set . These neurons were the only cells innervating predominantly the anterior and ventral layers of the FB with numerous processes ( Figure 9f ) . Intriguingly , some fibers appeared to pass the border between the FB and the EB , sending fine processes into the EB . While this needs to be confirmed by higher resolution data , PF1 cells might be uniquely innervating both the FB and the EB ( Figure 9d , f–g ) . Finally , a set of four related cell types was found in more ventral regions of the W , X , Y , Z bundles and traversed the FB in a shared fiber tract ( Figure 10 ) . Their main neurite was comparably thin when entering the FB dorsally , but substantially thickened when giving rise to projections inside the FB , making these cells clearly distinct from PFL neurons . As their main neurite was very thin when entering the PB , it could rarely be traced inside the PB , yielding only a few examples with verifiable branches within the PB ( Figure 10 ) . This leaves some doubt as to whether all these cell types indeed posses significant PB arborizations . For two of the four cell types , PB branches were confirmed by intracellular dye fills ( Figure 10; Figure 10—figure supplement 1 ) . Assuming that PB branches are a feature of all these cells , the four types were named PFx1-4 ( Figure 10 ) . Two of the four cell types ( PFx3 , 4 ) sent bilaterally reaching fibers toward the superior medial protocerebrum ( SMP ) and/or the anteriorly located portions of the inferior neuropils ( INP; Figure 10 ) . Similarly , but without bifurcation , PFx1 neurons leave the FB on the anterior side , but turn toward the contralateral anterior brain , wrapping around the anterior surface of the medial lobe of the mushroom body ( Figure 10 ) . The target of these cells was unclear and might either be the lateral complex , or the anterior regions of the inferior protocerebrum . In contrast , PFx2 cells clearly sent fibers toward the lateral complex and ventromedial protocerebrum ( VMP ) along a similar trajectory as PFL cells and EPG/PEG cells , with their axon passing the medial lobe of the mushroom body posteriorly . An additional , unusual type of cell included neurons with large diameter fibers that occurred in two individuals in one hemisphere and four in the other . These cells resembled PFL neurons in that they possessed very large neurites near the PB , which then dramatically thinned after passing the FB and turns toward the contralateral lateral complex . These neurons did not give rise to resolvable branches within the FB , but the existence of fibers below our resolution limit cannot be ruled out . Due to their resemblance with PFL cells , they are also assigned to the FB and were named PFLx neurons ( Figure 3b ) . Interestingly , this cell type shows a clear asymmetry in that one individual in column L1 ( projecting to the ipsilateral lateral complex ) does not have a counterpart on the contralateral side . The FB is the only CX neuropil which generally contains a system of interneurons that are confined solely within its boundaries . These neurons contain input fibers in one column and output fibers in a column in the opposite hemisphere ( Heinze and Homberg , 2008; Hulse et al . , 2020 ) . Historically , these cells have been termed pontine cells ( Hanesch et al . , 1989 ) , but have been renamed as hΔ neurons in the fly literature given their proposed role as neurons that horizontally shift ( i . e . horizontal-shift: hΔ ) neural activity from one column of the FB to another column in the opposite hemisphere . Here , we adhere to the new nomenclature proposed by Hulse et al . , 2020 and refer to these cells as hΔ ( Figure 11 ) . Similar to columnar cells , hΔ neurons possessed cell bodies residing in the dorsal-posterior regions surrounding the CX ( Figure 11 ) . We identified as many as 188 hΔ neurons that tile the FB into vertical columns , very similar to the 190 hΔ neurons identified in the fly by Hulse et al . , 2020 ( Figure 11 ) . This high number makes them the second most numerous cell type of the CX after the columnar PFN cells . Their numbers were surprisingly unevenly distributed across the bundles of origin . Interestingly , the X and Y bundles on both hemispheres contained almost exactly twice the number of hΔ cells compared to the W and Z bundles ( Figure 11 ) . This suggested an overall branching system consisting of twelve lateral sectors across the width of the FB . Indeed , the cells originating in the X and Y bundles could be segregated into two adjacent arborization domains each ( Figure 11 ) . Most hΔ neurons possessed a midline crossing neurite passing the FB posteriorly ( purple in Figure 11 ) , while other , rarer types passed the midline along a more dorsal ( orange ) or anterior course ( green; Figure 11 ) . The collective arborizations of the majority of these posterior cells followed the described twelve vertical column system across the FB ( Figure 11 ) . However , most cells that belonged to the dorsally ( Figure 11 ) and anteriorly projecting hΔ groups ( Figure 11 ) , as well as one posterior hΔ cell type ( Type 4 ) appear to form only eight vertical columns . These cells shared projections within the anteriormost layer of the FB . Based on their morphology , relative distribution , and their arborization layer within the FB , we identified a minimum of five hΔ types ( Figure 11 ) . hΔ Type 1 contained minute fibers that innervated only the posterior surface of the FB and arborized in the dorsalmost layers . hΔ Type 2 cells had slightly longer projections within the FB , but arborized in the ventralmost layers . hΔ Type 3 cells occupied layers between hΔ cell Types 1 and 2 , and had longer FB projections than both types . hΔ Type 4 cells had branches that spanned the entire posterior-anterior axis of the FB , and appeared to occupy the same horizontal domain as Type 3 cells . The branching fibers which innervated the FB furthest away from the location of their cell bodies , that is , their likely output fibers , were positioned anteriorly relative to the domains of their putative input fibers . Further , these cells had wider branches and collectively formed eight columns as opposed to the twelve formed by Types 1 , 2 , and 3 . hΔ Type 5 cells had wide-branching fibers in the dorsal regions of the FB . Interestingly , the putative input fibers of these neurons entered the FB posteriorly , but their putative outputs enter dorsally and are reminiscent in morphology of a claw crane ( Figure 11 ) . Lastly , all anterior hΔ cells were combined into Type 6 ( Figure 11 ) . Due to the limited resolution of our data set , we were unable to trace finer branches and could not determine the FB layer in which these cells send their terminal branches . Our analysis is therefore only a first approximation and it is highly likely that more types of hΔ cells exist .
Additionally to the low-resolution data , we collected higher resolution data sets of the NO from a different individual . This not only allowed us to trace the fine neurites of PFN cells , which have been proposed to integrate heading with distance information in bees and likely play a major role in path integration ( Stone et al . , 2017 ) , but also enabled us to estimate how many neurons we are likely missing when relying on low resolution data only . The comparison between the three data sets revealed that the lowest resolution data clearly misses all neurons with neurite diameters smaller than a certain cutoff , in our case 0 . 6 µm . This cutoff was not identical across all regions of the data , as especially toward the boundaries of the main image stack , resolution was much poorer compared to the center . Additionally , the inability to resolve fine details also led to the loss of all but the main branches of arborizations within neuropils . For instance , in the noduli , the PFN neurons as well as the LNO input cells possessed only fibers totalling a few µm in the 126 nm data , but possessed cable lengths of several orders of magnitude larger when reconstructed based on the higher resolution data . Synapses were not clearly visible in any of the three data sets . This not only has implications for detailed maps of overlapping arborizations , but limits cell type identification . For instance , E-PG/P-EG cells both occupy similar regions in the PB , EB , and gall , share projection patterns and morphology , but differ in polarity , with one cell type receiving input predominantly in the EB ( E-PG ) and the other in the PB ( P-EG ) . The information needed to resolve these cell types clearly lies in higher resolution data . Other examples are PFL1 versus PFL3 cells , as well as subtypes of PFN or hΔ cells , all of which , in Drosophila , are defined based on the innervated layer of the FB and , ultimately , by their underlying connectivity . Without higher resolution and the associated ability to reconstruct full branching trees , we are unable to discern the subtype identity of any of these cells in the bumblebee . Adding a different perspective , we also compared the skeletons of our low-resolution EM-based data to detailed neuron morphologies obtained by intracellular dye injections and confocal microscopy . This analysis not only demonstrated that regions of cells that lie outside of the imaged volume are obviously missing from our data , but that our low-resolution EM data can be supplemented by full arborization trees obtained by light microscopical data , as long as neurons are selectively labeled . The complete morphologies obtained with this method confirmed the result of the comparison between low- and high-resolution EM data and highlighted that neither polarity indicators ( fine versus beaded endings ) , nor the full size of branches were captured in the EM-based reconstructions based on low-resolution images . However , mapping of EM-based skeletons and neuron skeletons based on confocal data yielded surprisingly good matches , even in the occasional cases when data from M . genalis was mapped onto bumblebee data . This indicated that the gross morphology of neurons can be readily compared across these imaging modalities and future data based on single neuron dye fills can indeed be used to validate identity of neuron types in our projectome data . We conclude that , despite all the missing details , known cell types can be reliably identified based on our data . Sub-classifications of these main types that are based on either fiber polarity or details of innervated layers are , however , not possible and require additional data for cross validation . Mostly owing to the missing projections outside of the CX , the morphology of unknown types of neurons are also impossible to determine without additional verification based on dye-filled preparations . Nevertheless , by using the presented data , even an individual example of a dye filled neuron is sufficient to assign identity to all individuals of this type in our projectome . Each complete neuron morphology therefore automatically provides information about the entire isomorphic set of its kind . The existence of our projectome thus substantially increases the value of single neuron morphologies obtained in future studies . For decades CX research did not question that the CX contains 16 columns in the PB ( e . g . Hanesch et al . , 1989; Heinze and Homberg , 2008; El Jundi et al . , 2018 ) . Based on work in the locust , this assumption was used to draw functional conclusions for example about the basis of representing compass directions ( Heinze and Homberg , 2007 ) . More importantly , it was implied by generalizations across species , that the CX organization in one species ( e . g . the locust ) was equivalent to that of others . This was well justified by the developmental origin of the columnar CX neurons , which , across insects , are derived from eight neuroblasts , four on either side of the midline ( Boyan and Williams , 2011 ) . Each of these gives rise to one of the W , X , Y , Z bundles and thereby provides neurons for two PB columns ( Williams et al . , 2005; Williams and Boyan , 2008 ) . However , detailed analysis using genetic methods in the fruit fly revealed that the fly PB consists of nine instead of eight columns ( glomeruli; Wolff et al . , 2015 ) . This significantly challenged the notion that data between species could be easily generalized and raised the question of how to homologize neurons across species . Our data clearly revealed that , like the fly ( Wolff et al . , 2015 ) , the bumblebee PB also comprises nine columns in each PB hemisphere ( Figure 2c–d ) . As in most other insects , vertical columns are not visible using immunohistochemical labeling with antibodies against synapsin , neither in the FB , EB or PB , and a definite insight into this structural feature required EM based tracing of main fiber trajectories of the CX . Interestingly , the existence of nine columns in the EB was already reported by Williams , 1975 and was confirmed also for bees , e . g . by Hensgen et al . , 2021 . In each case , seven wide , medial columns are flanked by two narrow hemicolumns at the lateral edges of the EB , generating eight equal sized projection fields along the length of the EB . The division of the fly EB into eight equally wide segments ( referred to as ’tiles’ Wolff et al . , 2015 ) around the EB ring is directly equivalent and suggests that the existence of eight versus nine columns in the PB does not affect the overall structure of the CX ( Figure 12 ) . Rather , the ninth column likely evolved after the core circuitry of the CX . This hypothesis is consistent with the eightfold symmetry of the CX head direction circuit in flies ( Hulse et al . , 2020; Pisokas et al . , 2020 ) . It is also supported by our data , which show that most divergence in projection patterns between the fly and the bumblebee neurons was found at the innermost PB column ( Figure 2c ) , that is this additional column provides the substrate to evolve novel CX components without disrupting the existing functions based on the outer eight columns . This also suggests that the columns 2–9 in flies and bees are homologous to columns 1–8 in locusts and other insects . The consistent lack of PFN and PEN neurons in the innermost columns further supports this notion ( Figure 4c; Figure 6d ) . Underlining the high degrees of organizational conservation , we were able to identify four discrete columnar cell projection patterns ( Figure 2e–f ) , three of which were identical to the patterns produced by the corresponding neurons in the fly . Importantly , these projection patterns define the computations that are carried out within the EB and the FB . This resemblance in structure therefore suggests that highly conserved computational principles operate at the core of the CX across insects . For instance , functional recordings and connectomics analysis in the fly have led to a model in which navigation-related vector computations are carried out by columnar cells and interneurons in the FB ( Hulse et al . , 2020; Lu et al . , 2020; Lyu et al . , 2020 ) . Columnar cell projection patterns provide the structural backbone of these computations and would therefore define which vector operations can be achieved . Using the same argument , those patterns that were identified as being different , namely those of PFL2 neurons in bees and of PFL3 neurons in flies , directly suggest that distinct computations are implemented in either species , a finding that will be discussed in more detail below . On the level of neuropils , and in contrast to the strictly conserved columnar organization , the layers of the CB identified via immunohistochemical staining against serotonin ( 5HT ) and tyrosine hydroxylase ( TH , an enzyme required to produce dopamine ) , bear no close resemblance between bees and flies . We found at least three horizontal layers in the FB and two layers in the EB ( Figure 1d–e’’ ) , while , based on immunolabeling of synaptic markers alone , many more layers are present in flies ( Wolff et al . , 2015 ) , as well as in other insects , like locusts ( von Hadeln et al . , 2020 ) and butterflies ( Heinze and Reppert , 2012 ) . As the layers of the CB largely correspond to innervation by tangential input neurons , this suggests that which information is fed into the CX is highly species specific . In contrast , the computational algorithms ( defined by the columnar layout ) that use this input information are much more conserved . These structural differences are consistent with the idea that different sensory cues are relevant to different species , but that the decision about where to turn to in response to sensory information follows the same guiding principles ( Honkanen et al . , 2019 ) . The same argument applies to input into the FB that relays internal state or memory output from the mushroom bodies ( Hulse et al . , 2020 ) . Finally , the most pronounced difference between bees and flies was found in the organization of the NO . While the fly data , and in fact data on many other species ( locusts , butterflies , moths , beetles; Heinze and Homberg , 2008; Heinze and Reppert , 2012; Immonen et al . , 2017; Adden et al . , 2020 ) , reveal a highly structured organization into stacked layers , we found the bee noduli to be comparably disorganized . Our reconstructions only revealed three discrete territories , providing the possibilities for increased levels of cross-talk between the PFN neurons that remain segregated in other species ( Figure 4e-e' ) . Interestingly , and discussed in detail below , pronounced intra-columnar microcircuits among PFN cells was one prediction made by Stone et al . , 2017 for insects with highly developed path integration ability , such as bees . In total , we found 11 types of columnar cells ( Figure 3 ) . Importantly , the majority of these cell types have known homologues in other insects . These include EPG/PEGs , PENs , PFNs , hΔs , and putative motor output cells , PFL1 , 3 and PFL2 ( Homberg , 1985; Heinze and Homberg , 2008; Heinze et al . , 2013; Wolff et al . , 2015; Stone et al . , 2017; El Jundi et al . , 2018; Hensgen et al . , 2021 ) . Drawing comparisons to the fly data , we can distinguish four different categories of cross species resemblance: First , some cell types appear to be extremely conserved , including in quantity , distribution , morphology , and projectivity . In particular , EPG/PEG and PEN cells were indistinguishable between the species with respect to these parameters . Second , other cell types differed in numbers but maintain key characteristics , such as projection pattern , suggesting conserved roles in CX computations . This was the case for PFN and hΔ cells , at least when disregarding potential subtypes . Third , some cell types share quantities , but differ in projectivity . In this category we found the main output neurons of the CX in particular , that is , PFL1 , 2 , 3 cells ( discussed in detail below ) . Finally , some cell types appear to have no counterpart in the fly and might be unique to the bee . These include PFx1 , PFx2 , PFx3 , PFx4 ( Figure 10 ) , PF1 ( Figure 9 ) , and PFLx neurons ( Figure 3b ) , all of which project to the FB following the ’default’ pattern . In contrast to findings in the fruit fly , we did not observe any clear homologues for vΔ cells . In the fly , these cells have input and output branches constrained within single FB columns , but different layers , potentially shifting neural activity vertically ( i . e . across layers; Hulse et al . , 2020 ) . As these neurons have been implied in vector transformations based on fly connectomics data , the lack of such neurons in bees , if confirmed , can be expected to have important functional consequences . However , PF1 neurons in the bee , which only innervate the PB and FB and have branches across multiple layers in the FB may fulfill a similar role . Additionally , bee counterparts of vΔ cells might have small fiber diameters and could have been missed in our analysis . Interestingly , we also did not find any obvious candidates matching the fly FX cells , columnar neurons lacking arborizations in the PB ( Hulse et al . , 2020 ) . One potential candidate in the bee could be the PFx cells ( Figure 10 ) . Three of these cell types that have bilaterally projecting fibers toward regions in the superior medial protocerebrum ( SMP ) , similar to the fly FX cells ( Figure 10 ) . However , whereas FX cells in the fly only arborize in the FB , PFx neurons in the bee also possess arborizations in the PB ( Figure 10 ) . Intracelluar injection of one such cell revealed processes innervating both hemispheres of regions possibly within the CRE ( Figure 10; Figure 10—figure supplement 1c–d’ ) . Finally , a fourth FB columnar cell of the bee , PFx2 neurons , sends fibers contralaterally toward the LX , but follows the default projection pattern , making these cells clearly distinct from the PFL output neurons . Given these characteristics , these cells might correspond to Drosophila PFRa or PFG cells . Both cells receive input from PFN neurons in flies . This outlines a possible line of enquiry to establish equivalence despite diverging morphology by delineating the up- and downstream synaptic connections and hence place neurons in a corresponding computational context . Two columnar cell types , EPG and PEN cells , link the EB and the PB and form the core of the head direction circuit in the fly ( Turner-Evans et al . , 2020 ) . The numbers of both EPG and PEN cells per PB column identified in the bumblebee closely approximated corresponding numbers in the fly ( Figure 3a , Figure 3—figure supplement 1 ) and , importantly , their projection patterns , characteristically offset by one column in the EB between both cell types , were also conserved ( Figure 6e–f , Figure 6—figure supplement 1 ) . One major distinction between the bee and the fruit fly is the gross morphology of the EB . As the name suggests , the EB in the fly is toroidal , whereas in the bee the EB more closely resembles an eyebrow or a bent cylinder . In the fly , this morphology is a prime example of structure matching function . Because the EB is shaped as a torus , EPG cells from medialmost PB columns and EPGt cells from the lateralmost ninth columns have overlapping fibers within the EB , enabling an heading encoding activity bump to move in 360° with the same rotational freedom as a compass needle rotating within a compass ( Wolff et al . , 2015; Hulse et al . , 2020 ) . Based only on gross morphology , the bee EB would not seem to be well suited to serve this same function . However , our data revealed that in bumblebees the proposed head-direction circuit differs from its fly counterpart at lateral extremes of the EB , that is those positions that would correspond to the ring closure in Drosophila . Specifically , we found that EPG/PEG cells in the innermost PB columns were swapped in the bee relative to the fly . Unlike in all other columns , these cells project exclusively within the ipsilateral hemisphere and innervate the outermost EB column on the same side of the CX ( Figure 5d , e; Figure 6—figure supplement 1; Figure 12 ) . Additionally , the projections of the PEN cells in the lateralmost column of the EB covered twice the width as those PEN cells located in all remaining EB columns . These differences might enable a circuit solution to compensate for the morphologically open EB and produce a closed , ring-like circuit suited to encode smooth rotational movements in the bumblebee . Intriguingly , in a recent modeling study , Pisokas et al . , 2020 proposed that locusts have evolved a third solution to closing the loop , a solution that differs from both the bee and the fly ( Figure 12 ) . While most insects do not have a closed EB , evidence presented here and by Pisokas et al . , 2020 suggests that these different circuit modifications may fulfill the same functional purpose . It is worth noting however , that by modeling the head direction circuit in the locust and the fly , Pisokas et al . , 2020 found that other , more minor , morphological differences between the circuits of both species can significantly alter the dynamics of the ring attractor circuit underlying the head direction system in ways that could adapt circuit function in line with species specific behavioral demands . It will thus be interesting to explore the possible functional consequences and possible behavioral correlates of the detailed characteristics of the bumblebee head direction system . Several recent functional studies have implicated the NO as centers that relay idiothetic self-motion cues , including translational optic flow ( Lu et al . , 2020; Lyu et al . , 2020; Stone et al . , 2017 ) , rotational angular velocity ( Green et al . , 2017; Turner-Evans et al . , 2017 ) , and wind direction ( Currier et al . , 2020 ) to other regions within the CX . So far three neuron types that innervate the NO have been shown to facilitate the encoding of self-motion cues , with anatomical homologues having been identified across insect species ( Heinze and Homberg , 2008; Heinze et al . , 2013; Wolff et al . , 2015; Stone et al . , 2017; El Jundi et al . , 2018; von Hadeln et al . , 2020; Hensgen et al . , 2021 ) . These include PFN , PEN , and LNO tangential cells ( Figure 4e-f ) . LNO tangential neurons provide predominantly input to PFN and PEN neurons , carrying sensory information from other regions of the brain , mostly from the ipsilateral LX ( Stone et al . , 2017; Currier et al . , 2020; Lyu et al . , 2020; Lu et al . , 2020; Hulse et al . , 2020 ) . Different LNO types carry sensory information to the NO along multiple parallel channels . In the fly , this is reflected by the structural layout of the NO , in which the fibers of LNO , PFN , and PEN cells are neatly organized into discrete layers . Five subtypes of PFN cells project into five territories in the fly NO . With the addition of PENs that arborize in the dorsalmost layer , there are six distinct zones , each supplied by its own LNO input channel ( Wolff and Rubin , 2018; Hulse et al . , 2020 ) . Additionally , the fly NO contains numerous FB tangential cells ( FBt ) which overlap with and in some cases receive direct input from LNO cells ( Hulse et al . , 2020 ) . In the bee , we found three anatomical domains , NOs , NOm , and NOc , based on the input arborizations from tangential cells ( Figure 4e-g ) . Similar to the fly , PEN cells , likely encoding angular velocity , send their processes to a single zone , the NOs , where they receive isolated tangential cell input and do not overlap with PFN cells ( Figure 4e ) . In contrast to flies however , PFNs only arborize in two structurally discrete territories , the NOm and NOc . All PFNs that are associated with the same column in the PB display a surprising amount of overlap in the NO , with many spreading throughout the entire NOm or NOc domain ( Figure 4—figure supplement 1b ) . Similarly to the PFN organization , the LNO input cells associated with this region are equally overlapping . Besides suggesting fewer sensory input channels in bees than flies , with the notable exclusion of contralateral input , such an organization additionally offers the possibility of a large degree of intra-column recurrent connectivity in the bee NO relative to the fly , a prediction made by Stone et al . , 2017 to support path integration memory . However , connectivity data will be needed to verify such a claim . A second prediction of this model was that more PFN neurons would be required to increase the capacity and precision of vector memory in species with highly developed path integration . Given that we found more than twice the number of PFN cells in bees compared to the fly , this prediction is clearly met , even though its functional significance remains hypothetical . Even if all circuits present in flies are conserved in bees , the newly discovered circuit of PFNc neurons in the bee NOc provides an additional possibility for structured recurrent connections between PFN neurons as potential substrate for path integration memory that can be directly used for steering ( Stone et al . , 2017 ) . The fly connectome data revealed that in Drosophila neither highly structured intracolumnar recurrent connections exist , nor are PFN outputs to steering cells hemisphere specific ( as required by the Stone et al . , 2017 model ) . Both assumptions can in theory still be met in PFNc neurons , as long as they receive speed input indirectly , for example via PFNv neurons in the PB . Another specialization of the bee noduli compared to their fly counterparts is evident in the structure of the NO input . Interestingly , whereas the NOm contains putative input fibers from three LNO cells , the NOc receives no branches from LNO cells at all . Rather , the NOc is innervated by a group of 15–20 FB tangential cells that , based on projection , quantity , and lack of overlap with LNOs , have no equivalent in the Drosophila CX and were termed FB-NOc neurons . Based on one completely filled example from Megalopta , these cells have distant ventral cell bodies and fibers that extend dorsal-anteriorly , occupying regions in the vicinity of the SMP/CRE ( FB-NOc cells; Figure 4h , Figure 4—figure supplement 1d ) . While we cannot confirm identical morphologies of the neurons in the bumblebee and these cells might also comprise a family of related cell types rather than a group of identical cells , the fiber trajectories of all found neurons are fully consistent with the morphology of the Megalopta neuron . Our data therefore strongly suggest that the FB-NOc tangential neurons provide an additional input channel to the bee NO that specifically synapses onto the PFNc cells , the smallest and most numerous PFN cell type . The much higher cell count of PFN cells in the bumblebee compared to the fly therefore appears to be closely associated with the presence of an additional input channel . While the function of these FB-NOc tangential cells is unknown , their arborizations in the NO , the FB , and the SMP/CRE and their apparent uniqueness within the bee make them potential candidates to play a role in mediating the highly developed ability of bees for vector navigation . As suggested by Le Moël et al . , 2019 , and based on the computational model by Stone et al . , 2017 , complex navigation strategies , such as trap lining , could be explained by vector navigation , when enabling storage and retrieval of PFN population activity in long-term memory , for example in the mushroom body . Stored vectors from previous foraging segments could be compared to and combined with vectors representing the current state of the path integrator , enabling the choice of optimal foraging routes and novel shortcuts ( Le Moël et al . , 2019 ) . The key prediction of such a model is direct information flow between the mushroom body and the CX path integration memory , for example PFN neurons , both for storing and for recalling vectors . The found FB-NOc neurons are anatomically suited to serve this function , as their branches outside the CX coincide with brain regions innervated by MB output neurons ( Rybak and Menzel , 1993 ) . A prediction of this hypothesis is that the connection of these neurons to and from PFNc neurons is column specific , an idea that is directly testable via connectomics work and which is consistent with the high number of 15–20 FB-NOc neurons per hemisphere ( ca . two per PFN column ) . In summary , the specializations identified in the noduli of the bumblebee are consistent with the idea that this region plays a key role in path integration and that the highly developed abilities for path integration related behaviors in bees are indeed reflected in the circuitry of the noduli . Further analysis of the synaptic connectivity of the NOc circuit will have to test these predictions . PFL neurons relay signals to descending neurons in the lateral accessory lobes ( LAL ) , making them likely candidates for cells which propagate steering or other motor commands to the ventral nerve cord ( VNC ) ( Figure 7; Figure 8; Stone et al . , 2017; Steinbeck et al . , 2020; Rayshubskiy et al . , 2020; Hulse et al . , 2020 ) . Connectomic analysis of the fly CX revealed the presence of three types of PFL neurons that differ in their PB-FB projection offset and downstream partners . PFL1 cells were found to be offset in their PB-FB projections by one column ipsilaterally , PFL3 cells by two columns ipsilaterally , and PFL2 cells by four columns ipsilaterally ( Hulse et al . , 2020 ) . While PFL1 and PFL3 cells send output fibers to the contralateral LAL , PFL2 neurons contain bilaterally projecting output fibers that arborize in both LALs . In the bumblebee , we found evidence for a single cell type that shares characteristics from both fly PFL1 and PFL3 neurons ( Figure 7 ) . Interestingly , bee PFL1 , 3 neurons only have a one column ipsilateral PB-FB projection offset ( ’PB-Shifted’; Figure 7 , Figure 13 ) . Intracellular injection of a PFL1 , 3 cell revealed blebbed varicosities in the LAL and mixed branches in the FB , suggesting that this polarity is maintained across insects ( Figure 7a; Hensgen et al . , 2021 ) . Although detailed analysis of the FB projection fields of these neurons will have to confirm the identical offset patterns for PFL1 and PFL3 cells , our data suggests that the PFL3 projection pattern of the fly does not exist in bees . Additional discrepancies between bees and flies were identified in PFL2 neurons ( Figure 14 ) . With a four column ipsilateral PB-FB offset , fly PFL2 neurons project only to ipsilateral regions within the FB . In contrast , bee PFL2 neurons show only a three column ipsilateral offset between PF-FB columnar arborizations ( ’Both-shifted’; Figure 8d–e ) . Notably , there is one PFL2 neuron in both hemispheres that differs in its projection offset compared to the other PFL2 cells by being shifted one column further ipsilaterally , matching the fly projection offset ( Figure 8b; R3 and L3 ) . Further , the PFL2 neurons in the lateral most regions appear to project to the two furthest contralateral columns in the FB ( Figure 8d–e; Figure 14 ) , generating an offset pattern that is largely dependent on which PB column an individual PFL2 cell originates in . The very wide input domains in the PB further complicate functional interpretations for PFL2 neurons in bees , as they appear to be active across a much wider range of directions than other cells ( each PFL2 cell covers a range of at least 90° of head directions; Figure 14—figure supplement 1 ) . PFL neurons are thought to relay motor commands , such as steering ( PFL3 neurons; Rayshubskiy et al . , 2020; Hulse et al . , 2020 ) and forward velocity ( PFL2 neurons; Hulse et al . , 2020 ) , from the CX to the LALs ( Stone et al . , 2017; Steinbeck et al . , 2020 ) . Given that PFL1 , 3 neurons in the bee share characteristics with both fly PFL1 and PFL3 neurons , such an arrangement may suggest that the projection offset of these neurons evolutionarily diverged in the fly . This is supported by the fact that similar numbers of cells per column exist for all PFL1 + PFL3 neurons in the fly and all PFL1 , 3 neurons in the bee . Further , an identical bee-like projection offset for both PFL1 and PFL3 neurons is also consistent with data from dye filled neurons in the Monarch butterfly ( Heinze et al . , 2013 ) . PFL3s in the fly would be well suited to drive rotational movements when the fly is facing +/- 90° away from a goal location ( Hulse et al . , 2020 ) . Due to their offsets in the FB and bilaterally projecting output fibers in the LALs , PFL2s were suggested by Hulse et al . , 2020 to drive forward motion when the fly is facing in the direction opposite to their ’stored’ vector ( i . e . facing toward their starting point or goal vector ) . Assuming a corresponding function in the bee , PFL1 , 3 neurons would be most active ( i . e . propagate steering commands ) when bees are facing +/- 45° away from their ’stored’ vector and PFL2 neurons would drive forward velocity when they are facing +/- 135° away from their stored vector ( although see Figure 14—figure supplement 1 ) . Interestingly , if we assume direct and exclusively contralateral connections between PFN and PFL neurons to underlie steering as hypothesized by Stone et al . , 2017 , the opposite offset by PFN neurons would add 45° to these values , leading to a fly-like +/- 90° and +/- 180° away from their ’stored’ vector ( Figure 13 ) . But as the detailed projections of PFL cells , that is the extent of their inputs in the PB and their projection width and innervated layer in the FB , are unknown and partially differ between cells originating in different PB columns , conclusions about phase offsets and possible functional implications for behavior remain preliminary , especially without support from computational models . Nevertheless , the distinct anatomical offsets between flies and bees clearly suggest that bees have evolved a different mechanism for target driven steering , as strictly applying the fly-derived steering concept to bees would result in a 45° error relative to a stored vector . Synaptic resolution connectivity data are clearly required to resolve this interspecies discrepancy . Overall , the differences found at the level of the main output neurons of the CX are intriguing and suggest that the behavioral control of steering does depend on the species , possibly explaining divergent flight patterns and , combined with distinct sensory input channels , might possibly even account for different navigational strategies . In summary , while limited in resolution and missing the smaller cell types of the CX as well as the projections that leave the imaged volume , the projectome analysis of the bumblebee CX provides pioneering insights about the quantity and projectivity of all major CX neurons in the brain of the bumblebee . To the best of our knowledge , this project is the first attempt at comprehensively mapping all columnar and hΔ cells in a non-dipteran insect species , the projections of which define and constrain computations carried out by the CX .
In past studies , nomenclature used to identify cells and CX areas differed between fruit flies and other insects . Due to overwhelming support for CX homology across insects , the greater availability of functional and neuroanatomical data in Drosophila ( Hulse et al . , 2020 ) , and in an effort to make our anatomical descriptions more comparable to the Drosophila connectomics data , we adhere here to the short hand nomenclature used for Drosophila ( Ito et al . , 2014; Scheffer et al . , 2020; Hulse et al . , 2020 ) but have also provided alternative names historically used in other insects ( Table 1 ) . Briefly , columnar neurons are named with a three letter abbreviation according to the neuropils they innervate . In the fruit fly , these abbreviations are ordered by connectivity , with the first two letters being determined by the two neuropils the neuron receives the most input from , in order of which has the greatest amount of input first , and the third letter is given by the neuropil in which the neuron sends predominantly output to . For instance , a PEG cell receives mostly input in the PB and the EB and sends output to the gall . For a more in-depth description of fruit fly CX neuron nomenclature , see Hulse et al . , 2020 . Since connectivity information does not exist for neurons in our dataset , we define neuron homology based on corresponding morphology and arborization domains between species . As general connectivity patterns are likely conserved as well , based on the tight anatomical resemblance as well as functional conservation of CX neurons across species , we have adopted the fly naming scheme for the bumblebee . For neurons in which there are no clear homologues in the fly , we used the neuropils that they innervate within the imaged volume to define the neuron name . For example , PFx neurons innervate the PB , the FB , and an unknown tertiary region denoted here as ‘x’ . Lastly , we consider all CX neurons to fall into one of three classes: columnar neurons , FB interneurons ( i . e . , hΔ ) , and tangential neurons ( see Introduction ) . Neurons within these classes were defined as belonging to the same neuron type if they were anatomically indistinguishable from each other , yielding the columnar neuron types listed in Table 1 as well as several types of hΔ neurons . While some of these likely consist of several distinct subtypes based on connectivity and detailed projection areas ( e . g . PFN neurons ) , we were not able to unambiguously identify those due to the limits imposed by the resolution of our image data . Bumblebees ( Bombus terrestris ) , were obtained from a commercial supplier ( Koppert , Netherlands ) and kept in a greenhouse on campus at Lund University , Sweden . Bee colonies were maintained inside room-sized flight canvases with sugar water and pollen available from artificial feeders at all times . Only adult female workers were used in this study . A few of the intracellular dye injections were performed on female adult sweat bees ( Megalopta genalis ) and were used in this study to confirm single-cell morphologies where possible . Sweat bees were collected in the rainforest of Barro Colorado Island ( field station of the Smithsonian Tropical Research Institute , Panama ) . Sweat bees were collected with a light trap made using a white sheet and illuminated by a light source containing UV wavelengths . Traps were set during early morning twilight , a time when the bees are most active . Captured bees were kept in vials containing containing honey solution and water solution soaked into cotton balls and were processed within 2 weeks following capture . Whole mount synapsin immunolabeling followed a slightly modified version of the method described by Ott , 2008 . Bees were anesthetized to immobility over ice , after which their heads were removed and immediately placed into freshly made fixative containing 1% paraformaldehyde ( PFA ) , 0 . 25% zinc chloride , 0 . 79% sodium chloride , and 1 . 2% sucrose . Neural tissue was dissected free from the head capsule , removing as much of the neural sheath as possible , and left to fix overnight at 4°C . The next day , brain tissue was rinsed 8 x with HEPES buffered saline ( HBS ) over the course of an hour . Tissue was then rinsed 3 x for 10 min intervals in Tris-HCl ( pH 7 . 3 ) before being transferred into Dent’ s fixative containing 80:20 methanol to dimethyl sulfoxide ( DMSO ) were it was left to soak for 75 min on a gentle shake . Next , neural tissue was rinsed 3 x in Tris-HCl over 10 min intervals and placed into a blocking solution consisting of 5% normal donkey serum ( NDS ) in 0 . 01 M phosphate buffered saline with 0 . 5% Triton-X 100 ( PBST ) . After a 3 hr incubation , blocking solution was swapped with primary antibody solution containing 1% NDS , 0 . 5% PBST , and 1:25 mouse anti-synapsin monoclonal antibodies ( 3C11; Developmental Studies Hybridoma Bank; Iowa City , IA ) . In some experiments , a polyclonal antibody raised in rabbit against serotonin ( 5-HT; AB_572263; ImmunoStar; Hudson , WI ) was also included at a concentration of 1:2000 . Brains were left to incubate in this solution either in the fridge or at room temperature on a shaker for 4–6 days . The following day , brains were rinsed 6 x over the course of an hour in 0 . 1% PBST . Meanwhile , whole IgG secondary antibodies raised in goat and conjugated to Cy3 or Cy5 ( AB_2338003 and AB_2338713 , respectfully; Jackson ImmunoResearch; Philadelphia , PA ) were added to Eppendorf tubes containing 1% PBST at a concentration of 1:300 . Anti-mouse conjugated to Cy5 was used to detect synapsin and anti-rabbit conjugated to Cy3 to detect 5-HT , when applicable . Secondary-containing Eppendorf tubes were then placed in a centrifuge and spun for 5–10 min at 8000 g to separate potential antibody aggregates . The top 900 µl of secondary antibody solution was then added to the whole brains , which were subsequently left to incubate for 1–3 days either at 4°C , or on a shaker at room temperature . After secondary antibody labeling , brain tissue was washed 3 x 20 min with 0 . 1% PBST , 2 x 10 min with PBS , and then dehydrated through a series of increasing ethanol concentrations ( 50% , 70% , 90% , 2 x 100% ) at 10 min intervals . Brains were next transferred into a 1:1 mixture of ethanol to methyl salicylate for 15 min , and then cleared in 100% methyl salicylate for 75 min . Finally , brains were mounted in Permount mounting medium on slides between two coverslips and separated by spacers , where they were stored until imaging . For tyrosine hydroxylase ( TH ) immunolabeling , brains were dissected and fixed in 4% PFA in 0 . 1 M PBS containing 3% sucrose for 30–45 min . Fixation times lasting longer than an hour resulted in reduced or completely absent immunoreactivity , as has been previously reported in locusts ( Lange and Chan , 2008 ) . Whole brains were then rinsed 2 x 10 min in 0 . 1 M PBS , 3 x 10 min in 1% PBST , and subsequently blocked in 1% PBST containing 5% NGS for 3 hr on a gentle shake . Next , neural tissue was left to incubate for 2–3 days in primary antibody solution consisting of 1:250 mouse anti-TH ( AB_572268; ImmunoStar; Hudson , WI ) in 1% PBST with 1% NGS . Rinsing , secondary antibody labeling , dehydration , and mounting were then performed as described above . Intracellular dye injections were carried out during intracellular electrophysiology and followed the protocol described in Stone et al . , 2017 . In short , electrodes with a resistance of 50–150 MΩ were drawn from borosilicate glass capilaries ( Sutter P-97 puller ) . Bees were anesthetized over ice until immobile and then waved to a plastic holder . A frontal window was cut into the head cuticle and , where necessary , air sacs , fat , and neural sheath were removed using tweezers . Electrode tips were then filled with 4% neurobiotin ( Vector Laboratories ) in 1 M KCl , and backed with 1 M KCl . Silver wire placed in the ventral part of the head near the mandibles as a reference electrode and the main electrode inserted frontally in the brain and positioned using a micromanipulater ( Sensapex , stepping mode ) . Once a cell was successfully impaled , a depolarizing current of 1–3 nA was applied to iontophoretically inject neurobiotin . Injected brains were dissected free from the head capsule and fixed in a solution containing 4% PFA , 2% saturated picric acid , and 0 . 25% glutaraldehyde in 0 . 01 M PBS overnight at 4°C . The next morning , brains were washed 4 x 15 min in 0 . 01 M PBS and then left to incubate in 0 . 3% PBST containing 1:1000 Streptavidin conjugated to Cy3 for 3 days . Next , the brains were washed 4 x 20 min in 0 . 3% PBST and dehydrated in an increasing ethanol series , cleared in methyl salicylate and mounted in Permount as described in the previous section . Mass dye applications followed an identical protocol , but instead of intracellular iontophoretic injection , neurobiotin crystals were applied directly to the tissue . To that aim , the fine tip of an intracellular recording electrode was broken off and the remaining , coarse tip was dipped in petroleum jelly . The coated tip was used to pick up a few crystals of neurobiotin and then manually inserted into the target region of the exposed brain . Before injection , the brain was desheathed and all overlaying liquid was removed . After injection and careful rinsing with bee ringer solution , the dye was allowed to diffuse for 30 min , before the brain was removed from the head capsule and fixated in 4% PFA . To better determine location of arborization domains , injected brains were imaged , re-hydrated , and co-stained with immunohistochemical markers using the method reported by Heinze et al . , 2013 . Briefly , brain tissue mounted in Permount was freed by soaking in xylene for 2–3 hr before re-hydration in a decreasing ethanol series of 100% x 2 , 90% , 70% , and 50% ethanol at 15 min steps . Brains were then rinsed 3 x 15 min in 0 . 01 M PBS and transferred to 0 . 5% PBST . Next , brains were embedded in albumin gelatin ( 4 . 8% gelatin and 12% ovalbumin ) and post fixed with 4% formalin overnight at 4°C . The following day , brains were rinsed in 0 . 01 M PBS 3 x 15 min and vibratome sectioned at 140 µm . They were subsequently processed for immunohistochemistry as described in the previous section beginning at the blocking step , although with shorter incubation periods . Brain sections were left in blocking solution for 3 days , primary antibody solution for 4 days , and secondary antibody solution for 2 days . Rinse cycles and temperatures were the same . Following antibody incubation , sections were rinsed 4 x 15 min in 0 . 1% PBST , 2 x 15 min in 0 . 01 M PBS , then dehydrated in an increasing ethanol series at 10 min steps , cleared in methyl salicylate for 20 min and mounted in Permount between two coverslips separated by spacers . Methodology used here has been previously reported in Stone et al . , 2017 . Briefly , bee brains were dissected and fixed in a solution of 4% PFA and 2% glutaraldehyde in sodium cacodylate buffer overnight at 4°C . Neural tissue was then rinsed 4 x 15 min in 0 . 01 M PBS , embedded in albumin/gelatin , and post fixed overnight at 4°C . To image the CX , a single thick section ( 400 µm ) was cut from the albumin/gelatin block using a vibrating blade microtome and stored in 0 . 01 M PBS until further processing . This same technique was used for the noduli , using a smaller section thickness ( 200 µm ) . Large volume en bloc staining was then performed , beginning with osmification in a solution containing 2% osmium tetroxide and 1 . 5% potassium ferricyanide in double distilled water ( ddH20 ) for 1 hr at room temperature . Tissue was then washed 3 x 5 min in ddH20 and then sequentially immersed in 1% thiocarbohydrazide for 20 min and 2% osmium tetroxide for 30 min both steps being followed by 3 x 5 min rinses with ddH20 . Neural tissue was then left to incubate in 1% uranyl acetate overnight at 4°C . The next day , tissue was washed 3 x 5 min with ddH20 before being left to soak in lead aspartate for 60 min at 60°C . Lead aspartate solution was made by adding 0 . 066 g lead nitrate to 10 ml ddH20 , with a pH adjusted to 5 . 5 using KOH . Next , tissue was rinsed 3 x 5 min with ddH20 and dehydrated in an increasing ethanol series ( 20% , 50% , 70% , 90% , 2 x 100% ) at 5 min intervals . Samples were then slowly infiltrated with increasing concentrations of Durcupan resin to ethanol ( 25% , 50% , 75% ) at 2 hr intervals and left in 100% Durcupan overnight . The following day , tissue was transferred to fresh Durcupan for 2 hr and left to polymerize for 48 hr at 60°C . Finally , samples were trimmed and mounted onto aluminum stubs using two part conductive silver epoxy . Blocks of brain tissue were imaged using a Zeiss Sigma VP scanning electron microscope equipped with a Gatan 3View ultramicrotome . Three different scans were acquired for this study: an overview scan of the entire central complex imaged at a voxel size of 126 nm x 126 nm x 100 nm ( field of view 400 µm x 400 µm ) , a slightly higher resolution scan of the noduli at 100 nm x 100 nm x 100 nm ( field of view 95 µm x 95 µm ) , and a high-resolution scan of the noduli at 23 . 6 nm x 23 . 6 nm x 50 nm ( field of view 46 µm x 46 µm ) . All scans were obtained with a beam energy of 2 kV under high vacuum . Following image acquisition , image alignment and contrast optimization were carried out using Amira 5 . 3 . Image stacks were then down sampled to 8-bit depth , enabling the use of the Skeletonize plugin for Amira ( Schmitt et al . , 2004 ) as well as to perform image segmentation using Amira’ s segmentation editor . For segmentation of CX neuropils in each of the three SBEM data sets the image stacks were down-sampled to in Amira to 1 µm x 1 µm x 1 µm voxel size . A Leica SP8 DLS inverted confocal microscope was used to collect image data from immunohistochemically labeled brains as well as for individual neurons injected with neurobiotin . Image stacks of entire brains were generated using a 20x oil-immersion objective . To accomplish this , a mosaic of 8–10 image stacks ( voxel size of 0 . 76 µm x 0 . 76 µm x 1 µm ) were collected and stitched together in the x-y plane using Leica Application Suite X ( LAS X ) Navigator . Due to the relatively large brain size and limited working distance , brains had to be imaged both frontally and posteriorly with some overlap to allow for subsequent merging . Frontal and posterior image stacks were manually positioned in Amira using the transform tool and then registered using the affine registration tool with one stack as a reference . Once the transformations were applied to the stack being registered , both were exported and stitched together to form one stack of the full brain using the stitching plugin in FIJI ( Preibisch et al . , 2009; Schindelin et al . , 2012 ) . The resulting image stack was then down-sampled in Amira to 1 µm x 1 µm x 1 µm voxel size to enable reconstruction using Amira’ s segmentation editor . Neurons were manually reconstructed for all three SBEM data sets using both Amira ( Thermo Fisher Scientific ) and CATMAID software ( Saalfeld et al . , 2009 ) . We used Amira with the third-party plugin Skeletonize ( Schmitt et al . , 2004 ) at the beginning stages of this project . With this method , branch points ( nodes ) are placed along the length of a neuronal process . Nodes are automatically connected by a straight line ( edge ) , the diameter of which can be adjusted to approximate the diameter of the neuronal process that is being traced . Amira enabled the use of orthogonal slices along the xz and yz planes which contained a better resolution in the lowest resolution data set ( 126 nm x 100 nm compared to 126 nm x 126 nm ) . The Skeletonize plugin was also used to reconstruct and visualize neurons injected with neurobiotin . At later stages in the development of this study , we transferred all tracing data from Amira to CATMAID , an open-source and collaborative neural reconstruction software that enabled us to simultaneously trace EM image data over the web ( Saalfeld et al . , 2009 ) . We did so with the aid of natverse , a suite of R packages that include many functions for handling and analysing neuroanatomical data ( Bates et al . , 2020 ) . We specifically used the nat and catmaid packages to convert , re-scale , and upload neurons and neuropil surface data from Amira to CATMAID . Once in CATMAID , neuron skeletons were manually traced by placing nodes along the length of a neuronal fiber . CATMAID’s 3D Viewer was used to visualize neurons and generate many of the images used throughout this study . To increase tracing efficiency , we did not adjust node diameter to approximate fiber diameter when tracing in CATMAID as was done in Amira . Neuropils were reconstructed using the segmentation editor in Amira 5 . 3 . This was carried out by tracing key cross-sections in all three spatial planes using the paintbrush tool . Such a process results in a scaffold which was then turned into a surface by interpolation using the ’wrap’ function . This method was used to reconstruction neuropils in each of the three SBEM data sets as well as whole mount brains immunolabeled with synapsin , serotonin , or tyrosine hydroxylase . Immunolabeling of serotonin and tyrosine hydroxylase was used to aide in the visualization of layers within the CB ( available in the insect brain database , Heinze et al . , 2021 ) . Using Amiras segmentation editor , immunoreactive layers in the CX were reconstructed for each immunostained brain and were affine registered to the CB of the EM data . Doing so enabled us to establish a rough idea of layering within the FB and EB , and to visualize the trajectories of reconstructed neurons from the 126 nm data set relative to these immunopositive layers . Reconstructed neurons are available to view and download on the Insect Brain Database ( IBdb; Heinze et al . , 2021 ) : https://www . insectbraindb . org/app/connectomics;experiment=61;handle=EIN-0000061 . 1 and https://www . insectbraindb . org/app/connectomics;experiment=62;handle=EIN-0000062 . 1 . | Bumblebees forage widely for pollen and nectar from flowers , sometimes travelling kilometers away from their nest , but they can somehow always find their way home in a nearly straight line . These insects have been known to return to their nest from new locations almost 10 kilometers away . This homing ability is a complex neurological feat and requires the brain to combine several processes , including observing the external world , controlling bodily movements and drawing on memory . While the navigational behavior of bees has been well-studied , the neuronal circuitry behind it has not . Unfortunately , most of what is known about insects’ brain activity comes from studies in species such as locusts or fruit flies . In these species , a region of the brain known as the central complex has been shown to have an essential role in homing behaviors . However , it is unknown how similar the central complex of bumblebees might be to fruit flies’ or locusts’ , or how these differences may affect navigational abilities . Sayre et al . obtained images of thin slices of the bumblebee central complex using a technique called block-face electron microscopy , which produces high-resolution image volumes . These images were used to obtain a three-dimensional map of over 1300 neurons . This cellular atlas showed that key aspects of the central complex are nearly identical between flies and bumblebees , including the internal compass that monitors what direction the insect is travelling in . However , hundreds of millions of years of independent evolution have resulted in some differences . These were found in neurons possibly involved in forming memories of the directions and lengths of travelled paths , and in the circuits that use such vector memories to steer the insects towards their targets . Sayre et al . propose that these changes underlie bees’ impressive ability to navigate . These results help explain how the structure of insects’ brains can determine homing abilities . The insights gained could be used to develop efficient autonomous navigation systems , which are challenging to build and require a lot more processing power than offered by a small part of an insect brain . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2021 | A projectome of the bumblebee central complex |
In the fungus Ustilago maydis , sexual pheromones elicit mating resulting in an infective filament able to infect corn plants . Along this process a G2 cell cycle arrest is mandatory . Such as cell cycle arrest is initiated upon the pheromone recognition in each mating partner , and sustained once cell fusion occurred until the fungus enter the plant tissue . We describe that the initial cell cycle arrest resulted from inhibition of the nuclear transport of the mitotic inducer Cdc25 by targeting its importin , Kap123 . Near cell fusion to take place , the increase on pheromone signaling promotes Cdc25 degradation , which seems to be important to ensure the maintenance of the G2 cell cycle arrest to lead the formation of the infective filament . This way , premating cell cycle arrest is linked to the subsequent steps required for establishment of the infection . Disabling this connection resulted in the inability of fungal cells to infect plants .
Sexual reproduction is widely conserved within the eukaryotic life tree . The final outcome of this process is the fusion of two distinct haploid nuclei into a single diploid nucleus . To avoid an imbalance in the nuclear genetic information provided by each mating partner , the cell cycle status of both nuclei should be the same before karyogamy . In metazoans , this synchronization often occurs once the two partner nuclei are in the same zygotic cytoplasm ( Austin , 1978 ) . However , in simple eukaryotes such as fungi or unicellular algae , cell cycle synchronization occurs before cell fusion ( Hartwell , 1973 ) . In these organisms , cell cycle synchronization is mediated by the recognition of signals , often pheromones secreted by distinct mating partners . The paradigmatic and best studied case is the budding yeast Saccharomyces cerevisiae . In this fungus , pheromone recognition is mediated by plasma membrane-located receptors , which transmit the signal toward the cell cycle machinery using a widely conserved MAP kinase cascade ( Bardwell , 2005 ) . Pheromone recognition in budding yeast results in G1 cell cycle arrest , which is maintained throughout all cell fusion process , resulting in a diploid zygote that is able to resume the cell cycle or enter into meiosis ( Elion , 2000 ) . Premating cell cycle synchronization at G1 phase seems to be the rule , as other fungi such as Schizosaccharomyces pombe ( Davey , 1998 ) , diatoms ( Moeys et al . , 2016 ) , and -most likely- algae such as Chlamydomonas reinhardtii ( Joo et al . , 2017 ) and the slime mold Dictyostelium discoideum ( Ishida et al . , 2005 ) apply the same principle . However , there is one exception in the fungal maize smut pathogen Ustilago maydis , where premating cell cycle synchronization in response to secreted sexual pheromones occurs at G2 phase ( García-Muse et al . , 2003 ) . Paradoxically , in spite of the distinct cell cycle stage for arrest , the elements involved in the transmission process ( pheromone and receptors , MAPK cascade and transcription factors ) are similar to those described in fungi that undergo arrest at G1 phase ( Müller et al . , 2003; Vollmeister et al . , 2012 ) . This result strongly suggested that the differences were linked to alternative wiring of the U . maydis pheromone MAPK cascade with cell cycle regulators , although these connections were largely unknown . The reasons for the distinct cell cycle response to pheromone in U . maydis are likely related to the unusual developmental steps that mating triggers in this fungal system . In U . maydis , virulence and sexual development are intricately interconnected because the mating of two compatible budding haploid cells is the prerequisite to induce the infectious stage ( Brefort et al . , 2009; Vollmeister et al . , 2012 ) . Pathogenic development is mediated by two independent loci: the a-locus , which encodes a pheromone-receptor system , and the b-locus , which encodes a pair of homeoproteins ( bW and bE ) . On the plant surface , infection is initiated upon the recognition of mating pheromone secreted by haploid cells of the opposite mating type ( Bölker et al . , 1992 ) . This recognition induces G2 cell cycle arrest as well as the formation of long conjugation tubes ( García-Muse et al . , 2003; Spellig et al . , 1994 ) , which grow toward each other and fuse at their tips ( Snetselaar et al . , 1996 ) . Cytoplasmic fusion is not followed by karyogamy , resulting in a dikaryotic cell . After cell fusion on the plant surface , the G2 cell cycle arrest is sustained , and the single dikaryotic cell grows in a polar manner , producing the infective filament . This hypha expands , accumulating the cytoplasm at the tip of the filament , whereas the distal parts of the hypha become vacuolated and are sealed off by the insertion of regularly spaced septa , resulting in the formation of characteristic empty sections ( Steinberg et al . , 1998 ) . This growth mode enables the fungus to progress along the plant surface , most likely to find an appropriate point of entry . Eventually , the hyphae stop polar growth in response to an as yet unidentified signal , and their tips swell to form appressoria and penetrate the cuticle ( Snetselaar and Mims , 1992; Snetselaar and Mims , 1993 ) . Once the filament enters the plant , the cell cycle is reactivated , and the fungus proliferates inside the plant . During the differentiation process resulting in plant penetration , the presence of a sustained G2 cell cycle arrest is mandatory and the impairment of this cell cycle arrest resulted in the inability of the fungus to infect plants ( Castanheira and Pérez-Martín , 2015 ) . This cell cycle arrest is imposed first by the activation of the pheromone cascade and then maintained during the growth of the infective filament by a transcriptional regulator called b-factor , which is encoded in the b-locus and composed of two subunits ( bW and bE ) , provided by each mating partner . The mechanisms involved in this sustained cell cycle arrest have been described only partially . While the manner by which the presence of b-factor arrests the cell cycle is comprehended in detail , the molecular intricacies associated with the cell cycle arrest induced in response to pheromone are still unknown . G2/M transition in U . maydis is regulated by the presence of two distinct cyclin-dependent kinase ( CDK ) complexes: Cdk1-Clb1 and Cdk1-Clb2 ( Garcia-Muse , 2004 ) . Of these , the limiting step is provided by the activity of the Cdk1-Clb2 complex , which is controlled by the inhibitory phosphorylation of Cdk1 . The level of this phosphorylation depends on the relative activity of the Wee1 kinase ( which inhibits Cdk1 ) and the Cdc25 phosphatase ( which activates Cdk1 ) ( Sgarlata and Pérez-Martín , 2005a; Sgarlata and Pérez-Martín , 2005b ) . Not surprisingly , the mechanism by which the b-factor arrests the cell cycle at G2 during the growth of the dikaryotic infective filament relies on the increase of Cdk1 inhibitory phosphorylation: The b-factor activates the DNA damage response ( de Sena-Tomás et al . , 2011; Mielnichuk et al . , 2009 ) in the absence of DNA damage ( Tenorio-Gómez et al . , 2015 ) , resulting in the phosphorylation of Cdc25 , promoting thereby its interaction with 14-3-3 proteins , which in turn inactivates the phosphatase by its retention in the cytoplasm ( Mielnichuk and Pérez-Martín , 2008 ) ; at the same time , the b-factor represses the transcription of hsl1 , which encodes a kinase that downregulates Wee1 kinase , increasing as a consequence the level of inhibitory phosphorylation of Cdk1 ( Castanheira et al . , 2014 ) . Moreover , a second cell cycle brake is added during the formation of the infective filament since the b-factor also activates the transcription of biz1 , a transcriptional regulator that represses the transcription of clb1 , encoding the b cyclin required for the second Cdk1-cyclin complex involved in G2/M transition ( Flor-Parra et al . , 2006; Garcia-Muse , 2004 ) . Here we tried to uncover the elements required for the pheromone-induced cell cycle arrest . Although it was possible that the pheromone response shared the same regulatory scheme as the b-induced cell cycle arrest , we show that this is not entirely the case and that some of the elements involved are different . We report that the pheromone response MAPK cascade is distinctly wired to cell cycle regulation , resulting first in the inhibition of the nuclear localization of Cdc25 , via importin phosphorylation , and second , once the MAPK signaling reaches a threshold , in the degradation of the accumulated Cdc25 . These steps were found to be required to ensure the maintenance of cell cycle arrest during the infection process . Inability to do so resulted in a strong defect in virulence .
We sought to address the molecular mechanisms behind the G2 cell cycle arrest observed upon pheromone response in U . maydis . This response requires , in each mating partner , the recognition of the compatible pheromone by its cognate receptor and the transmission of the signal through a conserved MAPK cascade ( Müller et al . , 2003; Vollmeister et al . , 2012 ) . However , the expression of pheromone and pheromone receptor genes requires poor nutritional conditions , which enables the activity of the transcriptional regulator Prf1 that activates the promoters from the a-locus ( Hartmann et al . , 1999; Kaffarnik et al . , 2003 ) ( Figure 1—figure supplement 1A ) . Since changes in nutritional conditions could alter the cell cycle pattern in this fungus ( Pérez-Martín et al . , 2006 ) , we took advantage of the previous description of an activated allele of the pheromone cascade MAPKK Fuz7 ( fuz7DD , Figure 1—figure supplement 1B ) , whose expression faithfully recapitulates the pheromone response in U . maydis ( Müller et al . , 2003; Zarnack et al . , 2008 ) . In this way , we make the activation of the pheromone MAPK cascade independent of the elements located upstream of this cascade ( i . e . receptors and pheromones ) allowing us to focus on the connections between the pheromone response MAPK cascade and cell cycle regulators . When an ectopic copy of the fuz7DD allele was expressed under the control of the crg1 promoter ( induced by arabinose and repressed by glucose ) ( Figure 1—figure supplement 1C and D ) , it mimicked the G2 cell cycle arrest observed when pheromone is sensed by U . maydis ( García-Muse et al . , 2003 ) : cells accumulate 2C DNA content , carrying a single nucleus with an intact nuclear membrane ( U . maydis breaks down its nuclear envelope at mitosis; Straube et al . , 2005 ) ( Figure 1A and B ) . Furthermore , this cell cycle arrest was dependent on Kpp2 , the downstream MAPK , but independent of Prf1 ( Figure 1—figure supplement 1E ) . We observed that the expression of the fuz7DD allele was correlated with an increase in the level of inhibitory phosphorylation of Cdk1 , which has been reported to be associated with G2 cell cycle arrest in U . maydis ( Sgarlata and Pérez-Martín , 2005a; Sgarlata and Pérez-Martín , 2005b ) ( Figure 1C and Figure 1—figure supplement 2A ) . Moreover , the impairment of Cdk1 inhibitory phosphorylation , either by the expression of cdk1AF , an allele refractory to inhibitory phosphorylation , or by the downregulated expression of wee1 , the cognate kinase responsible for inhibitory phosphorylation ( Sgarlata and Pérez-Martín , 2005b ) , abrogated the fuz7DD-dependent cell cycle arrest ( Figure 1D and Figure 1—figure supplement 2B–F ) . These results supported the notion that the G2 cell cycle arrest associated with the activation of the pheromone cascade was dependent on the inhibitory phosphorylation of Cdk1 . The mechanism of cell cycle arrest induced by the b-factor , which is responsible for the sustained G2 arrest during the growth of the infective filament , also involves an increase in the level of Cdk1 inhibitory phosphorylation ( Mielnichuk et al . , 2009 ) . Moreover , in agreement with a previous report ( Zarnack et al . , 2008 ) , we observed that the transcription of hsl1 , a negative regulator of the Wee1 kinase , was strongly downregulated upon expression of the fuz7DD allele ( Figure 1—figure supplement 3A ) , as it occurs in b-dependent cell cycle arrest ( Castanheira et al . , 2014; Heimel et al . , 2010 ) . These findings prompted us to think that pheromone-dependent and b-dependent cell cycle arrests might share the same molecular mechanisms . To add further support to this idea , we analyzed the involvement in the fuz7DD-dependent cell cycle arrest of other elements required for b-dependent cell cycle arrest , like the Chk1 kinase . However , our results indicated that Chk1 was not involved in fuz7DD-dependent cell cycle arrest ( Figure 1—figure supplement 3B ) . In the same way , it was already reported that expression of the fuz7DD allele does not upregulate biz1 expression ( Flor-Parra et al . , 2006; Zarnack et al . , 2008 ) , suggesting different mechanisms for pheromone- and b-dependent cell cycle arrest . The target of Cdk1 inhibitory phosphorylation in U . maydis is the Cdk1-Clb2 complex , which is the main regulatory control for G2/M transition ( Garcia-Muse , 2004; Sgarlata and Pérez-Martín , 2005b ) . The amount of Cdk1 inhibitory phosphorylation depends on the relative activity levels of the Wee1 kinase and the Cdc25 phosphatase ( Sgarlata and Pérez-Martín , 2005a ) . We analyzed the effects of expression of the fuz7DD allele on the levels of these regulators ( Figure 1E ) . We observed that the protein levels of Cdc25 dropped abruptly upon expression of the fuz7DD allele , although this decrease in Cdc25 levels cannot be attributed to a decrease in the mRNA levels of its gene ( Figure 1—figure supplement 3A ) . Since the observed downregulation of Cdc25 levels could account for G2 cell cycle arrest ( Sgarlata and Pérez-Martín , 2005a ) , we aimed to bypass the fuz7DD-dependent cell cycle arrest by overexpression of cdc25 at the same time as expression of the fuz7DD allele . However , contrary to our expectations , overexpression of an ectopic copy of cdc25 does not abrogate cell cycle arrest , in spite of the presence of high protein levels of Cdc25 in these conditions ( Figure 1F and Figure 1—figure supplement 4 ) . This result not only strongly suggested that the decrease of Cdc25 levels was not the cause of the cell cycle arrest , but it also supported our view of distinct molecular mechanisms between pheromone- and b-factor-induced cell cycle arrest: While high levels of Cdc25 do not affect the ability to arrest the cell cycle by the pheromone cascade activation , in the b-dependent cell cycle arrest , Cdc25 is retained at the cytoplasm by Bmh1 ( 14-3-3 protein ) , and this repression can be overwhelmed by high levels of Cdc25 ( i . e . overexpressing Cdc25 ) ( Mielnichuk et al . , 2009 ) . In summary , these previous results indicated that some elements , like the downregulation of hsl1 ( and thereby the upregulation of Wee1 ) seemed to be shared by b- and pheromone cascade-induced cell cycle arrest . However , other elements , like the downregulation of biz1 and the Chk1-mediated retention of Cdc25 at cytoplasm seemed to be unique for b-dependent cell cycle arrest . In our search for elements connected to the pheromone cascade that could be involved in the induction of the cell cycle arrest , we recalled the gene pcl12 , which has been reported to be strongly induced by the pheromone MAPK ( Flor-Parra et al . , 2007 ) . This gene encodes a cyclin from the Pcl family ( Measday et al . , 1997 ) , and its ectopic expression under a regulatable promoter ( such as the crg1 promoter ) was sufficient to induce the formation of a cell structure resembling a conjugation tube , which was also arrested at G2 phase ( Flor-Parra et al . , 2007 ) . For that reason , we were curious about the involvement of this protein in the pheromone cascade-induced cell cycle arrest . Indeed , we found that Pcl12 was required for the induction of cell cycle arrest as well as for the observed decrease in Cdc25 levels upon the expression of the fuz7DD allele ( Figure 2A and B ) . Pcl12 forms a complex with the essential cyclin-dependent kinase Cdk5 ( Castillo-Lluva et al . , 2007 ) . The Pcl12-Cdk5 complex is required for sustained polar growth during the formation of the conjugation tubes in response to pheromone treatment ( Flor-Parra et al . , 2007 ) . However , we found that Cdk5 was not required for fuz7DD-induced cell cycle arrest ( Figure 2—figure supplement 1 ) . Since Pcl12 is a cyclin , this result suggested the existence of alternative partners ( most likely kinases ) for Pcl12 during pheromone cascade-induced cell cycle arrest . To identify these putative partners , we performed coimmunoprecipitation coupled with liquid chromatography-mass spectrometry ( LC/MS ) analysis of a GFP-tagged Pcl12 version in the presence of the expression of the fuz7DD allele . We found the kinase Crk1 among the major peptides copurifying with Pcl12 ( Figure 2—figure supplement 2 , Figure 2—source data 1 ) . Crk1 ( Cdk-Related Kinase 1 ) was previously described as a regulator of polar growth since its overexpression induces cell filamentation ( Garrido and Pérez-Martín , 2003 ) . In support of a functional role of the observed Pcl12 and Crk1 physical interaction , we found that loss of function of Crk1 abrogated fuz7DD-dependent cell cycle arrest as well as the decrease in Cdc25 levels in a similar manner as the pcl12 mutant did ( Figure 2A and B ) . Furthermore , we also observed that Crk1 was also required for cell cycle arrest upon ectopic expression of pcl12 ( Figure 2C and E ) . These results sustained the idea that Crk1 and Pcl12 were working together during the pheromone cascade-induced cell cycle arrest , most likely as a complex . Interestingly , Crk1 activation was previously described to be linked to the pheromone cascade: In order to promote hyperpolarized growth , Crk1 must be activated via phosphorylation of its T-loop by the MAPKK Fuz7 , as well as by phosphorylation of the C-terminal end by the MAPK Kpp2 ( Garrido et al . , 2004 ) . Therefore , we wondered whether these described connections between Crk1 and the pheromone cascade during the promotion of filamentous growth were also required for fuz7DD-induced cell cycle arrest . To address this question , we took advantage of previously described mutant alleles of crk1 refractory to phosphorylation by Fuz7 ( crk1AEF ) or by Kpp2 ( crk1AAA ) ( Figure 2D ) ( Garrido et al . , 2004 ) . Much to our surprise , we found that cells carrying crk1AEF or crk1AAA mutant alleles were able to arrest the cell cycle upon the expression of the fuz7DD allele ( Figure 2F ) . Moreover , we also found that the ectopic expression of pcl12 was still able to arrest the cell cycle in cells carrying different mutations affecting the phosphorylation of Crk1 by the MAPK cascade , such as the loss-of-function in fuz7 or kpp2 , or crk1 alleles refractory to phosphorylation ( Figure 2C and Figure 2—figure supplement 3 ) . These results indicated that the described activation of Crk1 via the pheromone cascade during filamentation played no role in the cell cycle arrest induced by the same MAPK cascade . To explain this apparent paradox , we propose that Crk1 , which is a kinase that have features of both CDK-like and MAPK-like kinases ( see discussion section for a detailed description of this hypothesis ) , can be activated by two distinct manners: either as a MAPK , by T-loop phosphorylation through the pheromone MAPK cascade , to support polar growth; or as a CDK , by interaction with Pcl12 cyclin , to support cell cycle arrest . Distinct activation mechanisms probably also involve distinct targets explaining the distinct outcomes . Both Crk1 and Pcl12 were previously described as regulators involved in morphogenesis of the conjugative tube ( Flor-Parra et al . , 2007; Garrido et al . , 2004 ) , although the relationships between these factors at this level were not studied . We observed that , in contrast to the cell cycle arrest , which is abrogated by a single mutation in either pcl12 or crk1 , the effects of each gene mutation on the morphology of the resulting filament upon expression of fuz7DD allele were distinct , and the double mutant was more affected than single mutants ( Figure 2—figure supplement 4A ) . On the basis of these genetic interactions , we propose that Pcl12 and Crk1 work together , most likely forming a complex , and acting on the pheromone cascade-induced cell cycle arrest . However , it seems that these proteins also work in parallel pathways during the control of conjugation tube morphogenesis , most likely through the Pcl12-Cdk5 complex in one pathway , and through Crk1 receiving MAPK cascade signals in the other ( Figure 2—figure supplement 4B ) . The results shown above can be included in a working model in which the pheromone cascade-dependent induction of the expression of pcl12 enables the formation of a Crk1-Pcl12 complex , which is responsible for the cell cycle arrest at G2 . In support of this view , we observed that the cell cycle arrest induced by ectopic expression of pcl12 was dependent on the inhibitory phosphorylation of Cdk1 ( Figure 3—figure supplement 1A and B ) , similarly to the fuz7DD-induced cell cycle arrest . However , in clear contrast with the observed drop in Cdc25 levels upon fuz7DD expression , the levels of Cdc25 did not decrease upon the ectopic expression of pcl12 ( Figure 3—figure supplement 1C ) . This result reinforced our conclusion that the decrease in Cdc25 levels was not responsible for pheromone cascade-induced cell cycle arrest at G2 . Strikingly , we also observed that in the filaments produced upon ectopic pcl12 expression , a GFP-Cdc25 fusion was excluded from the nucleus . Moreover , this exclusion was abrogated , as it was the cell cycle arrest , if the activity of Crk1 was eliminated ( Figure 3A and B ) . Since the phosphatase Cdc25 must be transported to the nucleus to activate mitosis entry ( Mielnichuk and Pérez-Martín , 2008 ) , we hypothesized that retaining Cdc25 in the cytoplasm could explain the observed G2 cell cycle arrest upon ectopic pcl12 expression and , most likely , also upon fuz7DD expression . This hypothesis was reminiscent of the b-induced retention of Cdc25 at the cytoplasm via interaction with 14-3-3 proteins . However , we considered unlikely to be the same mechanism , because for b-dependent cell cycle arrest such cytoplasmic retention can be saturated by overexpression of Cdc25 , and we observed that it was not the case for the pheromone cascade-induced cell cycle arrest ( Figure 1F ) . One of the interactors of Pcl12 obtained from the coimmunoprecipitation coupled with LC/MS analysis was the uncharacterized protein UMAG_15014 . Sequence phylogeny analysis indicated that this protein ( renamed Kap123 ) belongs to the family of β importins of class 3 and 4 ( Figure 3—figure supplement 2A ) . Since Pcl12 is a cytoplasmic protein ( Flor-Parra et al . , 2007 ) , we considered Kap123 unlikely to be involved in the transport of Pcl12 to the nucleus . A previous report showed that Sal3 from S . pombe , which is one of the members of the class 3 β importin family , was involved in the nuclear import of Cdc25 in this fungus ( Chua et al . , 2002 ) . Therefore , we decided to analyze whether U . maydis Kap123 was required for the nuclear localization of Cdc25 , and we found that it was ( Figure 3—figure supplement 2B and C ) . This result prompted us to hypothesize that Kap123 could be a target of the Crk1-Pcl12 complex and that the action of this complex could disable the interaction between Kap123 and Cdc25 upon pheromone signaling . In this way , interference with the nuclear localization of Cdc25 could explain pheromone cascade-induced G2 cell cycle arrest . Using GFP-trap beads , we analyzed the ability of a Kap123-GFP fusion to interact with Cdc25 under conditions for the ectopic expression of pcl12 in the presence or absence of functional Crk1 kinase . Additionally , to discard any effect of the expected interaction between Kap123 and Cdc25 as a consequence of the induced cell cycle arrest , we used the overexpression of Wee1 as a control for G2 cell cycle arrest ( Sgarlata and Pérez-Martín , 2005b ) . In support of our hypothesis , we observed that the presence of a Crk1-Pcl12 complex , but not G2 cell cycle arrest alone , disables the ability of Kap123 to interact with Cdc25 ( Figure 3C ) . We also observed that upon fuz7DD expression , the mobility of a Kap123-HA allele in SDS-acrylamide gels was reduced ( Figure 3D ) , and this decrease in mobility can be eliminated by λ phosphatase treatment of the protein immunoprecipitates ( Figure 3E ) . Moreover , we found similar electrophoretic mobility reduction upon ectopic expression of pcl12 ( Figure 3F ) . The decrease in the electrophoretic mobility of Kap123 upon expression of fuz7DD was dependent on the presence of functional alleles of crk1 , pcl12 , and kpp2 ( Figure 3G ) . In the case of ectopic expression of pcl12 , Crk1 was required for the decreased electrophoretic mobility of Kap123 , but neither fuz7 nor kpp2 were required ( Figure 3H ) . These results mirrored the same genetic requirements as the cell cycle arrest , suggesting a causal relationship between phosphorylation of Kap123 and cell cycle arrest . The previous results suggested that Kap123 could be phosphorylated by the Pcl12-Crk1 complex and that this modification could alter the ability of Kap123 to interact with Cdc25 . To support this hypothesis , we looked for Kap123 mutants that were refractory to the phosphorylation by Pcl12-Crk1 complex . Since Crk1 has been defined both as a CDK-like and as a MAPK , and no consensus sequence for its phosphorylation sites was known , we considered , as a first approach , the established phosphorylation consensus sequences for both CDK and MAPK . Kap123 carries in its sequence 2 and 5 putative phosphorylation sites for CDK and MAPK , respectively . Using distinct threonine- or serine-to-alanine mutants at predicted phosphorylation sites in the Kap123 sequence ( Figure 4—figure supplement 1A ) , we analyzed the electrophoretic mobility of HA-tagged Kap123 mutants in conditions of fuz7DD expression , and we found that the change of Thr867 to Ala ( one of the putative CDK sites ) resulted in abrogation of the electrophoretic mobility shift of Kap123 upon fuz7DD expression ( Figure 4—figure supplement 1B ) . The mutant allele Kap123T867A showed no reduced electrophoretic mobility upon fuz7DD expression and pcl12 expression ( Figure 4B and C ) . Encouragingly , the presence of this mutant allele makes the cells unable to undergo cell cycle arrest in response to the expression of either fuz7DD or pcl12 ( Figure 4D and E ) . Moreover , cells carrying the kap123T867A allele were unable to retain Cdc25 in the cytoplasm upon ectopic expression of pcl12 ( Figure 4F and G ) . Also , we found that the Kap123T867A-GFP allele , in contrast to the wild-type GFP-tagged allele , was able to interact with Cdc25 in conditions of the ectopic expression of pcl12 ( Figure 4H ) . Taken together , these data supported a model in which the pheromone cascade enables the formation of a Crk1-Pcl12 complex that inhibits the transport of Cdc25 to the nucleus by the phosphorylation of Kap123 importin , thereby promoting G2 cell cycle arrest . Following our model , we aimed to test directly the response to pheromone of cells carrying mutations that abolished the pheromone cascade-induced cell cycle arrest . Addition of synthetic pheromone to a culture of compatible mating type cells induced the formation of conjugative tubes that were cell cycle arrested ( Spellig et al . , 1994; Szabó et al . , 2002 ) . Therefore , we treated a1 mating type cells from U . maydis carrying an HA-tagged Kap123 allele , as well as its phosphorylation-defective version , with different concentrations of a2 synthetic pheromone , and analyzed the ability of cells to arrest the cell cycle . In concordance with our model , we found that cells carrying the kap123T867A allele were not able to arrest the cell cycle , even at the highest pheromone concentrations ( Figure 5A ) . We analyzed , by Western blot , samples from these cultures and we observed a shift in the electrophoretic mobility of the wild-type allele but not the kap123T867A allele at different pheromone concentrations ( Figure 5B ) . Furthermore , the presence of the kap123T867A allele makes the cells unable to keep Cdc25 out of the nucleus in response to synthetic pheromone ( Figure 5C and D ) . All these results added further support to our model about the process by which pheromone triggers G2 cell cycle arrest in U . maydis cells . The importance of cell cycle synchronization as a step prior to cell fusion during mating in unicellular eukaryotes was clearly illustrated more than 25 years ago by the defects in mating observed in the S . cerevisiae far1 mutants , which lack the main control to arrest the cell cycle upon pheromone response ( Chang and Herskowitz , 1990; Pope et al . , 2014 ) . Since the virulence of U . maydis is dependent on mating , we expected that the ability of U . maydis to infect corn plants would be affected in the absence of cell cycle arrest upon response to the pheromone . To test the hypothesis , we constructed compatible strains ( a1 b1 and a2 b2 mating type ) carrying the kap123T867A allele , and infected corn plants with compatible mixtures of control and mutant strains . We found that strains unable to undergo pre-mating cell cycle arrest were as virulent as wild-type strains ( Figure 6A ) . This unexpected result suggested a minor role of pheromone-induced cell cycle arrest in the virulence process , and therefore we were interested to understand the reasons for it . We wondered about the outcome from crosses of kap123T867A mutant strains . Mating in U . maydis is easily scored by the formation of dikaryotic filaments on charcoal plates , which can be observed as a white-appearing mycelial growth ( fuzzy phenotype ) ( Banuett and Herskowitz , 1989 ) . We observed that although slightly impaired , the mutant crosses showed a positive fuzzy phenotype ( i . e . , they were able to mate ) ( Figure 6B ) . Therefore , we were curious about the nuclear content of the filaments resulting from these crosses . To address this question , we crossed haploid strains expressing a GFP fusion to a nuclear localization signal under control of the b-dependent dik6 promoter ( Mielnichuk et al . , 2009 ) . In this way , only cells resulting from mating and therefore activating the transcriptional program dependent on b-factor ( which subunits are provided by each mating partner ) produced a fluorescent nuclear signal , so the mating-derived filaments can be distinguished from the cell population background ( frequently enriched in aberrant elongated cells ) . We found that the majority of b-dependent filaments from wild-type crosses carried two nuclei , whereas filaments obtained from mutant crosses frequently carried two ( 62% of filaments ) , three ( 35% ) and less frequently four ( 3% ) nuclei ( Figure 6C and D ) . We did not find filaments carrying more than four nuclei ( from a total of 100 filaments counted ) . To understand these results , it is worth remembering that once the cytoplasms of the compatible mating partners fuse , the formation of the heterodimeric b-factor activates the molecular mechanisms to induce a G2 cell cycle arrest . Indeed , we observed that the presence of the kap123T867A allele did not affect the ability of b-factor to arrest the cell cycle ( Figure 6—figure supplement 1 ) . The inability of cells carrying the kap123T867A allele to arrest the cell cycle in response to pheromone ( this is , to synchronize their cell cycles before cytogamy ) will result , most likely , in the fusion of mating partners at different stages of the cell cycle . However , once the two cytoplasms fuse , the establishment of a barrier at G2 by the b-factor would resynchronize the respective nuclear content in the resulting filament . In the case of mating partners fusing their cytoplasms during G1 to G2 phases ( i . e . providing a single nucleus each ) , the result will be a dikaryotic filament with nuclei arrested at G2 . Only for those partners that fuse during or immediately after mitosis ( and before cytokinesis ) , the result will be a filament carrying three nuclei ( only one partner carried two nuclei before fusion ) or four nuclei ( the two partners carried two nuclei each ) , with these nuclei arrested at G2 . In any case , the final result will be a cell cycle arrested filament ( by virtue of b-factor ) and therefore it will be proficient for infection , explaining the ability of mutant cells to infect corn plants . Interestingly , this ability to synchronize the cell cycle status of distinct nuclei in a common cytoplasm by homeodomain proteins of the b-factor family is the basis for accurate cell division in other basidiomycete dikaryotic fungi such as Coprinopsis cinerea ( de Sena-Tomás et al . , 2013 ) . We were curious about the observed dramatic decrease in Cdc25 levels upon expression of the fuz7DD allele and whether this observation bears any functional relevance . To link the observed downregulation of Cdc25 with the pheromone response , we analyzed the levels of Cdc25 in a1 mating type haploid cells in response to increasing concentrations of a2 synthetic pheromone ( Figure 7A ) . We found that the lowest pheromone concentration tested ( 0 . 25 μg/ml ) , although able to induce the formation of cell cycle arrested conjugation tubes ( Figure 5A ) , did not induce a decrease in the levels of Cdc25 . However , increasing the amount of pheromone ( up to 25 μg/ml ) was correlated with a decrease in Cdc25 levels , suggesting a dose-response relationship between pheromone cascade signaling and Cdc25 levels . We entertained the possibility that the observed downregulation of Cdc25 was a side effect of the permanent cytosolic retention of Cdc25 , which might promote its degradation by some way , not necessarily related to the pheromone response . In support of this explanation we observed that the downregulation of Cdc25 levels requires the retention of Cdc25 in the cytoplasm , since expression of the fuz7DD allele in a kap123T867A mutant strain does not result in a decrease in Cdc25 levels ( Figure 7B ) . However , we also observed that pcl12 ectopic expression , despite being able to arrest the cell cycle and retain Cdc25 in the cytoplasm , was unable to induce a decrease in the Cdc25 level ( Figure 3—figure supplement 1C ) . Furthermore , from the above described experiments using different pheromone doses , it was clear that the lowest pheromone concentration ( 0 . 25 μg/ml ) , was able to induce the phosphorylation of Kap123 ( Figure 5B ) as well as the retention of Cdc25 at cytoplasm ( Figure 5C and D ) , but not its downregulation ( Figure 7A ) . These observations suggested that although cytoplasmic retention was required , it was not sufficient to induce the downregulation of Cdc25 . We hypothesized that Cdc25 could be a target ( direct or indirect ) of the pheromone MAPK cascade , which induces its degradation only when Cdc25 is retained at the cytoplasm and the activation of the pheromone cascade reaches some threshold level ( as expected in the presence of high amounts of synthetic pheromone or expression of the activated fuz7DD allele ) . To uncover the involvement of the MAPK cascade in the downregulation of Cdc25 , we constructed a strain simultaneously expressing the fuz7DD allele and pcl12 under the crg1 promoter . In this way , we can induce cell cycle arrest , as well as the retention of Cdc25 in the cytoplasm ( as a consequence of the ectopic expression of pcl12 ) , even in conditions that potentially disable the MAPK cascade , such as the disruption of the kpp2 gene . In agreement with our hypothesis , we found that under conditions of cell cycle arrest and the retention of Cdc25 in the cytoplasm , Kpp2 was required for the decrease in Cdc25 levels upon fuz7DD expression ( Figure 7C ) . Since we considered the possibility that Cdc25 was a direct target of Kpp2 , we tried to detect physical interactions between Kpp2 and Cdc25 . We used a strain that expressed both the fuz7DD allele and pcl12 and carried a kinase-dead allele of Kpp2 fused to GFP ( Kpp2K50R ) ( Müller et al . , 2003 ) , as well as an HA-tagged Cdc25 allele . Using GFP-trap beads , we were able to detect interactions between Kpp2 and Cdc25 , which were dependent on the expression of the fuz7DD allele ( i . e . , high levels of MAPK signaling ) and the presence of a wild-type Kap123 allele ( i . e . , retainment of Cdc25 in the cytoplasm ) ( Figure 7D ) . Taken together , our results were consistent with the idea that activation of the pheromone cascade resulted in the negative regulation of Cdc25 levels by a two-step mechanism: First , Cdc25 is retained in the cytoplasm as a result of the formation of the Pcl12-Crk1 complex , and then Cdc25 is potentially phosphorylated by the Kpp2 MAPK , which probably promoted its degradation . Compared to other fungal Cdc25-like phosphatases , U . maydis Cdc25 carries an unusually long N-terminal extension of approximately 270 amino acids with no similarity to orthologues in the database ( Figure 7E ) . Previous attempts to ascribe a role to this domain were elusive because its deletion did not appear to affect the Cdc25 function in U . maydis growing under axenic conditions ( Sgarlata and Pérez-Martín , 2005a ) , although it affected the sensitivity of U . maydis cells to cell wall stressors ( Carbó and Pérez-Martín , 2010 ) . We found that the removal of this N-terminal extension resulted in the stabilization of Cdc25 upon expression of the fuz7DD allele ( Figure 7F ) . Based on sequence , this region contains at least three putative MAPK phosphorylation sites ( Ser7 , Ser23 and Ser239 , Figure 5E ) . With support from the observed physical interactions , we considered the possibility that Kpp2 phosphorylates Cdc25 in this region , promoting its degradation as described in mammalian Cdc25 in response to activation of the p38 MAPK ( Uchida et al . , 2009 ) . To investigate this possibility , the three candidate Ser residues were mutated to Ala ( cdc25AAA allele ) . Encouragingly , we found that the triple mutant was resistant to the downregulation observed in the wild-type allele upon expression of the fuz7DD allele ( Figure 7G ) . Moreover , in accordance with our model explaining how the cell cycle is arrested in response to pheromone , neither the deletion of the N-terminal extension nor the alanine mutations affected the ability of cells undergo cell cycle arrest ( Figure 7H ) as well as the retention of Cdc25 at cytoplasm ( Figure 7I ) . In summary , these data strongly suggested that the MAPK Kpp2 targets and downregulates the phosphatase Cdc25 , most likely promoting its degradation . We analyzed the capacity of sexually compatible strains ( a1 b1 and a2 b2 mating type ) carrying the cdc25AAA allele to infect plants . Strikingly , we found that the presence of this mutant allele strongly affected the ability of fungal cells to produce disease to corn plants ( Figure 8A ) . The defect in virulence could be attributed to problems related to the mating process itself or it might be related to subsequent steps , such as the ability of the fungus to proliferate within the plant , for example . To address this issue , we took advantage of the solopathogenic strain SG200 , which is a haploid strain that carries the genetic information from the two different mating types and , as a consequence , does not require cell fusion to produce the infective hypha ( Bölker et al . , 1995 ) . SG200-derived cells carrying the cdc25AAA allele were as virulent as the control strain ( Figure 8A ) , suggesting a problem in mating associated with the presence of the cdc25AAA allele . Moreover , a cdc25AAA mutant cross showed a weak fuzzy phenotype in charcoal plates , in comparison to wild-type control , indicating some kind of problem with the formation of the dikaryotic filament ( Figure 8B ) . We wondered whether the impaired fuzzy phenotype in the mutant crosses was caused by defects in cell fusion . To address that question , we used the Pdik6-NLS-GFP reporter described above . Since this reporter will be only active in the presence of a functional b-factor , it can be used as an indirect readout of cell fusion between compatible sexual partners , observing the appearance of nuclear GFP fluorescence in the cell mixture resulting from cross . We crossed mutant haploid cells carrying the dik6 reporter , and we were able to observe filaments carrying nuclear GFP signal , indicating that cell fusion was not affected . Interestingly , these filaments were multinucleated ( Figure 8C ) . In contrast to the multinucleated filaments found from crosses involving the Kap123T867A mutant , which seemed not to be able to bypass the G2 barrier imposed by the b-factor ( we never found more than four nuclei per filament , Figure 6D ) , a significant proportion of the filaments observed in cdc25AAA crosses showed more than five nuclei ( Figure 8D ) . Since cells carrying the cdc25AAA allele were able to be arrested in response to activation of the pheromone cascade ( Figure 7H ) , these observations prompted us to think that the presence of the cdc25AAA allele could affect the G2 cell cycle arrest activated in the presence of a functional b-factor . However , we observed that the presence of the cdc25AAA allele did not affect the ability of AB33 cells ( see Figure 6—figure supplement 1 for a description of this strain ) to undergo cell cycle arrest upon the expression of compatible b proteins ( Figure 8E ) . In summary , we observed that the absence of downregulation of Cdc25 levels in response to pheromone does not affect neither pheromone-induced cell cycle arrest nor de novo ( independent of previous pheromone-induced cell cycle arrest ) b-induced cell cycle arrest . However , the absence of downregulation of Cdc25 levels seems to affect the cell cycle arrest in b-dependent filaments originating from crosses ( i . e . , resulting from a mating process ) . These unexpected results can be explained keeping in mind that , in U . maydis , during a regular cell cycle Cdc25 accumulates at G2 phase; and , that in contrast to pheromone-dependent cell cycle arrest , the b-dependent cell cycle arrest can be bypassed by high levels of Cdc25 ( Mielnichuk et al . , 2009 ) . We believe that during the pheromone-dependent G2 cell cycle arrest , Cdc25 is accumulating in the cytoplasm of conjugation tubes , but that as the compatible conjugation tubes are getting close ( and therefore encountering higher pheromone concentrations ) the subsequent increase in the pheromone-cascade signaling , promotes a decrease in the Cdc25 levels that allows a proper b-dependent cell cycle arrest in the filament once the cells were fused . Inability to achieve this decrease ( as the case of stable Cdc25 alleles ) could result in impaired b-dependent cell cycle arrest and thereby defects in pathogenicity .
Smut fungi are a widespread group of plant pathogens whose sexual development is coupled with virulence . All smut species investigated so far have in common the need to undergo a successful mating reaction to form an infective dikaryotic filament before being able to infect their host plant ( Bakkeren et al . , 2008 ) . In the most studied member of this group , U . maydis , it has been described strong connections between the regulation of the cell cycle and the ability to infect plants: During the steps previous to plant infection , it is required a sustained G2 cell cycle arrest ( Perez-Martin , 2012 ) . The cell cycle arrest of the infective filament on plant surface observed in U . maydis seems to be more general , and it is also present in rust fungi like Uromyces phaseoli ( Heath and Heath , 1978 ) . The connections between cell cycle regulation and the virulence in plant fungal pathogens have been studied in detail in other systems , in addition to U . maydis , such as Magnaporthe oryzae and Colletotrichum orbiculare . In these systems , the infection depends on the formation of appressorium , a structure required for plant penetration . The appressoria from these fungi use a turgor-driven mechanical process to breach the plant cuticle . The functionality of this class of appressorium implies the increase of internal turgor pressure , which is linked to an elaborated program that includes from metabolic reprograming to morphological changes to produce a clearly defined structure with a thick , multilayered and highly melanized cell wall ( Ryder and Talbot , 2015 ) . The various steps required for the formation of appressorium in these phytopathogenic fungi are linked to the different cell cycle phases . There is a G1/S control to induce the formation of appressorium , first described in C . orbiculare ( Fukada and Kubo , 2015 ) , but also operating in M . oryzae ( Fukada et al . , 2019 ) ; and there is a S-phase checkpoint required for appressorium differentiation in M . oryzae ( Osés-Ruiz et al . , 2017; Osés-Ruiz and Talbot , 2017 ) . In contrast to appressoria from these phytopathogenic fungi , appressoria of U . maydis is not dependent on turgor pressure to penetrate the plant tissue . The appressorium directs the localized secretion of enzymes that weakens the plant cuticle , allowing the plant tissue penetration . For that reason , the morphogenesis process is less complex and appressoria are unmelanized , rather small swellings of the hyphal tip that form penetration structures that are less constricted ( Snetselaar and Mims , 1993 ) . However , in spite of this apparent lack of complexity , there is also a clear connection with cell cycle regulation . G2 cell cycle arrest is mandatory to promote the differentiation of the appressorium , and the impairment in cell cycle arrest resulted in a lack of virulence ( Castanheira et al . , 2014 ) . This obligation for cell cycle arrest at G2 phase is most likely related to the fact that mitosis and the morphogenetic program responsible for appressorium formation compete for the same cytoskeletal components . Because of this competition , it makes sense that cellular controls exist to force these two processes to be incompatible and therefore to prevent infective filaments from entering mitosis ( i . e . , to arrest the progression of the cell cycle at G2 ) ( Pérez-Martín et al . , 2016 ) . G2 cell cycle arrest is established before mating and is maintained once the dikaryotic infective filament is formed , until the fungus enters the plant . During this period , cell cycle arrest is sustained by two distinct regulatory networks: First , by the pheromone-recognition cascade in each mating partner and second , once cytoplasmic fusion occurs , by the presence of a homeodomain transcriptional regulator called b-factor ( which , among other targets , represses the a-locus , encoding the pheromone and receptors; Urban et al . , 1996 ) . Our results showed that the two regulatory networks recruit both shared and specific elements to induce and sustain cell cycle arrest ( Figure 9 ) . The G2/M transition in U . maydis is controlled by the level of inhibitory phosphorylation of the Cdk1-Clb2 complex , and both regulatory networks rely on an increase in Cdk1 inhibitory phosphorylation to arrest the cell cycle . To achieve this increase , both regulatory networks upregulate the kinase responsible for this inhibitory phosphorylation , the Wee1 kinase . This upregulation is indirect and is mediated by the transcriptional repression of hsl1 , which encodes a kinase that represses Wee1 activity . However , the repression of hsl1 is not enough to arrest the cell cycle at G2: loss-of-function mutants in Hsl1 are able to progress through the cell cycle , although they have an extended G2 phase ( Castanheira et al . , 2014 ) . To arrest the cell cycle , a second element is required , namely the downregulation of Cdc25 , the phosphatase that removes the Cdk1 inhibitory phosphorylation . Is at this step where the two regulatory networks differ . The pheromone cascade promotes the formation of a kinase-cyclin complex , Crk1-Pcl12 , which phosphorylates and inhibits the β-importin Kap123 , which is required for the nuclear translocation of Cdc25 . The b-factor , however , activates the cascade responsible for the DNA damage response ( DDR , composed of Atr1 and Chk1 kinases ) , resulting in phosphorylation of Cdc25 , which promotes its cytoplasmic retention via 14-3-3 proteins ( de Sena-Tomás et al . , 2011; Mielnichuk and Pérez-Martín , 2008; Mielnichuk et al . , 2009 ) . This downregulation of Cdc25 also provides a time window to load a third element ( not present in pheromone-induced cell cycle arrest ) consisting of the transcriptional repression ( via an intermediate regulator called Biz1 ) of clb1 , which encodes a second b-type cyclin that is also required for G2/M transition ( Flor-Parra et al . , 2006 ) . The reasons for using distinct mechanisms to retain Cdc25 in the cytoplasm are probably related to the degree of reversibility of each developmental step during the infection process . The pheromone recognition by each mating partner has to culminate with the fusion of the respective conjugation tubes . If for some reason this step does not occur , the unmated cells must be able to return to the previous vegetative status . In other words , the pheromone cell cycle arrest should be reversible if the stimulus ( pheromone ) disappears and mating was not successful . A control mediated by the phosphorylation of importin could probably be easily reversed by some phosphatase ( specific or not ) once the signaling through the pheromone cascade is abrogated . The formation of the dikaryotic filament , however , is a more terminal decision in the sense that once the mating partners fuse their respective cytoplasms , if they are compatible at the b-locus , the infective filament is committed to infecting the plant . The presence of two independent cell cycle brakes , provided by the retention of Cdc25 at cytoplasm by 14-3-3 proteins and by the transcriptional repression of clb1 could make this step less reversible . Our results also reveal the importance of the transition from pheromone-dependent to b-dependent cell cycle arrest once the respective partners fuse . Although the downregulation of Cdc25 activity consists of the retention of Cdc25 in the cytoplasm in both regulatory networks , there is one relevant difference: while the inhibition of nuclear translocation ( pheromone-dependent ) cannot be bypassed by high concentrations of Cdc25 , the retention of Cdc25 by 14-3-3 ( b-dependent ) is sensitive to the amount of Cdc25 ( probably because of saturation of the available retention proteins ) . This difference gains importance because Cdc25 most likely accumulates in the cytoplasm of conjugation tubes during G2 cell cycle arrest . Once the respective cytoplasms fuse , if the amount of Cdc25 has not been previously tuned , the 14-3-3 retention system will probably be overwhelmed , and the time window required for the establishment of permanent G2 cell cycle arrest will not be provided . For that reason , it seems logical that once the pheromone cascade reaches some signaling threshold ( probably reflecting a close encounter between compatible conjugation tubes ) , it triggers the downregulation of Cdc25 to levels that are most likely easily assimilable by the 14-3-3 proteins . The importance of adjusting the levels of Cdc25 is reflected by the impaired ability of cells carrying the cdc25AAA allele to infect plants . Central to the pheromone-induced cell cycle arrest in U . maydis was the complex formed by the kinase Crk1 and the cyclin Pcl12 . Both proteins were previously reported to be required for the proper morphogenesis of the infective filament in this fungus . Interestingly , in this function , they were part of distinct complexes or regulatory networks: Pcl12 complexed with the cyclin-dependent kinase Cdk5 ( Flor-Parra et al . , 2007 ) , while Crk1 was activated in a similar way to MAPK via T-loop phosphorylation by the upstream MAPK kinase ( Garrido et al . , 2004 ) . In this work , we showed that Pcl12 and Crk1 seem to form a complex with specific roles in cell cycle arrest upon pheromone recognition and that neither Cdk5 nor MAPK cascade phosphorylation are involved . How Pcl12 controls the activity of Crk1 in this process is beyond the scope of this work . However , we can anticipate some possibilities . Crk1 belongs to the family of RD kinases , which have a conserved arginine residue preceding an aspartate residue in the catalytic loop ( Johnson et al . , 1996 ) . For this reason , the members of this family require an activation segment ( T-loop ) that undergoes a conformational change to fold into an active catalytic site . This conformational change can be achieved either by the phosphorylation of specific residues ( such as TxY in MAP kinases ) or by interaction with other proteins ( such as cyclins in cyclin-dependent kinases ) ( Nolen et al . , 2004 ) . Crk1 carries a TEY signature at its T-loop , similar to a MAP kinase , but it is worth mentioning that it was cloned from a genetic screening devoted to uncovering CDK-like proteins in U . maydis and that the predicted three-dimensional folding of the catalytic domain of Crk1 fits the structure of Cdk2 ( Garrido and Pérez-Martín , 2003 ) . We propose that Crk1 can be activated either by T-loop phosphorylation ( through the pheromone MAPK cascade ) or by interaction with Pcl12 cyclin . The activation of Crk1 in the absence of Pcl12 , which is dependent on T-loop phosphorylation , resulted in the promotion of polar growth without affecting cell cycle regulation ( Garrido et al . , 2004 ) . However , cell cycle arrest by Crk1 was dependent on Pcl12 cyclin but independent of T-loop phosphorylation . The use of proteins unrelated to cyclins to activate CDKs is well described , including the so-called K-cyclins from viruses , p35 as a coactivator of mammalian Cdk5 and the family of RINGO/Speedy proteins ( Nebreda , 2006 ) . However , the use of a cyclin to activate a noncanonical cyclin-dependent kinase has not , to the best of our knowledge , been previously described . The importance of this alternative activation mechanism can be appreciated by considering that Crk1 belongs to a widely conserved family of fungal kinases , the founding member of which is Ime2 , a central regulator of meiosis in S . cerevisiae . The Ime2-related kinases exhibit amazing variety in controlling sexual developmental programs in fungi , although the targets and physiological inputs seem to be very species-specific ( Irniger , 2011 ) . Alternative mechanisms of activation could help to explain this plethora of roles .
Ustilago maydis strains are derived from FB1 and FB2 genetic backgrounds ( Banuett and Herskowitz , 1989 ) and are listed in Supplementary file 1 . Cells were grown in rich medium ( YPD ) , complete medium ( CMD or CMA ) or minimal medium ( MMD ) ( Holliday , 1974 ) . Controlled expression of genes under the crg1 and nar1 promoters was performed as described previously ( Brachmann et al . , 2001 ) . FACS analyses were described previously ( García-Muse et al . , 2003 ) . U . maydis strains used in the presented experiments are as follows: To construct the different strains , transformation of U . maydis protoplasts with the desired constructions was performed as described previously ( Tsukuda et al . , 1988 ) . Integration of the corresponding construction into the corresponding loci was verified in each case by diagnostic PCR and subsequent Southern blot analysis or RT-PCR analysis of transcripts depending on the type of integrated mutant allele . U . maydis DNA isolation was performed as previously described ( Tsukuda et al . , 1988 ) . Plasmid pGEM-T easy ( Promega ) and pJET1 . 2 ( Thermo Fisher ) was used for subcloning and sequencing of genomic fragments generated by PCR . Plasmids for C-terminus end tagging of endogenous loci from clb2 , cdc25 , wee1 , and kap123 with 3 HA or GFP as well as conditional alleles of kap123 and srp1 were performed using Golden Gate assembly , following published procedures ( Terfrüchte et al . , 2014 ) . Mutant alleles ( loss of function and allelic variants ) of the following genes were already described: crk1 ( Garrido and Pérez-Martín , 2003; Garrido et al . , 2004 ) ; pcl12 ( Flor-Parra et al . , 2007 ) ; kpp2 , fuz7 and prf1 ( Müller et al . , 2003 ) ; cdk1 and wee1 ( Sgarlata and Pérez-Martín , 2005b ) ; cdc25 and hsl1 ( Castanheira et al . , 2014 ) ; NLS-GFP ( Mielnichuk et al . , 2009 ) ; cut11-cherry ( Pérez-Martín , 2009 ) . Kap123 serine or threonine to alanine mutant variants were constructed as cassettes carrying the desired mutation ( associated with the appearance or loss of a restriction enzyme target for diagnostic purposes ) to be inserted by homologous recombination into the endogenous locus . The cassette also allowed the insertion at the C-terminus of a 3 HA epitope as well as an antibiotic resistance marker . For kap123T867A allele used in infections and mating assays , a specific cassette without 3 HA epitope was constructed , associated with a resistance to hygromycin , flanked by FRT sites to remove the resistance cassette once the endogenous locus was mutated , following published procedures ( Khrunyk et al . , 2010 ) . Cdc25 tagged at the N-terminus as well as the N-terminal Δ1–270 and cdc25AAA mutants , were constructed by inserting the corresponding mutant cassettes at the N-terminus associated with a resistance to hygromycin flanked by FRT sites to be removed upon transformation with a plasmid encoding a FLPase recombinase ( Khrunyk et al . , 2010 ) . Further details of the constructions explained above are available on request . Total RNA was extracted with acidic phenol solution . After extraction , the RNA was cleaned using the High Pure RNA Isolation Kit ( Roche Diagnostics GmbH ) . For RT- PCR , cDNA was synthesized using the High Capacity cDNA Reverse Transcription Kit ( Applied Biosystems ) employing 1 μg total RNA per sample . qRT-PCR was performed using the SsoAdvanced Universal SYBR Green Supermix ( BioRad ) in a CFX96 Real-Time PCR system ( BioRad ) . Reaction conditions were as follows: 3 min 95°C followed by 40 cycles of 10 s 95°C/10 s 60°C/30 s 72°C . Protein extracts were performed using an adapted chloroacetic acid ( TCA ) method . Briefly , cells from 5- to 10 ml aliquots of cultures were harvested , and 1 ml of 20% TCA was added . The supernatant was removed after centrifugation , and the pellet was resuspended in 100 μl of 20% TCA and stored at −80°C for 1 hr . Samples were thawed on ice , glass beads were added , and cells were broken using a FastPrep FP120 cell disrupter ( BIO 101 ThermoSavant , Obiogene , Carlsbad , CA ) . The lysate was recovered by punching a hole on the bottom of the tube , and the glass beads were further washed with 200 μl of 5% TCA . Lysates were centrifuged at 1000 × g for 3 min , and the pellet was thoroughly resuspended in 100 μl of 2 × Laemmli buffer and 50 μl of 2 M Tris base . After boiling for 5 min , 10–20 μl were loaded in the gels . For general purposes , TGX ( 4–20% ) gels from BioRad were used , at constant 100 v running conditions . To detect the phosphorylated forms of Kap123 , 8% acrylamide/0 . 1% bisacrylamide , pH 9 . 2 custom-made gels were used at constant 5mA running conditions . Western blots were repeated from at least three independent experiments in each case . To perform immunoprecipitations , crude protein extracts were prepared . Briefly , cells were harvested by centrifugation at 4°C and washed twice with ice-cold water . The cell pellet was resuspended in ice-cold HB buffer ( 25 mM MOPS pH 7 . 2 , 15 mM MgCl2 , 15 mM EGTA , 1% Triton X-100 , PhosSTOP [1 tablet per 10 mL] , Roche protease inhibitor cocktail [1 tablet per 10 mL] ) and cells were broken using a FastPrep FP120 cell disrupter . The lysate was recovered by punching a hole on the bottom of the tube , and the glass beads were further washed with ice-cold HB buffer , and the lysated was cleared by centrifugation ( 10 min/13000xg ) . For co-immunoprecipitation analysis , approximately 3 . 5 mg of total protein extracts ( 1 ml ) were incubated with 1 μg of the monoclonal antibody for 2 hr at 4°C and then prewashed G-protein coupled magnetic beads ( 50 μl ) were added and incubated for 30 min at 4°C with agitation . For GFP trap , 50 μl GFP trap beads were mixed with 3 . 5 mg of total protein extracts ( 1 ml ) for 2 hr at 4°C , with agitation . Immunoprecipitates were washed six times with 1 ml of HB buffer . Crude extracts were obtained and incubated with GFP-trap beads as explained above . After separation of the protein samples by SDS-PAGE and Coomassie Brilliant Blue R250 ( Serva , 17525 , Heidelberg , Germany ) staining and distaining of the SDS-Gel was performed and each lane was cut into 10 pieces . A digest with trypsin NB sequencing grade ( Serva , 37283 . 01 , Heidelberg , Germany ) was performed with each gel piece over night at 37°C . The eluates of the five upper and the five lower gel pieces were combined , suspended in sample buffer ( 98% H2O , 2% acetonitrile and 0 . 1% formic acid ) and analyzed by LC-MS . Peptides generated with trypsin on-bead digestion were subjected to LC-MS analysis: 1 . 5 µL of each peptide sample were separated with nano-flow LC using an RSLCnano Ultimate 3000 system ( Thermo Fisher Scientific ) . Peptides were loaded for 5 min with 0 . 07% trifluoroacetic acid on an Acclaim PepMap 100 pre-column ( 100 µm x 2 cm , C18 , 3 µm , 100 Å; Thermo Fisher Scientific ) with a flow rate of 20 µL/min . Separation of peptides was done with reverse phase chromatography on an Acclaim PepMap RSLC column ( 75 µm x 50 cm , C18 , 3 µm , 100 Å; Thermo Fisher Scientific ) at a flow rate of 300 nL/min . The solvent composition was gradually changed within a time period of 94 min from 96% solvent A ( 0 . 1% formic acid ) and 4% solvent B ( 80% acetonitrile , 0 . 1% formic acid ) to 10% solvent B within 2 min , to 30% solvent B within the following 58 min , to 45% solvent B within the next 22 min , and to 90% solvent B within the following 12 min . All solvents and acids had Optima LC/MS quality and were purchased from Thermo Fisher Scientific . Eluting peptides were on-line ionized with nano-electrospray ( nESI ) using the Nanospray Flex Ion Source ( Thermo Scientific ) at 1 . 5 kV ( liquid junction ) and on-line transferred into an Orbitrap VelosPro mass spectrometer ( Thermo Fisher Scientific ) . Full scans were recorded in a mass range of 300 to 1650 m/z at a resolution of 30 , 000 followed by data-dependent top 15 CID fragmentation ( dynamic exclusion enabled ) . LC-MS method programming and data acquisition was performed with the XCalibur 4 . 0 software ( Thermo Fisher Scientific ) . Pathogenic development of wild type and mutant strains was assayed by plant infections of the maize ( Zea mays ) variety Early Golden Bantam ( Olds seeds ) as described before ( Kämper et al . , 2006 ) . For mating assays , strains were crossed on charcoal-containing complete medium plates and incubated at 22°C ( Holliday , 1974 ) . Images were obtained using a Nikon Eclipse 90i fluorescence microscope with a Hamamatsu Orca-ER camera driven by Metamorph ( Universal Imaging , Downingtown , PA ) . Images were further processed with ImageJ software . To determine the presence of Cdc25 at nucleus , merged ( RGB ) images from cells carrying a cut11-cherry allele ( marking the nuclear membrane ) and a GFP-Cdc25 allele were used . In each nucleus , a line was traced covering all nucleus diameter and also surrounding cytoplasm using the ImageJ software and the plot intensity for each channel was obtained ( using the RGB line profile plugging from ImageJ ) . Those cells in which the signal inside the nucleus was above the cytoplasmic background were considered as positive for nuclear GFP . In the few cases of strong cytoplasmic background , the sorting or not of that particular nucleus was decided case by case , by looking the intensity of nuclear signal . To determine the statistical significance of differences a two-tailed Student t-test was used . P-Values were calculated with the GraphPad Prism 5 . 0 software . | Many fungi that cause diseases in plants need specialized structures to penetrate the plant’s tissues . To form these structures , the fungus must carefully control when and where its cells divide . As in other organisms , the sequence of events that lead to a fungal cell dividing in two are known as the cell cycle . Progress through the distinct steps in the cell cycle is regulated by enzymes including many that add or remove phosphate groups on other proteins . It remains unclear which regulatory enzymes allow any plant-infecting fungus to control its cell cycle when it forms an infection structure , but one fungus that could help answer this question is Ustilago maydis , the cause of a disease known as corn smut . The corn smut fungus forms infection structures after two different mating strains meet on the surface of the plant , stop dividing and then fuse . This implies that the cell cycles of both strains need to be coordinated to allow the fungus to infect the plant . The two strains recognize each other via chemical signals known as pheromones , and Bardetti et al . now show that pheromone recognition in the corn smut fungus results in an enzyme called Cdc25 being disabled , which in turn causes cell division to stop . Cdc25 is a phosphatase , meaning it removes phosphate groups from cell cycle regulators that are found in the nucleus of the cell . Specifically , Cdc25 targets phosphate groups that would otherwise inhibit the activity of these proteins . Further experiments showed that , following pheromone recognition , Cdc25 is disabled via a two-step process: first it is prevented from entering the nucleus which keeps it away from its targets , and then it is degraded . Bardetti et al . went on to show that this last step was required for the fungus to infect corn plants , since interfering with the breakdown of Cdc25 impairs its ability to stop the cell cycle and form an infection structure . Entry into plant tissue is a critical step for any parasites looking to invade a plant . Since it is difficult to reach the interior of plants with pesticides , most antimicrobial treatments in plants aim at prevention rather than cure . This means that increasing the delay between a fungus recognizing the surface of a plant and penetrating its tissues could give more time to prevent infections . These new findings represent a step towards achieving that goal , though more research is needed to better understand the molecular mechanisms required for the formation of infection structures in plant-infecting fungi . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2019 | Cytoplasmic retention and degradation of a mitotic inducer enable plant infection by a pathogenic fungus |
β-selection is the most pivotal event determining αβ T cell fate . Here , surface-expression of a pre-T cell receptor ( pre-TCR ) induces thymocyte metabolic activation , proliferation , survival and differentiation . Besides the pre-TCR , β-selection also requires co-stimulatory signals from Notch receptors - key cell fate determinants in eukaryotes . Here , we show that this Notch-dependence is established through antagonistic signaling by the pre-TCR/Notch effector , phosphoinositide 3-kinase ( PI3K ) , and by inositol-trisphosphate 3-kinase B ( Itpkb ) . Canonically , PI3K is counteracted by the lipid-phosphatases Pten and Inpp5d/SHIP-1 . In contrast , Itpkb dampens pre-TCR induced PI3K/Akt signaling by producing IP4 , a soluble antagonist of the Akt-activating PI3K-product PIP3 . Itpkb-/- thymocytes are pre-TCR hyperresponsive , hyperactivate Akt , downstream mTOR and metabolism , undergo an accelerated β-selection and can develop to CD4+CD8+ cells without Notch . This is reversed by inhibition of Akt , mTOR or glucose metabolism . Thus , non-canonical PI3K-antagonism by Itpkb restricts pre-TCR induced metabolic activation to enforce coincidence-detection of pre-TCR expression and Notch-engagement .
To generate a diverse T cell repertoire reactive against many pathogens , the T cell receptor ( TCR ) α and β chain genes somatically rearrange in developing thymocytes . TCR functionality is then assessed at various checkpoints . Thymocytes develop from bone marrow ( BM ) progenitors through successive CD4-CD8- 'double-negative' CD44+CD25-c-Kit+ DN1 , CD44+CD25+c-Kit+ DN2 , HSAhighc-Kit-CD44-CD25+ DN3 and HSAhighCD44-CD25- DN4 stages ( Petrie and Zuniga-Pflucker , 2007; Xiong et al . , 2011 ) . Productive rearrangement of one TCRβ-allele causes surface-expression of a pre-TCR comprised of TCRβ , pre-TCRα and signal-transducing CD3 subunits on DN3 cells ( Aifantis et al . , 2006 ) . At the first checkpoint , β-selection , ligand-independent pre-TCR signaling triggers DN3 cell metabolic activation , proliferation and survival . It also triggers allelic exclusion of the second TCRβ allele , initiation of TCRα gene-rearrangements and differentiation via CD8+HSAhighTCRβlow immature single positive ( ISP ) precursors into CD4+CD8+ 'double-positive' ( DP ) cells ( Petrie and Zuniga-Pflucker , 2007; Xiong et al . , 2011 ) . β-selection ensures that only DN3 cells expressing a functional TCRβ chain develop further . It is the major cell-fate determining event for αβ T cells . Defective β-selection causes a DN3 block and severe immunodeficiency ( Juntilla and Koretzky , 2008; Aifantis et al . , 2006 ) . pre-TCR signaling alone is insufficient for DN-to-DP cell differentiation without co-stimulation by thymic microenvironmental signals . In particular , ligand engagement of Notch on DN3/DN4 cells promotes nutrient receptor expression , glucose uptake , metabolism , growth , survival , proliferation and differentiation . But excessive Notch signaling causes thymocyte transformation and T cell acute lymphoblastic leukemia ( T-ALL ) . This is augmented by pre-TCR signals ( Ciofani et al . , 2004; Ciofani and Zuniga-Pflucker , 2005; Campese et al . , 2006; Fayard et al . , 2010; Taghon et al . , 2006; Aifantis et al . , 2006; Tussiwand et al . , 2011 ) . So , pre-TCR/Notch costimulation needs to be limited and elucidating the underlying mechanisms is of great importance . Both pre-TCR and Notch activate phosphatidylinositol 3-kinases ( PI3K ) ( Ciofani and Zuniga-Pflucker , 2005; Juntilla and Koretzky , 2008; Fayard et al . , 2010 ) . PI3K phosphorylate the membrane lipid phosphatidylinositol ( 4 , 5 ) bisphosphate ( PIP2 ) into phosphatidylinositol ( 3 , 4 , 5 ) trisphosphate ( PIP3 ) . PIP3 recruits and activates Itk/Tec- , Pdk1- , and Akt-family kinases by binding to their PH domains . PI3K are essential and rate-limiting for β-selection by promoting metabolism , proliferation , survival and differentiation ( Juntilla and Koretzky , 2008; Fayard et al . , 2010 ) . Itk promotes activation of phospholipase-Cγ1 ( PLCγ1 ) . PLCγ1 hydrolyzes PIP2 into the second messengers inositol ( 1 , 4 , 5 ) trisphosphate ( IP3 ) and diacylglycerol ( DAG ) , which then convey downstream signals ( Aifantis et al . , 2006 ) . Itk loss only subtly impairs β-selection ( Lucas et al . , 2007 ) . Pdk1 is required for DN3/DN4 cell differentiation mostly by activating Akt , and for thymocyte proliferation through other effectors ( Kelly et al . , 2007; Fayard et al . , 2010 ) . Akt kinases are required for β-selection by promoting DN3/DN4 cell glucose uptake , glycolysis , viability and differentiation ( Juntilla et al . , 2007; Fayard et al . , 2007; Mao et al . , 2007; Fayard et al . , 2010 ) . Recent studies suggest important roles for the Akt activator mTORC2 and possibly the Akt downstream-effector mTORC1 in β-selection ( Lee et al . , 2012; Tang et al . , 2012; Chou et al . , 2014 ) . Canonically , PI3K function is limited through PIP3-removal by the lipid-phosphatases Inpp5d/SHIP1 and Pten ( Juntilla and Koretzky , 2008; Fayard et al . , 2010 ) . Inpp5d/SHIP1-/- early thymocytes develop normally ( Kashiwada et al . , 2006 ) . Conditionally Pten-/- DN cells show constitutively active Akt and accelerated development to DP cells . They can generate DP cells without pre-TCR or Notch-signaling ( Hagenbeek et al . , 2004; Kelly et al . , 2007; Shiroki et al . , 2007; Wong et al . , 2012; Hagenbeek et al . , 2014 ) . Notch may promote DN3/DN4 cell survival and differentiation in part by repressing Pten ( Wong et al . , 2012 ) . So , limiting PI3K signaling is required for β-selection and its dependence on both pre-TCR and Notch . But many details about how pre-TCR and Notch cross-talk via PI3K are controversial , and it remains unclear why pre-TCR signaling alone is insufficient for β-selection ( Juntilla and Koretzky , 2008; Fayard et al . , 2010; Hagenbeek et al . , 2014 ) . IP3 is well known to mobilize Ca2+ but can also be phosphorylated into inositol ( 1 , 3 , 4 , 5 ) tetrakisphosphate ( IP4 ) by four mammalian IP3 3-kinases ( Sauer and Cooke , 2010 ) . Among these , we and others have identified Itpkb as an essential TCR effector . Thymocyte development in Itpkb-/- mice is blocked at the DP stage due to defective positive selection ( Huang et al . , 2007; Pouillon et al . , 2003; Wen et al . , 2004 ) . In thymocytes , TCR signaling activates Itpkb to produce IP4 , a soluble analog of the PH domain binding moiety of PIP3 . Itpkb-/- thymocytes have strongly reduced IP3 3-kinase activity and IP4 levels , but normal IP3 levels and Ca2+ mobilization ( Pouillon et al . , 2003; Wen et al . , 2004 ) . IP4 can bind to PH domains and control PIP3 binding ( Huang et al . , 2007; Jia et al . , 2007 ) . In NK cells , myeloid cells and hematopoietic stem cells ( HSC ) , IP4 competitively limits PIP3-binding to , and activation of Akt ( Jia et al . , 2008; 2007; Sauer et al . , 2013; Siegemund et al . , 2015 ) . Thus , besides PIP3-turnover by Inpp5d/SHIP1 and Pten , IP3 3-kinases can limit PI3K function through a non-canonical mechanism , IP4 antagonism with PIP3 . Here , we present data which suggest that this non-canonical mechanism restricts pre-TCR induced pro-metabolic PI3K/Akt signaling to limit the kinetics and enforce the Notch-dependence of β-selection . Itpkb-/- DN3 cells were pre-TCR hyperresponsive with Akt/mTOR hyperactivation and evidence for metabolic hyperactivity . They showed an accelerated and Notch independent , but pre-TCR dependent differentiation to the DP stage . Pharmacologic inhibition of Akt , mTOR or glucose metabolism restored wildtype ( WT ) developmental kinetics and Notch-dependence of Itpkb-/- DN3 cells .
DN3 cells from Itpkb+/+Rag2-/- but not Itpkb-/-Rag2-/- mice express Itpkb ( Figure 1 ) . To study if Itpkb is required for DN3 cell development , we analyzed DN cell subsets in Itpkb+/+ ( WT ) vs . Itpkb-/- mice by flow-cytometry . For enhanced sensitivity , we gated out lineage-marker positive ( Lin+ ) non-T cells , γδ T cells and HSAlow mature DN αβ T cells ( Bruno et al . , 1996 ) . Compared to controls , Itpkb-/- mice had increased DN3 and reduced DN4 cell proportions with a ~three-fold increased DN3:DN4 ratio ( Figure 2A , B ) . Blocked β-selection usually increases this ratio via accumulation of pre-selection DN3 cells and loss of DN4 cells and descendants , ultimately reducing thymic cellularity ( Michie and Zuniga-Pflucker , 2002 ) . Surprisingly , Itpkb-/- mice had WT-like total thymic cellularity and numbers of DN3 cells and CD25int 'DN3-4' intermediates between DN3 and DN4 cells ( Xiong et al . , 2011 ) ( Figure 2C ) . 10 . 7554/eLife . 10786 . 003Figure 1 . Itpkb protein is expressed in DN3 cells from Itpkb+/+ but not Itpkb-/- mice . Shown are immunoblots ( IB ) of Itpkb immunoprecipitates ( IP , top ) or whole cell lysates ( WCL , bottom ) from Itpkb+/+ or Itpkb-/- DN3 cell-enriched Rag2-/- thymocytes ( left ) or sorted CD53- DP thymocytes ( right ) , resolved via SDS-PAGE and probed with antibodies against Itpkb ( top ) or PLCγ1 ( bottom , loading control ) as in ( Miller et al . , 2007; Huang et al . , 2007 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10786 . 00310 . 7554/eLife . 10786 . 004Figure 2 . Altered β-selection in Itpkb-/- mice . ( A ) Flow cytometric profiles of thymocytes from Itpkb+/+ and Itpkb-/- littermate mice . Top , CD4 and CD8 expression . Upper center , HSA and mature lineage marker ( Lin = CD11b , CD11c , CD19 , B220 , CD49b , Gr-1 , Ter119 , TCRγ ) expression on CD4-CD8- ( DN ) cells . Lower center , CD44 and CD25 expression on HSAhighLin- DN cells . The bottom gates denote DN3 , transitional DN3-4 and DN4 cells from right to left . Bottom two panels , TCRβ expression on DN4 cells , and HSA and TCRβ expression on CD4-CD8+ cells with ISP ( HSAhighTCRβlow ) and mature CD8 T cell ( HSAlowTCRβhigh ) gates . Numbers denote % cells per gate . Representative of at least 7 independent experiments . ( B ) Mean ± SEM %DN3 cells , %DN4 cells or %DN3:%DN4 cell ratio in Itpkb+/+ or Itpkb-/- 5–7 week old littermates . Statistical significance of genotype differences was analyzed by unpaired two-tailed Student's t-tests ( n = 7 ) . ( C ) Total numbers of thymocytes , Lin-HSAhigh CD44-CD25+ DN3 , CD44-CD25int DN3-4 , TCRβ-CD44-CD25- DN4 , CD8α+HSAhighTCRβlow ISP or CD4+CD8+ DP cells in individual Itpkb+/+ or Itpkb-/- mice . Horizontal lines denote means ± SEM . Significance of genotype differences was analyzed as in ( B ) . nWT = 9 , nItpkb-/- = 11 . ns , no significant difference . ( D ) Histograms of surface or total cellular TCRβ or CD3 levels on the indicated thymocyte populations from Itpkb+/+ ( black ) or Itpkb-/- ( gray ) mice . Representative of ≥3 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 10786 . 004 Itpkb-/- DN3 , DN3-4 and DN4 cells contained WT amounts of total TCRβ and CD3 protein ( Figure 2D ) . Upon successful TCRβ-rearrangement , DN3 cells express intracellular TCRβ protein which is then transported to the cell surface ( Aifantis et al . , 2006 ) . Due to constitutive endocytosis , only small surface TCRβ/CD3 amounts are detectable on DN3 and DN3-4 cells ( Panigada et al . , 2002 ) . These were similar between genotypes ( Figure 2D ) . In contrast , the proportion of Lin-HSAhigh DN4 cells expressing surface TCRβ/CD3 was reduced in Itpkb-/- vs . WT mice . These HSAhighLin ( including DX5 ) -CD4-CD8-CD44-CD25- TCRβ+ cells are not β-selection intermediates , but rather comprise post-selection mature precursor thymocytes of CD4-CD8αβ- αβ T cells , found mainly in the gut epithelium ( Pobezinsky et al . , 2012; Gangadharan et al . , 2006 ) ( and Hilde Cheroutre , personal communication ) . Their reduction reflects the mature T cell deficiency in Itpkb-/- mice ( Huang et al . , 2007; Pouillon et al . , 2003; Wen et al . , 2004 ) . Surface TCRβ- DN4 cells have all hallmarks of 'true' DN4 cells: They proliferate highly , are metabolically active and efficiently generate DP cells in vitro ( Petrie et al . , 1990; Levelt et al . , 1993; Panigada et al . , 2002; Kelly et al . , 2007; Yuan et al . , 2011 ) . So , we used TCRβ- DN4 cells to further characterize DN4 cell phenotypes . Compared to WT mice , Itpkb-/- mice had significantly less TCRβ- DN4 and ISP cells , but similar DP cell numbers and an increased ( DN3 + DN3-4 ) :TCRβ- DN4 cell ratio ( Figure 2C ) . pre-TCR expression correlates with upregulated surface CD2 , pre-TCR signaling upregulates surface CD5 , and surface CD27 upregulation is one of the earliest markers of β-selection ( Taghon et al . , 2006; Patra et al . , 2006 ) . Compared to WT controls , Itpkb-/- DN3-4 cells and later stages had increased CD2 levels and normal to increased surface levels of CD5 , CD27 and CD71 ( Figure 3A ) . Also , Itpkb-/- DN3-4 and DN4 cells had elevated surface-levels of the costimulatory chemokine-receptor and Notch-target CXCR4 ( Trampont et al . , 2010; Xie et al . , 2013 ) . In contrast , Itpkb-/- and WT DN3 , DN3-4 and TCRβ- DN4 cells each expressed comparable surface CD28 , CD127 and CD98 . Normal to elevated activation markers , and normal DN3 cell and total thymocyte numbers argue against a pre-TCR signaling defect and β-selection block in Itpkb-/- mice but might rather suggest pre-TCR hyper-responsiveness . To further test this , we bred our mice to Nr4a1/Nur77-GFP transgenics . Here , Nr4a1/Nur77-GFP expression is a highly sensitive readout for TCR signal intensity ( Moran et al . , 2011 ) . Supporting pre-TCR hyper-responsiveness , DN3 and later stages of thymocyte development expressed more Nr4a1/Nur77-GFP in Itpkb-/- than WT mice ( Figure 3B ) . 10 . 7554/eLife . 10786 . 005Figure 3 . Surface marker expression , steady-state proliferation and viability of Itpkb+/+ and Itpkb-/- thymocytes . ( A ) Surface levels of the indicated markers for activation or β-selection on thymocyte subpopulations from Itpkb+/+ ( solid ) or Itpkb-/- ( hatched ) mice . ( B ) Nr4a1/Nur77-GFP expression in Itpkb+/+ or Itpkb-/- Nr4a1/Nur77-GFP transgenic ( solid or hatched black , respectively ) or non-transgenic ( gray ) mice . Representative of ≥3 independent experiments . ( C ) Steady-state proliferative status of the indicated thymocyte subpopulations in Itpkb+/+ ( black ) or Itpkb-/- ( gray ) mice was analyzed by Ki67 stain ( top , representative of 3 independent experiments ) or BrdU incorporation assay ( bottom , representative of 2 independent experiments ) . Thin open histograms , Itpkb+/+ isotype or BrdU-uninjected , respectively , negative control . TCRβlow DP cells were analyzed as they represent the majority of DP cells and Itpkb-/- mice lack TCRβhigh DP cells ( Wen et al . , 2004 ) . ( D ) Steady-state viability of the indicated thymocyte subpopulations in Itpkb+/+ ( black ) or Itpkb-/- ( gray ) mice was analyzed by AnnexinV ( AnnV ) stain . Representative of 4 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 10786 . 005 We next analyzed whether the reduced DN4 and ISP cell numbers in Itpkb-/- mice might reflect reduced proliferation or viability . But similar Ki67-staining and in vivo BrdU incorporation suggest comparable steady-state proliferation of all thymocyte subsets between genotypes ( Figure 3C ) . Similar AnnexinV staining suggests comparable viability ( Figure 3D ) . To explore if the altered β-selection is caused by thymocyte-intrinsic Itpkb-loss , we injected a 1:1 mix of mature T/B cell-depleted CD45 . 1 WT and CD45 . 2 Itpkb-/- BM into lethally irradiated CD45 . 1/CD45 . 2-congenic hosts and analyzed reconstituted thymocyte subsets 6–8 weeks later . We distinguished WT vs . Itpkb-/- donor-derived cells by CD45 allelic expression . Compared to WT controls , Itpkb-/- donor-derived thymocytes reproduced the published cell-intrinsic block at the DP stage ( Wen et al . , 2004 ) and the increased DN3 and reduced DN4 cell proportions , partial loss of TCRβ- DN4 and ISP cells , and increased ( DN3 + DN3-4 ) :TCRβ- DN4 cell ratio of Itpkb-/- mice ( Figure 4A–C ) . Moreover , Itpkb-/- vs . WT donor-derived DN3-4 and DN4 cells overexpressed CD2 and tended to upregulate CXCR4 ( Figure 4D ) . Ki67 staining was again similar between genotypes ( Figure 4E ) . Thus , Itpkb-/- thymocytes show a cell-intrinsically altered β-selection not rescued by a WT environment and the presence of WT thymocytes . 10 . 7554/eLife . 10786 . 006Figure 4 . Itpkb controls β-selection cell-autonomously . B/T cell-depleted BM from CD45 . 1 Itpkb+/+ and CD45 . 2 Itpkb-/- mice was mixed at a 1:1 ratio and injected into CD45 . 1/CD45 . 2 lethally irradiated hosts . 7 weeks later , thymocytes were analyzed by FACS . ( A ) Top , thymocyte expression of CD45 . 1 and CD45 . 2 . The other panels show expression of the indicated markers on CD45 . 1+CD45 . 2-Itpkb+/+ or CD45 . 1-CD45 . 2+Itpkb-/- donor-derived thymocytes , using the gating strategy in Figure 2A . Numbers denote % cells per gate . ( B ) Chimerism of the indicated thymocyte subpopulations , expressed as mean ± SEM ratio of CD45 . 1-CD45 . 2+Itpkb-/- to CD45 . 1+CD45 . 2-Itpkb+/+ donor-derived thymocytes . ( C ) Mean ± SEM ratio of total DN3 cell numbers to TCRβ- DN4 cell numbers in Itpkb+/+ or Itpkb-/- donor-derived thymocytes . Significance of the indicated comparisons was analyzed as in Figure 2 ( n = 3 ) . ( D ) CD2 and CXCR4 expression on Itpkb+/+ ( solid ) and Itpkb-/- ( hatched ) thymocyte subsets in mixed BM chimeras . Representative of 3 independent hosts . ( E ) Ki67 expression in Itpkb+/+ ( black ) and Itpkb-/- ( gray ) DN3 , DN3-4 and DN4 cells in mixed BM chimeras . Open histogram , Itpkb+/+ isotype staining negative control . Representative of 3 independent hosts . DOI: http://dx . doi . org/10 . 7554/eLife . 10786 . 006 The above data suggest that a developmental block , hypoproliferation or increased death do not cause the loss of DN4 and ISP cells in Itpkb-/- mice . Another possibility is that these subsets are depleted by accelerated development to DP cells . Indeed , mathematical modeling suggests that a two fold or larger increase in the rate constants for the successive transitions from DN3 to DN4 to ISP to DP cells with unaltered rates for progenitor development to DN3 cells , and for thymocyte turnover due to proliferation and death , can cause similarly reduced DN4 cell and ISP numbers but normal DN3 and DP cell numbers as seen in Itpkb-/- vs . WT mice ( Figure 5A , B ) . Consistent with faster thymocyte development , Itpkb-/- fetal thymi had reduced overall DN and DN4 cell proportions but higher DP cell proportions and total numbers than WT controls on embryonic day 16 . 5 ( E16 . 5 ) where DP cells are first detectable in WT mice , and on E17 . 5 despite 'catching up' WT DP cells ( Figure 5C , Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 10786 . 007Figure 5 . Accelerated differentiation of Itpkb-/- DN3 thymocytes . ( A ) In silico analysis of β-selection kinetics in Itpkb-/- and Itpkb+/+ mice . Scheme of intrathymic DP cell development from progenitors . The velocities of relevant developmental transitions are characterized by rate constants K ( identical between genotypes ) and K1 , K2 and K3 ( development from DN3 to DP cells , set to over two-fold higher in Itpkb-/- vs . WT mice , Table 1 ) . Rate constants Kd1-Kd4 for subset turnover via proliferation and death were considered identical between genotypes . ( B ) Predicted steady-state numbers of the indicated thymocyte populations in Itpkb+/+ ( black ) or Itpkb-/- ( red ) mice . ( C ) Left , CD4/CD8 expression on embryogenesis day ( E ) 16 . 5 or 17 . 5 fetal thymocytes . Center , CD44/CD25 expression on CD4-CD8-HSAhighLin-TCRγ- cells ( Figure 5—figure supplement 1A ) . Numbers denote % cells per gate . CD4-CD8+ fetal thymocytes are ≥92% ISP ( Figure 5—figure supplement 1B ) . Right , mean ± SEM DP cell number ( # ) or % in E16 . 5 or E17 . 5 Itpkb+/+ or Itpkb-/- thymi . Significance of genotype differences was analyzed as in Figure 2 ( nE16 . 5 = 2 , nE17 . 5 = 3 ) . E16 . 5 data with t-test from another experiment in Figure 5—figure supplement 1C , D . DOI: http://dx . doi . org/10 . 7554/eLife . 10786 . 00710 . 7554/eLife . 10786 . 008Figure 5—figure supplement 1 . Raw and replicate data related to Figure 5C . ( A , B ) HSA and combined Lin/TCRγ expression on the DN ( A ) and CD4-CD8+ ( B ) cells in Figure 5C , left panel . The HSAhighLin-TCRγ- gates in ( A ) were analyzed for DN cell subsets in Figure 5C , center panel . The HSAhighLin-TCRγ- gates in ( B ) denote ISP . ISP comprise ≥92% of CD4-CD8+ cells in E16 . 5 and E17 . 5 fetal thymi in both Itpkb+/+ and Itpkb-/- mice . ( C ) CD4 and CD8 expression on Itpkb+/+ or Itpkb-/- fetal thymocytes from embryos harvested on day 16 . 5 of embryogenesis ( E16 . 5 ) from the same mother . Numbers denote % cells per gate . ( D ) Mean ± SEM numbers or % of DP cells in E16 . 5 Itpkb+/+ or Itpkb-/- fetal thymi . Significance of the indicated comparisons was analyzed as in Figure 2 ( nWT = 3 , nKO = 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10786 . 008 To corroborate these findings in an in vivo system that allows one to kinetically study β-selection of a synchronized DN3 cell population , we generated Itpkb-/-Rag2-/-mice . Rag2-loss causes a developmental arrest at the DN3 stage due to blocked TCRβ expression . Injected anti-CD3 antibodies ( α-CD3 ) can crosslink CD3ε on Rag-/- DN3 cells and trigger their differentiation to DP cells ( Campese et al . , 2006 ) . Our mathematical model predicted immediately increasing DN4 cell numbers , and DP cell accumulation at ≥2 days post α-CD3 injection in Rag2-/- mice ( Figure 6A , B ) . Simulating over two fold faster DN3-to-DP development in Rag2-/-Itpkb-/- mice predicted transiently increased DN4 cell accumulation resulting in earlier DP cell accumulation . Experiments confirmed the predictions: Thymocyte development was blocked at the DN3 stage in Itpkb-/-Rag2-/-and Rag2-/- control mice ( Figure 6C , D ) . α-CD3 injection triggered progressive accumulation of DN4 , ISP and DP cells 2 and 3 days later in both genotypes ( Figure 6C–F ) . However , Itpkb-/-Rag2-/- mice accumulated larger proportions of DN4 ( 33% vs . 18% ) , ISP ( 3% vs . 1% ) and DP cells ( 20% vs . 1% ) than Rag2-/- mice on day 2 when Rag2-/- mice barely had any DP cells . Itpkb-/- DN4 and ISP cell numbers tended to be increased on day 2 but this was not statistically significant ( Figure 6E ) . Itpkb-/-Rag2-/- mice continued to accumulate more DP cells towards an ~four fold excess over Rag2-/- controls on day 3 ( Figure 6F ) . This resulted mostly from faster development , as both genotypes showed similar thymocyte subset viability and proliferation post α-CD3 injection ( Figure 6G , H ) . 10 . 7554/eLife . 10786 . 009Figure 6 . Accelerated differentiation of Itpkb-/- DN3 thymocytes . ( A , B ) Mathematically predicted numbers of DN4 ( A ) and DP cells ( B ) in Rag2-/-Itpkb+/+ ( black ) or Rag2-/-Itpkb-/- ( red ) mice on the indicated days post α-CD3 antibody injection . >98% of thymocytes in Rag2-/- mice are DN3 cells ( C , D ) , and progenitor influx within 3 days is negligible . Thus , we set K = 0 in our model for ( A , B ) . ( C , D ) CD4/CD8 expression on thymocytes ( C ) and CD44/CD25 expression on DN cells ( D ) from Itpkb+/+Rag2-/- or Itpkb-/-Rag2-/- mice before ( day 0 ) , or 2 or 3 days post α-CD3 antibody injection . Gates in ( D ) denote DN1 , DN2 , DN3 , DN3-4 and DN4 cells in clock-wise order , numbers % cells in the DN3 and DN4 gates . Representative of 7 independent experiments . ( E ) Measured mean ± SEM DN4 cell ( upper panel ) and ISP ( lower panel ) numbers in Itpkb+/+Rag2-/- ( solid line ) or Itpkb-/-Rag2-/- ( hatched ) mice before ( day 0 ) or 1 , 2 or 3 days after α-CD3 injection . Significance of the indicated comparisons was analyzed as in Figure 2 . For DN4 cell numbers , n = 4 , 4 , 6 or 6 Rag2-/- and 4 , 4 , 5 or 5 Itpkb-/-Rag2-/- mice , respectively . For ISP numbers , n = 6 , 5 , 7 or 7 Rag2-/- and 4 , 4 , 6 or 5 Itpkb-/-Rag2-/- mice , respectively . ( F ) Mean ± SEM DP cell % ( upper panel ) or number ( lower panel ) in Itpkb+/+Rag2-/- ( solid line ) or Itpkb-/-Rag2-/- ( hatched ) mice before ( day 0 ) or 1 , 2 or 3 days after α-CD3 antibody injection . Significance of genotype differences per day was analyzed as in Figure 2 . n = 6 , 5 , 7 or 7 Rag2-/- and 4 , 4 , 6 or 5 Itpkb-/-Rag2-/- mice , respectively . ( G ) Annexin V ( AnnV ) staining of DN3 , DN3-4 , DN4 , HSAhiCD8+ ISP or DP cells from uninjected Rag2-/- ( open histograms ) or α-CD3 antibody injected Itpkb+/+Rag2-/- ( black filled histograms ) or Itpkb-/-Rag2-/- ( gray filled histograms ) mice two ( left ) or three ( right ) days post antibody injection . Representative of 3 independent experiments and 3–4 mice per genotype . Uninjected Rag2-/- mice contain dying cells in the DN4 , CD8-ISP and DP cell gates due to failed β-selection at the DN3 stage . These serve as positive controls for the Annexin V stain . ( H ) Ki67 expression in DN3 , DN3-4 , DN4 , HSAhiCD8+ ISP or DP cells from α-CD3 antibody injected Itpkb+/+Rag2-/- ( black filled histograms ) or Itpkb-/-Rag2-/- ( gray filled histograms ) mice two ( left ) or three ( right ) days post antibody injection . Representative of 2 independent experiments and 3 mice per genotype . Open histograms , day 0 WT isotype control . ( I ) CD2 , CD5 , CD71 and CXCR4 surface-levels on , and transgenic Nr4a1/Nur77-GFP expression in DN3 or DN4 thymocytes from uninjected Rag2-/- ( open histograms ) or α-CD3 injected Itpkb+/+Rag2-/- ( black ) or Itpkb-/-Rag2-/- ( gray ) mice 2 days post injection . The <1% CD44-CD25- negative control cells in uninjected Rag2-/- mice are non-T cells . Representative of ≥3 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 10786 . 009 Further supporting accelerated development , Itpkb-/- sorted DN3 cells generated larger proportions of DP cells than WT DN3 cells after 4-day co-culture on OP9DL1 stroma cells ( Ciofani and Zuniga-Pflucker , 2005; Ciofani et al . , 2004 ) ( Figure 7A , B , Figure 9—figure supplement 1 ) . Finally , Itpkb-/- E15 . 5 fetal thymic organ cultures ( FTOC ) produced more DP cells than WT controls ( Figure 7C , D ) . 10 . 7554/eLife . 10786 . 010Figure 7 . Itpkb-loss in DN3 cells causes accelerated , Notch-independent development to the DP stage . ( A , B ) Sorted DN3 cells from 6 . 5 week old Itpkb+/+ or Itpkb-/- mice were seeded onto Delta-like 1 Notch ligand-expressing OP9DL1 or Notch ligand-free OP9 stroma cells and analyzed for CD4/CD8 expression 4 days later . ( A ) Representative FACS data from input ( day 0 ) or day 4 cultures . The numbers indicate % cells in the DP or DN gates , respectively . Representative of 5 independent experiments . ( B ) Bar-graphs showing mean ± SEM Itpkb+/+ ( black bars ) or Itpkb-/- ( open bars ) % DP cells after 4-day culture on OP9DL1 or OP9 cells , averaged from 4 independent experiments . Significance for genotype differences was analyzed as in Figure 2 ( n = 4 ) . ( C , D ) Fetal thymic lobes from Itpkb+/+ or Itpkb-/- embryos harvested on day 15 . 5 of embryogenesis ( E15 . 5 ) from the same mother were cultured in the presence of ethanol ( vehicle ) or 20 μM rapamycin for 4 days , harvested and analyzed . ( C ) Representative FACS plots of CD4/CD8 expression on total thymocytes . Numbers denote % cells in the respective gate . ( D ) Bar graph of mean ± SEM % DP cells for each condition and genotype from 3 independent experiments . Significance of the indicated comparisons was analyzed as in Figure 2 ( n = 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10786 . 010 Activation marker upregulation and faster DN-to-DP cell development suggest increased pre-TCR signaling in Itpkb-/- mice . To confirm this , we analyzed the α-CD3 induced upregulation of pre-TCR activation markers in the Rag2-/- mouse system . Compared to uninjected Rag2-/- controls , α-CD3 injection upregulated surface CD2 , CD5 , CD71 and CXCR4 on DN3 and , more pronounced , on DN4 cells in Rag2-/- and Itpkb-/- Rag2-/- mice 2 days later ( Figure 6I ) . All markers reached higher surface-levels on Itpkb-/-Rag2-/- than Rag2-/- DN3 and DN4 cells . Thus , Itpkb-/- DN3/DN4 cells are hyperresponsive to CD3-crosslinking . To determine if signaling is increased , we injected α-CD3 into Nr4a1/Nur77-GFP transgenic Rag2-/- and Itpkb-/-Rag2-/- mice . Higher Nr4a1/Nur77-GFP induction in Itpkb-/- Rag2-/- than Rag2-/- mice confirmed increased CD3 signaling ( Figure 6I ) . Conditionally Pten-/- DN cells show constitutively active Akt and accelerated development to DP cells ( Hagenbeek et al . , 2004; Kelly et al . , 2007; Shiroki et al . , 2007; Wong et al . , 2012 ) . Itpkb can limit receptor-mediated Akt activation through IP4/PIP3 antagonism ( Sauer and Cooke , 2010 ) . To test if this mechanism restricts pre-TCR responses , we compared signaling via PI3K , Akt and downstream mTOR in Itpkb-/- and WT thymocyte subsets . Compared to negative controls and positive controls treated with the phosphatase-inhibitor Calyculin A ( Pozuelo-Rubio et al . , 2010 ) , WT and Itpkb-/- TCRβ- pre-selection DN3 cells had significant but similar basal amounts of phosphorylated active Akt , mTOR and downstream S6 protein ( Figure 8A ) . TCRβ+ DN3 cells undergoing β-selection upregulated phospho-Akt , -mTOR , -S6 and control -Erk , indicating pre-TCR mediated activation . Importantly , and contrasting with a WT-like total Akt protein content in Itpkb-/- cells , phospho-Akt , -mTOR and -S6 but not -Erk levels were higher in Itpkb-/- than WT TCRβ+ DN3 and DN3-4 cells ( Figure 8A ) . This genotype difference disappeared as signaling was downregulated in later developmental stages or in mature TCRβ+ 'DN4-phenotype' gut T cell precursors . 10 . 7554/eLife . 10786 . 011Figure 8 . Increased pre-TCR signaling via PI3K/Akt/mTOR in Itpkb-/-DN3 cells . ( A , B ) We analyzed ( A ) cellular content of T308-phosphorylated active Akt ( pAktT308 ) , S2481-phosphorylated mTOR ( pmTORS2481 ) , S235/S236-phosphorylated ribosomal protein S6 ( pS6S235/S236 ) , Glut1 protein , T202/Y204-phosphorylated Erk ( pErkT202/Y204 ) and Akt protein , and ( B ) cell size via side/forward-scatter analysis ( SSC-A/FSC-A ) in the indicated thymocyte populations of Itpkb+/+ ( black histograms ) or Itpkb-/- ( gray histograms ) mice by FACS . Thin open histograms , Itpkb+/+ isotype or second antibody stained negative controls . Bold open histograms , Calyculin A-treated positive controls . Arrowheads show gate positions . In ( B ) , numbers indicate % cells per large cell gate . Representative of at least 2 ( pS6S235/S236 ) , 3 ( total Akt ) or 8 ( else ) independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 10786 . 011 Akt/mTOR-activation by the pre-TCR and Notch promotes DN3/DN4 cell metabolism in part by increasing nutrient import through upregulation or activation of the glucose-transporter Glut1 , the L-amino acid transporter CD98 and the transferrin receptor CD71 on the cell surface . This results in an increased cell size ( Ciofani and Zuniga-Pflucker , 2005; Kelly et al . , 2007; Fayard et al . , 2010; Juntilla and Koretzky , 2008; Janas et al . , 2010 ) . Metabolic activation appeared increased in Itpkb-/- TCRβ+ DN3 and DN3-4 cells , because they showed Glut-1 hyperinduction and increased large cell proportions over WT controls ( Figure 8A , B ) . Moreover , α-CD3 injection hyperinduced surface CD71 on DN3 and DN4 cells in Itpkb-/-Rag2-/- versus Rag2-/- mice ( Figure 6I ) . Taken together , these data suggest that Itpkb limits PI3K/Akt/mTOR signaling in , and metabolic activation of surface TCRβ+ DN3 and DN3-4 cells . To determine if the Akt/mTOR and metabolic hyperactivation of Itpkb-/- pre-TCR+ DN3 cells causes their accelerated development to DP cells , we next studied if treatment with inhibitors of Akt ( Akt-inhibitor VIII ) , mTOR ( rapamycin ) or glucose-metabolism ( 2-deoxy-D-glucose , 2DG ) could reverse the increased DP cell generation from equal numbers of sorted Itpkb-/- versus Itpkb+/+ DN3 cells on OP9DL1 stroma . Strikingly , all three treatments yielded Itpkb-/- DP cell proportions similar to those of untreated or carrier-treated WT controls after 4-day co-culture ( Figure 9A , Figure 9—figure supplement 1A , C , E ) . As expected , the treatments also reduced WT DP cell generation below untreated controls . Their reduced efficacy towards Itpkb-/- DN cells is expected , as these had increased amounts of the respective active inhibitor-targets ( Figure 8A ) . Similarly complete rapamycin reversal of the accelerated DN-to-DP development in Itpkb-/- versus Itpkb+/+ FTOC confirmed these important findings in a less reductionist system ( Figure 7C , D ) . These data suggest that the hyperactive Akt , mTOR and glucose metabolism of Itpkb-/- DN3 cells contribute to their accelerated DN-to-DP development . 10 . 7554/eLife . 10786 . 012Figure 9 . Itpkb renders β-selection Notch-dependent . ( A , B ) Addition of inhibitors of Akt , mTOR or glucose metabolism reverses the accelerated development of Itpkb-/- DN3 cells and re-establishes Notch-dependence . Shown are mean ± SEM Itpkb+/+ ( solid black bars ) or Itpkb-/- ( pen bars ) % DP cells after 4-day culture on OP9DL1 ( A ) or OP9 ( B ) cells without or with addition of carrier ( solid gray bars; DMSO , ethanol or PBS , respectively ) , 500 nM Akt-inhibitor VIII in DMSO ( Akt-I , added once on day 0 ) , 4 μM rapamycin in ethanol ( Rapa , added once on day 0 ) or 500 μM 2-deoxy-D-glucose in PBS ( 2DG , added once daily ) , averaged from 3 ( Akt-I ) , 4 ( rapamycin ) or 2 ( 2DG ) independent experiments . Significance of the indicated comparisons was analyzed as in Figure 2 . Replicate numbers are indicated above each bar . Representative FACS-data in Figure 9—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10786 . 01210 . 7554/eLife . 10786 . 013Figure 9—figure supplement 1 . Raw FACS data from one representative experiment included in the averaged data in Figure 9 . Itpkb+/+ or Itpkb-/- sorted DN3 cells were cultured for 4 days on OP9DL1 ( A , C , E ) or OP9 ( B , D , F ) stroma cells without ( left ) or with once on day 0 ( Akt-I , rapamycin ) or once-daily ( 2DG ) addition of carrier ( center of each panel ) , 500 nM Akt-inhibitor VIII in DMSO ( Akt-I ) , 4 μM rapamycin in ethanol or 500 μM 2-deoxy-D-glucose in PBS ( 2DG ) . Numbers denote % cells in the respective DP or DN gates . Representative of 3 ( A , B ) , 4 ( C , D ) or 2 ( E , F ) independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 10786 . 013 Constitutive Akt activity promotes glucose metabolism and allows DN3-to-DP cell maturation without Notch signaling ( Ciofani and Zuniga-Pflucker , 2005; Lee et al . , 2012; Fayard et al . , 2010; Juntilla and Koretzky , 2008 ) . To test if the Akt/mTOR hyperactivity in Itpkb-/- TCRβ+ DN3 cells has the same effect , we co-cultured sorted Itpkb-/- or WT DN3 cells with OP9 stroma lacking Notch ligands ( Figures 7A , B , 9B , Figure 9—figure supplement 1B , D , F ) . As previously described ( Ciofani and Zuniga-Pflucker , 2005; Xiong et al . , 2011 ) , WT DN3 cells much less efficiently generated DP cells without than with Notch ( Figures 7A , B , 9A , B ) . Inhibition of Akt , mTOR or Glucose-metabolism further reduced WT DP cell production consistent with known Akt , mTOR and glycolysis requirements for DN thymocyte survival and differentiation ( Lee et al . , 2012; Fayard et al . , 2010; Juntilla and Koretzky , 2008 ) . In striking contrast , Itpkb-/- DN3 cells efficiently generated DP cells on OP9 stroma ( Figures 7A , B , 9B , Figure 9—figure supplement 1B , D , F ) . Inhibition of Akt , mTOR or glycolysis strongly reduced DP cell output . Thus , Itpkb-loss renders the DN3-to-DP transition Notch-independent in an at least partially Akt , mTOR and glycolysis-dependent manner . Notch signaling depends on its cleavage by cellular γ-secretases ( Wong et al . , 2004 ) . To corroborate our findings in vivo , we thus analyzed DN3 cell maturation in Itpkb-/-Rag2-/- versus Rag2-/- mice treated p . o . with vehicle or the γ-secretase inhibitor LY-411 , 575 for two days post α-CD3 injection . P . o . administered 10 mg/kg LY-411 , 575 potently inhibited γ-secretase function in mice and impaired DN thymocyte maturation into αβ T cells with a particularly profound loss of DP cells ( Wong et al . , 2004 ) . We found that LY-411 , 575 strongly impaired the α-CD3 induced DN cell development into ISP and DP cells in Rag2-/- but not Itpkb-/-Rag2-/- mice ( Figure 10 ) . Hence , Itpkb-loss reduces the Notch-dependence of DN thymocyte development to DP cells both in vitro and in vivo . 10 . 7554/eLife . 10786 . 014Figure 10 . Itpkb-loss reduces the Notch-dependence of DN thymocyte development to DP cells in vivo . Shown are ( A ) CD4/CD8 expression on total thymocytes and ( B ) HSA/TCRβ expression on CD4-CD8+ thymocytes from Rag2-/- and Rag2-/-Itpkb-/- mice two days post α-CD3 antibody injection . Starting 3–4 hr before α-CD3 injection , the mice were treated once daily with orally administered γ-secretase inhibitor LY-411 , 575 or vehicle ( Wong et al . , 2004 ) . Numbers indicate % cells per respective gate . The gates in ( B ) denote CD8+HSAhigh ISP ( Petrie and Zuniga-Pflucker , 2007; Xiong et al . , 2011 ) . Representative of two independent experiments ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10786 . 014
Here , we identify Itpkb as a novel pre-TCR effector which restricts the kinetics of β-selection and establishes its Notch-dependence . Itpkb-/- mice show a cell-autonomously accelerated and Notch independent , but pre-TCR dependent DN3-to-DP cell differentiation associated with DN3 cell pre-TCR hyperresponsiveness , Akt/mTOR hyperactivation and evidence for metabolic hyperactivity . Pharmacologic inhibition of Akt , mTOR or glucose metabolism restored WT kinetics and Notch-dependence of Itpkb-/- DN3-to-DP cell development . In thymocytes , TCR engagement activates Itpkb to produce IP4 . Itpkb-/- thymocytes had strongly reduced IP3 3-kinase activity and IP4 levels , but normal IP3 levels and Ca2+ mobilization ( Huang et al . , 2007; Pouillon et al . , 2003; Wen et al . , 2004 ) . IP4 competitively limits PIP3-binding to the Akt PH domain and Akt activation in NK cells , myeloid cells and HSC ( Jia et al . , 2008; 2007; Sauer et al . , 2013; Siegemund et al . , 2015 ) . Thus , we propose that pre-TCR induced IP4/PIP3 antagonism governs β-selection by restricting PI3K/Akt/mTOR signaling and metabolic activation . We derive a model where Itpkb controls pre-TCR/Notch crosstalk through combined restriction of pre-TCR induced and Notch induced PI3K signaling via Akt ( Figure 11 ) . Itpkb enforced coincidence detection of pre-TCR surface expression and Notch-engagement ensures that Akt is only activated to the degree needed for β-selection and only in an appropriate context , pre-TCR+ DN3 cells interacting with Notch-ligand expressing subcapsular stromal cells ( Petrie and Zuniga-Pflucker , 2007 ) . This prevents premature differentiation . Similarly accelerated DN-to-DP cell development of Itpkb-/- and Pten-/- thymocytes ( Hagenbeek et al . , 2004; Shiroki et al . , 2007 ) and enhanced DP cell production from DN3 cells expressing a dominant-active mutant version of the class I PI3K regulatory subunit p85α/Pik3r1 ( p65PI3K transgenic mice ) ( Rodriguez-Borlado et al . , 2003 ) or dominant-active , myristoylated Akt1 ( myr-Akt transgenic mice ) ( Lee et al . , 2012 ) highlight the importance of restricting PI3K signaling via Akt for proper β-selection kinetics , even though the specific purpose of delaying DP cell maturation remains unknown . 10 . 7554/eLife . 10786 . 015Figure 11 . Antagonistic signaling by PI3K and Itpkb controls the kinetics and Notch-dependence of β-selection . ( A ) We propose a model in which pre-TCR and Notch signaling both activate PI3K to produce PIP3 in DN3/DN3-4 cells . PIP3 then recruits and activates Akt to increase glucose metabolism via the Akt/mTOR pathway . This is required for DN3-to-DP cell differentiation . However , pre-TCR signaling also activates Itpkb to produce IP4 , which competes with PIP3 for Akt PH domain binding and limits Akt recruitment , Akt and mTOR activation in pre-TCR expressing DN3/DN3-4 cells . IP4 may have additional effectors , indicated by the question mark . By limiting downstream glucose metabolism , this "IP4 brake" delays the kinetics of β-selection and renders this process dependent on Notch costimulation . ( B ) Without Itpkb , IP4 no more dampens Akt activation and pre-TCR signaling alone sufficiently activates Akt/mTOR signaling to trigger DP cell development in the absence of Notch engagement . ( C ) In the presence of Notch-signals , Akt is now hyperactivated and causes an accelerated DN3-to-DP cell differentiation . DOI: http://dx . doi . org/10 . 7554/eLife . 10786 . 015 The increased DP cell production from Itpkb-/- versus WT DN3 thymocytes in several different in vivo and in vitro models without concomitantly increased proliferation or viability suggests accelerated developmental kinetics , consistent with our mathematical simulations . We propose that this is caused by pre-TCR hyperresponsiveness based on the Akt/mTORC1 hyperactivation , increased Nr4a1/Nur77-GFP expression and hyperinduction of activation markers in Itpkb-/- versus WT DN3 and later stage thymocytes , and on the phenotype reversal by Akt/mTORC1 and metabolic inhibitors . Importantly , the Nr4a1/Nur77-GFP and activation marker hyperinduction in Itpkb-/- DN3 cells indicate increased transcriptional responses . This might accelerate development by inducing required amounts of cell fate determinants earlier in Itpkb-/- DN3 cells than in WT cells . Alternatively or in addition , Itpkb-loss might increase the number of DN3 cells responding to pre-TCR signals or developmental cues present even in the reductionist OP9-DL1 cell co-culture system . Our present data do not allow us to discern the relative contributions of accelerated kinetics of cellular signaling events versus increased proportions of responding cells due to lowered pre-TCR signaling thresholds or enhanced sensitivity to developmental cues upon Itpkb-loss . Distinguishing between these possibilities will require future detailed studies of the effects of Itpkb-loss on the sizes ( amplitudes and proportions of cells responding ) , kinetics ( rate constants ) and shapes ( analog , digital ) of pre-TCR signaling events , transcriptional and functional responses in populations of individually analyzed DN3 cells , combined with mathematical simulations . pre-TCR and Notch signaling both promote DN3 cell proliferation , survival and differentiation in part by activating PI3K/Akt/mTOR ( Janas et al . , 2010; Ciofani et al . , 2004; Ciofani and Zuniga-Pflucker , 2005; Taghon et al . , 2006; Lee et al . , 2012; Kelly et al . , 2007 ) . The dependence of β-selection on combined pre-TCR and Notch signaling relies on dampened PI3K/Akt signaling ( Juntilla and Koretzky , 2008; Fayard et al . , 2010 ) : Conditionally Pten-/- DN cells have constitutively active Akt and generate DP cells without pre-TCR or Notch-signaling ( Hagenbeek et al . , 2004; Kelly et al . , 2007; Shiroki et al . , 2007; Wong et al . , 2012 ) . And constitutive Akt activity can substitute for Notch or mTORC2 to promote DN cell glycolysis , survival and differentiation . It allows DN3-to-DP cell development without pre-TCR or Notch-signaling , but not without both ( Mao et al . , 2007; Ciofani and Zuniga-Pflucker , 2005; Kelly et al . , 2007; Lee et al . , 2012 ) . Although many details of how Notch and PI3K intersect remain unclear , Notch may promote β-selection in part by inducing HES1 to repress Pten , and c-Myc to promote proliferation ( Wong et al . , 2012 ) . Contrasting with Pten-loss , Notch or Akt hyperactivity , Itpkb-loss accelerates DN3 cell differentiation without significant effects on proliferation and viability , and overcomes the Notch dependence but not the pre-TCR dependence of β-selection . This is evidenced by the lack of accumulating intracellular TCRβ- DP cells in Itpkb-/- mice , and of post-DN3 cells in Itpkb-/-Rag2-/- mice ( Figures 2D , 6C , D ) . We speculate that this reflects the need for TCR signals to activate Itpkb and produce IP4 ( Chamberlain et al . , 2005; Wen et al . , 2004; Pouillon et al . , 2003 ) . By abrogating pre-TCR induced IP4-inhibition of pre-TCR and Notch induced Akt/mTOR signaling , Itpkb-loss mimics the effects of Pten-loss or dominant active Akt1 expression . This overcomes Notch-requirements and accelerates differentiation but not proliferation , because Notch-induction of c-Myc is PI3K-independent ( Wong et al . , 2012 ) . The surprising lack of increased DN3/DN4 cell viability in Itpkb-/- mice might reflect differing degrees of Akt/mTOR hyperactivation in Pten-/- , dominant active Akt1-expressing and Itpkb-/- DN3/DN4 cells , consistent with unaltered development of Inpp5d/SHIP1-/- thymocytes ( Kashiwada et al . , 2006 ) . Altogether , the largely restored developmental kinetics and Notch-dependence of Itpkb-/- DN3 cells by treatment with Akt , mTORC1 or metabolic inhibitors support contributing roles for the Akt/mTOR-hyperactivity . Future studies with sub-optimal Akt/mTOR inhibitor concentrations not affecting WT thymocytes but still reversing the Itpkb-/- phenotype , with complex genetic models and with inhibitors of β-selection effectors unaffected by Itpkb will be needed to more conclusively distinguish between specific causative roles for the Akt/mTORC1 and metabolic hyperactivity and mere remaining sensitivity of Itpkb-/- DN thymocytes to inhibition of this particular pathway . Such studies can also address whether additional mechanisms contribute to the β-selection phenotype of Itpkb-/- mice . Contrasting with dominant active Akt1 expression or loss of Pten , which has high constitutive PIP3-phosphatase activity ( Leslie and Foti , 2011 ) , Itpkb-loss cannot replace pre-TCR signals because Itpkb is inactive without them , so its loss has no further effect . Itpkb-loss might also reduce less essential positive Itpkb roles in pre-TCR signaling , such as augmenting PLCγ1/Erk activation by Itk ( Huang et al . , 2007 ) . Indeed , TCRβ+ DN3 cells from Itpkb-/- vs . WT mice tended to have mildly reduced Erk activity ( Figure 8A ) . Erk signaling is required for DN cell proliferation and differentiation ( Kortum et al . , 2013 ) . The mild defects in Itpkb-/- mice are consistent with the only minor role of Itk in β-selection ( Lucas et al . , 2007 ) and the unaltered DN cell proliferation . Hyper-upregulation of Glut1 , CD71 and cell-size in Itpkb-/- TCRβ+ DN3 cells and reversal of their accelerated , Notch-independent differentiation by the glycolytic inhibitor 2DG suggest that Itpkb controls β-selection by ultimately restricting DN3 cell metabolic activation . Similar Akt-inhibitor and rapamycin effects indicate a causative role for Akt/mTOR hyperactivation . Akt promotes metabolism by increasing Glut1 expression and activity , regulating enzymes in glucose and lipid metabolism and promoting mTOR-dependent protein translation ( Juntilla et al . , 2007 ) . In DN cells , Pdk1/Akt/mTORC1 also upregulate surface CD71 and CD98 downstream of pre-TCR and Notch ( Kelly et al . , 2007; Fayard et al . , 2010 ) . Thus , upregulated iron uptake , glucose and amino acid metabolism and protein biosynthesis might all contribute to the accelerated , Notch-independent development of Itpkb-/- DN3 cells . Excessive Notch signaling causes thymocyte transformation and T-ALL . This is augmented by pre-TCR signals ( Campese et al . , 2006; Fayard et al . , 2010 ) . Excessive Akt activity in thymocytes due to PI3K hyperactivity , Pten inactivation or dominant-active Akt1 expression causes leukemia/lymphoma ( Aifantis et al . , 2006; Fayard et al . , 2010 ) . The intermediate β-selection phenotype of Itpkb-loss between those of Pten-loss ( Hagenbeek et al . , 2004; Kelly et al . , 2007; Shiroki et al . , 2007; Wong et al . , 2012 ) and Inpp5d/SHIP1-loss ( Kashiwada et al . , 2006 ) raises the possibility that IP3 3-kinases could have tumor suppressor functions by limiting Akt signaling . But we have not seen thymocyte neoplasia or accumulation of intracellular TCRβ- DP cells in Itpkb-/- mice . One possible explanation consistent with low residual IP3 3-kinase activity and IP4-production in Itpkb-/- thymocytes ( Wen et al . , 2004; Pouillon et al . , 2003 ) is partial Itpkb redundancy with other IP3 3-kinases . Moreover , their premature lethality due to infections ( Pouillon et al . , 2003 ) and anemia ( Siegemund et al . , 2015 ) limits aging studies with Itpkb-/- mice . It will be important to re-assess in a germ-free vivarium whether conditional Itpkb-disruption in thymocytes causes T-ALL as the mice age , or on a sensitized Trp53-/- background as seen for p65PI3K transgenics ( Borlado et al . , 2000 ) . Then again , IP4-loss might simply not augment PIP3 cellular activity sufficiently to transform thymocytes , reminiscent of Inpp5d/SHIP1-/--loss ( Kashiwada et al . , 2006 ) . Also , the potential reduction of required Akt-unrelated IP4 functions such as promoting Itk/Erk signaling ( Huang et al . , 2007 ) might prevent thymocyte transformation in Itpkb-/- mice . Clearly , more studies are needed to assess the tumor suppressor potential of Itpks . By unveiling Itpkb antagonism with PI3K/Akt/mTOR signaling as a key determinant of the kinetics and Notch-dependence of thymocyte β-selection , our findings expand our limited knowledge about physiological IP3 3-kinase functions ( Sauer and Cooke , 2010 ) . They unveil a novel molecular mechanism that integrates pre-TCR signaling with costimulatory Notch signaling to specifically restrict DN3 cell differentiation uncoupled from proliferation and survival . Broad expression of IP3 3-kinases and PI3Ks , IP4 detection in multiple tissues ( Sauer and Cooke , 2010 ) and common PI3K implication in costimulation raise the possibility that 'metabokinetic' control and costimulation-enforcement through non-canonical PI3K antagonism by IP3 3-kinases are broadly relevant .
Our C57BL/6 Itpkb-/- mice were described in ( Sauer et al . , 2013 ) . All animal studies were approved by the Scripps Research Institute animal care and use committee and conform to all relevant regulatory standards . Mixed bone marrow chimeras were generated as in ( Sauer et al . , 2013 ) . For in vivo induction of Rag2-/- DN3 cell differentiation , 10 μg anti-CD3 antibodies ( BD Biosciences , San Jose , CA , clone 145-2C11 ) were injected i . p . 1–3 days later , the mice were euthanized and analyzed . Where indicated , the mice were treated orally once daily with 10 mg/kg LY-411 , 575 ( Wong et al . , 2004 ) or vehicle ( 5% polyethylene glycol , 3% propylene glycol , 1% ethanol , 0 . 4% methylcellulose ) . The first dose was administrated 3–4 hr prior to α-CD3 injection . For BrdU incorporation assays , we injected mice i . p . with 100 µl BrdU [10 mg/ml] and analyzed thymi 4 hr later . For preparation of thymocyte suspensions , thymi were placed in M199 medium/2% FCS/1x penicillin/streptomycin/glutamate at room temperature and single-cell suspensions prepared by passage through a 40 μm mesh ( BD Biosciences ) . Thymocytes were stained with fluorochome-conjugated antibodies against CD2 ( clone RM2-5 ) , CD3 ( 145-2C11 , eBiosciences ) , CD4 ( GK1 . 5 ) , CD5 ( 53–7 . 3 ) , CD8α ( 53–6 . 7 ) , CD24/HSA ( M1/69 ) , CD25 ( 3C7 ) , CD27 ( LG . 3A10 ) , CD44 ( IM7 ) , CD71 ( R17217 ) , CD98 ( RL388 , eBiosciences ) , CD127 ( A7R34 ) , TCRβ ( H57-597 ) or CXCR4 ( 2B11 , eBiosciences ) . Our lineage ( Lin ) cocktail included biotinylated antibodies against CD11b ( M1/70 ) , CD11c ( N418 ) , CD19 ( 6D5 ) , B220 ( 30-F11 ) , CD49b ( DX5 ) , Gr-1 ( RB6-8C5 ) , Ly-76/Ter119 ( TER-119 ) and TCRγ/δ ( GL3 ) . Unless indicated otherwise , all antibodies were from Biolegend . For intracellular staining , cells were permeabilized with BD Cytofix/Cytoperm kits or 0 . 3% Triton X100 and stained with antibodies against Ki-67 ( B56 , BD Biosciences ) , phospho-Akt T308 ( C31E5E , Cell Signaling Technology ) , phospho-mTORC1 S2481 ( poly6517 , Biolegend ) ( Soliman et al . , 2010 ) , phospho-ribosomal protein S6 S235/S236 ( D57 . 2 . 2E , Cell Signaling Technology ) , Glut1 ( Fitzgerald Industries ) , phospho-Erk T202/Y204 ( D13 . 14 . 4E , Cell Signaling Technology ) , Akt ( Cell Signaling Technology ) or isotype controls ( Cell Signaling Technology ) followed by anti-rabbit IgG secondary antibodies ( Cell Signaling Technology ) if needed . Calyculin A ( Cell Signaling ) was used according to the manufacturer's protocol . Annexin V staining was performed using eBiosciences Annexin V apoptosis detection kits . BrdU incorporation was assayed with BD Biosciences BrdU-FITC kits , using 0 . 8 μl anti-BrdU-FITC antibodies per 106 cells . All data were acquired on an LSRII flow cytometer ( BD Biosciences ) and analyzed using FlowJo software . For DN3 cell purification , thymocytes were first immunomagnetically depleted of CD3 , CD4 or CD8α positive cells using biotinylated antibodies against CD3 ( 145-2C11 , Biolegend ) , CD4 ( GK1 . 5 , eBiosciences ) and CD8α ( 53–6 . 7 , eBiosciences ) , anti-biotin microbeads and LS columns ( Miltenyi Biotec ) following the manufacturer’s protocol , and then stained with anti-CD44-FITC ( IM7 , eBiosciences ) , anti-CD25-PerCp-Cy5 . 5 ( 3C7 , Biolegend ) and Streptavidin-APC ( Life Technologies ) . CD44-CD25+SA- DN3 cells were sorted on a BD FACS Aria cell sorter . CD53- DP thymocytes were purified as in ( Huang et al . , 2007 ) . Different lobes from the same thymus of an embryonic day 15 . 5 ( E15 . 5 ) Itpkb+/+ or Itpkb-/- embryo were cultured on gelfoam sponges in complete DMEM-10 with vehicle ( 100% ethanol ) or 20 µM rapamycin in ethanol ( BIOTANG/TSZCHEM ) . New vehicle or rapamycin were added on culture days 1 , 2 and 3 . The lobes were analyzed on day 4 . OP9 or OP9DL1 cells ( Ciofani and Zuniga-Pflucker , 2005; Ciofani et al . , 2004 ) were seeded at 8000 cells per well and incubated for 24 hr in OP9 Culture Media ( alpha-MEM , Life Technology/15% FCS/1x Penicillin and Streptomycin ) , followed by addition of 70 , 000–100 , 000 sorted DN3 cells per well together with 1 ng/ml recombinant mouse IL-7 ( PeproTech ) and once on day 0 ( rapamycin , Akt-I ) or once-daily ( 2DG ) addition of carrier , 500 nM Akt-inhibitor VIII ( Sauer et al . , 2013 ) ( Akt-I , Calbiochem ) in DMSO , 4 μM rapamycin ( BIOTANG/TSZCHEM ) in ethanol or 500 μM 2-deoxy-D-glucose ( Wang et al . , 2011 ) ( 2DG , SIGMA ) in PBS ( all concentrations final ) . Thymocytes were lysed as previously described ( Huang et al . , 2007 ) in 1% Triton X-100/60 mM octylglucoside/150 mM NaCl/25 mM Tris-HCl , pH 7 . 5/1 mM EDTA containing Roche Complete Mini Protease Inhibitor and PhosSTOP Phosphatase Inhibitor Cocktails . Lysates were incubated for 20 min at 4°C , then cleared by centrifugation at 14 , 000 g for 10 min at 4°C . For immunoprecipitations , pre-cleared lysates were incubated for 1 . 5 hr with anti-Itpkb antibodies ( G-20 , Santa-Cruz Biotechnology ) followed by incubation with Protein G-conjugated beads for 1 . 5 hr . Beads were washed 3 times with 1x lysis buffer , denaturated in 1x sample buffer at 99°C for 10 min and analyzed via SDS-PAGE/immunoblot . For immunoblot analysis , nitrocellulose membranes were incubated overnight at 4°C with anti-Itpkb ( #AP8167b , Abgent ) or anti-PLCγ1 ( #2822 , Cell Signaling Technology ) antibodies and then for 45 min with anti-rabbit-HRP secondary antibodies ( Bio-Rad Laboratories ) in TBS . Bound antibodies were detected by enhanced chemiluminescence ( ECL kit , GE Healthcare ) . The kinetics of DN thymocyte differentiation were modeled by a set of linear ordinary differential equations ( ODE ) . In these , the rate constants for DN3 cell generation and for thymocyte subset turnover were similar between genotypes . The rate constants for DN3-to-DP cell differentiation were increased over two2--fold for Itpkb-/- cells . The ODE were solved by pen and paper calculations and results verified using BIONETGEN . Aggregated results are shown as mean ± SEM . p values for the indicated comparisons were calculated by two-tailed unpaired Student's t-test . Group sizes are described in the figure legends . Significant p values are denoted by asterisks: *p<0 . 05; **p<0 . 01; ***p<0 . 001; ****p<0 . 0001 . All statistical analyses were performed in Prism . | T cells defend our body against cancer and infectious agents such as viruses . However , they can also cause rheumatoid arthritis and other autoimmune diseases by attacking healthy tissue . T cells recognize target cells via receptor proteins on their surface . To maximize the variety of infections and cancers our immune system can recognize , we generate millions of T cells with different T cell receptors every day . To ensure T cells work correctly , T cell receptors are tested at various checkpoints . The first checkpoint involves a process called beta ( β ) selection , during which T cells produce their first T cell receptor – the so-called pre-T cell receptor . This receptor causes T cells to divide and mature , and sets their future identity or “fate” . To complete β-selection , T cells must also receive signals from another surface receptor – one that belongs to the Notch family , which determines cell fate in many different tissues . The Notch receptor and the pre-T cell receptor both activate an enzyme called PI3K – a key mediator of β-selection . But the pre-T cell receptor also activates another enzyme called Itpkb that is required for T cell development . Westernberg , Conche et al . have now investigated how these different proteins and signaling processes work and interact during β-selection , using mice that lack several immune genes , including the gene that produces Itpkb . The results of the experiments show that during β-selection , Itpkb limits the ability of PI3K to activate some of its key target proteins . This “dampened” PI3K signaling ensures that both the pre-T cell receptor and the Notch receptor must be activated to trigger T cell maturation . Without Itpkb , β-selection can occur in the absence of Notch signaling . As Notch signaling is important for determining the fate of many different cell types , Westernberg , Conche et al . ’s findings raise the possibility that Itpkb might also regulate cell fate determination in other tissues . Moreover , Itpkb may suppress tumor development , because excessive PI3K signaling drives many cancers . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"immunology",
"and",
"inflammation"
] | 2016 | Non-canonical antagonism of PI3K by the kinase Itpkb delays thymocyte β-selection and renders it Notch-dependent |
The steps from HIV-1 cytoplasmic entry until integration of the reverse transcribed genome are currently enigmatic . They occur in ill-defined reverse-transcription- and pre-integration-complexes ( RTC , PIC ) with various host and viral proteins implicated . In this study , we report quantitative detection of functional RTC/PIC by labeling nascent DNA combined with detection of viral integrase . We show that the viral CA ( capsid ) protein remains associated with cytoplasmic RTC/PIC but is lost on nuclear PIC in a HeLa-derived cell line . In contrast , nuclear PIC were almost always CA-positive in primary human macrophages , indicating nuclear import of capsids or capsid-like structures . We further show that the CA-targeted inhibitor PF74 exhibits a bimodal mechanism , blocking RTC/PIC association with the host factor CPSF6 and nuclear entry at low , and abrogating reverse transcription at high concentrations . The newly developed system is ideally suited for studying retroviral post-entry events and the roles of host factors including DNA sensors and signaling molecules .
The early phase of human immunodeficiency virus ( HIV-1 ) replication involves reverse transcription of the viral RNA genome in the cytoplasm followed by nuclear entry of the linear double-stranded viral cDNA and integration into a host cell chromosome . Virus entry by membrane fusion releases the viral capsid containing genomic RNA and viral structural and replication proteins into the cytoplasm . While much is known about the biochemistry and inhibition of reverse transcription ( Hu and Hughes , 2012 ) , the processes of uncoating , genome replication , and nuclear trafficking are currently not well understood in infected cells . Several viral and host factors have been implicated in these steps , but their precise contributions are largely unknown . Different models have been proposed for genome uncoating , which is defined as the loss of the viral capsid structure that consists of homomultimers of CA ( capsid ) proteins: ( i ) immediate dissociation after fusion followed by reverse transcription within the remaining nucleoprotein complex; ( ii ) gradual dissociation during reverse transcription regulated by sequential contact with host factors; ( iii ) reverse transcription in an intact or largely intact capsid structure that only dissociates at the nuclear pore complex ( reviewed in Arhel ( 2010 ) ) . Several recent studies indicated that mutations in CA as well as small molecule compounds targeting CA can affect reverse transcription and nuclear import ( reviewed in Matreyek and Engelman ( 2013 ) ) , disfavoring the model of immediate capsid dissociation . Furthermore , various host factors interacting with CA have been implicated in the early phase of HIV-1 replication , again arguing for a functional contribution of CA at this stage ( reviewed in Matreyek and Engelman ( 2013 ) ; Ambrose and Aiken ( 2014 ) ; Hilditch and Towers ( 2014 ) ) . Reverse transcription in the shielded capsid environment could prevent HIV-1 from recognition by cytoplasmic DNA sensors with subsequent capsid uncoating at the nuclear pore considered to be required for nuclear entry ( Matreyek and Engelman , 2013 ) . Productive retroviral reverse transcription complexes ( RTC ) and pre-integration ( PIC ) complexes have been defined biochemically as high molecular weight complexes that after isolation from cells support endogenous reverse transcription ( Fassati and Goff , 1999 ) and integrate viral cDNA into heterologous target DNA ( Brown et al . , 1987 ) , respectively . However , the lack of methods for robust and reliable detection so far prevented direct determination of the association of functional RTC/PIC with specific viral or cellular proteins at different post-entry stages . The majority of incoming HIV-1 particles fails to transform into productive replication complexes in most commonly used cell lines ( Thomas et al . , 2007b ) . Accordingly , detection of viral constituents alone does not allow identifying functional structures over the large background of non-functional complexes . Furthermore , HIV-1 replication complexes are inherently transient and unstable complicating biochemical analysis , and different constituents were reported in different biochemical studies ( Farnet and Haseltine , 1991; Bukrinsky et al . , 1993; Miller et al . , 1997; Fassati and Goff , 2001 ) . Identifying functional HIV-1 RTC/PIC by imaging techniques would allow in situ investigation but requires the detection of nascent HIV-1 cDNA in association with viral constituents . A pioneering study applied microinjection of fluorescent nucleotides and subsequent HIV-1 infection , detecting microtubule association of a replication complex by correlative light and negative stain electron microscopy ( McDonald et al . , 2002 ) . However , no further information on the RTC could be derived in this study , and the method did not permit reliable identification of larger numbers of RTC/PIC . Accordingly , no follow-up studies applying this technique have been published . Specific detection of HIV-1 cDNA could be achieved by fluorescence in situ hybridization , and this method has been used to visualize integrated HIV-1 cDNA ( Lusic et al . , 2013 ) . Processing steps during FISH interfere with immunostaining ( Solovei and Cremer , 2010 ) , however , making detection of productive HIV-1 RTC/PIC difficult . Alternatively , viral cDNA may be detected by non-specific labeling . Classically , DNA synthesis has been visualized by bromo-deoxyuridine incorporation ( Gratzner et al . , 1975 ) , which again requires denaturation . More recently , labeling nascent DNA with a nucleoside analogue carrying an alkyne group for bio-orthogonal ‘click-labeling’ with a fluorophore carrying an azide group has been introduced ( Salic and Mitchison , 2008 ) . Click-labeling can be directly performed without denaturation or other processing steps . This novel approach has recently been used successfully for the investigation of SV40 disassembly using pre-labeled particles ( Kuksin and Norkin , 2012 ) and for the detection of replication factories of several DNA viruses ( Strang et al . , 2012; Wang et al . , 2013 ) . Many viral genomes are rapidly produced in a confined space in these cases , however . In contrast , HIV-1 reverse transcription produces only a single cDNA copy from the ∼10 kb RNA genome , thus requiring a high sensitivity of detection . Here , we describe a robust strategy for detecting HIV-1 RTC/PIC using click-labeling of 5-ethynyl-2′-deoxyuridine ( EdU ) incorporated into nascent viral DNA . Co-localization of a cytoplasmic EdU signal with GFP-tagged HIV-1 integrase ( IN ) could be used to reliably detect HIV-1 RTC/PIC in a reporter cell line and primary human macrophages . Specific immunostaining revealed the presence of the HIV-1 CA and NC ( nucleocapsid ) proteins on almost all cytoplasmic RTC/PIC , while CA was absent from nuclear PIC in a HeLa-derived reporter cell line . In contrast , almost all nuclear PIC were positive for CA in primary human macrophages , while CA detection on cytoplasmic RTC/PIC was more variable in this case . EdU labeling further allowed determination of a dual mode of action for the recently described HIV-1 inhibitor PF-3450074 ( PF74; [Shi et al . , 2011] ) targeting the CPSF6 binding interface of CA . The new method for identification of functional retroviral replication complexes will be widely applicable for studying the RTC/PIC pathway and the roles of host factors , DNA sensors and signaling molecules , as well as association with the nuclear import machinery .
Infectious HIV-1 particles carrying IN . eGFP were produced by co-transfection of HEK 293T cells with the respective plasmids . The majority of virions contained microscopically detectable IN . eGFP ( Figure1—figure supplement 1 ) . IN . eGFP was incorporated in similar amounts as virus-encoded IN ( Figure1—figure supplement 2 ) , and incorporation of IN . eGFP reduced infectivity only weakly ( Figure 1—figure supplement 3A ) . Detection of ( potentially weak ) cytoplasmic DNA signals may be obscured by very strong labeling of replicating nuclear DNA . Indeed , initial EdU labeling experiments revealed bright nuclear fluorescence that prevented analysis of fluorescent signals in the extranuclear region of the same and neighboring cells ( Figure 1—figure supplement 4 ) . This problem was overcome by infecting cells in the presence of the DNA polymerase α/δ inhibitor aphidicolin ( APC ) . Cell cycle arrest by APC is compatible with HIV-1 replication , and APC caused only a modest reduction of infectivity ( Figure 1—figure supplement 3B ) since HIV-1 does not require nuclear envelope breakdown for genome integration ( Suzuki and Craigie , 2007 ) . Infecting cells at APC concentrations that did not exhibit toxic effects during the incubation period led to a strong reduction of the nuclear EdU signal , now allowing detection of cytoplasmic EdU signals ( Figure 1—figure supplement 4 ) . The remaining nuclear EdU signal reflects residual nuclear DNA replication at this APC concentration . HeLa-based reporter cells ( TZM-bl ) were pre-incubated with HIV-1 ( IN . eGFP ) at a high multiplicity of infection ( m . o . i . = 25 ) for 30 min at 16°C to synchronize the infection . Cells were subsequently shifted to 37°C and infection was performed in the presence of EdU for 4 hr , followed by fixation and click-labeling using Alexa Fluor 647 ( Figure 1A ) . Immunostaining of cytochrome C ( CC ) was carried out in addition to mark mitochondria . Spinning disc confocal microscopy ( SDCM ) revealed large numbers of intracellular eGFP-labeled particles , lacking EdU or CC labeling ( Figure 1A , large panels , green signal ) . Most of these objects probably represent virions taken up by non-specific heparan sulfate-mediated endocytosis known to occur efficiently in HeLa-derived cells ( Lampe et al . , 2007 ) . Importantly , some eGFP-labeled particles co-localized with distinct , punctate EdU signals in the extranuclear region without overlapping mitochondria labeling ( Figure 1A , insets and enlargements i–iii; compare Figure 1—figure supplement 5 for staining of mitochondrial DNA synthesis ) . Co-localization was defined as objects in which Alexa Fluor 647- and eGFP signals overlapped in the major part of the pixel area in several adjacent frames of the confocal z-stack . 10 . 7554/eLife . 04114 . 003Figure 1 . HIV-1 RTC/PIC detection in TZM-bl cells . ( A ) Identification of candidate RTC/PIC based on co-localization of EdU with IN . eGFP . TZM-bl cells were infected with HIV-1 ( IN . eGFP ) for 4 hr in the presence of APC and EdU , followed by fixation and click-labeling . Mitochondria were detected by cytochrome C immunostaining ( CC; white ) . z-stacks covering the whole-cell volume were acquired by SDCM and analyzed for distinct co-localizing EdU-AlexaFluor647 ( red ) and IN . eGFP ( green ) signals . Left panels show z-sections through representative cells . Boxed regions ( i–iii ) mark three co-localizing objects from each cell displayed as enlargements on the right . Images from the optimal focal plane of each object are shown in the enlargements . Scale bars: 5 μm ( overviews ) , 1 μm ( enlargements ) . ( B ) Super-resolution microscopy confirming co-localization between EdU-AlexaFluor647 ( red ) and IN . mEos3 . 2 ( green ) . TZM-bl cells were infected with HIV-1 ( IN . mEos3 . 2 ) for 4 hr before click-labeling and analyzed by dual-color PALM/dSTORM . The overview shows an image recorded in TIRF mode; boxed regions i–ii are enlarged below . Enlargements iii–iv show additional examples of co-localization recorded by PALM/dSTORM in epifluorescence mode . Scale bars: 1 μm ( overview ) , 100 nm ( enlargements ) . ( C ) Inhibition of reverse transcriptase prevents RTC/PIC formation . TZM-bl cells were infected with HIV-1 ( IN . eGFP ) for 4 . 5 hr in the presence of DMSO or 5 μM EFV , respectively , followed by click-labeling and immunostaining as in ( A ) . z-stacks covering the whole-cell volume were acquired for randomly selected cells and the number of RTC/PIC per cell was determined . The graph shows pooled results from three independent experiments . Statistical significance was assessed using the Mann–Whitney test . See Figure 1—figure supplement 1–3 for characterization of labeled virus particles and effect of APC on infectivity , Figure 1—figure supplement 4 for reduction of nuclear EdU signal with APC treatment , Figure 1—figure supplement 5 for detection of mitochondrial DNA in the cytoplasm , Figure 1—figure supplement 6 for analysis of detection sensitivity , and Figure 1—figure supplement 7 for summary of number of RTC/PIC detected per cell . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 00310 . 7554/eLife . 04114 . 004Figure 1—figure supplement 1 . Analysis of labeled virus particles by immunofluorescence . HIV-1 ( IN . eGFP ) particles were produced as described in ‘Materials and methods’ . Particles were adhered to glass cover slips , fixed and immunostained using antiserum against HIV-1 CA . The figure shows a representative confocal image recorded in the red ( CA ) and green ( IN . eGFP ) channel , respectively . Scale bar: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 00410 . 7554/eLife . 04114 . 005Figure 1—figure supplement 2 . Analysis of labeled virus particles by immunoblot . Pelleted HIV-1 and HIV-1 ( IN . eGFP ) virus particles were separated by SDS-PAGE and analyzed by immunoblot using antisera against CA or IN , respectively . Positions of molecular mass standards ( in kDa ) are indicated on the left , HIV-1 specific proteins are marked on the right . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 00510 . 7554/eLife . 04114 . 006Figure 1—figure supplement 3 . Quantification of relative infectivity of labeled virions . ( A ) Comparison of wt virus and particles carrying exogenous IN . eGFP . TZM-bl cells were infected with serial dilutions of HIV-1 and HIV-1 ( IN . eGFP ) . At 48 hr p . i . , cell lysates were harvested and infection was quantified by luciferase assay . Values were normalized to the amount of virus measured by p24 ELISA . The graph shows mean values and SD from three independent experiments . RLU , relative light units . ( B ) Effect of APC treatment on HIV-1 infection . TZM-bl cells pre-incubated overnight with DMEM containing DMSO or 6 µM APC , respectively , were infected with HIV-1 or HIV-1 ( IN . eGFP ) with an MOI of 0 . 1 for 4 hr in the presence of DMSO or 6 µM APC . The medium was subsequently changed to DMEM and incubation was continued . At 48 hr p . i . , the percentage of infected cells was quantified by immunostaining for HIV-1 CA followed by flow cytometry . The graph shows mean values and SD from two independent experiments performed in duplicate . Black bars: DMSO treated cells; white bars: APC treated cells . p-values were calculated using a Student's t test . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 00610 . 7554/eLife . 04114 . 007Figure 1—figure supplement 4 . Reduction of nuclear EdU signal by APC treatment . TZM-bl cells were pre-incubated in medium without ( A ) or with ( B ) 6 µM APC over night . Cells were then infected with HIV-1 ( IN . eGFP ) in the presence of 6 μM APC and 10 µM EdU . At 5 hr p . i . , cells were fixed , click-labeled , and analyzed . Images show z-sections through the nuclear region of cells recorded using identical microscope settings . Boxed areas i–iii in ( B ) show punctate EdU-Alexa-647 signals detected in the cytoplasm; these regions are shown enlarged at the right . Scale bars: 5 µm ( overviews ) , 1 µm ( enlargements ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 00710 . 7554/eLife . 04114 . 008Figure 1—figure supplement 5 . Detection of mitochondrial DNA synthesis in the cytoplasm . TZM-bl cells were pre-incubated in medium containing 6 µM APC overnight and then infected with HIV-1 ( IN . eGFP ) in the presence of 10 µM EdU and 6 µM APC . At 4 hr p . i . , cells were fixed , click-labeled ( red ) , immunostained with antiserum against cytochrome C ( white ) , and imaged . The upper panel shows a representative z-section through the nuclear region . EdU signals overlapping with CC-stained areas ( i ) may result from mitochondrial DNA synthesis , whereas distinct IN . eGFP-EdU co-localizing objects overlapping with mitochondrial staining ( ii ) were classified as RTC/PIC . Scale bars: 5 µm ( overview ) , 1 µm ( enlargements ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 00810 . 7554/eLife . 04114 . 009Figure 1—figure supplement 6 . Sensitivity of EdU-Alexa Fluor647 detection . ( A ) Sensitivity of the microscopy setup . 10-fold serial dilutions were prepared starting from a solution of 3 . 5 mM Alexa-647 azide ( Click-iT EdU-Alexa Fluor 647 Imaging Kit; Life technologies ) in PBS . Solutions were imaged with the SDCM setup used throughout this study directly upon distribution into LabTek chamber slide wells . The graph shows intensity values recorded for the different dye concentrations; mean values and SD from three independent dilution experiments are displayed . An intensity value of 65 , 535 reflects camera saturation . Intensity clearly above background was detected for a 35 nM solution ( dotted line ) , corresponding to ∼2 dye molecules per confocal volume ( 250 nm × 250 nm × 900 nm , x–y–z ) . ( B ) Sensitivity of RT product detection . The figure shows exemplary RTC/PIC from TZM-bl cells transduced with retroviral vector particles produced as outlined in ‘Materials and methods’ ( i and ii ) or infected with HIV-1 ( IN . eGFP ) ( iii ) . Vector particles packaged RNA derived from plasmid pRRL . PPT . SF . GFPpre ( i ) or plasmid pWPI ( ii ) , respectively . At 4 . 5 hr after particle addition , cells were fixed and click-labeled as in Figure 1A . Samples were immunostained using antiserum raised against HIV-1 CA ( blue ) . CA staining in ( i ) was weaker than in the other two cases; the image has been enhanced for comparable visibility . The approximate length of the respective full-length cDNA , as well as the approximate number of T residues in the complete cDNA is indicated at the right . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 00910 . 7554/eLife . 04114 . 010Figure 1—figure supplement 7 . Numbers of cytoplasmic RTC/PIC detected per cell . Infection of TZM-bl cells with HIV-1 ( IN . eGFP ) , click-labeling , immunostaining with antibody against cytochrome C , and single cell imaging were performed as described in Figure 1 . The binned histogram summarizes pooled data from 69 individual cells from 3 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 010 To validate the co-localization of labeled IN with newly synthesized EdU-labeled DNA in subviral complexes , we performed super-resolution microscopy . Cells were infected with virions carrying a photo-switchable IN . mEos3 . 2 fusion protein ( Zhang et al . , 2012 ) , suitable for detection with photoactivated localization microscopy ( PALM ) ( Betzig et al . , 2006 ) and click-labeled with EdU . Dual-color PALM/dSTORM ( Heilemann et al . , 2008 ) yielded a lateral resolution of 46 nm and 32 nm for mEos3 . 2 and Alexa Fluor 647 , respectively . In order to ensure high resolution in all three dimensions , samples were imaged in total internal reflection mode . This method limits detection to the close proximity ( ∼200 nm ) of the ventral plasma membrane , where only a small proportion of all RTC/PIC present in the cell is expected to be located . Nevertheless , in two independent experiments we detected 13 clear EdU/IN co-localizing objects that displayed partial overlap of the two signals at super-resolution precision ( Figure 1B , i , ii ) . Such objects were also observed deeper in the cytoplasm when employing PALM/dSTORM in epifluorescence mode ( Figure 1B , iii , iv ) . To determine whether IN . eGFP/EdU co-localization is suitable for identifying productive HIV-1 RTC/PIC , we performed experiments in the presence of the HIV-1 reverse transcriptase inhibitor efavirenz ( EFV ) . Approximately 68% of cells infected in the absence of EFV displayed IN . eGFP/EdU-co-localizing objects at 4 . 5 hr p . i . , with up to 30 objects in individual cells and an average of five co-localizing objects per cell ( Figure 1C ) . Control experiments verified that individual viral cDNA molecules and also partially reverse transcribed intermediates should be easily detectable in our setup ( Figure 1—figure supplement 6 ) . The wide variation in the number of co-localizing eGFP- and EdU-positive objects per cell ( Figure 1—figure supplement 7 ) , observed even within one experiment , reflects the stochasticity of the infection process . Infection in the presence of EFV did not affect the uptake of IN . eGFP particles but dramatically reduced the number of IN . eGFP/EdU co-localizing objects to 0 . 2 per cell ( Figure 1C ) . We therefore defined IN . eGFP/EdU co-localization as the signature of productive HIV-1 replication complexes . Since the transition from RTC to PIC is defined biochemically and our imaging analyses do not allow discriminating between these complexes in the cytoplasm , we designated IN . eGFP/EdU-positive subviral particles in the cytoplasm of infected cells as RTC/PIC . We subsequently determined the time course of RTC/PIC formation . RTC/PIC were detected in almost half of the cells at 2 hr after infection , with the majority of positive cells containing a single object ( Figure 2 ) . The proportion of RTC/PIC containing cells , as well as the average number of complexes per cell , increased significantly when the infection period was extended to 3 hr . Infection for 4 hr led to a further increase , while no significant additional RTC/PIC accumulation was detected at later time points ( Figure 2 ) . Based on this result , an incubation time of 4–5 hr was chosen for all further experiments in HeLa-derived cells . 10 . 7554/eLife . 04114 . 011Figure 2 . Time course of RTC/PIC detection . TZM-bl cells were infected with HIV-1 ( IN . eGFP ) as in Figure 1 . Cells were fixed at the indicated time points after infection and analyzed as in Figure 1C . Numbers of RTC/PIC detected per cell were counted for randomly selected cells in each incubation interval . The graph shows pooled results from two independent experiments . Mean values of RTC/PIC per cell are indicated by black lines . Statistical significance was assessed using the Mann–Whitney test . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 011 We subsequently analyzed whether the main HIV-1 structural proteins MA ( matrix ) , CA , or NC remain associated with productive RTC/PIC in the early phase of HIV-1 replication . Cells were infected with HIV-1 ( IN . eGFP ) for 4 hr in the presence of EdU followed by click-labeling and immunostaining ( Figure 3A ) . Numerous intracellular particles exhibiting co-localization of IN . eGFP with the viral structural proteins , but lacking an EdU signal , were observed in all cases ( Figure 3A ) . Analysis of >100 RTC/PIC per condition revealed co-localization with easily detectable NC and CA signals in ∼95% of both cases ( Figure 3B ) . In contrast , MA co-localization was only detected in 6% of the RTC/PIC analyzed ( Figure 3B ) , while the MA signal was easily detectable on most intracellular IN . eGFP-positive objects lacking the EdU signal ( Figure 3A ) . Intensity of CA and NC staining on RTC/PIC was variable . Quantitative comparison of the CA signals on >100 RTC/PIC with an equal number of EdU-negative , but IN . eGFP and CA-positive objects ( many of which are likely to correspond to complete particles in endosomes ) revealed a slightly higher average intensity on RTC/PIC ( Figure 3C ) . This may be due to different exposure of conformational epitopes that can be recognized by our polyclonal antiserum . IN . eGFP signals were almost equal in both cases . 10 . 7554/eLife . 04114 . 012Figure 3 . Association of HIV-1 Gag derived proteins with RTC/PIC . ( A ) Immunofluorescence analysis . TZM-bl cells were infected with HIV-1 ( IN . eGFP ) as in Figure 1 . At 4 hr p . i . samples were fixed , click-labeled , and immunostained using antisera against HIV-1 MA ( i ) , CA ( ii ) , or NC ( iii ) , respectively . z-stacks covering the whole-cell volume were acquired by SDCM , RTC/PIC were identified as in Figure 1 and co-localization with each Gag derived protein was analyzed . The panel shows two representative RTC/PIC per antiserum used . Scale bar: 1 μm . ( B ) Quantitative analysis of co-localization with viral proteins . For each staining condition , n > 100 individual randomly selected RTC/PIC , identified as in ( A ) , were analyzed for detection of the respective Gag derived protein . 6% , 97% , and 94% of RTC/PIC were positive for MA , CA , or NC , respectively . ( C ) Quantification of relative signal intensity . A total of 114 intracellular CA-IN . eGFP objects not containing detectable EdU signal ( 1 ) and 114 RTC/PIC ( 2 ) , respectively , were identified in cells from three independent experiments . Signal intensity of CA ( circles ) and IN . eGFP ( triangles ) was calculated for each object as described in ‘Materials and methods’ . Within each experiment , values were normalized to the mean intensity of CA or IN . eGFP signal , respectively , that was obtained for the CA-IN . eGFP objects lacking EdU . Each data point in the graph represents a single object; mean intensity values are indicated by red lines . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 012 The observation that almost all cytoplasmic RTC/PIC co-localized with a clearly detectable CA signal indicated that reverse transcription occurs within the incoming capsid or a CA-containing structure in this cell type . To validate this co-localization , we performed three-color PALM/dSTORM super-resolution microscopy detecting ( i ) IN . mEos3 . 2 , ( ii ) EdU-Alexa Fluor 647 , and ( iii ) CA , using antibodies labeled with Alexa Fluor 532 . This approach yielded a localization precision of 20 nm for the Alexa Fluor 532 label . Co-localization of the CA-specific signal with RTC/PIC could indeed be validated at super-resolution ( Figure 4A ) , with 16 co-localizing objects observed in eight different cells . We measured an average diameter of ∼155 nm for the CA signal and ∼133 nm for the mEos3 . 2 signal with modest variation and no clear outliers ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 04114 . 013Figure 4 . Co-localization of HIV-1 CA with RTC/PIC at different intracellular localizations . ( A ) Three-color PALM/dSTORM analysis of RTC/PIC . TZM-bl cells were infected with HIV-1 ( IN . mEos3 . 2 ) as in Figure 1 . At 4 hr p . i . , cells were fixed , click-labeled , and immunostained using antiserum against HIV-1 CA . Three-color PALM/dSTORM images were acquired in TIRF mode as described in ‘Materials and methods’ . The panel shows an example of an EdU-negative IN . mEos3 . 2/CA co-localizing object that possibly represents an intracellular virion ( i ) , as well as three examples of triple-labeled complexes classified as RTC/PIC ( ii–iv ) . Scale bar: 100 nm . ( B ) RTC/PIC detected in the cytoplasmic and perinuclear regions . TZM-bl cells were infected with HIV-1 ( IN . eGFP ) as in Figure 1 . At 5 hr p . i . , samples were fixed , click-labeled , and immunostained using an antibody against nuclear pore proteins ( white ) and antiserum against HIV-1 CA ( blue ) . The top panel shows a z-section through an individual cell displaying RTC/PIC in the peripheral cytoplasmic region ( i ) , close to the nucleus ( ii ) , and overlapping or in close proximity of the nuclear membrane ( iii ) . The lower panels show enlargements of the insets with images selected from the optimal focal plane of each RTC/PIC . Scale bars: 5 μm ( overview ) , 1 μm ( enlargements ) . ( C ) PIC detected inside the nucleus . The left panel shows a z-section through two individual cells , selected for very low levels of nuclear EdU labeling . Immunostaining was performed as in ( B ) . The boxed area shows an IN . eGFP/EdU co-localizing object within the nucleus . Enlargements of this object in the individual channels are shown on the right . Scale bars: 5 μm ( overview ) , 1 μm ( enlargements ) . See Figure 4—figure supplement 1 for analysis of cluster sizes of RTC/PIC detected in three-color PALM/dSTORM analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 01310 . 7554/eLife . 04114 . 014Figure 4—figure supplement 1 . Size distribution of RTC/PIC associated CA and IN . mEos clusters derived from PALM/dSTORM analyses . Diameters of CA and IN . mEos clusters were measured using images from 3-color PALM/dSTORM analysis . A total of 16 clusters detected in eight individual cells were analyzed . The graph shows mean values and SD . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 014 The subcellular localization of HIV-1 reverse transcription is not fully established due to a lack of experimental approaches detecting productive RTC . Our analysis revealed IN/EdU-positive objects in the cell periphery and deeper in the cytoplasm ( Figure 1A , B ) . To determine whether RTC/PIC in the cell periphery and in the perinuclear region exhibit differences in CA association , we performed SDCM using an antibody against nucleoporins to mark the nuclear envelope . RTC/PICs in the periphery ( Figure 4B , i ) , close to the nucleus ( Figure 4B , ii ) or co-localizing with nuclear pore staining ( Figure 4B , iii ) all contained CA . Nuclear HIV-1 specific EdU-positive particles ( corresponding to PIC ) could not be unambiguously identified in most cells , since EdU signals from residual chromosomal DNA replication prevented their detection , even under APC treatment ( Figure 1—figure supplement 4 ) . Occasional cells did not show EdU labeling of chromosomal DNA , however ( presumably due to the absence of cellular DNA synthesis during the incubation period ) . IN . eGFP/EdU-positive nuclear PIC could be detected in these cells , and the nuclear PIC exhibited no detectable CA staining in almost all ( 23/24 ) cases ( Figure 4C ) . Thus , cytoplasmic RTC/PIC are clearly associated with CA independent of their subcellular localization , while nuclear PIC have lost most or all of their CA content , at least in this HeLa-derived cell line . The small molecule inhibitor PF74 has been reported to abrogate HIV-1 replication prior to the stage of cDNA synthesis ( Blair et al . , 2010; Shi et al . , 2011 ) . PF74 binds a groove in the N-terminal domain of HIV-1 CA and has been proposed to induce premature uncoating with consequent loss of reverse transcription ( Shi et al . , 2011; Matreyek et al . , 2013 ) . Furthermore , PF74 has been reported to interfere with association of the RTC with the host proteins CPSF6 and Nup153 ( Matreyek et al . , 2013 ) . We confirmed that PF74 completely blocks HIV-1 infection at a concentration of 10 µM ( Figure 5A ) as previously shown ( Shi et al . , 2011 ) . Inhibition affected a stage preceding cDNA synthesis: detection of RTC/PIC was dramatically reduced in cells infected for 4 . 5 hr in the presence of 10 µM PF74 ( Figure 5B ) , and this result was confirmed by quantitative determination of HIV-1 specific cDNA using PCR ( Figure 5C ) . The observation that the remaining RTC/PIC observed at this PF74 concentration were significantly less frequently positive for CA ( 65% vs 97% in the control sample; p < 0 . 01 ) is consistent with the proposed effect of PF74 on capsid uncoating . 10 . 7554/eLife . 04114 . 015Figure 5 . Effect of PF74 treatment on RTC/PIC formation and infection of host cells . ( A ) Inhibition of productive infection . TZM-bl cells were infected with serial dilutions of HIV-1 produced in the presence of DMSO or the indicated concentrations of PF74 , respectively . Cell lysates were harvested at 48 hr p . i . , and infectivity was determined by measuring luciferase reporter gene activity . The graph shows mean values and standard deviation ( SD ) from three replicate infections . ( B ) Inhibition of RTC/PIC formation . TZM-bl cells were infected with HIV-1 ( IN . eGFP ) in the presence of DMSO or two different concentrations of PF74 , respectively . At 4 . 5 hr p . i . , cells were fixed , click labeled , immunostained with antiserum against HIV-1 CA , and analyzed . Numbers of RTC/PIC per cell were determined for randomly selected cells from z-stacks covering the whole-cell volume , and the proportion of complexes co-localizing with CA immunostaining was determined . The figure shows pooled results from three independent experiments . Statistical significance was assessed using the Mann–Whitney test . ( C ) Quantification of HIV-1 DNA . MT-2 cells were infected with HIV-1 ( IIIB ) in the presence of DMSO or of the indicated concentrations of the inhibitors EFV , EVG , or PF74 , respectively . Late RT products ( top ) and 2-LTR circles ( bottom ) in cell lysates were quantified by qPCR as described in ‘Materials and methods’ using appropriate primer sets . The graph shows mean values and SD from three replicate infections . Statistical significance of the difference between each treatment and the DMSO control ( p-values determined by Student's t test ) is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 015 Titration of PF74 had revealed an EC50 for inhibition of HIV-1 infection of ∼0 . 5 µM in MT-2 cells ( Blair et al . , 2010 ) . In HeLa-derived cell lines , 80–90% reduction of infectivity was observed at a concentration of 1 or 2 µM PF74 , respectively ( Shi et al . , 2011 ) . 2 µM PF74 reduced HIV-1 infectivity >95% in our experimental system , even at the high multiplicity of infection used . Unexpectedly , however , RTC/PIC formation was completely unaffected at this PF74 concentration , despite almost complete loss of infectivity . EdU-positive RTC/PIC were readily detected in cells infected with HIV-1 for 4 . 5 hr in the presence of 2 µM PF74 with no apparent difference compared to control infections ( 7 . 7 vs 6 . 8 RTC/PIC per cell on average , respectively; Figure 5B ) . Furthermore , infection in the presence of 2 µM PF74 did not affect association of RTC/PIC with CA: 98% and 97% of RTC/PIC exhibited a positive CA signal when cells were infected in the presence or absence of 2 µM PF74 , respectively ( Figure 5B ) . The result that reverse transcription was unaffected by ∼2 µM PF74 was confirmed by quantitative PCR analysis . No significant difference was observed for late reverse transcription products when cells were infected with HIV-1 in the presence or absence of 2 . 7 µM PF74 , while these products were lost upon infection in the presence of EFV or at a higher PF74 concentration , respectively ( Figure 5C ) . A different result was observed when HIV-1 specific 2-LTR circles , indicative of nuclear import of viral DNA , were quantified . Infection in the presence of 2 . 7 µM PF74 almost completely abolished the formation of 2-LTR circles despite normal synthesis of late reverse transcription products , indicating a specific defect in nuclear import of viral cDNA . As expected , EFV or 8 . 1 µM PF74 abolished 2-LTR circles as well , while incubation in the presence of the IN inhibitor elvitegravir ( EVG ) , which blocks chromosomal integration of the HIV-1 cDNA without affecting reverse transcription or nuclear import ( Shimura et al . , 2008 ) , yielded a strong increase of 2-LTR circles ( Figure 5C ) . Taken together these results indicate a dose-dependent bimodal effect of PF74 on early HIV-1 infection , targeting reverse transcription and PIC nuclear import , respectively . The nuclear import defect at low concentrations of PF74 may be caused by binding of this compound to a reactive groove on the viral capsid , leading to competitive inhibition of capsid interaction with the host proteins Nup153 and CPSF6 ( Matreyek et al . , 2013; Price et al . , 2014 ) . Both of these proteins have been implicated in HIV-1 PIC nuclear import ( Matreyek and Engelman , 2013 ) . We therefore investigated the association of RTC/PIC with CPSF6 in the presence and absence of PF74 . Immunostaining of CPSF6 revealed an almost exclusively nuclear localization of this protein with very weak cytoplasmic staining in TZM-bl cells ( Figure 6—figure supplement 1A ) , consistent with previously published reports ( De Iaco et al . , 2013; Fricke et al . , 2013 ) . Nevertheless , a clear CPSF6 signal co-localizing with the RTC/PIC was detected in 22% of all cases in TZM-bl cells ( Figure 6A; 92 RTC/PIC analyzed ) . Given the very weak cytoplasmic CPSF6 signal , we considered the possibility that this relatively low number of co-localizing structures may be due to insufficient sensitivity of our detection system . To overcome this obstacle , we made use of a HeLa-derived cell line with a stable knock-down of transportin-3 ( TNPO3 ) ( Thys et al . , 2011 ) . This protein functions as a nuclear import factor for CPSF6 , andTNPO3 knock-down has been shown to lead to cytoplasmic accumulation of CPSF6 ( De Iaco et al . , 2013 ) . Accordingly , an increased cytoplasmic CPSF6 signal was detected in the TNPO3 knock-down cell line but not in a control cell line expressing a scrambled shRNA ( Figure 6—figure supplement 1B , C ) . Consistent with our hypothesis , we observed that 87% of all RTC/PIC were positive for CPSF6 upon infection of the TNPO3 knock-down cell line ( Figure 6B; 87 RTC/PIC analyzed ) . Treatment of TNPO3 knock-down cells with 2 µM PF74 during infection strongly reduced the level of CPSF6 association with RTC/PIC to 18% ( Figure 6C; 74 RTC/PIC analyzed ) . These results provide direct evidence for the association of cytoplasmic CPSF6 with the incoming viral capsid and for competitive inhibition of this interaction by the small molecule inhibitor PF74 . 10 . 7554/eLife . 04114 . 016Figure 6 . Co-localization of CPSF6 with RTC/PIC in the cytoplasm . The figure shows co-localization of CPSF6 with RTC/PIC in the cytoplasm of TZM-bl ( A ) and TNPO3KD ( B and C ) cells in the absence ( B ) or presence ( C ) of 2 µM PF74 . Cells were infected with HIV-1 ( IN . eGFP ) , and click-labeled as described in Figure 1A , followed by immunostaining against CPSF6 ( blue ) and SDCM analyses . The RTC/PIC in the boxed area ( i ) are shown enlarged in the top rows at the right; the middle and bottom rows at the right show examples of RTC/PIC from other cells . Arrows in enlargements indicate positions of RTC/PIC without detectable CPSF6 . To visualize weak signals for proper visualization , greyscale enlargement images were auto-contrasted in ImageJ . Scale bars: 5 μm ( overviews ) , 1 μm ( enlargements ) . See Figure 6—figure supplement 1 for localization of CPSF6 in different cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 01610 . 7554/eLife . 04114 . 017Figure 6—figure supplement 1 . CPSF6 localization in TZM-bl ( A ) , TNPO3Scr ( B ) , and TNPO3KD ( C ) cells . Cells were immunostained with antibody against CPSF6 and imaged by SDCM using the same microscope settings . Panels show representative z-sections through the middle region of cells . Scale bar: 5 µM . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 017 All imaging experiments described so far were performed in a HeLa-derived cell line , which is well suited for microscopic analyses and allows robust detection of RTC/PIC . These cells do not represent natural target cells of HIV-1 , however . We therefore extended our analysis to primary human monocyte-derived macrophages ( MDM ) . MDM were chosen because they are natural targets of HIV-1 infection with a morphology well suited for microscopy , and they are post-mitotic cells lacking nuclear DNA synthesis . The experiments with MDM were performed using HIV-1 carrying Env proteins with a tropism for the CCR5 co-receptor . IN . eGFP/EdU-positive RTC/PIC could be detected in HIV-1 infected MDM ( Figure 7 ) , albeit at a much lower frequency compared to TZM-bl cells . HIV-1 uptake into macrophages is more specific , but much less efficient than in HeLa-derived cells . Furthermore , the cellular restriction factor SAMHD1 inhibits HIV-1 replication in primary macrophages at the stage of reverse transcription ( Hrecka et al . , 2011 ) . Accordingly , the average number of RTC/PIC detected in macrophages observed at 24 hr or 48 hr after infection was 10- to 50-fold lower compared to HeLa-derived cells infected for 4 hr and showed significant donor dependent variation ( Figure 7—source data 1 ) . Analyzing a total of 567 MDM from eight different donors in five independent experiments , we observed a total of 135 RTC/PIC in the cytoplasm of these cells . HIV-1 CA was found to be associated with only 44% of these RTC/PIC ( Table 1 ) , while >97% of cytoplasmic RTC/PIC were CA-positive in the case of TZM-bl cells . The proportion of CA-positive RTC/PIC in MDM decreased over time with ∼50% CA-positive at 24 hr and ∼30% CA-positive at 48 hr ( Table 1 ) , indicating time-dependent loss of CA from cytoplasmic RTC/PIC . 10 . 7554/eLife . 04114 . 018Table 1 . Cytoplasmic and nuclear subviral complexes in MDMDOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 018CytoplasmNucleusProportion of nuclear complexes [%]Number of cells imagedRTC/PIC ( n ) CA-positive RTC/PIC ( n ) CA-positive RTC/PIC [%]PIC ( n ) CA-positive PIC ( n ) CA-positive PIC [%]24 hr90465199100936748 hr45132961599758200Total135594470689734567The table summarizes data from primary MDM infected with NL4-3-R5 ( IN . eGFP ) or NL4-3-4059 ( IN . eGFP ) for 24 hr or 48 hr , respectively . Infected MDM were immunostained with antibodies against nuclear pore complexes and CA . Data were obtained as outlined in Figure 7 in cells from eight different donors in five independent experiments ( cells from three donors were infected for both 24 hr and 48 hr ) . As post-mitotic cells , MDM lack nuclear DNA synthesis and therefore exhibited almost no nuclear EdU signal . Lack of cellular DNA synthesis in the nucleus greatly facilitated visualization of nuclear HIV-1 PIC in these cells ( Figure 7B , C ) , and we detected 70 nuclear PIC in the total of 567 MDM that were analyzed in this study . The percentage of nuclear PIC ( over all RTC/PIC detected ) increased from 9% at 24 hr to 58% at 48 hr ( Table 1 ) . Surprisingly , almost all nuclear PIC ( 68 of 70 ) exhibited a clearly detectable CA signal ( Figure 7B , C; Table 1 ) , suggesting that the intact capsid or a capsid-derived structure remains associated with the replication complex during and after nuclear import . 10 . 7554/eLife . 04114 . 019Figure 7 . Detection of RTC/PIC in primary macrophages . ( A ) Detection of RTC/PIC in MDM . Human MDM were prepared and infected with an R5-tropic variant of HIV-1 ( IN . eGFP ) ( NL4-3-R5 ( IN . eGFP ) ) for 24 hr in the presence of 5 μM EdU . Cells were fixed , click-labeled , and immunostained for CC and HIV-1 CA . The left panel shows a z-section through part of a representative cell , displaying immunostaining for CC ( white ) and HIV-1 CA ( blue ) together with EdU ( red ) and IN . eGFP ( green ) . Boundaries of the cell ( white line ) and nucleus ( dashed line ) are indicated . The RTC/PIC in the boxed area ( i ) , as well as two other exemplary RTC/PIC detected in cells from two other donors ( ii–iii ) are shown enlarged on the right . To visualize weak signals for proper visualization , greyscale enlargement images of example ( i ) were auto-contrasted in ImageJ . Scale bar: 5 μm ( overview ) , 1 μm ( enlargements ) . ( B and C ) CA signals associated with nuclear PIC . MDM were prepared and infected for 48 hr , followed by click-labeling , and immunostaining using an antibody against nuclear pore complexes ( white ) and antiserum against HIV-1 CA ( blue ) . ( B ) Channel overlay of a z-section through part of a representative cell , together with images of the individual channels ( EdU , red; IN . eGFP , green; CA , blue ) . Arrows indicate an example of a cytoplasmic RTC/PIC ( left ) and a nuclear PIC ( right ) , respectively . Scale bar: 5 μm . ( C ) Additional examples of nuclear PIC detected at 48 hr p . i . in MDM from two other donors . Scale bars: 5 μm . See Figure 7—source data 1 for a summary of infected MDM immunostained with CC and CA antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 01910 . 7554/eLife . 04114 . 020Figure 7—source data 1 . RTC/PIC detected in MDM . Human MDM were infected with NL4-3-R5 ( IN . eGFP ) for 24 hr , fixed , click-labeled , and stained with CC and CA antibodies . RTC/PIC detected were counted from z-stacks covering the whole-cell volume . The table summarizes data from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 04114 . 020
In this study , we describe a robust strategy for detecting and characterizing productive HIV-1 post-entry complexes by identifying objects with nascent DNA synthesis co-localizing with viral replication proteins in the cytoplasm or in the nucleus of infected cells . EdU click-labeling has no specificity for viral DNA , but the observation of IN . eGFP/EdU-positive objects in the extranuclear region , not overlapping with mitochondria and almost completely blocked by EFV , clearly identified functional HIV-1 RTC/PIC in proliferating reporter cells . Nuclear PIC were only detectable in cells that had not undergone DNA synthesis during the labeling period in several HeLa-derived cell lines , while lack of cellular DNA synthesis in post-mitotic macrophages allowed easier detection of nuclear PIC in this case . EdU-positive HIV-1 complexes mainly accumulated between 2 hr and 4 hr after infection of HeLa-derived cells , consistent with detection of reverse transcription products by quantitative PCR in this cell type at early time points ( Thomas et al . , 2011 ) . RTC/PIC were detected much later upon infection of primary MDM and accumulated over time; this slower formation is again consistent with previous PCR results ( Ambrose et al . , 2012 ) . Furthermore , nuclear PIC mainly accumulated between 24 hr and 48 hr in MDM , indicating the time frame of nuclear import in this cell type . In contrast to PCR-based detection of RT products or nuclear 2-LTR circles , which has to be performed on bulk extracts , the imaging method established here allows identification of individual RTC/PIC in the physiological environment of the infected cell . EdU is incorporated in the position of thymidine in the viral cDNA , and we can thus expect a maximum of 5588 EdU molecules for the entire double-stranded cDNA of HIV-1 ( NL4-3 ) . Obviously , the actual number is expected to be much lower as EdU competes with the pool of cellular dTTP . Our experimental system operates at almost single molecule sensitivity ( detection limit 1–2 dye molecules per confocal volume; Figure 1—figure supplement 6A ) , but RTC/PIC detection also depends on relative EdU incorporation and click-labeling efficiency . Using lentiviral vectors of different lengths , we could identify nascent cDNA for vector genomes of 5 . 4 kb and 3 . 7 kb ( Figure 1—figure supplement 6B ) , indicating that detection of short reverse transcription products , as suggested for abortive infection of resting primary T-cells ( Doitsh et al . , 2014 ) , may be possible with this method . Accumulation of RTC/PIC over several hours may reflect asynchronous cell entry and initiation of reverse transcription in combination with the relatively slow process of minus-strand cDNA synthesis in infected cells ( Thomas et al . , 2007a ) . The vast majority of cytoplasmic RTC/PIC in HeLa-derived cells were associated with the viral structural proteins CA and NC but lacked MA . As a membrane-associated protein , MA can be expected to separate from the incoming capsid upon virus–cell fusion , but some previous studies reported residual MA retained on HIV-1 RTC ( Bukrinsky et al . , 1993; Gallay et al . , 1995; Miller et al . , 1997; Iordanskiy et al . , 2006 ) . Clearly , lack of MA detection by immunofluorescence does not rule out the presence of some MA molecules on RTC/PIC , but the bulk of MA appears to be lost . NC is a nucleic-acid binding protein that coats the viral RNA genome , and NC mutations have been shown to affect HIV-1 integration ( Cimarelli et al . , 2000; Thomas et al . , 2011 ) . The association of NC with almost all RTC/PIC may thus not be surprising , but NC had not been detected in several studies where RTC and/or PIC were biochemically fractionated ( Farnet and Haseltine , 1991; Miller et al . , 1997 ) . Association of almost all cytoplasmic RTC/PIC with CA in HeLa-derived cells clearly indicates that reverse transcription occurs in the viral capsid or a capsid-derived structure in these cells . CA association of cytoplasmic RTC/PIC was independent of subcellular localization , arguing against a gradual loss of CA during trafficking to the nuclear pore . Consistent with this hypothesis , we observed a similar or slightly higher CA labeling intensity on RTC/PIC compared with EdU-negative intracellular HIV-1 particles . Several previous studies reported a time-dependent loss of CA signals on cytoplasmic structures positive for the HIV-1 accessory protein Vpr in HeLa- or HOS-derived cell lines infected with VSV-G pseudotyped VLPs ( McDonald et al . , 2002; Hulme et al . , 2011 ) , suggesting gradual capsid uncoating during reverse transcription . Alternatively , the reported loss of CA signal may have been due to the inaccessibility of the reactive epitope due to conformational changes and/or host protein binding after cytoplasmic entry , especially since monoclonal antibodies were used in these studies . An influence of the entry pathway may also be considered as these prior studies used VSV-G pseudotyped particles entering via the endosomal route . RTC/PIC co-localizing with nuclear pore proteins in TZM-bl cells also exhibited a strong CA signal , while CA was absent from all but one nuclear PIC indicating that CA is lost at the nuclear pore or shortly after nuclear entry in this cell type . This hypothesis is consistent with the described interaction of CA with nucleoporins 153 and 358 ( Schaller et al . , 2011; Matreyek et al . , 2013 ) and suggests that cDNA within the intact viral capsid or a capsid-derived structure docks to the nuclear pore complex followed by dissociation of CA and nuclear import . A very different picture was observed for CA association of RTC/PIC in primary human MDM . Less than half of the RTC/PIC showed a detectable CA signal in this case , while >97% of all RTC/PIC were CA-positive in HeLa-derived cells . This difference is unlikely to be caused by earlier CA uncoating of productive RTC/PIC in MDM , since almost all nuclear PIC ( 97% ) were CA-positive in this cell type . We even observed a CA-positive nuclear PIC and a CA-negative cytoplasmic RTC/PIC in the same cell ( Figure 7B ) . These observations suggest that the intact or remodeled viral capsid is imported through the intact nuclear pore in post-mitotic MDM following complete or partial reverse transcription . This is completely different from the results observed in HeLa-derived cells and is not predicted by any of the current models for early HIV-1 replication . The fact that nuclear PIC in MDM were clearly CA-positive , while the majority of cytoplasmic RTC/PIC were CA-negative in the same cells , and CA appeared to be lost over time suggests that many of the cytoplasmic RTC/PIC detected in MDM were not predecessors of nuclear PIC but may rather represent dead-end products . These ( at least partially ) uncoated cytoplasmic structures contain incomplete or complete HIV-1 cDNA , which can be exposed to cytoplasmic DNA sensors . They may thus contribute to the strong innate immune response in MDM ( Lahaye et al . , 2013; Rasaiyaah et al . , 2013 ) . It is difficult to envision , however , how the CA-negative cytoplasmic RTC/PIC could give rise to nuclear PIC , since almost all nuclear PIC were strongly CA-positive . Our results therefore suggest that the capsid or a capsid-derived structure remains intact not only during reverse transcription but also during nuclear import , at least in MDM . Lack of a CA signal on nuclear PIC in the proliferating HeLa-derived reporter cell line may indicate the loss of the capsid at the nuclear pore , but prior to nuclear import , or may be due to more rapid removal of CA after nuclear import in these cells compared to MDM . It appears likely that the capsid has to be removed before proviral integration in all productively infected cells , and the two CA-negative nuclear PIC in MDM in the present study may thus reflect a later stage after nuclear entry . Accordingly , CA-negative nuclear PIC were only detected 48 hr after infection , while all nuclear PIC were CA-positive at 24 hr after infection . Detailed investigation of the time course of events will require detection of RTC/PIC in living cells , however , which is not possible with the current experimental system . Imaging productive HIV-1 RTC/PIC allowed us to unravel the dose-dependent bimodal inhibitory mechanism of the CA-directed antiviral PF74 and to directly prove association of CPSF6 with RTC/PIC . Consistent with previous reports ( Shi et al . , 2011 ) , we observed loss of reverse transcription products and of RTC/PIC when infecting cells in the presence of 10 µM PF74 . This concentration is much higher than the IC90 of PF74 , however , and inhibition at lower concentrations , while still blocking HIV-1 infection , did not affect RTC/PIC formation or CA association , but rather blocked nuclear import . PF74 binds the same pocket in the N-terminal domain of CA as the cellular proteins CPSF6 and Nup153 , both of which have been implicated in nuclear import of the PIC ( Matreyek et al . , 2013 ) . Recently , Price et al . revealed that the CA hexamer of the mature lattice ( and not monomeric CA ) constitutes the optimal interface for Nup153 and CPSF6 , which have overlapping binding sites on the capsid that are lost upon disassembly ( Price et al . , 2014 ) . It appears likely , therefore , that PF74 acts as a competitive inhibitor of host factor binding on the intact ( or semi-intact ) capsid structure at limiting drug concentrations , thereby blocking CPSF6 association and nuclear import . Accordingly , we observed a strong reduction of CPSF6 association with RTC/PIC in the presence of 2 µM PF74 when cytoplasmic CPSF6 concentration was increased by knock-down of its nuclear import factor TNPO3 . Saturating concentrations of PF74 may lead to dissociation of the entire capsid structure with consequent loss of reverse transcription , but this does not appear to be the primary mode of action of this antiviral . Many host cell proteins including Trim5α , cyclophilin A , CPSF6 , MX-2 , TNPO3 , as well as nucleoporins have been implicated in early HIV-1 post-entry stages in a CA-dependent way ( Matreyek and Engelman , 2013; Ambrose and Aiken , 2014; Hilditch and Towers , 2014 ) . CA mutations affecting many of these interactions are available , and differential effects of these mutations on infectivity have been reported depending on the host cell ( Matreyek and Engelman , 2013 ) . Direct identification of individual productive HIV-1 RTC/PIC by microscopy will be ideally suited to determine the role of these factors in HIV-1 post-entry stages in various host cells including MDM . The association of nuclear PIC with CA in primary human MDM challenges the current hypothesis concerning early HIV-1 replication and shows the power of this direct imaging-based approach . Detection of productive RTC/PIC and their association with viral and host factors can thus be expected to find wide applications in the future and will contribute to unraveling the current mysteries of HIV-1 post-entry events .
TZM-bl cells ( Wei et al . , 2002 ) and human embryonic kidney 293T cells were grown at 37°C in the presence of 5% CO2 in Dulbecco's modified Eagle's medium ( DMEM; Life Technologies , Germany ) , supplemented with 10% fetal calf serum ( FCS; Biochrom , Germany ) , 100 U/ml penicillin , and 100 µg/ml streptomycin . HeLaP4 derived cell lines stably transduced with shRNA targeting TNPO3 ( referred to here as TNPO3KD ) or a scrambled shRNA control ( referred to here as TNPO3Scr ) , respectively , were kindly provided by Zeger Debyser ( University of Leuven , Belgium ) ( Thys et al . , 2011 ) . Cells were propagated in complete DMEM supplemented with 1 µg/ml puromycin ( Life technologies ) . MT-2 cells ( Harada et al . , 1985 ) were grown in RPMI medium with the same supplements . Human peripheral blood mononuclear cells ( PBMCs ) were isolated from buffy coats of healthy blood donors by Ficoll density gradient centrifugation . PBMCs were seeded in 8-well LabTek chamber slides ( Thermo Fisher Scientific , Waltham , MA ) in DMEM containing 10% heat inactivated FCS and incubated at 37°C for 6 hr . Subsequently , floating lymphocytes were removed and adherent monocytes were washed followed by cultivation in DMEM containing 10% heat inactivated FCS and 10% human AB serum ( Sigma Aldrich , St . Louis , MO ) for 7 days to allow differentiation into macrophages . The HIV-1 proviral plasmid pNL4-3 has been described ( Adachi et al . , 1986 ) . Plasmid pNL4-3-delEnv carries a 2 bp fill-in in the env ORF resulting in a frameshift and premature stop codon . Plasmid pNLC4-3-R5 was cloned by exchanging a StuI/XhoI fragment comprising part of the Env coding region with the corresponding fragment from plasmid pCAGGS . NL4-3R5 which carries mutations in the V3-loop coding region conferring CCR5 tropism ( Bozek et al . , 2012 ) . Plasmid pEnv-4059 expressing an Env protein from a primary HIV-1 isolate ( Schnell et al . , 2011 ) was kindly provided by R Swanstrom ( University of North Carolina , USA ) . Plasmid pVpr . IN . eGFP ( Albanese et al . , 2008 ) encoding a Vpr . IN . eGFP fusion protein with an HIV-1 protease recognition site between Vpr and IN was kindly provided by Anna Cereseto ( CIBIO , Mattareo , Italy ) . To generate plasmid pVpr . IN . mEos3 . 2 , a BamHI/NotI fragment of pVpr . IN . eGFP comprising the eGFP coding region was replaced with a PCR fragment of the mEos3 . 2 coding sequence from pCMVmEos3 . 2 ( Zhang et al . , 2012 ) flanked by BamHI/NotI cleavage sites . PCR primers used were: forward primer CGCGGATCCACCGGTCGCCACCATGAGTGCGATTAAGCCAG; reverse primer CGCGCGGCCGCTTATCGTCTGGCATTGTCAG . Plasmid pRRL . PPT . SF . GFPpre ( Schambach et al . , 2006 ) was kindly provided by Jens Bohne ( Hannover Medical School , Germany ) . Plasmid pAdVAntage was from Promega , plasmids pWPI , pMD2 . G , and psPAX2 were generated in the lab of Didier Trono ( EPFL , Lausanne , Switzerland ) and obtained through Addgene . Rabbit polyclonal antisera against HIV-1 CA , MA , NC , or IN were raised against purified recombinant proteins . Mouse monoclonal antibody 414 , recognizing phenylalanine–glycine repeats of nuclear pore complex proteins , was from Abcam ( UK ) ; mouse monoclonal antibody against cytochrome C was from BD Biosciences ( Franklin , NJ ) ; affinity purified rabbit antibody against CPSF6 was obtained from Sigma ( HPA039973 ) . Secondary antibodies Alexa Fluor 405 Goat Anti-Mouse IgG , Alexa Fluor 532 Goat Anti-Rabbit IgG , and Alexa Fluor 568 Goat Anti-Rabbit IgG were from Life Technologies . Stock solutions of 6 mM aphidicolin ( Sigma A0781 ) , 5 mM efavirenz ( obtained through the AIDS Research and Reference Reagent Program , Division of AIDS , NIAID ) , 10 mM PF74 ( Gilead Sciences , Foster City , CA ) , and 10 mM elvitegravir ( Gilead Sciences ) , respectively , were prepared in dimethyl sulfoxide ( DMSO ) and stored at −20°C . HEK 293T cells were transfected with pNL4-3 or pNLC4-3-R5 ( for producing NL4-3-R5 ( IN . eGFP ) ) , respectively , using a standard CaPO4 transfection procedure . For producing viruses containing IN . eGFP or IN . mEos3 . 2 , the respective expression vector was co-transfected with the proviral plasmid at a molar ratio of 1:4 . 5 . R5-tropic virus ( NL4-3-4059 ( IN . eGFP ) ) carrying Env-4059 from a primary patient isolate ( Schnell et al . , 2011 ) was produced by co-transfecting HEK293T cells with pVpr . IN . eGFP , pEnv-4059 , and pNL4-3-delEnv in a molar ratio of 1:1:4 . 5 . For production of lentiviral vector particles , cells were transfected with pMD2 . G , psPAX2 , pAdVAntage , and the lentiviral vector pRRL . PPT . SF . GFPpre or pWPI , respectively , using a molar ratio of 2 . 8:2 . 8:1:4 . Virus or vector particle containing supernatants were harvested at 36 hr post-transfection , filtered through 0 . 45 µM filters and virus was concentrated by ultracentrifugation through a 20% ( wt/wt ) sucrose cushion . Particles were resuspended in PBS containing 10% FCS and 10 mM HEPES ( pH7 . 5 ) and stored in aliquots at −80°C . For immunoblot analyses , samples were separated by SDS-PAGE ( 12 . 5% ) and proteins were transferred to a polyvinylidene-difluoride membrane by semi-dry blotting . Detection was performed using a LiCor Odyssey instrument , using the indicated primary antisera with corresponding LiCor secondary antibodies . Relative infectivity of virus was analyzed by titration on TZM-bl indicator cells . At 48 hr post-infection , cells were lysed and luciferase activity was quantitated using the SteadyGlo assay kit ( Promega ) according to the manufacturer's instructions . Values obtained were normalized to the amounts of virus particles as assessed by p24 ELISA using an in-house protocol . Determination of virus titers was performed by titration on TZM-bl indicator cells followed by detection of beta-lactamase expressing cells as described previously ( Wei et al . , 2002 ) . MT-2 cells were infected at an m . o . i . of 10 with HIV-1 IIIB pelleted from infected H9 cells ( Advanced Biotechnologies , Eldersburg , MD; catalog number 10-124-000 ) in the presence of 6 µg/ml polybrene by mutation for 3 hr at 37°C . Cells were washed and seeded onto 6-well plates ( 1 × 106 per well ) in a total volume of 2 ml RPMI medium supplemented with FCS . DMSO or inhibitors were added and cells were incubated at 37°C for 12 hr for late RT product quantification or 24 hr for 2-LTR circle product quantification , respectively . Viral DNA was isolated using a QIAamp DNA mini kit ( Qiagen , Germany ) and quantified using TaqMan real-time PCR using the ABI Prism 7900HT sequence detection system ( Life Technologies ) . Primer-probe sets used were: Late RT products , PBS-F ( 5′- TTTTAGTCAGTG TGGAAAATCTCTAGC-3′ ) , PBS-R ( 5′-TTGGCGTACTCACCAGTCGCC-3′ ) , and PBS probe ( 5′-6FAMTCGACGCAGGACTCGGCTTGCT-6TAMSp-3′ ) . Primer-probe sets for 2-LTR circles and the host β-globin gene ( used to normalize for cell number ) were as previously described ( Butler et al . , 2001 ) . 2 × 105 and 4 × 105 TZM-bl cells per well were seeded in 6-well plates and pre-incubated at 37°C with complete DMEM containing DMSO or 6 µM APC for 24 hr . Cells were infected with HIV-1 or HIV-1 ( IN . eGFP ) , respectively , with an m . o . i . of 0 . 1 in the presence of DMSO or 6 µM APC . At 4 hr p . i . , medium was replaced by DMEM and incubation was continued . At 48 hr p . i . , cells were fixed with 3% PFA , permeabilized and probed with phycoerythrin-conjugated monoclonal anti-HIV CA antibody KC57-RD1 ( Beckman Coulter , Germany ) . The proportion of infected cells was quantified by flow cytometry using a BD FACSVerse instrument . TZM-bl cells were seeded in 8-well LabTek chamber slides in DMEM , 10% FCS containing 6 µM APC . On the following day , virus infection was performed at an m . o . i . of 25 in the same medium composition and 10 µM EdU ( Life Technologies ) was added . Cells were pre-incubated at 16°C for 30 min and then shifted to 37°C for 2 hr . Subsequently , infection was either stopped , or the infection mixture was replaced by pre-warmed medium containing 6 µM APC and 10 µM EdU , and incubation at 37°C was continued for the specified time periods . To stop infection , cells were washed with PBS and fixed with 3% paraformaldehyde ( PFA ) at room temperature for 30 min . Cells were washed and permeabilized with 0 . 2% ( vol/vol ) Triton X-100 . Click-labeling was performed for 40 min at room temperature using the Click-iT EdU-Alexa Fluor 647 Imaging Kit ( Life Technologies ) according to the manufacturer's instructions , followed by immunostaining with the indicated antisera . In the case of MDM infection , APC incubation was omitted and infection was performed in the presence of 5 µM EdU . Virus corresponding to 75 ng CA was added to each well and incubated with cells for 24 hr or 48 hr . The amount of virus particles applied yielded an m . o . i . of 50 or 25 on TZM-bl cells for NL4-3-R5 ( IN . eGFP ) or NL4-3-4059 ( IN . eGFP ) viruses , respectively . Cell-fixation , click-labeling , and immunofluorescence staining were performed as described for TZM-bl cells . Fixed and permeabilized cells were blocked with 3% BSA–PBS , washed and incubated with the respective first and secondary antisera for 1 hr each at room temperature . Signal quantification was performed using ImageJ . Equal amounts of objects representing cytoplasmic RTC/PIC or viral particles lacking EdU signals were randomly selected from images taken in three independent experiments . Background signal measured in the extracellular area was subtracted in each channel . Sum signal intensity was calculated for each object in the CA and IN . eGFP channels . Signal intensities of CA or IN . eGFP of RTC/PIC were normalized for the mean values of CA or IN . eGFP obtained from all CA–IN . eGFP objects lacking EdU labeling from the same experiment . Multi-channel 3D image series were acquired with a PerkinElmer UltraVIEW VoX 3D SDCM using a 60× or 100× oil immersion objective ( NA 1 . 49 ) ( Perkin Elmer , Waltham , MA ) , with a z-spacing of 200 nm . Images were recorded in the 405 , 488 , 561 , and 640 nm channels . For quantitation of co-localizing objects , cells were randomly selected for imaging . Co-localization of signals detected in different channels was determined manually on each plane of the image series . Data were analyzed using GraphPad Prism . Super-resolution microscopy was performed using a custom-built microscope setup described earlier ( Nanguneri et al . , 2012 ) . Briefly , a multi-line argon–krypton laser ( Innova70C , Coherent , Santa Clara , CA ) and a 405 nm diode laser ( Cube , Coherent ) were coupled into an inverted microscope ( IX71 , Olympus ) equipped with a 63× oil immersion objective ( PlanApo 63× , NA 1 . 45 , Olympus ) suitable for total internal reflection fluorescence ( TIRF ) imaging . The excitation and emission beams were separated using appropriate dichroic mirrors and filters ( AHF , Germany ) . The fluorescence emission was detected by an EM-CCD camera ( Ixon , Andor , UK ) . PALM/dSTORM imaging was performed using an imaging buffer which is suitable for both photoswitching the fluorescent protein mEos3 . 2 as well as the organic fluorophores Alexa Fluor 532 or Alexa Fluor 647 ( Endesfelder et al . , 2011 ) . Briefly , the cells were imaged in oxygen-depleted hydrocarbonate buffer ( pH 8 ) supplemented with 50 mM mercaptoethylamine ( MEA ) . First , Alexa Fluor 647 was reversibly photoswitched by irradiation with 488 nm ( photoactivation ) and 647 nm ( read-out ) . For each channel , 8000 images were recorded with an integration time of 50 ms . Then mEos3 . 2 was photoactivated by irradiation with 405 nm and imaged using an excitation wavelength of 568 nm . Alexa Fluor 532 was photoactivated by irradiation with 514 nm . Single-molecule localization and image reconstruction was performed using the rapidSTORM software ( Wolter et al . , 2011 ) . The localization accuracy of single-molecule super-resolution microscopy was evaluated experimentally as described earlier using a custom written software ( coordinate based localization precision estimator , provided as supplement to Endesfelder et al . ( 2014 ) ) . For each localized fluorophore , the distance to its nearest neighbor fluorophore in an adjacent frame was calculated . As the majority of fluorophores are detected in multiple adjacent frames , the maximum of the nearest neighbor distance distribution represents the error of localization . A pre-requisite for using this approach is a statistically significant number of events ( n ∼ 4000 ) . Images were recorded by adding multi-spectral beads ( Life Technologies ) to the sample and post-aligning the individual images ( Malkusch et al . , 2012 ) . | Major advances in the treatment of HIV have been made possible by carefully studying the virus and its interaction with the host cell . The virus consists of two strands of RNA—representing the genetic information of the virus—contained in a protein coat called capsid . Scientists have learned that the virus' RNA is used to create viral DNA in the cytoplasm of an infected cell , in a process called reverse transcription . This viral DNA then enters the cell's nucleus and becomes incorporated into the cell's DNA , and the cell unwittingly begins to help the virus reproduce . It is less clear what happens to the capsid after the virus enters a cell . Some researchers have suggested that it is lost shortly after entry or during reverse transcription . However , some recent studies have found that damaging the capsid hampers reverse transcription and significantly impairs the entry of viral DNA into the cell's nucleus . This suggests that the capsid might continue to protect the viral genome when the RNA is converted into DNA . To learn more about what happens during reverse transcription and when the viral DNA enters the nucleus , it is important to watch individual events as they occur . Until recently , it had been hard to do this without changing the DNA or RNA in ways that might affect their properties . Recently , a technique called click-labeling has been developed that can add a fluorescent label to DNA or RNA without potentially damaging this genetic material . This label allows the movement of the DNA or RNA to be followed when the cell is viewed under a microscope . Peng et al . used this new technique to watch reverse transcription , how viral DNA enters the cell nucleus and what happens to the capsid when HIV invades different kinds of cells . When the virus entered a type of cell often used in laboratory research called HeLa cells , the capsid protected the viral genetic material when it was in the cell's cytoplasm but disappeared before or shortly after the viral DNA entered the cell's nucleus . However , HeLa cells are not natural targets of HIV; when Peng et al . looked at the behavior of the capsid in the immune cells that the virus normally invades , the capsid was present in both the cytoplasm and the nucleus of these cells . Peng et al . also observed what happens in HIV-infected cells treated with a chemical called PF74 that interferes with the capsid . This revealed that low concentrations of PF74 make it hard for the viral DNA to enter the nucleus , probably by blocking the interaction of the capsid with a protein from the host cell . At high concentrations , the drug prevented reverse transcription . The approach used by Peng et al . allows direct visualization of how HIV replicates and how this DNA is imported into the nucleus of cells naturally targeted by the virus . This will aid our understanding of how the virus selects where in the host genome it should insert its DNA , which is important for establishing a permanent infection in the cell . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"Methods"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2014 | Quantitative microscopy of functional HIV post-entry complexes reveals association of replication with the viral capsid |
We describe Cleavage Under Targets and Release Using Nuclease ( CUT&RUN ) , a chromatin profiling strategy in which antibody-targeted controlled cleavage by micrococcal nuclease releases specific protein-DNA complexes into the supernatant for paired-end DNA sequencing . Unlike Chromatin Immunoprecipitation ( ChIP ) , which fragments and solubilizes total chromatin , CUT&RUN is performed in situ , allowing for both quantitative high-resolution chromatin mapping and probing of the local chromatin environment . When applied to yeast and human nuclei , CUT&RUN yielded precise transcription factor profiles while avoiding crosslinking and solubilization issues . CUT&RUN is simple to perform and is inherently robust , with extremely low backgrounds requiring only ~1/10th the sequencing depth as ChIP , making CUT&RUN especially cost-effective for transcription factor and chromatin profiling . When used in conjunction with native ChIP-seq and applied to human CTCF , CUT&RUN mapped directional long range contact sites at high resolution . We conclude that in situ mapping of protein-DNA interactions by CUT&RUN is an attractive alternative to ChIP-seq .
The action of transcription factors ( TFs ) at their binding sites on DNA drives gene expression patterns , and so genome-wide TF mapping has become a central goal both for individual researchers and large-scale infrastructural projects . TF profiling is most commonly carried out using chromatin immunoprecipitation ( ChIP ) , a protocol that has changed little since it was first introduced over 30 years ago ( Solomon and Varshavsky , 1985 ) . Cells are crosslinked with formaldehyde , chromatin is fragmented and solubilized , an antibody is added , and the antibody-bound chromatin is recovered for DNA extraction . Successive advances in DNA mapping technologies have revolutionized the use of X-ChIP ( formaldehyde crosslinking ChIP ) , and with ChIP-seq , base-pair resolution mapping of TFs became feasible ( Rhee and Pugh , 2011; Skene and Henikoff , 2015; He et al . , 2015 ) . Improvements to ChIP-seq retain the crosslinking step to preserve the in vivo pattern while the entire genome is fragmented to create a soluble extract for immunoprecipitation . However , crosslinking can promote epitope masking and can generate false positive binding sites ( Teytelman et al . , 2013; Park et al . , 2013; Jain et al . , 2015; Baranello et al . , 2016; Meyer and Liu , 2014 ) . ChIP can also be performed without crosslinking , using ionic conditions that do not disrupt electrostatic contacts ( Kasinathan et al . , 2014 ) . ‘Native’ ChIP provides a map of direct protein-DNA interactions with sensitivity and specificity trade-offs that compare favorably with X-ChIP methods . Native ChIP also minimizes problems with epitope masking and improves efficiency relative to X-ChIP , making it more amenable to low starting numbers of cells ( O'Neill et al . , 2006; Brind'Amour et al . , 2015 ) . But problems remain with incomplete extraction efficiency of protein-DNA complexes and the potential loss of binding . The uncertainties caused by systematic biases and artifacts in ChIP emphasize the need for methods based on different principles . An important class of non-ChIP mapping methods involves tethering of an enzyme to a DNA-binding protein by a chimeric fusion and action of the enzyme on DNA in the local vicinity . For example , in DamID ( van Steensel et al . , 2001 ) and related methods ( Southall et al . , 2013; Hass et al . , 2015 ) , Escherichia coli Dam methyltransferase is tethered to the TF and catalyzes N6-methylation of adenine at GATC sites in vivo . Sites can be mapped genome-wide using an N6-methyl-directed restriction enzyme . However , as the resolution of DamID is limited by the distribution of GATC sites , DamID cannot obtain the high resolution that is potentially attainable using a sequencing read-out ( Aughey and Southall , 2016 ) . An alternative enzyme tethering method , chromatin endogenous cleavage ( ChEC ) tethers the endo-exonuclease Micrococcal Nuclease ( MNase ) to the TF ( Schmid et al . , 2004 ) . In ChEC , MNase is activated by permeabilizing cells and adding calcium for controlled cleavage . We recently applied an Illumina sequencing read-out to ChEC ( ChEC-seq ) , achieving near base-pair resolution ( Zentner et al . , 2015 ) . Enzyme tethering methods fundamentally differ from ChIP because they are carried out in vivo ( DamID ) or in situ ( ChEC ) , with extraction of DNA directly from live or permeabilized cells , thus eliminating the need to solubilize and recover chromatin . Both DamID and ChEC require that a different chimeric fusion construct be produced for each TF to be mapped , limiting their transferability , for example to animal models , patient biopsies and post-translational modifications . In the original chromatin immunocleavage ( ChIC ) method , crude nuclei from crosslinked cells are first treated with a TF-specific antibody , followed by addition of a chimeric fusion between Protein A and MNase ( pA-MN ) and activation by calcium ( Schmid et al . , 2004 ) . Protein A binds specifically to Immunoglobulin G , which obviates the need for a fusion protein . Here we report a major development of ChIC that retains the advantages of enzyme tethering methods , while extending its applicability and ease-of-use to a point that it replaces other existing methodologies . A key feature of our protocol is that in the absence of crosslinking , seconds after calcium-induced MNase cleavage on both sides of the TF , the TF-DNA complex is released into solution , allowing for recovery of pure TF-bound DNA fragments for sequencing simply by centrifugation and DNA extraction . By carrying out the procedure on magnetic beads , our ‘Cleavage Under Targets and Release Using Nuclease’ ( CUT&RUN ) technique is simpler than ChIP-seq while retaining the advantages of in situ methods . Targeted digestion by CUT&RUN greatly reduces background relative to complete genomic fragmentation for ChIP , requiring only ~1/10th the sequencing depth of standard ChIP methods . Simple spike-in controls allow accurate quantification of protein binding not possible by other methods . Furthermore , the method allows low starting cell numbers , and robotic automation is possible by performing the reaction on magnetic beads .
Chromatin Immuno-Cleavage ( ChIC ) has the advantage of using TF-specific antibodies to tether MNase and cleave only at binding sites . To adapt ChIC for deep sequencing , it was necessary to reduce the representation of background breaks in DNA that otherwise dominate deep sequencing libraries . The CUT&RUN modifications of ChIC were motivated by the observation that light MNase treatment of nuclei liberates mononucleosomes and TF-DNA complexes , leaving behind oligonucleosomes ( Sanders , 1978; Teves and Henikoff , 2012 ) . We expected that targeted cleavage on both sides of a TF would release the TF-DNA complex into the supernatant , leaving the remainder of the genome in the pelleted nuclei . By performing brief digestion reactions on ice , we would recover TF-DNA complexes in the supernatant before TF-bound MNase diffuses around the genome and cleaves accessible chromatin . Based on this rationale , we developed a simple CUT&RUN protocol ( Figure 1A ) . Unfixed nuclei are ( 1 ) immobilized on lectin-coated magnetic beads , ( 2 ) successively incubated with antibodies and protein A-MNase ( pA-MN ) followed by minimal washing steps , ( 3 ) mixed with Ca++ on ice to initiate the cleavage reaction then stopped seconds-to-minutes later by chelation , and ( 4 ) centrifuged to recover the supernatant containing the released TF-DNA complexes . DNA is then extracted from the supernatant and used directly for sequencing library preparation . 10 . 7554/eLife . 21856 . 003Figure 1 . CUT&RUN produces limit digestion TF-DNA complexes . ( A ) Schematic diagram of the CUT&RUN strategy . Nuclei attached to magnetic beads can be treated successively with an antibody ( or optionally with a primary and secondary antibody ) and Protein A-MNase ( pA-MN ) , which diffuse in through the nuclear pores . After Ca++ addition to activate MNase cleavage , fragments are released and diffuse out of the nucleus . DNA extracted from the supernatant is used to prepare libraries for paired-end sequencing . ( B ) CUT&RUN cleaves and releases chromatin particles into the supernatant . S . cerevisiae nuclei in which the endogenous H2A genes were replaced with H2A-3XFLAG were subjected to CUT&RUN and incubated at 0°C in Ca++ for the indicated times . DNA extracted from both the insoluble ( ins ) and soluble ( sol ) fractions was electrophoresed on a 1% agarose gel . The No 1o Ab control was digested for 10 min in parallel but without having added the primary mouse anti-FLAG antibody . ( C ) Size distributions of mapped paired-end reads from sequencing of indicated TF samples . An H2A size distribution is included for comparison . Data are normalized such that the sum of all points at each length step in base pairs equals 1 . ( D ) Time-course profiles for Abf1 and Reb1 samples ( ~2–3 million mapped paired-end reads per track ) showing ≤120 bp ( blue ) and ≥150 bp ( brown ) fragment length classes , compared to ORGANIC ChIP-seq ( ~20–30 million mapped paired-end reads ) and standard ChIP-seq ( Paul et al . , 2015 ) ( ~5 million Abf1 and ~126 million Reb1 mapped single-end 50 bp reads ) . A negative control track shows the result of leaving out the primary antibody ( No 1o Ab ) . Within each TF and fragment size group , the Y-axis scale is autoscaled by IGV showing normalized counts and the fragment size classes are superimposed . Ticks mark the location of significant Abf1 ( upper ) and Reb1 ( lower ) motifs . This region was chosen as having the largest cluster of Abf1 motifs on Chromosome 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 00310 . 7554/eLife . 21856 . 004Figure 1—figure supplement 1 . CUT&RUN and ORGANIC ChIP produce qualitatively similar TF occupancy profiles . Representative examples of Abf1 and Reb1 profiles for CUT&RUN data pooled from the 1” to 32” and the 64” and 128” time-course samples and ORGANIC for ≤120 bp and ≥150 bp fragment lengths , and standard ChIP-seq . An MNase-seq profile is shown in grey . Ticks mark the location of significant Abf1 ( upper ) and Reb1 ( lower ) motifs . The Y-axis was autoscaled within each region by IGV . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 00410 . 7554/eLife . 21856 . 005Figure 1—figure supplement 2 . Kinetics of CUT&RUN DNA release . ( A ) Electrophoresis on a 1% agarose gel of DNA from the pellet fractions ( 10 µL per sample ) over a 1–128 s digestion time series at 0°C for the two yeast TFs described in this study . As these sites are on average ~10 kb apart in the yeast genome a gradual decrease in fragment size can be observed with time of digestion from 1 s to 128 s for both Abf1 and Reb1 . The average distance between CTCF sites in the human genome is too large to observe cleavages using a conventional gel assay . ( B ) Percent release of DNA based on Picogreen fluorescence measurements: [Supn]/ ( [Supn]+[Pellet] ) *100 . Total yield ~500 ng/sample . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 00510 . 7554/eLife . 21856 . 006Figure 1—figure supplement 3 . Quantitative recovery of bound TFs in supernatants . ( A ) A comparison of Abf1 and Reb1 profiles of CUT&RUN data from a single experiment as described in the legend to Figure 1D , except comparing the supernatant fraction ( Soluble ) to Total DNA after removal of large fragments on AMPure beads . ( B ) Expanded region of high TF occupancy in ( A ) . ( C ) Heat map alignments of CUT&RUN ≤120 bp digestion data to motifs and ordering by TF occupancy was performed as described in the legend to Figure 2 , except with log scaling and Contrast = 5 , centered on 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 00610 . 7554/eLife . 21856 . 007Figure 1—figure supplement 4 . Abf1 and Reb1 motifs based on CUT&RUN and ORGANIC ChIP-seq are similar . The MEME motif-finding program was applied to ( A–B ) 1”−32” pooled CUT&RUN ≤120 bp data and ( C ) 600 mM Abf1 and ( D ) 80 mM Reb1 ORGANIC data , and log-odds sequence logos are shown . Note the close correspondence between motifs determined using CUT&RUN and ORGANIC . ( E ) Percentage of peak calls with motifs . For each ≤120 bp dataset , peaks were called using thresholds set to recover similar numbers of peaks ( stringent ~650 and relaxed ~1100 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 007 We first performed CUT&RUN using crude yeast nuclei . To rigorously compare CUT&RUN and ChIP-seq , we used the same FLAG-tagged TF strains , the same nuclear preparation protocol , the same mouse anti-FLAG monoclonal antibody and the same procedure for Illumina library preparation and paired-end sequencing ( Kasinathan et al . , 2014 ) . As mouse Protein A binds only weakly to mouse IgG , we used a rabbit anti-mouse secondary antibody for CUT&RUN . To test the efficiency of CUT&RUN , we used a Saccharomyces cerevisiae strain expressing 3XFLAG-tagged histone H2A , which would be expected to release nucleosomal fragments genome-wide . Indeed , over a 100-fold digestion time course at 0°C , we observed gradual cleavage and release of fragments down to mononucleosomal size , completely dependent on the presence of the primary antibody ( Figure 1B ) . We next applied CUT&RUN to two structurally distinct S . cerevisiae TFs , ARS binding factor 1 ( Abf1 ) , and rDNA enhancer binding protein 1 ( Reb1 ) , obtaining ~2–3 million mapped paired-end reads per sample . We found that the size distributions of mapped fragments were virtually superimposable below ~150 bp for time points between 4 s and 128 s ( Figure 1C ) . This close concordance between time points over a 32-fold range suggests that limit digestion of TF-bound fragments occurs rapidly upon addition of Ca++ , and demonstrates that digestion time is not a critical parameter . Mapped TF fragment sizes peaked at ~100 bp , in contrast to H2A fragments , which peaked at ~150 bp . As we expect TF complexes to be smaller than ~100 bp , and nucleosomes are ~150 bp , we mapped ≤120 bp and ≥150 bp fragments separately . Time-point profiles show crisp CUT&RUN peaks within the ≤120 bp size class for each TF motif in each region ( Figure 1D and Figure 1—figure supplement 1 ) . Except for a slow monotonic increase in peak occupancy when normalized to the spike-in control ( Figure 1—figure supplement 2 ) , no consistent differences between time points were observed within the 1 s to 128 s interval , confirming that gradual release of TF-DNA complexes yields limit digestion reactions . Total DNA extraction and purification of small fragments produced nearly identical results ( Figure 1—figure supplement 3 ) , which demonstrates that extraction of DNA from the supernatant quantitatively recovers TF-bound fragments . To verify that the ≤120 bp fragments represent cleavages around TF binding sites , we first identified all significant Abf1 and Reb1 motifs in the genome and found that motifs based on CUT&RUN data and motifs based on ORGANIC data were nearly identical ( Figure 1—figure supplement 4A–D ) . We use ORGANIC-derived motifs to scan the yeast genome , which provided us with a comprehensive list of 1899 Abf1 and 1413 Reb1 motifs determined completely independently of CUT&RUN . We confirmed that the majority of peak calls overlapped the motif for each dataset , with somewhat better performance for CUT&RUN than ORGANIC for Abf1 and vice versa for Reb1 ( Figure 1—figure supplement 4E ) . We then aligned the ≤120 bp and ≥150 bp profiles centered over these motifs and constructed heat maps . When rank-ordered by occupancy over the 2 kb interval centered over each Abf1 and Reb1 motif , we observed that >90% of the TF sites were occupied by fragments over the corresponding motif relative to flanking regions ( Figure 2 and Figure 2—figure supplement 1 , upper panels ) , representing likely true positives . CUT&RUN occupancies over Abf1 and Reb1 motifs showed high dynamic range relative to nuclease accessibility ( Figure 2—figure supplement 1 , lower panels ) , seen in heat maps as higher contrast above background for CUT&RUN . In contrast , Abf1 fragments showed negligible occupancy at non-overlapping Reb1 sites and vice-versa for Reb1 fragments at non-overlapping Abf1 sites ( Figure 2 and Figure 2—figure supplement 1 , middle panels ) . The almost complete correspondence between the presence of a TF motif and occupancy of the TF , and the general absence at sites of a different TF , imply that CUT&RUN is both highly sensitive and specific for TF binding . 10 . 7554/eLife . 21856 . 008Figure 2 . CUT&RUN accuracy and robustness compares favorably with ChIP-seq . Abf1 ( A ) and Reb1 ( B ) heat maps of CUT&RUN datasets from a single experiment ( 20160630 ) , pooling 1” to 32” time-course samples , and separating into ≤120 bp and ≥150 bp size classes ( left ) . Also shown is the ORGANIC ChIP-seq ≤120 bp size class ( middle ) and standard ChIP-seq datasets ( right ) . Abf1 has two DNA-binding domains spaced ~10 bp apart ( Cho et al . , 1995 ) , whereas Reb1 has a single Myb-like DNA-binding domain ( Morrow et al . , 1990 ) . Solubilization of Abf1 chromatin after MNase digestion required 600 mM NaCl to obtain the best trade-off between specificity and sensitivity , whereas for Reb1 , 80 mM gave the best results ( Kasinathan et al . , 2014 ) , and these are the datasets used for comparison . As in our previous comparison of ORGANIC to ChIP-exo and ChIP-chip ( Kasinathan et al . , 2014 ) , we consider the set of all statistically significant Abf1 and Reb1 motifs as the ‘gold standard’ for judging sensitivity ( occupancy of sites by the correct TF ) and specificity ( exclusion from sites of an incorrect TF ) . Aligned profiling data were centered and oriented over the motif for the same TF ( top ) and for the other TF ( bottom ) for display ( removing 81 sites where Abf1 and Reb1 sites were within 50 bp of one another ) and were ordered by average pixel density over the −1 kb to +1 kb span of the ≤120 bp datasets using Java Treeview with log2 scaling and contrast = 5 . Ordering was performed independently for CUT&RUN ( based on ≤120 bp fragments ) and ChIP-seq , in which case the approximate fraction of sites occupied relative to flanking regions becomes evident , and comparison of the top panel ( correct TF ) to the bottom panel ( incorrect TF ) reflects the sensitivity/specificity tradeoff for a dataset . Sites were determined by MAST searching of the S . cerevisiae genome using the position-specific scoring matrices ( PSSMs ) based on ChIP-seq data ( Figure 1—figure supplement 4 ) , but similar results were obtained using MAST with PSSMs based on CUT&RUN data ( not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 00810 . 7554/eLife . 21856 . 009Figure 2—figure supplement 1 . CUT&RUN reveals cleavage kinetics in situ . A comparison of Abf1 ( left ) and Reb1 ( right ) heat maps of CUT&RUN data from a single experiment ( 20160630 ) , pooling the 1” to 32” and the 64” and 128” time-course samples , and separated into ≤120 bp ( left ) and ≥150 bp ( right ) size classes . Alignments to motifs and ordering by TF occupancy was performed as described in the legend to Figure 3 , except that Treeview was used with log scaling and contrast = 3 . Note that with increased digestion time , more of the TFs are released , deepening the ‘hole’ of ≥150 bp fragments without any noticeable change in dynamic range . CUT&RUN shows a much higher dynamic range than MNase-seq for particle detection ( compare top panels with bottom panels ) . MNase-seq data are from Henikoff et al . , 2011 . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 009 To directly compare CUT&RUN to high-resolution ChIP-seq , we similarly lined up ‘ORGANIC’ ChIP-seq data over Abf1 and Reb1 motifs . As previously reported ( Kasinathan et al . , 2014 ) , ORGANIC ChIP-seq detected the large majority of Abf1 true positive motifs and nearly all Reb1 motifs throughout the genome ( Figure 2 , upper middle panels ) . The best Reb1 data were obtained with 80 mM NaCl extraction , and the best Abf1 data were obtained with 600 mM NaCl , although the dynamic range for Reb1 was always better than that for Abf1 with frequent false positive occupancy ( Figure 2 , lower middle panels ) . In contrast , CUT&RUN showed the same dynamic range for both TFs over the same range of digestion time points with ~10 fold fewer paired-end reads , demonstrating that CUT&RUN is more robust than ORGANIC ChIP-seq . Relative to these high-resolution methods ( Kasinathan et al . , 2014 ) , standard ChIP-seq using crosslinking and sonication showed inferior sensitivity and specificity ( Figure 2 , right panels ) . Thus , CUT&RUN provides robust TF occupancy maps with improved sensitivity/specificity trade-offs relative to ChIP-seq . To estimate the resolution of CUT&RUN , we plotted the ‘footprint’ of each TF as the average density of fragment ends around the motif midpoint . For both Abf1 and Reb1 , we observed sharp 20 bp wide footprints , indicating that these transcription factors protect ~20 bp centered over the motif with near base-pair resolution ( Figure 3A ) . Interestingly , upstream and downstream ‘slopes’ in the cleavage maps show a sawtooth pattern on either side of both Abf1 and Reb1 motifs , with distances between 'teeth' ~10 bp apart over >100 bp , and confirmed by autocorrelation analysis to be independent of base composition ( Figure 3B ) . Such 10 bp periodic cleavage preferences match the 10 bp/turn periodicity of B-form DNA , which suggests that the DNA on either side of these bound TFs is spatially oriented such that tethered MNase has preferential access to one face of the DNA double helix . Tethering of MNase to a TF constrains it to cleave nearby DNA even on the surface of a nucleosome , suggesting flexibility of the chromatin fiber ( Figure 3C ) . Thus , the very rapid kinetics that we observe at 0°C are due to immobilized MNase poised for cleavage near the tethering site . 10 . 7554/eLife . 21856 . 010Figure 3 . CUT&RUN maps TF binding sites at high resolution . ( A ) Mapping of fragment ends reveals a deep ‘hole’ and steep ‘walls’ for Abf1 and Reb1 CUT&RUN datasets averaged at their oriented and aligned motifs genome-wide , plotting all normalized base-pair counts from combined 1”−32” datasets ( Figure 2 ) . Sawtooth patterns with an apparent ~10 bp periodicity on the upstream and downstream ‘slopes’ are confirmed by ( B ) autocorrelation analysis of the difference between the 1 bp resolution profile shown in ( A ) and the same profile smoothed with an 11 bp sliding window , which also shows that there is no corresponding periodicity in average G+C content ( thin lines ) . ( C ) Same as ( A ) , but subject to smoothing with an 11 bp sliding window and displayed at larger scale . The fact that the slopes around Reb1 show depressions at +150 and −150 likely reflects the presence of phased nucleosomes , shown below ( Nucls , Y-axis arbitrary ) based on the ≥150 bp size class from ORGANIC input data ( Kasinathan et al . , 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 010 High-resolution mapping of mobile components of the chromatin landscape can be challenging for ChIP-based methods . For example , the ~1 megadalton 17-subunit RSC nucleosome remodeling complex dynamically slides a nucleosome that it transiently engulfs ( Lorch et al . , 2010; Ramachandran et al . , 2015 ) , and the Mot1 DNA translocase dynamically removes TATA-binding protein ( TBP ) from sites of high-affinity binding ( Zentner and Henikoff , 2013; Auble et al . , 1997 ) . Although X-ChIP crosslinks nucleosome remodeling complexes to their nearest nucleosomes , native ChIP successfully captures yeast chromatin remodelers at their sites of action , both in nucleosome-depleted regions ( NDRs ) and on nucleosomes ( Zentner et al . , 2013 ) . For CUT&RUN to profile such large chromatin-associated complexes we found it necessary to extract total DNA rather than chromatin solubilized by CUT&RUN in situ , which may be too large to diffuse through nuclear pores . Therefore , we extracted all DNA and preferentially removed large DNA fragments with AMPure beads . When this modified protocol was applied to Mot1 over a >2 order-of-magnitude digestion range , we observed chromatin profiles that were very similar to those obtained using ORGANIC profiling , but with only ~15% the number of paired-end reads ( Figure 4A ) . As expected , Mot1 peaks on the upstream side of TBP binding sites are seen for both CUT&RUN and ORGANIC profiles , confirming that Mot1 approaches TBP from the upstream side in vivo ( Zentner and Henikoff , 2013 ) as it does in vitro ( Wollmann et al . , 2011 ) . Heat map and average plot analyses show that the ≤120 bp fragments track closely with TBP sites , whereas the ≥150 bp fragments are diffusely distributed in the local vicinity , perhaps representing Mot1 translocation dynamics ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 21856 . 011Figure 4 . CUT&RUN precisely maps large mobile chromatin complexes . ( A ) Representative tracks showing a Mot1 CUT&RUN time-course experiment ( average ~3 million paired-end reads per sample ) , including a no primary antibody ( No 1o Ab ) negative control , aligned with Mot1 ORGANIC data for two MNase digestion time points ( 2 . 5’ and 10’ , average 22 million reads per sample ) ( Zentner and Henikoff , 2013 ) . TBP sites shown as dotted lines reveal that Mot1 peaks are just upstream of TBP peak maxima . ( B ) Occupancy profiles for Sth1 CUT&RUN digestion over a 120-fold range , spike-in normalized , showing absolute quantitation . ( C ) Sth1 ORGANIC profiles ( ~15 million reads ) show concordance with the CUT&RUN 5 s sample ( ~2 million reads ) . Note that the same CUT&RUN 5 s ≤120 bp profile is shown in both panels ( B ) and ( C ) , but at different scales . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 01110 . 7554/eLife . 21856 . 012Figure 4—figure supplement 1 . CUT&RUN and ORGANIC profiles for Mot1 . ( A ) Heat maps of two CUT&RUN and two ORGANIC time points aligned around TBP sites and ordered by increasing Mot1 occupancy over the 2 kb region surrounding each site . ( B ) Occupancy profiles for Mot1 CUT&RUN digestion over a 120-fold range , spike-in normalized , showing absolute quantitation . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 01210 . 7554/eLife . 21856 . 013Figure 4—figure supplement 2 . CUT&RUN and ORGANIC profiles for Sth1 . ( A ) Length distributions of Sth1 CUT&RUN AMPure-bead filtered total DNA fragments normalized such that there is equal areas under the curves . Uniform digestion and release is observed over the time-course . Data are combined from two biological replicates . No anti-FLAG primary antibody ( No Ab ) ( B ) Tracks of the Gal1-Gal4 region ( ChrII:276 , 000–281 , 000 ) showing concordance with the mapping of RSC to the Gal4 UAS ( UASg , Floer et al . , 2010 . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 013 We also applied CUT&RUN to Sth1 , the catalytic component of the RSC complex . RSC acts to slide nucleosomes at NDRs , and so we aligned yeast genes at the inferred dyad axis of the +1 nucleosome just downstream of the transcriptional start site ( Ramachandran et al . , 2015 ) . We observed uniform digestion over a 5 s to 30 min time course ( Figure 4—figure supplement 2A ) , and confirmation of the abundance of RSC directly over the GAL4 UAS ( Figure 4—figure supplement 2B ) ( Floer et al . , 2010 ) . Sth1 peaks are most abundant in the NDRs , where CUT&RUN profiles show a gradual increase in yield with digestion times between 5 s and 10 min ( Figure 4B ) , indicating that quantitative limit digestions are obtained using our protocol . Importantly , we observed a nearly flat line for our negative control derived from 3XFLAG-Sth1 nuclei treated in parallel for the maximum digestion time , but where the primary anti-FLAG antibody was omitted . Our results for Sth1 CUT&RUN are similar to results for Sth1 ORGANIC profiling ( Ramachandran et al . , 2015 ) , but with much higher yield ( Figure 4C ) . We conclude that CUT&RUN provides efficient high-resolution mapping of chromatin-associated complexes , even those that are very large and dynamic . Abf1 and Reb1 are relatively abundant TFs , but many DNA-binding proteins of interest are rare , and so can be challenging to profile by ChIP . In budding yeast , there is only only one centromeric nucleosome per chromosome , which is only ~1% the molar abundance of Abf1 or Reb1 . An additional challenge to studying the centromeric nucleosome , which contains the CenH3 ( Cse4 ) histone variant in place of H3 , is that it is part of the multi-megadalton kinetochore complex throughout the cell cycle ( Akiyoshi et al . , 2010 ) , which renders it highly insoluble ( Krassovsky et al . , 2012 ) . To profile the Cse4 nucleosome by CUT&RUN , we split the samples after digestion , extracting just the supernatant from one aliquot , and total DNA from the other . In this way we could compare the recovery of the soluble and the insoluble kinetochore complex . In parallel , we similarly profiled histone H2A . By taking the difference between total and soluble chromatin , we can infer the occupancy of each histone in the insoluble pellet . As expected for the insoluble kinetochore , the highest Cse4 occupancy on the chromosome is seen at the centromere ( Figure 5A ) . Strikingly , occupancy of insoluble H2A , which is present in every nucleosome througout the genome , is also maximum at the centromere . Indeed , at all 16 yeast centromeres we observe very similar enrichments of Cse4 and H2A confined to the ~120 bp functional centromere over the digestion time-course , with resolution that is 4-fold better than that of standard X-ChIP ( Figure 5B ) . We also extracted total DNA from the bead-bound chromatin derived from cells that had been formaldehyde crosslinked prior to applying CUT&RUN , with similar results ( Figure 5C ) . Interestingly , crosslinking results in a more distinct profile and the appearance of phased nucleosomes on either side , which we interpret as a reduction in chromatin flexiblity with crosslinking , while demonstrating that the basic strategy can be applied to crosslinked cells . 10 . 7554/eLife . 21856 . 014Figure 5 . CUT&RUN maps the rare highly insoluble S . cerevisiae kinetochore complex . ( A ) After stopping digestion for the indicated times , samples were split in half and both the soluble fraction and total DNA were extracted . Large fragments were removed from total DNA with AMPure beads before library preparation . Normalized counts are shown for S . cerevisiae Centromere 1 , where Cse4 and H2A tracks are on the same Y-axis scale . Similar maxima over centromeres was also seen genome-wide . ( B ) Same as ( A ) but zoomed in over the 5 kb interval at the centromere . ( C ) Occupancies of insoluble Cse4 and H2A , where we define log2 ( Insoluble ) = log2 ( Total ) – log2 ( Soluble ) = log2 ( Total/Soluble ) for the medians of all 16 S . cerevisiae centromeres aligned around their midpoints . A published X-ChIP-seq profile ( Pekgöz Altunkaya et al . , 2016 ) is shown on the same scale for comparison ( left ) . Asterisk: log2 ( ChIP/Input ) averaged over two replicates . ( D ) Normalized count profile of Cse4 and H2A CUT&RUN applied to formaldehyde cross-linked cells digested for the indicated times . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 01410 . 7554/eLife . 21856 . 015Figure 5—figure supplement 1 . CUT&RUN maps the rare highly insoluble S . cerevisiae kinetochore complex . ( A ) After antibody and pA-MN addition , samples were split in half , pA-MN was activated with calcium , and the reaction stopped with either the standard 100 mM NaCl buffer ( - ) or a buffer containing 2 M NaCl ( + ) . Tracks are displayed for Chromosome one using spike-in normalization to reflect absolute recovery . ( B ) Close-up views of Cse4 and log-ratios of Cse4 and H2A high-salt versus low-salt extracted fragments . ( C ) Log-ratios of high-salt versus low-salt extracted fragments for the medians of all 16 S . cerevisiae centromeres aligned around their midpoints . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 015 To confirm that the differences we observed between the CUT&RUN supernatant and total DNA were due to differential solubility of kinetochore chromatin , we split the samples before digestion , and for one aliquot we stopped the cleavage reaction with 2 M NaCl and recovered the supernatant for sequencing . We obtained similar results for the high-salt fraction as for total DNA ( Figure 5—figure supplement 1 ) . The unequivocal presence of insoluble H2A in the centromeric nucleosome directly addresses the continuing controversy over its composition ( Wisniewski et al . , 2014; Henikoff et al . , 2014; Aravamudhan et al . , 2013; Shivaraju et al . , 2012 ) . Moreover , as the yeast centromeric nucleosome wraps DNA that is >90% A+T ( Krassovsky et al . , 2012 ) , the intactness of the centromeric particle over a >100 fold digestion time-course ( Figure 5 ) demonstrates that CUT&RUN is not biased by the inherent preference of MNase for AT-rich DNA ( Chung et al . , 2010; McGhee and Felsenfeld , 1983 ) . We conclude that CUT&RUN can map large DNA-binding complexes , even those that are rare , insoluble and AT-rich . Examination of the ≥150 bp profiles ( Figure 1D and Figure 4 ) reveals broad peaks around the binding sites , sometimes with ‘notches’ corresponding to the sites themselves that deepen with time of digestion ( Figure 2—figure supplement 1 ) . We interpret this pattern as representing the gradual release of fragments with one end resulting from cleavage around the TF-DNA complex and a second cleavage that is close enough to the TF-bound site to produce a soluble fragment . Heat map analysis of the ≥150 bp fragments also showed occupancy of Abf1 and Reb1 fragments over their respective binding motifs , extending ~0 . 5 kb on either side ( Figure 2 and Figure 2—figure supplement 1 ) . Such extended local cleavage is reminiscent of the >1 kb reach of DamID ( van Steensel et al . , 2001 ) , which suggests that the flexibility of the tether results in probing of nearby chromatin . Having established proof-of-principle in a simple well-studied genome , we next applied CUT&RUN to CCCTC-binding factor ( CTCF ) in human K562 cells . To directly compare the efficiency of various methods , we randomly selected 10 million reads for each technique and plotted the raw scores as an indication of information content per sequenced read . As was the case for yeast TFs , CTCF CUT&RUN showed higher dynamic range than other profiling methods , including standard X-ChIP-seq and ChIP-exo ( Figure 6A ) . When aligned to CTCF motifs found within DNaseI hypersensitive sites or previously identified binding sites , CUT&RUN and X-ChIP-seq CTCF heat maps showed strong concordance , with CUT&RUN having a higher dynamic range ( Figure 6B ) . A no antibody control showed undetectable background ( Figure 6—figure supplement 1 ) when CUT&RUN is performed at low temperature ( Figure 6—figure supplement 2 ) . As was the case for budding yeast TFs , we observed release of the neighboring fragments , which correspond to phased nucleosomes immediately adjacent to CTCF sites . By plotting just the end positions of the short CUT&RUN fragments that are the cleavage positions of the tethered MNase , we observed pronounced ‘tram-tracks’ separated by 44 bp at defined positions relative to the CTCF motif . Furthermore , the exact cleavage pattern is consistent over a ~300 fold time-course digestion range , with a predominant single base-pair cut site on either side of the CTCF-bound site , highlighting the limit digest obtained ( Figure 6C ) . This pattern indicates that the cleavage positions are precise and highly homogeneous within the population of cells . Our results suggest that CUT&RUN accurately maps both the TFs and their flanking chromatin in the same experiment . 10 . 7554/eLife . 21856 . 016Figure 6 . CUT&RUN maps high-resolution footprints of CTCF . ( A ) Representative signal over a genomic locus for 10 million randomly sampled reads from ENCODE CTCF ChIP ( GSM749690 ) , CTCF ChIP-exo , and CUT&RUN . In the top panel , the y-axis is the same for all datasets indicating the higher dynamic range for CUT&RUN . In the bottom panel , the y-axis is individually set . ( B ) Heat maps of CUT&RUN pooled datasets ( 7 . 5 min to 45 min ) separated into ≤120 bp ( including fragment ends ) and ≥150 bp size classes and of ENCODE X-ChIP-seq and high resolution X-ChIP-seq ( Skene and Henikoff , 2015 ) for CTCF in human K562 cells . Sites were determined by an unbiased approach in which the data were centered and oriented on CTCF motifs that were found within DNaseI hypersensitive sites and ordered by genomic location . Asymmetric release of the upstream and downstream nucleosome likely comes from epitope location controlling access to nucleosomes either side of the motif . ( C ) Mean plots of end positions from ≤120 bp fragments resulting from a CUT&RUN digestion time-course centered over sites as above . Data are represented as a percentage of the maximum signal within the ±1 kb flanking region . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 01610 . 7554/eLife . 21856 . 017Figure 6—figure supplement 1 . CUT&RUN recapitulates X-ChIP-seq but with higher dynamic range . For a direct comparison of genome wide dynamic range at previously identified CTCF binding sites , 10 million reads were randomly selected from ENCODE CTCF X-ChIP-seq ( GSM749690 ) and CUT&RUN datasets and plotted at ENCODE peak called sites ( GSM749690_narrowPeak ) . The upper plot shows the mean average of raw counts over these sites and heat maps below are ordered by genomic location . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 01710 . 7554/eLife . 21856 . 018Figure 6—figure supplement 2 . CUT&RUN has low background when performed on ice . During protocol optimization , we performed the cleavage reactions over a range of temperatures . ( A ) We initially used 37°C as is often done for MNase reactions . Careful analysis of the data , however , showed that despite clearly mapping CTCF at its true sites with a low density genome-wide background , we also had a specific background at random DNase1 sites . We rationalized that specific background arose from the liberated chromatin complexes that are still bound by Protein A-MNase diffusing around the nucleus and cutting accessible regions of chromatin . ( B ) To test this hypothesis , after the CTCF antibody and Protein A-MNase had bound in situ , we disrupted the nuclear envelope with limited sonication to release the chromatin into the large reaction volume . When CUT&RUN was performed under disrupted conditions , we no longer observed this specific background . ( C ) We therefore tried to limit the diffusion of these chromatin complexes by performing the cleavage reaction at room temperature . We observed that the signal-to-noise ratio started low , but increased over time and by 8 min the noise was indistinguishable from the signal . ( D ) However , by keeping the reaction on ice the signal-to-noise ratio was hi and independent of time . Therefore , by controlling the temperature for the cleavage reaction , we can robustly maintain a low background . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 01810 . 7554/eLife . 21856 . 019Figure 6—figure supplement 3 . The high signal-to-noise ratio of CUT&RUN allows robust identification of DNA binding sites not possible with X-ChIP-seq . CUT&RUN was performed for Myc and Max in K562 cells and compared to ENCODE X-ChIP-seq datasets ( GSM935410; GSM935539 ) . For each dataset 10 million reads were randomly selected and ( A ) a typical genomic region is shown . Note for Myc different antibodies were used and therefore quantitative comparison is not possible . ( B ) Proportional Venn diagrams displaying the overlap between Myc and Max peak called sites identified by CUT&RUN or previously by ENCODE . ( C ) Heat maps showing CUT&RUN and ENCODE X-ChIP-seq signal plotted at peak called sites identified by Max CUT&RUN ( n = 20146 ) . Sites were ranked by Max CUT&RUN score , note the change in the dynamic range of the heat maps . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 01910 . 7554/eLife . 21856 . 020Figure 6—figure supplement 4 . CUT&RUN can map compacted chromatin with a high dynamic range . CUT&RUN was performed for H3K27me3 in K562 cells either by extracting all the DNA after digestion followed by size selection or allowing cut fragments to diffuse out of the nuclei . For comparison an ENCODE H3K27me3 X-ChIP-seq ( GSM733658 ) dataset was analyzed . For each dataset 10 million reads were randomly selected and a typical genomic region is shown with the upper panels equally scaled and the lower panel rescaled for the ENCODE dataset . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 020 CTCF has 11 zinc fingers and therefore may represent an unusually stable protein–DNA interaction . We therefore tested CUT&RUN using Myc and Max which are basic-loop-helix proteins that bind to a short E-box motif and have b residence times ( Phair et al . , 2004 ) . CUT&RUN successfully mapped both Myc and Max at high resolution ( Figure 6—figure supplement 3A ) . In the case of Max , a quantitative comparison with ENCODE ChIP-seq data is possible as the same antibody was used , and here CUT&RUN had a much higher dynamic range and therefore was able to robustly identify a much larger number of Max binding sites ( Figure 6—figure supplement 3B ) . To bind DNA at E-boxes , Myc forms a heterodimer with Max ( Blackwood et al . , 1991 ) but in addition Max has other binding partners ( Ayer and Eisenman , 1993 ) , As expected , we see very high overlap with Max present at almost all Myc binding sites . In contrast , there is poor overlap between previously identified binding sites by ENCODE X-ChIP-seq for Myc and Max , as 10-fold fewer Max sites were identified . However , when we lined up Max ENCODE X-ChIP-seq data over Max CUT&RUN sites , we saw high occupancy ( Figure 6—figure supplement 3C ) , suggesting that the lower dynamic range of X-ChIP-seq relative to CUT&RUN was responsible for the failure to identify these Max binding sites by X-ChIP-seq . We also considered the possibility that antibody-tethered MNase may be excluded from highly compacted heterochromatic regions in higher eukaryotes and as such CUT&RUN might be limited to analysis of protein-DNA interactions in euchromatic regions . We therefore performed CUT&RUN for the repressive histone mark H3K27me3 . By analyzing 10 million reads from CUT&RUN and ENCODE X-ChIP-seq , we observed similar H3K27me3 landscapes , but at a much higher dynamic range for CUT&RUN , which demonstrates that Protein A-MNase is able to access compacted chromatin ( Figure 6—figure supplement 4 ) . Furthermore , H3K27me3 cleaved chromatin is readily released from the intact nuclei into the soluble fraction , indicating that CUT&RUN is applicable to probing protein-DNA interactions in compacted chromatin . As nucleosome-sized fragments adjacent to TFs are released together with TF-containing fragments , we wondered whether 3D adjacencies might also be subject to cleavage and release . Chromosome-Conformation-Capture ( 3C ) methods , such as Hi-C and ChIA-PET ( Tang et al . , 2015; Lieberman-Aiden et al . , 2009 ) , are the preferred techniques for mapping 3D genome-wide contacts . These methods use the same formaldehyde crosslinking protocol as X-ChIP to identify 3D interactions , such as between a TF bound at an enhancer and its contact with a promoter via co-activators . In this example the binding sites for a protein identified by X-ChIP will include both the promoter and the enhancer , even though one of the interactions is via indirect protein-protein interactions crosslinked by formaldehyde . But in both X-ChIP and 3C-based mapping there is no systematic way to distinguish between direct and indirect sites . We therefore attempted to map CTCF binding sites using native ChIP , which we have previously shown results in mapping only direct binding sites containing the TF-specific DNA-binding motif , due to the transient nature of protein-protein interactions ( Kasinathan et al . , 2014 ) . We developed a new native ChIP protocol ( Appendix 1 ) , and achieved near-complete protein extraction with no evidence of protein redistribution ( Figure 7—figure supplement 1 ) . Under native conditions , we identified 2298 sites with high motif scores . In contrast , CUT&RUN mapping of CTCF detected ~22 , 000 sites that were also present in X-ChIP ( Figure 6—figure supplement 1 ) , with a diverse range of motif scores ( Figure 7—figure supplement 2 ) . As expected , all sites identified by native ChIP also were robustly detected by CUT&RUN and X-ChIP , showing a similar signal distribution ( Figure 7A ) . CUT&RUN sites lacking a significant native ChIP signal nevertheless showed a robust footprint in the native ChIP input with a similar cumulative distribution of counts ( Figure 7B ) , indicating the presence of unknown bound factors , as would be expected for 3D genomic interactions . This suggests that CUT&RUN , as with X-ChIP , can discover both direct ( native CTCF peak ) and indirect ( CUT&RUN peak only ) chromatin interactions at high resolution . 10 . 7554/eLife . 21856 . 021Figure 7 . CTCF directly binds a subset of CUT&RUN peaks despite a robust footprint at all sites . ( A ) Chromatin was fragmented and solubilized under native conditions and either directly sequenced as native input or CTCF bound chromatin was immunoprecipitated and sequenced . ENCODE X-ChIP-seq was analyzed for comparative purposes . Peaks of CTCF binding under native conditions were identified and centered on the best match to the CTCF motif ( JASPAR database MA0139 . 1 , http://jaspar . genereg . net/ ) . Data were plotted over these sites ( −1 to +1 kb ) as heat maps for native ChIP DNA fragments ( 20–75 bp ) and CUT&RUN ( ≤120 bp ) and ordered by native CTCF ChIP occupancy ( sum over the center region ( −30 to +30 bp ) minus the sum over the flanks ( −1000 to −700 and +700 to+1000 bp ) . The graph below shows the cumulative percent of sequencing counts for the different techniques over peak-called sites ( −30 to +30 bp ) and ranked by similarity to the CTCF motif . This shows the high concordance between the chromatin profiling techniques at native ChIP peaks . Note that the dynamic range scales for Native ChIP and CUT&RUN are ~30–40 fold higher than those for Native Input and ENCODE X-ChIP , which was needed to show the input and ENCODE patterns . ( B ) Data plotted over CUT&RUN peak-called sites , with processing as per ( A ) . The cumulative distribution shows the shift to lower motif scores for CUT&RUN sites ( see Figure 7—figure supplement 2 and the separation between CUT&RUN and native ChIP . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 02110 . 7554/eLife . 21856 . 022Figure 7—figure supplement 1 . A modified native ChIP protocol allows complete protein extraction . ( A ) Western blotting to test the extraction efficiency of RNA polymerase II ( RNAPII ) and CTCF under native conditions with varying SDS and sonication conditions . ( B ) To test for potential redistribution of CTCF under native conditions , extracts were incubated with 95 bp DNA probes with a high scoring motif ( positive ) , or a shuffled sequence ( negative ) , or the 601 nucleosome positioning sequence at 1000 copies per cell . Following the ChIP and DNA extraction , quantitative PCR was used to test for CTCF binding to a native peak in the genome ( genomic ) or to the DNA probes . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 02210 . 7554/eLife . 21856 . 023Figure 7—figure supplement 2 . Peaks identified by CUT&RUN have a more diverse range of motif scores than peaks from native ChIP . Peak calling was performed on native CTCF ChIP ( false positives were removed that did not contain a clear peak ) and CUT&RUN . The underlying DNA sequence was extended in both directions by 100 bp and the best match and score to the JASPAR position frequency matrix ( MA0139 . 1 ) calculated . Histograms plot the distribution of motif scores . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 023 To confirm that CTCF CUT&RUN sites not observed by native ChIP correspond to contact sites , we compared direct and indirect sites to contact sites observed by ChIA-PET . CTCF ChIA-PET identifies interacting genomic regions mediated through CTCF , but cannot discern between directly CTCF bound regions and the interacting indirectly bound region . We find that for a typical ~1 Mb genomic region all high-scoring ChIA-PET fragments overlap with direct and indirect sites ( Figure 8A ) . Whereas mapped CTCF ChIA-PET fusion fragments are in the several kb range , determined by the distance between sites for the 6-cutter restriction enzyme used , both direct and indirect CUT&RUN CTCF sites are mapped with near base-pair resolution . Moreover , 91% of the direct sites are present in CTCF ChIA-PET data , with 43% of these ChIA-PET fragments interacting with an indirect site , and the remainder contained a high CUT&RUN signal ( Figure 8C ) , which suggests these are indirect sites involved in multiple contacts , just below the peak calling threshold . 10 . 7554/eLife . 21856 . 024Figure 8 . CUT&RUN in combination with native ChIP can discern direct and indirect 3D contact sites . ( A ) Typical genomic region displaying CUT&RUN ( ≤120 bp ) , native ChIP ( 20–75 bp ) data for CTCF and CTCF ChIA-PET fragments ( GSM1872886; score ≥15 ) . ChIA-PET fragments were ascribed as a direct interaction ( overlapping a native ChIP peak ) or an indirect interaction ( overlapping a CUT&RUN peak only ) . ( B ) Peak called sites were separated into either direct ( present in native ChIP ) or indirect ( only present in CUT&RUN ) . Hi-C fragments that intersect with direct sites or an equal number of random genomic locations were identified . The Hi-C interacting fragment was then intersected with the indirect sites and the CUT&RUN signal compared to Hi-C raw signal . Data were ranked by CUT&RUN score and plotted as a moving average with a window size of 1500 . ( C ) ChIA-PET fragments that contained a direct site were identified and the interacting fragment intersected with direct peaks , indirect peaks or random locations as above . Interacting fragments that did not overlap with these sites were classed as uncategorized . Boxplots indicate the CUT&RUN score for the observed contacts at the interacting fragment . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 024 As further evidence that CUT&RUN can detect indirect contact interactions , we found a high frequency of Hi-C interactions between direct sites and indirect sites and a quantitative correlation between Hi-C score and CUT&RUN signal at the indirect sites ( Figure 8B ) . Therefore , by comparing CUT&RUN and native ChIP it is possible to map contact sites at near base-pair resolution , to distinguish direct from indirect protein binding sites that result from long-range genomic interactions , and to determine the directionality to these contacts , not feasible by other methods . Typical ChIP-seq experiments require large numbers of cells , and low cell number ChIP has been limited to abundant proteins ( Kasinathan et al . , 2014; Brind'Amour et al . , 2015 ) . We performed CTCF CUT&RUN with starting K562 cell numbers ranging from 600 , 000 to 10 million . To compare absolute occupancies between datasets , we used a simple spike-in strategy ( see Materials and methods ) , allowing accurate quantitative measurements of protein occupancy . When normalized to spike-in DNA , we observed that the number of cleavage events is proportional to the starting cell number ( Figure 9 ) . Furthermore , when the data are normalized to the total number of reads aligning to the human genome , there is no clear difference in the samples , suggesting that high data quality is maintained with low input material . 10 . 7554/eLife . 21856 . 025Figure 9 . CUT&RUN allows simple quantification of protein-DNA interactions . ( A ) A digestion timecourse of CUT&RUN was performed for CTCF in K562 cells . To allow quantification of released fragments , 1 ng of Drosophila DNA was added after the cleavage reaction . Mean plots of ≤120 bp sequenced fragments were centered over CTCF motifs found within DNaseI sites . Data were normalized either to the number of fly reads ( Spike-in normalization ) or to the total number of human reads ( Standard normalization ) . ( B ) A titration of starting material was used to map CTCF binding genome-wide . Heat maps and mean plots were generated for the ≤120 bp sequenced fragments using Spike-in or Standard normalization . Data were centered over CTCF motifs found within DNaseI sites . DOI: http://dx . doi . org/10 . 7554/eLife . 21856 . 025
CUT&RUN is based on the ChIC antibody-tethered nuclease strategy of Laemmli and co-workers ( Schmid et al . , 2004 ) . To adapt ChIC into a genome-wide profiling method , we made five critical modifications . First , we immobilize permeabilized cells or crude nuclei to magnetic beads , allowing for rapid and efficient solution changes , so that CUT&RUN is performed in a day and is suitable for automation . Second , we bind antibodies and pA-MNase to native unfixed nuclei , where epitopes are preserved and accessible . Third , as cleavage by immobilized MNase is a zero-order reaction , we performed digestion at ice-cold temperature , which limits diffusion of released fragments , thus reducing background . Fourth , we use native chromatin , which allows us to fractionate cleaved fragments based on solubility ( Sanders , 1978; Teves and Henikoff , 2012; Jahan et al . , 2016 ) to enrich specifically for the released chromatin complex . We simply remove the insoluble bulk chromatin , and only chromatin fragments with cleavages on both sides of a particle enter the supernatant . Fifth , after DNA extraction we use these soluble fragments for Illumina library preparation and paired-end DNA sequencing . We find that CUT&RUN performs as well as or better than ChIP-seq with respect to simplicity , resolution , robustness , efficiency , data quality , and applicability to highly insoluble complexes . CUT&RUN requires only ~1/10th of the sequencing depth of other high resolution methodologies due to the intrinsically low background achieved by performing the reaction in situ . As nuclei are intact when MNase is activated , CUT&RUN can probe the local environment around the targeted site . Indeed , we find that CUT&RUN recovers sites of 3D contacts with base-pair resolution at relatively low sequencing depth in human cells . Although ChIC was described as a basic mapping method using Southern blotting 12 years ago , we are unaware of a single publication using it . Meanwhile , ChIP-seq alone has been mentioned in ~30 , 000 publications for profiling almost every type of chromatin component , including histone modifications , transcription factors and chromatin-associated proteins . Like ChIP , CUT&RUN is antibody-based , so that it can be applied to any epitope on chromatin , making it a general method for chromatin profiling that takes advantage of the large antibody production infrastructure developed for ChIP . CUT&RUN provides quantitative occupancy profiles with standard and spike-in normalization options implemented by our custom software for processing and comparing ChIP-seq and CUT&RUN datasets . The only non-standard feature of CUT&RUN is the requirement for the pA-MN fusion protein , which can be produced and purified in a batch from bacterial culture that yields enough pA-MN for profiling >100 , 000 samples . As CUT&RUN is based on different principles from ChIP , it can resolve crosslinking- , sonication- and solubilization-related issues . Backgrounds are low with CUT&RUN , because cleavages occur only around binding sites , whereas ChIP first pulverizes the entire genome , and these fragments contribute to a genome-wide background noise that must still be sequenced . The near absence of detectable background under the brief low-temperature conditions that we used , the lack of preference for accessible or AT-rich DNA , and the recovery of essentially all Abf1 and Reb1 motifs in the yeast genome , suggests that CUT&RUN is not subject to the types of artifacts that sometimes have plagued ChIP ( Teytelman et al . , 2013; Park et al . , 2013; Jain et al . , 2015; Baranello et al . , 2016; Kasinathan et al . , 2014 ) . Furthermore , CUT&RUN antibody binding takes place in an intact nuclear environment resembling conditions for immunofluorescence microscopy , so that it should be successful for antibodies that are validated cytologically , even those that fail in ChIP . As CUT&RUN solubilizes chromatin only after the targeted cleavage reaction , it should be appropriate for extending classical chromatin salt-fractionation ( Sanders , 1978; Teves and Henikoff , 2012; Jahan et al . , 2016 ) to specific TFs and chromatin complexes . A consequence of using intact nuclei for CUT&RUN is that the long reach of antibody-tethered MNase can probe the local environment . In yeast , we observed cleavages on one surface of the DNA flanking the TF , gradually decreasing with distance . In human cells , we observed cleavages at sites previously identified as contact sites for CTCF . Recently , Hi-C contact sites have been predicted computationally with a high degree of confidence given sites of CTCF binding ( Sanborn et al . , 2015 ) . As CUT&RUN maps both CTCF binding sites and interactions , and our native ChIP protocol identifies those sites that are directly TF-bound , it can provide a complete high-resolution 1D map of a genome while enriching its 3D contact map with high-resolution distinctions between direct and indirect TF binding sites . ChIP-seq analysis typically includes normalization to compensate for different numbers of reads between samples . In ChIP-seq , whole genome fragmentation leads to a constant low density genome-wide background that provides a basis for normalization , for example , in comparing wildtype and knockdown cell lines . Although normalization fails on abundant proteins , this can be corrected by the use of spike-in controls ( Bonhoure et al . , 2014; Chen et al . , 2015; Orlando et al . , 2014 ) . However , a rigorous spike-in strategy requires the addition of cells from a different species , and quantification is reliant upon antibody cross-reactivity ( Orlando et al . , 2014 ) . To normalize between samples despite the low background in CUT&RUN , we find that simply adding a constant low amount of fragmented spike-in DNA from a different species suffices and allows accurate quantification of protein occupancy . The low background levels of cleavage with CUT&RUN require fewer reads to crisply define peaks . For example only ~10 million paired-end reads were required for each CTCF time point , similar to the requirement for low-resolution ChIP-seq , and many fewer than for ChIP-exo , which required ~100 million reads for CTCF ( Rhee and Pugh , 2011 ) . Furthermore , in the cases of Max and H3K27me3 , 10 million reads for CUT&RUN provided very high dynamic range , but 10 million reads were insufficient for calling peaks from Max ENCODE X-ChIP-seq . This cost-effectiveness makes CUT&RUN attractive as a replacement for ChIP-seq especially where depth of sequencing is limiting . We might attribute the high efficiency of CUT&RUN to fundamental differences between in situ profiling and ChIP: CUT&RUN retains the in vivo 3D conformation , so antibodies access only exposed surfaces in a first-order binding reaction , whereas in ChIP , antibodies interact with the solubilizable genome-wide content of the pulverized cells or nuclei . Furthermore , CUT&RUN cleavage is effectively a zero-order reaction , resulting in steady particle release during the brief low temperature time-course for all bound epitopes in the genome . Accounting for epitope abundances , we estimate that our mapping of ~22 , 000 direct and indirect CTCF sites with 600 , 000 cells is comparable to the sensitivity of ultra-low input ChIP-seq protocols that are typically restricted to abundant histone modifications , such as H3K27me3 , with ~5000 cells ( Brind'Amour et al . , 2015 ) . Whereas ultra-low-input ChIP provides only ~2 kb resolution , CUT&RUN provides near base-pair resolution . The inherent robustness , high information content , low input and sequencing requirements and suitability for automation of our method suggests that CUT&RUN profiling of CTCF and other TFs might be applied to epigenome diagnostics . In summary , CUT&RUN has a number of practical advantages over ChIP and its derivatives: with low background resulting in low sequence depth requirements , the ease of use makes it amenable to robotic automation , while allowing accurate quantification with a simple spike-in strategy . We conclude that in all important respects CUT&RUN provides an attractive alternative to ChIP-based strategies .
W1588-4C S . cerevisiae strains carrying Flag-tagged H2A ( SBY2688 ) , Cse4 ( SBY5146 ) , Abf1 and Reb1 under the control of their respective endogenous promoters were previously described ( Kasinathan et al . , 2014; Krassovsky et al . , 2012; Gelbart et al . , 2001 ) . Yeast nuclei were prepared as described ( Kasinathan et al . , 2014 ) , flash frozen in 0 . 5–0 . 6 ml aliquots and stored at −80°C . Human K562 cells were cultured under standard conditions . Standard protocols were used for electrophoretic gel analysis and immunoblotting . Antibodies used were Mouse anti-FLAG ( M2 , Sigma , St . Louis , MO , Catalog #F1804 ) , Rabbit anti-mouse ( Abcam , Cambridge , UK , Catalog #ab46540 ) , CTCF ( Millipore Billerica , MA , Catalog #07–729 ) , H3K27me3 ( Millipore Catalog #07-449 ) , c-Myc ( Cell Signaling Technology Beverly , MA , Catalog #D3N8F ) , Max ( Santa Cruz Biotechnology , Dallas , TX , Catalog #sc-197 ) and RNA Pol II ( 8WG16 , Abcam Catalog #ab817 ) . The pK19pA-MN plasmid was a gift from Ulrich Laemmli and pA-MN protein was prepared from E . coli cells as described ( Schmid et al . , 2004 ) . CUT&RUN begins with crude nuclei prepared according to published procedures . The following protocol is provided in step-by-step format ( Appendix 2 ) . Nuclei from ~5 × 108 cells at OD600 ~0 . 7 were prepared as described ( Orsi et al . , 2015 ) , divided into 10 600 µL aliquots , snap-frozen and held at −80°C , then thawed on ice before use . Bio-Mag Plus Concanavalin A ( lectin ) coated beads were equilibrated with HNT ( 20 mM HEPES pH7 . 5 , 100 mM NaCl and 0 . 1% Tween 20 ) that was supplemented with 1 mM each MgCl2 , CaCl2 and MnCl2 . Only Ca++ and Mn++ are needed to activate lectins , and omitting MgCl2 had no effect on binding of permeabilized cells to beads . The beads ( 300 µL ) were rapidly mixed with a thawed nuclei aliquot and held at room temperature ( RT ) ≥5 min , placed on a magnet stand to clear ( <1 min ) , and decanted on a magnet stand . The beads were then incubated 5 min RT in HNT supplemented with protease inhibitors ( Roche Complete tablets ) and 1 mM phenylmethylsulfonyl fluoride ( PMSF ) ( =HNT-PPi ) containing 3% bovine serum albumen ( BSA ) and 2 mM EDTA pH 8 , then incubated 5’ with HNT-PPi +0 . 1% BSA ( blocking buffer ) , using the magnet stand to decant . The beads were incubated 2 hr at 4°C with mouse anti-FLAG antibody ( 1:200–1:350 ) , decanted , washed once in HNT + PMSF , then incubated 1 hr at 4°C with rabbit anti-mouse IgG antibody ( 1:200 ) in blocking buffer . The beads were washed once in HNT + PMSF , then incubated 1 hr at 4°C with pA-MN ( 600 µg/ml , 1:200 ) in blocking buffer . The beads were washed twice in HNT + PMSF and once in 20 mM HEPES pH 7 . 5 , 100 mM NaCl ( Digestion buffer ) , optionally including 10% polyethylene glycol 8000 for Sth1 and Mot1 . The beads were brought up in 1 . 2 ml Digestion buffer , divided into 8 × 150 µL aliquots , equilibrated to 0°C , then quickly mixed with CaCl2 , stopping the reaction with 150 µL 2XSTOP [200 mM NaCl , 20 mM EDTA , 4 mM EGTA , 50 µg/ml RNase A ( Thermo Scientific , Waltham , MA , Catalog #EN0531 ) and 40 µg/ml glycogen ( Sigma , Catalog #10901393001 ) , containing 5-50 pg/ml heterologous mostly mononucleosome-sized DNA fragments extracted from formaldehyde crosslinked MNase-treated Drosophila chromatin as a spike-in] . After incubating at 37°C for 20 min , the beads were centrifuged 5 min at 13 , 000 rpm at 4°C , the supernatant was removed on a magnet stand and mixed with 3 µL 10% SDS and 2 µL Proteinase K ( Invitrogen , Carlsbad , CA , Catalog #25530049 ) , incubated at 70°C 10 min , then extracted at room temperature once with buffered phenol-chloroform-isoamyl alcohol ( 25:24:1 , Sigma P2069 ) , transferred to a phase-lock tube ( Qiagen , Hilden , Germany , Catalog #129046 ) , re-extracted with one vol CHCl3 , transferred to a fresh tube containing 2 µL 2 mg/ml glycogen , precipitated by addition of 2–2 . 5 vol ethanol , chilled in ice and centrifuged 10 min at 13 , 000 rpm at 4°C . The pellet was rinsed with 100% ethanol , air-dried and dissolved in 25 µL 0 . 1 x TE8 ( =1 mM Tris pH 8 , 0 . 1 mM EDTA ) . To extend CUT&RUN for high-salt extraction , digestions were performed in a 50 µL volume , stopped with 50 µL 2XSTOP , omitting RNase and substituting the standard 200 mM NaCl with 4 M NaCl . After 20 min at 37°C , 200 µL 67 µg/ml RNase A was added , incubated 20 min , then centrifuged 13 , 000 rpm to clarify the supernatant . Human K562 cells were purchased from ATCC ( Manassas , VA , Catalog #CCL-243 ) . CUT&RUN was performed using a centrifugation-based protocol . Ten million cells were harvested by centrifugation ( 600 g , 3 min in a swinging bucket rotor ) and washed in ice cold phosphate-buffered saline ( PBS ) . Nuclei were isolated by hypotonic lysis in 1 ml NE1 ( 20 mM HEPES-KOH pH 7 . 9; 10 mM KCl; 1 mM MgCl2; 0 . 1% Triton X-100; 20% Glycerol ) for 5 min on ice followed by centrifugation as above . ( We have found that nucleases in some cells cause Mg++-dependent degradation of DNA , in which case 0 . 5 mM spermidine can be substituted for 1 mM MgCl2 . ) Nuclei were briefly washed in 1 . 5 ml Buffer 1 ( 20 mM HEPES pH 7 . 5; 150 mM NaCl; 2 mM EDTA; 0 . 5 mM Spermidine; 0 . 1% BSA ) and then washed in 1 . 5 ml Buffer 2 ( 20 mM HEPES pH 7 . 5; 150 mM NaCl; 0 . 5 mM Spermidine; 0 . 1% BSA ) . Nuclei were resuspended in 500 µl Buffer 2 and 10 µl antibody was added and incubated at 4°C for 2 hr . Nuclei were washed 3 x in 1 ml Buffer 2 to remove unbound antibody . Nuclei were resupended in 300 µl Buffer 2 and 5 µl pA-MN added and incubated at 4°C for 1 hr . Nuclei were washed 3 x in 0 . 5 ml Buffer 2 to remove unbound pA-MN . Tubes were placed in a metal block in ice-water and quickly mixed with 100 mM CaCl2 to a final concentration of 2 mM . The reaction was quenched by the addition of EDTA and EGTA to a final concentration of 10 mM and 20 mM respectively and 1 ng of mononucleosome-sized DNA fragments from Drosophila DNA added as a spike-in . Cleaved fragments were liberated into the supernatant by incubating the nuclei at 4°C for 1 hr , and nuclei were pelleted by centrifugation as above . DNA fragments were extracted from the supernatant and used for the construction of sequencing libraries . We have also adapted this protocol for use with magnetic beads ( Appendix 3 ) . Genome-wide background in TF ChIP-seq datasets is typically sufficiently high to provide a constant background level for normalization to compensate for variations between samples in library preparation and sequencing . For standard normalization , the number of fragment ends corresponding to each base position in the genome was divided by the total number of read ends mapped . However , the inherently low background levels of CUT&RUN necessitate a spike-in control for quantitative comparisons ( Hu et al . , 2014 ) . For spike-in normalization of human CUT&RUN , we added a low constant amount of Drosophila melanogaster DNA to each reaction . We mapped the paired-end reads to both human and fly genomes , normalizing human profiles to the number of fly reads ( Figure 9 ) . Using internal normalization , we observed no increase in cleavages over a digestion time-course . However , by normalizing to the fly spike-in DNA , we observed an ~4 fold increase in cleavage level over time . As such , CUT&RUN is amenable to accurate quantification of protein-DNA interactions . Sequencing libraries were prepared from DNA fragments as described ( Kasinathan et al . , 2014; Henikoff et al . , 2011 ) but without size-selection , following the KAPA DNA polymerase library preparation kit protocol ( https://www . kapabiosystems . com/product-applications/products/next-generation-sequencing-2/dna-library-preparation/kapa-hyper-prep-kits/ ) and amplifying for eight or more cycles . To deplete total DNA samples of large fragments originating from insoluble chromatin , samples were mixed with ½ volume of Agencourt AMPure XP beads , held 5–10 min , placed on a magnet stand , and the supernatant was retained , discarding the beads . To reduce the representation of the remaining large fragments , the number of PCR cycles using the KAPA polymerase library preparation method was increased to 14 cycles and adapter concentrations were increased accordingly: Increasing the number of PCR cycles favors exponential amplification of shorter fragments over linear amplification of fragments that are too long for polymerase to completely transit . Libraries were sequenced for 25 cycles in paired-end mode on the Illumina HiSeq 2500 platform at the Fred Hutchinson Cancer Research Center Genomics Shared Resource . Paired-end fragments were mapped to the sacCer3/V64 genome and build and to release r5 . 51 ( May 2013 ) of the D . melanogaster genomic sequence obtained from FlyBase using Novoalign ( Novocraft ) as described to generate SAM files . For human samples , paired-end fragments were mapped to hg19 using Bowtie2 . Custom scripts for data processing are provided in Supplementary Software and can be downloaded from https://github . com/peteskene . For comparative analyses , publicly available datasets downloaded from the NCBI SRA archive were: ERR718799 ( Abf1 ) , SRR2568522 ( Reb1 ) , GSM749690 ( CTCF; 150 bp sliding window at a 20 bp step across the genome with a false discovery rate of 1% ) , and the CTCF ChIP-exo BAM file was kindly provided by Frank Pugh . To obtain sets of TF-specific motifs without biasing towards CUT&RUN peaks , we applied the MEME motif-finding program to yeast ORGANIC ChIP-seq peak calls . We used the resulting log-odds position-specific scoring matrix ( PSSM ) for MAST searching of the S . cerevisiae genome to identify sites with significant log-odds motif scores . This identified 1899 Abf1 sites and 1413 Reb1 sites . Following previous studies , we use correspondence of a yeast TF binding site to the motif for that TF to be the ‘gold-standard’ for a true-positive call ( Rhee and Pugh , 2011; Kasinathan et al . , 2014; Zentner et al . , 2015; Ganapathi et al . , 2011 ) . MEME was used to construct log-odds PSSMs from peaks called using the threshold method of Kasinathan et al . ( Kasinathan et al . , 2014 ) . Peak-calling cut-off was the 99 . 5th percentile of normalized counts for pooled 1 s−32 s ≤120 bp Abf1 and Reb1 datasets , where the interpeak distance = 100 , minimum peak width = 50 , and maximum peak width = 1000 . To compare CUT&RUN and ORGANIC motif recovery , peak-call thresholds were adjusted to report similar numbers of peaks . Log-odds sequence logos were produced using PWMTools ( http://ccg . vital-it . ch/pwmtools/ ) . Track screen shots were produced using IGV ( Thorvaldsdottir et al . , 2013 ) Sequencing data have been deposited in GEO at National Center for Biotechnology Information under the accession number GSE84474 . | The DNA in a person’s skin cell will contain the same genes as the DNA in their muscle or brain cells . However , these cells have different identities because different genes are active in skin , muscle and brain cells . Proteins called transcription factors dictate the patterns of gene activation in the different kinds of cells by binding to DNA and switching nearby genes on or off . Transcription factors interact with other proteins such as histones that help to package DNA into a structure known as chromatin . Together , transcription factors , histones and other chromatin-associated proteins determine whether or not nearby genes are active . Sometimes transcription factors and other chromatin-associated proteins bind to the wrong sites on DNA; this situation can lead to diseases in humans , such as cancer . This is one of the many reasons why researchers are interested in working out where specific DNA-binding proteins are located in different situations . A technique called chromatin immunoprecipitation ( or ChIP for short ) can be used to achieve this goal , yet despite being one of the most widely used techniques in molecular biology , ChIP is hampered by numerous problems . As such , many researchers are keen to find alternative approaches . Skene and Henikoff have now developed a new method , called CUT&RUN ( which is short for “Cleavage Under Targets & Release Using Nuclease” ) to map specific interactions between protein and DNA in a way that overcomes some of the problems with ChIP . In CUT&RUN , unlike in ChIP , the DNA in the starting cells does not need to be broken up first; this means that protein-DNA interactions are more likely to be maintained in their natural state . With CUT&RUN , as in ChIP , a specific antibody identifies the protein of interest . But in CUT&RUN , this antibody binds to the target protein in intact cells and cuts out the DNA that the protein is bound to , releasing the DNA fragment from the cell . This new strategy allows the DNA fragments to be sequenced and identified more efficiently than is currently possible with ChIP . Skene and Henikoff showed that their new method could more accurately identify where transcription factors bind to DNA from yeast and human cells . CUT&RUN also identified a specific histone that is rarely found in yeast chromatin and the technique can be used with a small number of starting cells . Given the advantages that CUT&RUN offers over ChIP , Skene and Henikoff anticipate that the method will be viewed as a cost-effective and versatile alternative to ChIP . In future , the method could be automated so that multiple analyses can be performed at once . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"tools",
"and",
"resources"
] | 2017 | An efficient targeted nuclease strategy for high-resolution mapping of DNA binding sites |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.